diff --git a/README.md b/README.md
index e70392c..b05610f 100644
--- a/README.md
+++ b/README.md
@@ -134,6 +134,11 @@ Running the pre-trained models on sample images can the easily be done via:
 dune exec examples/pretrained/predict.exe path/to/resnet18.ot images/tiger.jpg
 ```
 
+## Internals
+
+ocaml-torch uses extensive code generation to produce bindings to thousands of torch C++ functions.
+Read [internals.md](./internals.md) for details.
+
 ## Acknowledgements
 
 Many thanks to [@LaurentMazare](https://github.com/LaurentMazare) for the [original
diff --git a/images/codegen_graph.dot b/images/codegen_graph.dot
new file mode 100644
index 0000000..a673e4f
--- /dev/null
+++ b/images/codegen_graph.dot
@@ -0,0 +1,59 @@
+digraph {
+  subgraph cluster_binding_gen {
+    style=filled
+    color=lightgrey
+    label="binding generation"
+    declarations [label="declarations (yaml)"]
+    bindinggen [label="binding gen exe"];
+  }
+  subgraph cluster_bindings {
+    style=filled
+    color=lightgrey
+    label="stub generation and bindings"
+    bindings [label="bindings (manual)"]
+    bindingsg [label="bindings (generated)"]
+    stubgen [label="ctypes stub gen exe"];
+  }
+  subgraph cluster_wrapper {
+    style=filled
+    color=lightgrey
+    label="wrapper"
+    {rank=same;
+    stubsml [label="OCaml stubs (manual)", group=g1];
+    stubsmlg [label="OCaml stubs (generated)", group=g2];
+    }
+    {rank=same;
+    stubsc [label="C stubs (manual)", group=g1];
+    stubscg [label="C stubs (generated)", group=g2];
+    }
+    {rank=same;
+    apiml [label="OCaml wrapper (manual)", group=g1];
+    apimlg [label="OCaml wrapper (generated)", group=g2];
+    }
+    {rank=same;
+    apic [label="C/C++ API (manual)", group=g1];
+    apicg [label="C/C++ API (generated)", group=g2];
+    }
+  }
+
+
+  // GENERATION
+  bindinggen -> bindingsg [penwidth=2];
+  bindinggen -> apimlg [penwidth=2];
+  bindinggen -> apicg [penwidth=2];
+  stubgen -> stubscg [penwidth=2];
+  stubgen -> stubsmlg [penwidth=2];
+
+  // DEPENDENCY
+  declarations -> bindinggen[style="dashed"];
+  bindings -> stubgen[style="dashed"];
+  bindingsg -> stubgen[style="dashed"];
+  apic -> stubsc[style="dashed"];
+  apic -> stubscg[style="dashed"];
+  apicg -> stubscg[style="dashed"];
+  stubsc -> stubsml[style="dashed"];
+  stubscg -> stubsmlg[style="dashed"];
+  stubsml -> apiml[style="dashed"];
+  stubsml -> apimlg[style="dashed"];
+  stubsmlg -> apimlg[style="dashed"];
+}
diff --git a/internals.md b/internals.md
new file mode 100644
index 0000000..6f4d0e6
--- /dev/null
+++ b/internals.md
@@ -0,0 +1,96 @@
+# ocaml-torch internals
+
+ocaml-torch faces several challenges, including:
+* binding to thousands of functions
+* avoiding any minor memory leaks in these functions
+* quickly cleaning up the memory allocations of tensors when OCaml is done with them
+
+In order to solve this, we have 2 steps of code generation. In this diagram, solid arrows
+represent the code generation DAG and dashed arrows represent the code dependency DAG:
+
+![code generation DAGs](./images/codegen_graph.png)
+
+At a high level,
+
+* Declarations.yaml contains the function signatures for the whole Torch C++ API.
+* Custom binding generation reads all the declarations, and whenever possible, generating
+  * glue code for crossing between C/C++ (the generated C/C++ API),
+  * glue code for using the (yet to be generated) OCaml `foreign` functions in OCaml (the generated OCaml wrapper),
+  * and `ctypes` bindings.
+* Stub generation uses the `ctypes` library, reading the bindings and generating C and
+  OCaml stubs. These are just glue code to handle C/OCaml FFI. Note that we have some
+  manually-written C++ functions and bindings that get generated stubs.
+* There are an extremely small number of manually-written stubs (just 1 as of writing)
+  that ctypes cannot handle.
+* A combination of the generated OCaml wrapper and manually written wrapper provide an
+  actually usable OCaml API. These are further built upon in the main library (not
+  pictured).
+
+# Memory management
+
+A large part of this complexity is driven by memory management.
+
+## Avoiding memory leaks
+
+It is challenging to write manual FFI stubs without memory leaks or race conditions. We
+use `ctypes` to make sure we get this right on the vast majority of functions. Although it
+requires a second code generation step, this spares us from reinventing stub generation.
+
+## Cleaning up tensors
+
+We ensure that tensors are freed when OCaml garbage collects them. To do this, each Tensor
+is equipped with a custom finalizer. This could be done on either the C++ or OCaml side.
+However, the API to inform OCaml of a tensor's true size in memory only exists in C++ (the
+custom block API). Without this, OCaml would not know when to garbage collect on CPU and
+would OOM easily.
+
+
+Note that:
+
+* We have not yet informed OCaml of each tensor's true size, but this is coming soon.
+* OCaml is unaware of GPU memory usage. GPU users may need to manually free tensors or
+  manaully garbage collect.
+
+### Raw tensors and GC tensors
+
+One wrinkle in this setup is that ctypes cannot handle custom blocks. Since we want the
+bulk of our stubs to be generated by ctypes, we create a distinction between `raw_tensor`s
+and `gc_tensor`s.
+
+|                    | raw tensor | GC tensor   |
+|--------------------|------------|-------------|
+| has finalizer?     | no         | yes         |
+| GC knows its size? | no         | coming soon |
+| FFI input for C?   | no         | yes         |
+| FFI output from C? | yes        | no          |
+| ctypes type        | void ptr   | void ptr    |
+
+The only way to convert from a `raw_tensor` to `gc_tensor` is with the hand-written,
+non-ctypes function `with_tensor_gc`. It is used copiously in the generated OCaml wrapper
+code to ensure we only surface GC tensors to the user.
+
+The lifecycle of each tensor looks like this:
+
+1. Some wrapper function `let t = Tensor.foo ()` gets invoked, which makes its way into C++.
+2. C++ returns a `raw_tensor` that goes through a regular ctypes stub and makes its way
+   back to the OCaml `Tensor.foo` call.
+3. Still in `Tensor.foo`, `with_tensor_gc` gets invoked. This goes back into C++ and
+   copies the pointer (but not the data) of the tensor to a new custom block. It now has
+   known off-heap size and a finalizer to free its memory. This gets returned to OCaml
+   with the same memory layout ctypes uses but without going through ctypes.
+4. Now `let () = Tensor.bar t` gets invoked. This goes through usual ctypes stubs, since
+   `t` looks just like a regular `void ptr` to ctypes.
+5. Eventually `t` gets garbage collected. OCaml traverses its blocks and runs the
+   finalizer on each one, freeing the tensor's data.
+
+The memory of each tensor (raw or GC) looks like this:
+
+```
+             block 1              block2
+        ------------------      ----------
+root -> | ctypes fat ptr |----> | void * |----> tensor
+        ------------------      ----------
+
+```
+
+For GC tensors, `block2` is the one with finalizer and off-heap memory.
diff --git a/src/bindings/dune b/src/bindings/dune
new file mode 100644
index 0000000..396ef14
--- /dev/null
+++ b/src/bindings/dune
@@ -0,0 +1,6 @@
+(library
+ (name torch_bindings)
+ (public_name torch.bindings)
+ (libraries ctypes.stubs)
+ (preprocess
+  (pps ppx_jane)))
diff --git a/src/bindings/torch_bindings.ml b/src/bindings/torch_bindings.ml
new file mode 100644
index 0000000..1dc33c5
--- /dev/null
+++ b/src/bindings/torch_bindings.ml
@@ -0,0 +1,284 @@
+open Ctypes
+module Type_defs = Type_defs
+
+module C (F : Cstubs.FOREIGN) = struct
+  open Type_defs
+  open F
+
+  let manual_seed = foreign "at_manual_seed" (int64_t @-> returning void)
+  let free = foreign "free" (ptr void @-> returning void)
+  let get_num_threads = foreign "at_get_num_threads" (void @-> returning int)
+  let set_num_threads = foreign "at_set_num_threads" (int @-> returning void)
+
+  module Tensor = struct
+    let new_tensor = foreign "at_new_tensor" (void @-> returning raw_tensor)
+
+    let tensor_of_data =
+      foreign
+        "at_tensor_of_data"
+        (ptr void
+         (* data *)
+         @-> ptr int64_t
+         (* dims *)
+         @-> int
+         (* ndims *)
+         @-> int
+         (* element size in bytes *)
+         @-> int
+         (* kind *)
+         @-> returning raw_tensor)
+    ;;
+
+    let copy_data =
+      foreign
+        "at_copy_data"
+        (gc_tensor
+         (* tensor *)
+         @-> ptr void
+         (* data *)
+         @-> int64_t
+         (* numel *)
+         @-> int
+         (* element size in bytes *)
+         @-> returning void)
+    ;;
+
+    let copy_ =
+      foreign "at_copy_" (gc_tensor (* dst *) @-> gc_tensor (* src *) @-> returning void)
+    ;;
+
+    let set_data =
+      foreign
+        "at_set_data"
+        (gc_tensor (* dst *) @-> gc_tensor (* src *) @-> returning void)
+    ;;
+
+    let float_vec =
+      foreign
+        "at_float_vec"
+        (ptr double (* values *)
+         @-> int (* num values *)
+         @-> int (* kind *)
+         @-> returning raw_tensor)
+    ;;
+
+    let int_vec =
+      foreign
+        "at_int_vec"
+        (ptr int64_t
+         (* values *)
+         @-> int
+         (* num values *)
+         @-> int
+         (* kind *)
+         @-> returning raw_tensor)
+    ;;
+
+    let device = foreign "at_device" (gc_tensor @-> returning int)
+    let defined = foreign "at_defined" (gc_tensor @-> returning bool)
+    let num_dims = foreign "at_dim" (gc_tensor @-> returning int)
+    let shape = foreign "at_shape" (gc_tensor @-> ptr int (* dims *) @-> returning void)
+    let scalar_type = foreign "at_scalar_type" (gc_tensor @-> returning int)
+    let backward = foreign "at_backward" (gc_tensor @-> int @-> int @-> returning void)
+    let requires_grad = foreign "at_requires_grad" (gc_tensor @-> returning int)
+    let grad_set_enabled = foreign "at_grad_set_enabled" (int @-> returning int)
+    let get = foreign "at_get" (gc_tensor @-> int @-> returning raw_tensor)
+
+    let double_value =
+      foreign
+        "at_double_value_at_indexes"
+        (gc_tensor @-> ptr int @-> int @-> returning float)
+    ;;
+
+    let int64_value =
+      foreign
+        "at_int64_value_at_indexes"
+        (gc_tensor @-> ptr int @-> int @-> returning int64_t)
+    ;;
+
+    let double_value_set =
+      foreign
+        "at_set_double_value_at_indexes"
+        (gc_tensor @-> ptr int @-> int @-> float @-> returning void)
+    ;;
+
+    let int64_value_set =
+      foreign
+        "at_set_int64_value_at_indexes"
+        (gc_tensor @-> ptr int @-> int @-> int64_t @-> returning void)
+    ;;
+
+    let fill_double = foreign "at_fill_double" (gc_tensor @-> float @-> returning void)
+    let fill_int64 = foreign "at_fill_int64" (gc_tensor @-> int64_t @-> returning void)
+    let print = foreign "at_print" (gc_tensor @-> returning void)
+    let to_string = foreign "at_to_string" (gc_tensor @-> int @-> returning string)
+    let free = foreign "at_free" (gc_tensor @-> returning void)
+
+    let run_backward =
+      foreign
+        "at_run_backward"
+        (ptr gc_tensor
+         @-> int
+         @-> ptr gc_tensor
+         @-> int
+         @-> ptr raw_tensor
+         @-> int
+         @-> int
+         @-> returning void)
+    ;;
+  end
+
+  module Scalar = struct
+    let int = foreign "ats_int" (int64_t @-> returning scalar)
+    let float = foreign "ats_float" (float @-> returning scalar)
+    let free = foreign "ats_free" (scalar @-> returning void)
+  end
+
+  module Serialize = struct
+    let save = foreign "at_save" (gc_tensor @-> string @-> returning void)
+    let load = foreign "at_load" (string @-> returning raw_tensor)
+
+    let save_multi =
+      foreign
+        "at_save_multi"
+        (ptr gc_tensor @-> ptr (ptr char) @-> int @-> string @-> returning void)
+    ;;
+
+    let load_multi =
+      foreign
+        "at_load_multi"
+        (ptr raw_tensor @-> ptr (ptr char) @-> int @-> string @-> returning void)
+    ;;
+
+    let load_multi_ =
+      foreign
+        "at_load_multi_"
+        (ptr gc_tensor @-> ptr (ptr char) @-> int @-> string @-> returning void)
+    ;;
+
+    let load_callback =
+      foreign
+        "at_load_callback"
+        (string
+         @-> static_funptr Ctypes.(string @-> raw_tensor @-> returning void)
+         @-> returning void)
+    ;;
+  end
+
+  module Optimizer = struct
+    let adam =
+      foreign
+        "ato_adam"
+        (float @-> float @-> float @-> float @-> float @-> returning optimizer)
+    ;;
+
+    let rmsprop =
+      foreign
+        "ato_rmsprop"
+        (float
+         (* learning rate *)
+         @-> float
+         (* alpha *)
+         @-> float
+         (* eps *)
+         @-> float
+         (* weight decay *)
+         @-> float
+         (* momentum *)
+         @-> int
+         (* centered *)
+         @-> returning optimizer)
+    ;;
+
+    let sgd =
+      foreign
+        "ato_sgd"
+        (float
+         (* learning rate *)
+         @-> float
+         (* momentum *)
+         @-> float
+         (* dampening *)
+         @-> float
+         (* weight decay *)
+         @-> bool
+         (* nesterov *)
+         @-> returning optimizer)
+    ;;
+
+    let add_parameters =
+      foreign "ato_add_parameters" (optimizer @-> ptr gc_tensor @-> int @-> returning void)
+    ;;
+
+    let set_learning_rate =
+      foreign "ato_set_learning_rate" (optimizer @-> float @-> returning void)
+    ;;
+
+    let set_momentum = foreign "ato_set_momentum" (optimizer @-> float @-> returning void)
+    let zero_grad = foreign "ato_zero_grad" (optimizer @-> returning void)
+    let step = foreign "ato_step" (optimizer @-> returning void)
+    let free = foreign "ato_free" (optimizer @-> returning void)
+  end
+
+  module Cuda = struct
+    let device_count = foreign "atc_cuda_device_count" (void @-> returning int)
+    let is_available = foreign "atc_cuda_is_available" (void @-> returning int)
+    let cudnn_is_available = foreign "atc_cudnn_is_available" (void @-> returning int)
+    let set_benchmark_cudnn = foreign "atc_set_benchmark_cudnn" (int @-> returning void)
+  end
+
+  module Ivalue = struct
+    let to_int64 = foreign "ati_to_int" (ivalue @-> returning int64_t)
+    let to_bool = foreign "ati_to_bool" (ivalue @-> returning int)
+    let to_double = foreign "ati_to_double" (ivalue @-> returning double)
+    let to_tensor = foreign "ati_to_tensor" (ivalue @-> returning raw_tensor)
+    let tuple_length = foreign "ati_tuple_length" (ivalue @-> returning int)
+    let list_length = foreign "ati_list_length" (ivalue @-> returning int)
+
+    let to_tuple =
+      foreign "ati_to_tuple" (ivalue @-> ptr ivalue @-> int @-> returning void)
+    ;;
+
+    let to_tensor_list =
+      foreign "ati_to_tensor_list" (ivalue @-> ptr raw_tensor @-> int @-> returning void)
+    ;;
+
+    let to_generic_list =
+      foreign "ati_to_generic_list" (ivalue @-> ptr ivalue @-> int @-> returning void)
+    ;;
+
+    let to_string = foreign "ati_to_string" (ivalue @-> returning string)
+    let none = foreign "ati_none" (void @-> returning ivalue)
+    let bool = foreign "ati_bool" (int @-> returning ivalue)
+    let tensor = foreign "ati_tensor" (gc_tensor @-> returning ivalue)
+    let int64 = foreign "ati_int" (int64_t @-> returning ivalue)
+    let double = foreign "ati_double" (float @-> returning ivalue)
+    let tuple = foreign "ati_tuple" (ptr ivalue @-> int @-> returning ivalue)
+
+    let tensor_list =
+      foreign "ati_tensor_list" (ptr gc_tensor @-> int @-> returning ivalue)
+    ;;
+
+    let string = foreign "ati_string" (string @-> returning ivalue)
+    let tag = foreign "ati_tag" (ivalue @-> returning int)
+    let free = foreign "ati_free" (ivalue @-> returning void)
+  end
+
+  module Module = struct
+    let load = foreign "atm_load" (string @-> int @-> returning module_)
+    let load_str = foreign "atm_load_str" (string @-> int @-> int @-> returning module_)
+
+    let forward =
+      foreign "atm_forward" (module_ @-> ptr gc_tensor @-> int @-> returning raw_tensor)
+    ;;
+
+    let forward_ =
+      foreign "atm_forward_" (module_ @-> ptr ivalue @-> int @-> returning ivalue)
+    ;;
+
+    let named_buffers = foreign "atm_named_buffers" (module_ @-> returning ivalue)
+    let free = foreign "atm_free" (module_ @-> returning void)
+  end
+
+  module Generated = Torch_bindings_generated.C (F)
+end
diff --git a/src/bindings/torch_bindings_generated.ml b/src/bindings/torch_bindings_generated.ml
new file mode 100644
index 0000000..ca88d0e
--- /dev/null
+++ b/src/bindings/torch_bindings_generated.ml
@@ -0,0 +1,17947 @@
+(* THIS FILE IS AUTOMATICALLY GENERATED, DO NOT EDIT BY HAND! *)
+
+open Ctypes
+
+module C0 (F : Cstubs.FOREIGN) = struct
+  open F
+  open Type_defs
+
+  let stubs___and__ = foreign "atg___and__" (gc_tensor @-> scalar @-> returning raw_tensor)
+
+  let stubs___and__tensor_ =
+    foreign "atg___and__tensor_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs___iand__ =
+    foreign "atg___iand__" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs___iand__tensor_ =
+    foreign "atg___iand__tensor_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs___ilshift__ =
+    foreign "atg___ilshift__" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs___ilshift__tensor_ =
+    foreign "atg___ilshift__tensor_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs___ior__ = foreign "atg___ior__" (gc_tensor @-> scalar @-> returning raw_tensor)
+
+  let stubs___ior__tensor_ =
+    foreign "atg___ior__tensor_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs___irshift__ =
+    foreign "atg___irshift__" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs___irshift__tensor_ =
+    foreign "atg___irshift__tensor_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs___ixor__ =
+    foreign "atg___ixor__" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs___ixor__tensor_ =
+    foreign "atg___ixor__tensor_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs___lshift__ =
+    foreign "atg___lshift__" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs___lshift__scalar_out_ =
+    foreign
+      "atg___lshift__scalar_out_"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs___lshift__tensor_ =
+    foreign "atg___lshift__tensor_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs___lshift__tensor_out_ =
+    foreign
+      "atg___lshift__tensor_out_"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs___or__ = foreign "atg___or__" (gc_tensor @-> scalar @-> returning raw_tensor)
+
+  let stubs___or__tensor_ =
+    foreign "atg___or__tensor_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs___rshift__ =
+    foreign "atg___rshift__" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs___rshift__scalar_out_ =
+    foreign
+      "atg___rshift__scalar_out_"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs___rshift__tensor_ =
+    foreign "atg___rshift__tensor_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs___rshift__tensor_out_ =
+    foreign
+      "atg___rshift__tensor_out_"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs___xor__ = foreign "atg___xor__" (gc_tensor @-> scalar @-> returning raw_tensor)
+
+  let stubs___xor__tensor_ =
+    foreign "atg___xor__tensor_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__adaptive_avg_pool2d =
+    foreign
+      "atg__adaptive_avg_pool2d"
+      (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__adaptive_avg_pool2d_backward =
+    foreign
+      "atg__adaptive_avg_pool2d_backward"
+      (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__adaptive_avg_pool2d_backward_out =
+    foreign
+      "atg__adaptive_avg_pool2d_backward_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__adaptive_avg_pool2d_out =
+    foreign
+      "atg__adaptive_avg_pool2d_out"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__adaptive_avg_pool3d =
+    foreign
+      "atg__adaptive_avg_pool3d"
+      (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__adaptive_avg_pool3d_backward =
+    foreign
+      "atg__adaptive_avg_pool3d_backward"
+      (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__adaptive_avg_pool3d_backward_out =
+    foreign
+      "atg__adaptive_avg_pool3d_backward_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__adaptive_avg_pool3d_out =
+    foreign
+      "atg__adaptive_avg_pool3d_out"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__add_batch_dim =
+    foreign
+      "atg__add_batch_dim"
+      (gc_tensor @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs__add_relu =
+    foreign "atg__add_relu" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__add_relu_ =
+    foreign "atg__add_relu_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__add_relu_out =
+    foreign
+      "atg__add_relu_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__add_relu_scalar =
+    foreign "atg__add_relu_scalar" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs__add_relu_scalar_ =
+    foreign "atg__add_relu_scalar_" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs__add_relu_scalar_out =
+    foreign
+      "atg__add_relu_scalar_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs__addmm_activation =
+    foreign
+      "atg__addmm_activation"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__addmm_activation_out =
+    foreign
+      "atg__addmm_activation_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__aminmax =
+    foreign "atg__aminmax" (ptr raw_tensor @-> gc_tensor @-> returning void)
+  ;;
+
+  let stubs__aminmax_dim =
+    foreign
+      "atg__aminmax_dim"
+      (ptr raw_tensor @-> gc_tensor @-> int64_t @-> int @-> returning void)
+  ;;
+
+  let stubs__aminmax_dim_out =
+    foreign
+      "atg__aminmax_dim_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs__aminmax_out =
+    foreign
+      "atg__aminmax_out"
+      (ptr raw_tensor @-> gc_tensor @-> gc_tensor @-> gc_tensor @-> returning void)
+  ;;
+
+  let stubs__amp_update_scale =
+    foreign
+      "atg__amp_update_scale"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> double
+       @-> int64_t
+       @-> returning void)
+  ;;
+
+  let stubs__amp_update_scale_ =
+    foreign
+      "atg__amp_update_scale_"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> double
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__amp_update_scale_out =
+    foreign
+      "atg__amp_update_scale_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> double
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__assert_tensor_metadata =
+    foreign
+      "atg__assert_tensor_metadata"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs__autocast_to_full_precision =
+    foreign
+      "atg__autocast_to_full_precision"
+      (gc_tensor @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__autocast_to_reduced_precision =
+    foreign
+      "atg__autocast_to_reduced_precision"
+      (gc_tensor @-> int @-> int @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__cast_byte =
+    foreign "atg__cast_byte" (gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__cast_char =
+    foreign "atg__cast_char" (gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__cast_double =
+    foreign "atg__cast_double" (gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__cast_float =
+    foreign "atg__cast_float" (gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__cast_half =
+    foreign "atg__cast_half" (gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__cast_int =
+    foreign "atg__cast_int" (gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__cast_long =
+    foreign "atg__cast_long" (gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__cast_short =
+    foreign "atg__cast_short" (gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__cdist_backward =
+    foreign
+      "atg__cdist_backward"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__cdist_backward_out =
+    foreign
+      "atg__cdist_backward_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__cholesky_solve_helper =
+    foreign
+      "atg__cholesky_solve_helper"
+      (gc_tensor @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__cholesky_solve_helper_out =
+    foreign
+      "atg__cholesky_solve_helper_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__coalesce = foreign "atg__coalesce" (gc_tensor @-> returning raw_tensor)
+
+  let stubs__coalesce_out =
+    foreign "atg__coalesce_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__coalesced =
+    foreign "atg__coalesced" (gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__coalesced_ =
+    foreign "atg__coalesced_" (gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__coalesced_out =
+    foreign "atg__coalesced_out" (gc_tensor @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__compute_linear_combination =
+    foreign
+      "atg__compute_linear_combination"
+      (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__compute_linear_combination_out =
+    foreign
+      "atg__compute_linear_combination_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__conj = foreign "atg__conj" (gc_tensor @-> returning raw_tensor)
+  let stubs__conj_copy = foreign "atg__conj_copy" (gc_tensor @-> returning raw_tensor)
+
+  let stubs__conj_copy_out =
+    foreign "atg__conj_copy_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__conj_physical =
+    foreign "atg__conj_physical" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__conj_physical_out =
+    foreign "atg__conj_physical_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__conv_depthwise2d =
+    foreign
+      "atg__conv_depthwise2d"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__conv_depthwise2d_out =
+    foreign
+      "atg__conv_depthwise2d_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__convert_indices_from_coo_to_csr =
+    foreign
+      "atg__convert_indices_from_coo_to_csr"
+      (gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__convert_indices_from_coo_to_csr_out =
+    foreign
+      "atg__convert_indices_from_coo_to_csr_out"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__convert_indices_from_csr_to_coo =
+    foreign
+      "atg__convert_indices_from_csr_to_coo"
+      (gc_tensor @-> gc_tensor @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__convert_indices_from_csr_to_coo_out =
+    foreign
+      "atg__convert_indices_from_csr_to_coo_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__convolution =
+    foreign
+      "atg__convolution"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__convolution_deprecated =
+    foreign
+      "atg__convolution_deprecated"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__convolution_mode =
+    foreign
+      "atg__convolution_mode"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> string
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__convolution_out =
+    foreign
+      "atg__convolution_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__copy_from =
+    foreign "atg__copy_from" (gc_tensor @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__copy_from_and_resize =
+    foreign "atg__copy_from_and_resize" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__copy_from_and_resize_out =
+    foreign
+      "atg__copy_from_and_resize_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__copy_from_out =
+    foreign
+      "atg__copy_from_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__cslt_compress =
+    foreign "atg__cslt_compress" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__cslt_sparse_mm =
+    foreign
+      "atg__cslt_sparse_mm"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__ctc_loss =
+    foreign
+      "atg__ctc_loss"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs__ctc_loss_backward =
+    foreign
+      "atg__ctc_loss_backward"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__ctc_loss_backward_out =
+    foreign
+      "atg__ctc_loss_backward_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__ctc_loss_backward_tensor =
+    foreign
+      "atg__ctc_loss_backward_tensor"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__ctc_loss_out =
+    foreign
+      "atg__ctc_loss_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs__ctc_loss_tensor =
+    foreign
+      "atg__ctc_loss_tensor"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs__ctc_loss_tensor_out =
+    foreign
+      "atg__ctc_loss_tensor_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs__cudnn_ctc_loss =
+    foreign
+      "atg__cudnn_ctc_loss"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs__cudnn_ctc_loss_out =
+    foreign
+      "atg__cudnn_ctc_loss_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> int
+       @-> returning void)
+  ;;
+end
+
+module C1 (F : Cstubs.FOREIGN) = struct
+  open F
+  open Type_defs
+
+  let stubs__cudnn_ctc_loss_tensor =
+    foreign
+      "atg__cudnn_ctc_loss_tensor"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs__cudnn_init_dropout_state =
+    foreign
+      "atg__cudnn_init_dropout_state"
+      (double @-> int @-> int64_t @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__cudnn_init_dropout_state_out =
+    foreign
+      "atg__cudnn_init_dropout_state_out"
+      (gc_tensor @-> double @-> int @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs__cudnn_rnn =
+    foreign
+      "atg__cudnn_rnn"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> ptr gc_tensor
+       @-> int
+       @-> int64_t
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> int
+       @-> double
+       @-> int
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> gc_tensor
+       @-> returning void)
+  ;;
+
+  let stubs__cudnn_rnn_flatten_weight =
+    foreign
+      "atg__cudnn_rnn_flatten_weight"
+      (ptr gc_tensor
+       @-> int
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__cudnn_rnn_flatten_weight_out =
+    foreign
+      "atg__cudnn_rnn_flatten_weight_out"
+      (gc_tensor
+       @-> ptr gc_tensor
+       @-> int
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__cudnn_rnn_out =
+    foreign
+      "atg__cudnn_rnn_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr gc_tensor
+       @-> int
+       @-> int64_t
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> int
+       @-> double
+       @-> int
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> gc_tensor
+       @-> returning void)
+  ;;
+
+  let stubs__debug_has_internal_overlap =
+    foreign "atg__debug_has_internal_overlap" (gc_tensor @-> returning int64_t)
+  ;;
+
+  let stubs__dim_arange =
+    foreign "atg__dim_arange" (gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs__dimi = foreign "atg__dimi" (gc_tensor @-> returning int64_t)
+  let stubs__dimv = foreign "atg__dimv" (gc_tensor @-> returning int64_t)
+
+  let stubs__dirichlet_grad =
+    foreign
+      "atg__dirichlet_grad"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__dirichlet_grad_out =
+    foreign
+      "atg__dirichlet_grad_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__efficient_attention_backward =
+    foreign
+      "atg__efficient_attention_backward"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> gc_tensor
+       @-> double
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int
+       @-> double
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs__efficientzerotensor =
+    foreign
+      "atg__efficientzerotensor"
+      (ptr int64_t @-> int @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__efficientzerotensor_out =
+    foreign
+      "atg__efficientzerotensor_out"
+      (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__embedding_bag =
+    foreign
+      "atg__embedding_bag"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> gc_tensor
+       @-> int
+       @-> int64_t
+       @-> returning void)
+  ;;
+
+  let stubs__embedding_bag_backward =
+    foreign
+      "atg__embedding_bag_backward"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> gc_tensor
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__embedding_bag_dense_backward =
+    foreign
+      "atg__embedding_bag_dense_backward"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int
+       @-> int64_t
+       @-> gc_tensor
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__embedding_bag_dense_backward_out =
+    foreign
+      "atg__embedding_bag_dense_backward_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int
+       @-> int64_t
+       @-> gc_tensor
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__embedding_bag_forward_only =
+    foreign
+      "atg__embedding_bag_forward_only"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> gc_tensor
+       @-> int
+       @-> int64_t
+       @-> returning void)
+  ;;
+
+  let stubs__embedding_bag_forward_only_out =
+    foreign
+      "atg__embedding_bag_forward_only_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> gc_tensor
+       @-> int
+       @-> int64_t
+       @-> returning void)
+  ;;
+
+  let stubs__embedding_bag_out =
+    foreign
+      "atg__embedding_bag_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> gc_tensor
+       @-> int
+       @-> int64_t
+       @-> returning void)
+  ;;
+
+  let stubs__embedding_bag_per_sample_weights_backward =
+    foreign
+      "atg__embedding_bag_per_sample_weights_backward"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__embedding_bag_per_sample_weights_backward_out =
+    foreign
+      "atg__embedding_bag_per_sample_weights_backward_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__embedding_bag_sparse_backward =
+    foreign
+      "atg__embedding_bag_sparse_backward"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int
+       @-> int64_t
+       @-> gc_tensor
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__empty_affine_quantized =
+    foreign
+      "atg__empty_affine_quantized"
+      (ptr int64_t @-> int @-> int @-> int @-> double @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs__empty_affine_quantized_out =
+    foreign
+      "atg__empty_affine_quantized_out"
+      (gc_tensor @-> ptr int64_t @-> int @-> double @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs__empty_per_channel_affine_quantized =
+    foreign
+      "atg__empty_per_channel_affine_quantized"
+      (ptr int64_t
+       @-> int
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__empty_per_channel_affine_quantized_out =
+    foreign
+      "atg__empty_per_channel_affine_quantized_out"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__euclidean_dist =
+    foreign "atg__euclidean_dist" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__euclidean_dist_out =
+    foreign
+      "atg__euclidean_dist_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__fake_quantize_learnable_per_channel_affine =
+    foreign
+      "atg__fake_quantize_learnable_per_channel_affine"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> double
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__fake_quantize_learnable_per_channel_affine_backward =
+    foreign
+      "atg__fake_quantize_learnable_per_channel_affine_backward"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> double
+       @-> returning void)
+  ;;
+
+  let stubs__fake_quantize_learnable_per_channel_affine_out =
+    foreign
+      "atg__fake_quantize_learnable_per_channel_affine_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> double
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__fake_quantize_learnable_per_tensor_affine =
+    foreign
+      "atg__fake_quantize_learnable_per_tensor_affine"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> double
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__fake_quantize_learnable_per_tensor_affine_backward =
+    foreign
+      "atg__fake_quantize_learnable_per_tensor_affine_backward"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> double
+       @-> returning void)
+  ;;
+
+  let stubs__fake_quantize_learnable_per_tensor_affine_out =
+    foreign
+      "atg__fake_quantize_learnable_per_tensor_affine_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> double
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__fake_quantize_per_tensor_affine_cachemask_tensor_qparams =
+    foreign
+      "atg__fake_quantize_per_tensor_affine_cachemask_tensor_qparams"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> returning void)
+  ;;
+
+  let stubs__fake_quantize_per_tensor_affine_cachemask_tensor_qparams_out =
+    foreign
+      "atg__fake_quantize_per_tensor_affine_cachemask_tensor_qparams_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> returning void)
+  ;;
+
+  let stubs__fft_c2c =
+    foreign
+      "atg__fft_c2c"
+      (gc_tensor @-> ptr int64_t @-> int @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__fft_c2c_out =
+    foreign
+      "atg__fft_c2c_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__fft_c2r =
+    foreign
+      "atg__fft_c2r"
+      (gc_tensor @-> ptr int64_t @-> int @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs__fft_c2r_out =
+    foreign
+      "atg__fft_c2r_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__fft_r2c =
+    foreign
+      "atg__fft_r2c"
+      (gc_tensor @-> ptr int64_t @-> int @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__fft_r2c_out =
+    foreign
+      "atg__fft_r2c_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__fill_mem_eff_dropout_mask_ =
+    foreign
+      "atg__fill_mem_eff_dropout_mask_"
+      (gc_tensor @-> double @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs__flash_attention_backward =
+    foreign
+      "atg__flash_attention_backward"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> double
+       @-> int
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs__foobar =
+    foreign "atg__foobar" (gc_tensor @-> int @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__foobar_out =
+    foreign
+      "atg__foobar_out"
+      (gc_tensor @-> gc_tensor @-> int @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__functional_assert_async =
+    foreign
+      "atg__functional_assert_async"
+      (gc_tensor @-> string @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__functional_sym_constrain_range =
+    foreign
+      "atg__functional_sym_constrain_range"
+      (scalar
+       @-> int64_t
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__functional_sym_constrain_range_for_size =
+    foreign
+      "atg__functional_sym_constrain_range_for_size"
+      (scalar
+       @-> int64_t
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__fused_adam =
+    foreign
+      "atg__fused_adam"
+      (ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> double
+       @-> double
+       @-> double
+       @-> double
+       @-> double
+       @-> int
+       @-> int
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> returning void)
+  ;;
+
+  let stubs__fused_adam_ =
+    foreign
+      "atg__fused_adam_"
+      (ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> double
+       @-> double
+       @-> double
+       @-> double
+       @-> double
+       @-> int
+       @-> int
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> returning void)
+  ;;
+
+  let stubs__fused_adam_tensor_lr_ =
+    foreign
+      "atg__fused_adam_tensor_lr_"
+      (ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> gc_tensor
+       @-> double
+       @-> double
+       @-> double
+       @-> double
+       @-> int
+       @-> int
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> returning void)
+  ;;
+
+  let stubs__fused_adam_tensor_lr_out =
+    foreign
+      "atg__fused_adam_tensor_lr_out"
+      (ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> gc_tensor
+       @-> double
+       @-> double
+       @-> double
+       @-> double
+       @-> int
+       @-> int
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> returning void)
+  ;;
+
+  let stubs__fused_adamw =
+    foreign
+      "atg__fused_adamw"
+      (ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> double
+       @-> double
+       @-> double
+       @-> double
+       @-> double
+       @-> int
+       @-> int
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> returning void)
+  ;;
+
+  let stubs__fused_adamw_ =
+    foreign
+      "atg__fused_adamw_"
+      (ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> double
+       @-> double
+       @-> double
+       @-> double
+       @-> double
+       @-> int
+       @-> int
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> returning void)
+  ;;
+
+  let stubs__fused_adamw_tensor_lr_ =
+    foreign
+      "atg__fused_adamw_tensor_lr_"
+      (ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> gc_tensor
+       @-> double
+       @-> double
+       @-> double
+       @-> double
+       @-> int
+       @-> int
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> returning void)
+  ;;
+
+  let stubs__fused_adamw_tensor_lr_out =
+    foreign
+      "atg__fused_adamw_tensor_lr_out"
+      (ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> gc_tensor
+       @-> double
+       @-> double
+       @-> double
+       @-> double
+       @-> int
+       @-> int
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> returning void)
+  ;;
+
+  let stubs__fused_dropout =
+    foreign
+      "atg__fused_dropout"
+      (ptr raw_tensor @-> gc_tensor @-> double @-> returning void)
+  ;;
+
+  let stubs__fused_dropout_out =
+    foreign
+      "atg__fused_dropout_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> returning void)
+  ;;
+
+  let stubs__fused_moving_avg_obs_fq_helper =
+    foreign
+      "atg__fused_moving_avg_obs_fq_helper"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> int
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs__fused_moving_avg_obs_fq_helper_functional =
+    foreign
+      "atg__fused_moving_avg_obs_fq_helper_functional"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> int
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs__fused_moving_avg_obs_fq_helper_out =
+    foreign
+      "atg__fused_moving_avg_obs_fq_helper_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> int
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs__fused_sdp_choice =
+    foreign
+      "atg__fused_sdp_choice"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> returning int64_t)
+  ;;
+
+  let stubs__fw_primal =
+    foreign "atg__fw_primal" (gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs__fw_primal_copy =
+    foreign "atg__fw_primal_copy" (gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs__fw_primal_copy_out =
+    foreign
+      "atg__fw_primal_copy_out"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs__gather_sparse_backward =
+    foreign
+      "atg__gather_sparse_backward"
+      (gc_tensor @-> int64_t @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__grid_sampler_2d_cpu_fallback =
+    foreign
+      "atg__grid_sampler_2d_cpu_fallback"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__grid_sampler_2d_cpu_fallback_backward =
+    foreign
+      "atg__grid_sampler_2d_cpu_fallback_backward"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs__grid_sampler_2d_cpu_fallback_out =
+    foreign
+      "atg__grid_sampler_2d_cpu_fallback_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__has_compatible_shallow_copy_type =
+    foreign
+      "atg__has_compatible_shallow_copy_type"
+      (gc_tensor @-> gc_tensor @-> returning bool)
+  ;;
+
+  let stubs__has_same_storage_numel =
+    foreign "atg__has_same_storage_numel" (gc_tensor @-> gc_tensor @-> returning bool)
+  ;;
+
+  let stubs__histogramdd_bin_edges =
+    foreign
+      "atg__histogramdd_bin_edges"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr double
+       @-> int
+       @-> gc_tensor
+       @-> int
+       @-> returning (ptr raw_tensor))
+  ;;
+
+  let stubs__histogramdd_bin_edges_out =
+    foreign
+      "atg__histogramdd_bin_edges_out"
+      (ptr gc_tensor
+       @-> int
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr double
+       @-> int
+       @-> gc_tensor
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs__histogramdd_from_bin_cts =
+    foreign
+      "atg__histogramdd_from_bin_cts"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr double
+       @-> int
+       @-> gc_tensor
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__histogramdd_from_bin_cts_out =
+    foreign
+      "atg__histogramdd_from_bin_cts_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr double
+       @-> int
+       @-> gc_tensor
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__histogramdd_from_bin_tensors =
+    foreign
+      "atg__histogramdd_from_bin_tensors"
+      (gc_tensor @-> ptr gc_tensor @-> int @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__histogramdd_from_bin_tensors_out =
+    foreign
+      "atg__histogramdd_from_bin_tensors_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr gc_tensor
+       @-> int
+       @-> gc_tensor
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__index_put_impl =
+    foreign
+      "atg__index_put_impl"
+      (gc_tensor
+       @-> ptr gc_tensor
+       @-> int
+       @-> gc_tensor
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__index_put_impl_ =
+    foreign
+      "atg__index_put_impl_"
+      (gc_tensor
+       @-> ptr gc_tensor
+       @-> int
+       @-> gc_tensor
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__index_put_impl_out =
+    foreign
+      "atg__index_put_impl_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr gc_tensor
+       @-> int
+       @-> gc_tensor
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__indices = foreign "atg__indices" (gc_tensor @-> returning raw_tensor)
+
+  let stubs__indices_copy =
+    foreign "atg__indices_copy" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__indices_copy_out =
+    foreign "atg__indices_copy_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__int_mm =
+    foreign "atg__int_mm" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__int_mm_out =
+    foreign
+      "atg__int_mm_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__is_all_true = foreign "atg__is_all_true" (gc_tensor @-> returning raw_tensor)
+  let stubs__is_any_true = foreign "atg__is_any_true" (gc_tensor @-> returning raw_tensor)
+  let stubs__is_zerotensor = foreign "atg__is_zerotensor" (gc_tensor @-> returning bool)
+
+  let stubs__linalg_check_errors =
+    foreign "atg__linalg_check_errors" (gc_tensor @-> string @-> int @-> returning void)
+  ;;
+
+  let stubs__linalg_det =
+    foreign "atg__linalg_det" (ptr raw_tensor @-> gc_tensor @-> returning void)
+  ;;
+
+  let stubs__linalg_det_result =
+    foreign
+      "atg__linalg_det_result"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> returning void)
+  ;;
+
+  let stubs__linalg_eigh =
+    foreign
+      "atg__linalg_eigh"
+      (ptr raw_tensor @-> gc_tensor @-> string @-> int @-> returning void)
+  ;;
+
+  let stubs__linalg_eigh_eigenvalues =
+    foreign
+      "atg__linalg_eigh_eigenvalues"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> string
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs__linalg_slogdet =
+    foreign "atg__linalg_slogdet" (ptr raw_tensor @-> gc_tensor @-> returning void)
+  ;;
+
+  let stubs__linalg_slogdet_sign =
+    foreign
+      "atg__linalg_slogdet_sign"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> returning void)
+  ;;
+end
+
+module C2 (F : Cstubs.FOREIGN) = struct
+  open F
+  open Type_defs
+
+  let stubs__linalg_solve_ex =
+    foreign
+      "atg__linalg_solve_ex"
+      (ptr raw_tensor @-> gc_tensor @-> gc_tensor @-> int @-> int @-> returning void)
+  ;;
+
+  let stubs__linalg_solve_ex_result =
+    foreign
+      "atg__linalg_solve_ex_result"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs__linalg_svd =
+    foreign
+      "atg__linalg_svd"
+      (ptr raw_tensor @-> gc_tensor @-> int @-> int @-> string @-> returning void)
+  ;;
+
+  let stubs__linalg_svd_u =
+    foreign
+      "atg__linalg_svd_u"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> int
+       @-> string
+       @-> returning void)
+  ;;
+
+  let stubs__log_softmax =
+    foreign "atg__log_softmax" (gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__log_softmax_backward_data =
+    foreign
+      "atg__log_softmax_backward_data"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__log_softmax_backward_data_out =
+    foreign
+      "atg__log_softmax_backward_data_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__log_softmax_out =
+    foreign
+      "atg__log_softmax_out"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__logcumsumexp =
+    foreign "atg__logcumsumexp" (gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs__logcumsumexp_out =
+    foreign
+      "atg__logcumsumexp_out"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs__lstm_mps =
+    foreign
+      "atg__lstm_mps"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> int
+       @-> int64_t
+       @-> double
+       @-> int
+       @-> int
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs__lstm_mps_out =
+    foreign
+      "atg__lstm_mps_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> int
+       @-> int64_t
+       @-> double
+       @-> int
+       @-> int
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs__lu_with_info =
+    foreign
+      "atg__lu_with_info"
+      (ptr raw_tensor @-> gc_tensor @-> int @-> int @-> returning void)
+  ;;
+
+  let stubs__make_dep_token =
+    foreign "atg__make_dep_token" (int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__make_dual =
+    foreign "atg__make_dual" (gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs__make_dual_copy =
+    foreign
+      "atg__make_dual_copy"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs__make_dual_copy_out =
+    foreign
+      "atg__make_dual_copy_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs__make_per_channel_quantized_tensor =
+    foreign
+      "atg__make_per_channel_quantized_tensor"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs__make_per_channel_quantized_tensor_out =
+    foreign
+      "atg__make_per_channel_quantized_tensor_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__make_per_tensor_quantized_tensor =
+    foreign
+      "atg__make_per_tensor_quantized_tensor"
+      (gc_tensor @-> double @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs__make_per_tensor_quantized_tensor_out =
+    foreign
+      "atg__make_per_tensor_quantized_tensor_out"
+      (gc_tensor @-> gc_tensor @-> double @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs__masked_scale =
+    foreign
+      "atg__masked_scale"
+      (gc_tensor @-> gc_tensor @-> double @-> returning raw_tensor)
+  ;;
+
+  let stubs__masked_scale_out =
+    foreign
+      "atg__masked_scale_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> double @-> returning raw_tensor)
+  ;;
+
+  let stubs__masked_softmax =
+    foreign
+      "atg__masked_softmax"
+      (gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__masked_softmax_backward =
+    foreign
+      "atg__masked_softmax_backward"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__masked_softmax_backward_out =
+    foreign
+      "atg__masked_softmax_backward_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__masked_softmax_out =
+    foreign
+      "atg__masked_softmax_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__mkldnn_reshape =
+    foreign
+      "atg__mkldnn_reshape"
+      (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__mkldnn_reshape_out =
+    foreign
+      "atg__mkldnn_reshape_out"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__mkldnn_transpose =
+    foreign
+      "atg__mkldnn_transpose"
+      (gc_tensor @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs__mkldnn_transpose_ =
+    foreign
+      "atg__mkldnn_transpose_"
+      (gc_tensor @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs__mkldnn_transpose_out =
+    foreign
+      "atg__mkldnn_transpose_out"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs__mps_convolution =
+    foreign
+      "atg__mps_convolution"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__mps_convolution_out =
+    foreign
+      "atg__mps_convolution_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__mps_convolution_transpose =
+    foreign
+      "atg__mps_convolution_transpose"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__mps_convolution_transpose_out =
+    foreign
+      "atg__mps_convolution_transpose_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__native_batch_norm_legit =
+    foreign
+      "atg__native_batch_norm_legit"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> double
+       @-> double
+       @-> returning void)
+  ;;
+
+  let stubs__native_batch_norm_legit_functional =
+    foreign
+      "atg__native_batch_norm_legit_functional"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> double
+       @-> double
+       @-> returning void)
+  ;;
+
+  let stubs__native_batch_norm_legit_no_stats =
+    foreign
+      "atg__native_batch_norm_legit_no_stats"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> double
+       @-> double
+       @-> returning void)
+  ;;
+
+  let stubs__native_batch_norm_legit_no_stats_out =
+    foreign
+      "atg__native_batch_norm_legit_no_stats_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> double
+       @-> double
+       @-> returning void)
+  ;;
+
+  let stubs__native_batch_norm_legit_no_training =
+    foreign
+      "atg__native_batch_norm_legit_no_training"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> double
+       @-> returning void)
+  ;;
+
+  let stubs__native_batch_norm_legit_no_training_out =
+    foreign
+      "atg__native_batch_norm_legit_no_training_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> double
+       @-> returning void)
+  ;;
+
+  let stubs__native_batch_norm_legit_out =
+    foreign
+      "atg__native_batch_norm_legit_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> double
+       @-> double
+       @-> returning void)
+  ;;
+
+  let stubs__native_multi_head_attention =
+    foreign
+      "atg__native_multi_head_attention"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs__native_multi_head_attention_out =
+    foreign
+      "atg__native_multi_head_attention_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs__neg_view = foreign "atg__neg_view" (gc_tensor @-> returning raw_tensor)
+
+  let stubs__neg_view_copy =
+    foreign "atg__neg_view_copy" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__neg_view_copy_out =
+    foreign "atg__neg_view_copy_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__nested_from_padded =
+    foreign
+      "atg__nested_from_padded"
+      (gc_tensor @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__nested_from_padded_and_nested_example =
+    foreign
+      "atg__nested_from_padded_and_nested_example"
+      (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__nested_from_padded_and_nested_example_out =
+    foreign
+      "atg__nested_from_padded_and_nested_example_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__nested_from_padded_out =
+    foreign
+      "atg__nested_from_padded_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__nested_select_backward =
+    foreign
+      "atg__nested_select_backward"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs__nested_sum_backward =
+    foreign
+      "atg__nested_sum_backward"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__nested_view_from_buffer =
+    foreign
+      "atg__nested_view_from_buffer"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__nested_view_from_buffer_copy =
+    foreign
+      "atg__nested_view_from_buffer_copy"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__nested_view_from_buffer_copy_out =
+    foreign
+      "atg__nested_view_from_buffer_copy_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__new_zeros_with_same_feature_meta =
+    foreign
+      "atg__new_zeros_with_same_feature_meta"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs__new_zeros_with_same_feature_meta_out =
+    foreign
+      "atg__new_zeros_with_same_feature_meta_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs__nnpack_available = foreign "atg__nnpack_available" (void @-> returning bool)
+
+  let stubs__nnpack_spatial_convolution =
+    foreign
+      "atg__nnpack_spatial_convolution"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__nnpack_spatial_convolution_out =
+    foreign
+      "atg__nnpack_spatial_convolution_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__nnz = foreign "atg__nnz" (gc_tensor @-> returning int64_t)
+
+  let stubs__pack_padded_sequence =
+    foreign
+      "atg__pack_padded_sequence"
+      (ptr raw_tensor @-> gc_tensor @-> gc_tensor @-> int @-> returning void)
+  ;;
+
+  let stubs__pack_padded_sequence_backward =
+    foreign
+      "atg__pack_padded_sequence_backward"
+      (gc_tensor @-> ptr int64_t @-> int @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__pack_padded_sequence_out =
+    foreign
+      "atg__pack_padded_sequence_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs__pad_circular =
+    foreign
+      "atg__pad_circular"
+      (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__pad_enum =
+    foreign
+      "atg__pad_enum"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__pad_packed_sequence =
+    foreign
+      "atg__pad_packed_sequence"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> scalar
+       @-> int64_t
+       @-> returning void)
+  ;;
+
+  let stubs__pdist_backward =
+    foreign
+      "atg__pdist_backward"
+      (gc_tensor @-> gc_tensor @-> double @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__pdist_backward_out =
+    foreign
+      "atg__pdist_backward_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__pin_memory =
+    foreign "atg__pin_memory" (gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__pin_memory_out =
+    foreign
+      "atg__pin_memory_out"
+      (gc_tensor @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__prelu_kernel =
+    foreign "atg__prelu_kernel" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__prelu_kernel_backward =
+    foreign
+      "atg__prelu_kernel_backward"
+      (ptr raw_tensor @-> gc_tensor @-> gc_tensor @-> gc_tensor @-> returning void)
+  ;;
+
+  let stubs__propagate_xla_data =
+    foreign "atg__propagate_xla_data" (gc_tensor @-> gc_tensor @-> returning void)
+  ;;
+
+  let stubs__remove_batch_dim =
+    foreign
+      "atg__remove_batch_dim"
+      (gc_tensor @-> int64_t @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs__reshape_alias =
+    foreign
+      "atg__reshape_alias"
+      (gc_tensor @-> ptr int64_t @-> int @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__reshape_alias_copy =
+    foreign
+      "atg__reshape_alias_copy"
+      (gc_tensor @-> ptr int64_t @-> int @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__reshape_alias_copy_out =
+    foreign
+      "atg__reshape_alias_copy_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__reshape_copy =
+    foreign
+      "atg__reshape_copy"
+      (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__reshape_from_tensor =
+    foreign "atg__reshape_from_tensor" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__resize_output =
+    foreign
+      "atg__resize_output"
+      (gc_tensor @-> ptr int64_t @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__resize_output_ =
+    foreign
+      "atg__resize_output_"
+      (gc_tensor @-> ptr int64_t @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__resize_output_out =
+    foreign
+      "atg__resize_output_out"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__rowwise_prune =
+    foreign
+      "atg__rowwise_prune"
+      (ptr raw_tensor @-> gc_tensor @-> gc_tensor @-> int @-> returning void)
+  ;;
+
+  let stubs__sample_dirichlet =
+    foreign "atg__sample_dirichlet" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__sample_dirichlet_out =
+    foreign "atg__sample_dirichlet_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__saturate_weight_to_fp16 =
+    foreign "atg__saturate_weight_to_fp16" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__scaled_dot_product_attention_math =
+    foreign
+      "atg__scaled_dot_product_attention_math"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> int
+       @-> gc_tensor
+       @-> double
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs__scaled_dot_product_efficient_attention =
+    foreign
+      "atg__scaled_dot_product_efficient_attention"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs__scaled_dot_product_flash_attention_backward =
+    foreign
+      "atg__scaled_dot_product_flash_attention_backward"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> double
+       @-> int
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs__scaled_mm =
+    foreign
+      "atg__scaled_mm"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> returning void)
+  ;;
+
+  let stubs__scaled_mm_out =
+    foreign
+      "atg__scaled_mm_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> returning void)
+  ;;
+
+  let stubs__scatter_reduce =
+    foreign
+      "atg__scatter_reduce"
+      (gc_tensor
+       @-> int64_t
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> string
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__scatter_reduce_ =
+    foreign
+      "atg__scatter_reduce_"
+      (gc_tensor
+       @-> int64_t
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> string
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__scatter_reduce_two_out =
+    foreign
+      "atg__scatter_reduce_two_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> string
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__segment_reduce_backward =
+    foreign
+      "atg__segment_reduce_backward"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> string
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> scalar
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__segment_reduce_backward_out =
+    foreign
+      "atg__segment_reduce_backward_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> string
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> scalar
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__shape_as_tensor =
+    foreign "atg__shape_as_tensor" (gc_tensor @-> returning raw_tensor)
+  ;;
+end
+
+module C3 (F : Cstubs.FOREIGN) = struct
+  open F
+  open Type_defs
+
+  let stubs__slow_conv2d_backward =
+    foreign
+      "atg__slow_conv2d_backward"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs__sobol_engine_draw =
+    foreign
+      "atg__sobol_engine_draw"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs__sobol_engine_ff_ =
+    foreign
+      "atg__sobol_engine_ff_"
+      (gc_tensor
+       @-> int64_t
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__sobol_engine_initialize_state_ =
+    foreign
+      "atg__sobol_engine_initialize_state_"
+      (gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs__sobol_engine_scramble_ =
+    foreign
+      "atg__sobol_engine_scramble_"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs__softmax =
+    foreign "atg__softmax" (gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__softmax_backward_data =
+    foreign
+      "atg__softmax_backward_data"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__softmax_backward_data_out =
+    foreign
+      "atg__softmax_backward_data_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__softmax_out =
+    foreign
+      "atg__softmax_out"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__sparse_addmm =
+    foreign
+      "atg__sparse_addmm"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__sparse_addmm_out =
+    foreign
+      "atg__sparse_addmm_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__sparse_broadcast_to =
+    foreign
+      "atg__sparse_broadcast_to"
+      (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__sparse_broadcast_to_copy =
+    foreign
+      "atg__sparse_broadcast_to_copy"
+      (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__sparse_broadcast_to_copy_out =
+    foreign
+      "atg__sparse_broadcast_to_copy_out"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__sparse_bsc_tensor_unsafe =
+    foreign
+      "atg__sparse_bsc_tensor_unsafe"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__sparse_bsr_tensor_unsafe =
+    foreign
+      "atg__sparse_bsr_tensor_unsafe"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__sparse_compressed_tensor_unsafe =
+    foreign
+      "atg__sparse_compressed_tensor_unsafe"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__sparse_coo_tensor_unsafe =
+    foreign
+      "atg__sparse_coo_tensor_unsafe"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__sparse_coo_tensor_with_dims =
+    foreign
+      "atg__sparse_coo_tensor_with_dims"
+      (int64_t
+       @-> int64_t
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__sparse_coo_tensor_with_dims_and_tensors =
+    foreign
+      "atg__sparse_coo_tensor_with_dims_and_tensors"
+      (int64_t
+       @-> int64_t
+       @-> ptr int64_t
+       @-> int
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__sparse_coo_tensor_with_dims_and_tensors_out =
+    foreign
+      "atg__sparse_coo_tensor_with_dims_and_tensors_out"
+      (gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> ptr int64_t
+       @-> int
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__sparse_coo_tensor_with_dims_out =
+    foreign
+      "atg__sparse_coo_tensor_with_dims_out"
+      (gc_tensor @-> int64_t @-> int64_t @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__sparse_csc_tensor_unsafe =
+    foreign
+      "atg__sparse_csc_tensor_unsafe"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__sparse_csr_prod =
+    foreign
+      "atg__sparse_csr_prod"
+      (gc_tensor @-> ptr int64_t @-> int @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__sparse_csr_prod_dim_dtype_out =
+    foreign
+      "atg__sparse_csr_prod_dim_dtype_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__sparse_csr_sum =
+    foreign
+      "atg__sparse_csr_sum"
+      (gc_tensor @-> ptr int64_t @-> int @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__sparse_csr_sum_dim_dtype_out =
+    foreign
+      "atg__sparse_csr_sum_dim_dtype_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__sparse_csr_tensor_unsafe =
+    foreign
+      "atg__sparse_csr_tensor_unsafe"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__sparse_log_softmax =
+    foreign
+      "atg__sparse_log_softmax"
+      (gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__sparse_log_softmax_backward_data =
+    foreign
+      "atg__sparse_log_softmax_backward_data"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__sparse_log_softmax_backward_data_out =
+    foreign
+      "atg__sparse_log_softmax_backward_data_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__sparse_log_softmax_int =
+    foreign
+      "atg__sparse_log_softmax_int"
+      (gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__sparse_log_softmax_out =
+    foreign
+      "atg__sparse_log_softmax_out"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__sparse_mask_projection =
+    foreign
+      "atg__sparse_mask_projection"
+      (gc_tensor @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__sparse_mask_projection_out =
+    foreign
+      "atg__sparse_mask_projection_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__sparse_mm =
+    foreign "atg__sparse_mm" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__sparse_mm_reduce =
+    foreign
+      "atg__sparse_mm_reduce"
+      (gc_tensor @-> gc_tensor @-> string @-> returning raw_tensor)
+  ;;
+
+  let stubs__sparse_mm_reduce_impl =
+    foreign
+      "atg__sparse_mm_reduce_impl"
+      (ptr raw_tensor @-> gc_tensor @-> gc_tensor @-> string @-> returning void)
+  ;;
+
+  let stubs__sparse_semi_structured_linear =
+    foreign
+      "atg__sparse_semi_structured_linear"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> string
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__sparse_softmax =
+    foreign "atg__sparse_softmax" (gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__sparse_softmax_backward_data =
+    foreign
+      "atg__sparse_softmax_backward_data"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__sparse_softmax_backward_data_out =
+    foreign
+      "atg__sparse_softmax_backward_data_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__sparse_softmax_int =
+    foreign
+      "atg__sparse_softmax_int"
+      (gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__sparse_softmax_out =
+    foreign
+      "atg__sparse_softmax_out"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__sparse_sparse_matmul =
+    foreign "atg__sparse_sparse_matmul" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__sparse_sparse_matmul_out =
+    foreign
+      "atg__sparse_sparse_matmul_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__sparse_sum = foreign "atg__sparse_sum" (gc_tensor @-> returning raw_tensor)
+
+  let stubs__sparse_sum_backward =
+    foreign
+      "atg__sparse_sum_backward"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__sparse_sum_backward_out =
+    foreign
+      "atg__sparse_sum_backward_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__sparse_sum_dim =
+    foreign
+      "atg__sparse_sum_dim"
+      (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__sparse_sum_dim_dtype =
+    foreign
+      "atg__sparse_sum_dim_dtype"
+      (gc_tensor @-> ptr int64_t @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__sparse_sum_dim_out =
+    foreign
+      "atg__sparse_sum_dim_out"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__sparse_sum_dtype =
+    foreign "atg__sparse_sum_dtype" (gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__spdiags =
+    foreign
+      "atg__spdiags"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__spdiags_out =
+    foreign
+      "atg__spdiags_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__stack =
+    foreign "atg__stack" (ptr gc_tensor @-> int @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs__stack_out =
+    foreign
+      "atg__stack_out"
+      (gc_tensor @-> ptr gc_tensor @-> int @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs__standard_gamma =
+    foreign "atg__standard_gamma" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__standard_gamma_grad =
+    foreign "atg__standard_gamma_grad" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__standard_gamma_grad_out =
+    foreign
+      "atg__standard_gamma_grad_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__standard_gamma_out =
+    foreign "atg__standard_gamma_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__test_ambiguous_defaults =
+    foreign
+      "atg__test_ambiguous_defaults"
+      (gc_tensor @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs__test_ambiguous_defaults_b =
+    foreign
+      "atg__test_ambiguous_defaults_b"
+      (gc_tensor @-> int64_t @-> string @-> returning raw_tensor)
+  ;;
+
+  let stubs__test_autograd_multiple_dispatch =
+    foreign "atg__test_autograd_multiple_dispatch" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__test_autograd_multiple_dispatch_fullcoverage_out =
+    foreign
+      "atg__test_autograd_multiple_dispatch_fullcoverage_out"
+      (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__test_autograd_multiple_dispatch_ntonly =
+    foreign
+      "atg__test_autograd_multiple_dispatch_ntonly"
+      (gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__test_autograd_multiple_dispatch_view =
+    foreign
+      "atg__test_autograd_multiple_dispatch_view"
+      (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__test_autograd_multiple_dispatch_view_copy =
+    foreign
+      "atg__test_autograd_multiple_dispatch_view_copy"
+      (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__test_autograd_multiple_dispatch_view_copy_out =
+    foreign
+      "atg__test_autograd_multiple_dispatch_view_copy_out"
+      (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__test_check_tensor =
+    foreign "atg__test_check_tensor" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__test_functorch_fallback =
+    foreign
+      "atg__test_functorch_fallback"
+      (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__test_functorch_fallback_out =
+    foreign
+      "atg__test_functorch_fallback_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__test_optional_filled_intlist =
+    foreign
+      "atg__test_optional_filled_intlist"
+      (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__test_optional_filled_intlist_out =
+    foreign
+      "atg__test_optional_filled_intlist_out"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__test_optional_floatlist =
+    foreign
+      "atg__test_optional_floatlist"
+      (gc_tensor @-> ptr double @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__test_optional_floatlist_out =
+    foreign
+      "atg__test_optional_floatlist_out"
+      (gc_tensor @-> gc_tensor @-> ptr double @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__test_optional_intlist =
+    foreign
+      "atg__test_optional_intlist"
+      (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__test_optional_intlist_out =
+    foreign
+      "atg__test_optional_intlist_out"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__test_serialization_subcmul =
+    foreign
+      "atg__test_serialization_subcmul"
+      (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__test_string_default =
+    foreign
+      "atg__test_string_default"
+      (gc_tensor @-> string @-> string @-> returning raw_tensor)
+  ;;
+
+  let stubs__test_warn_in_autograd =
+    foreign "atg__test_warn_in_autograd" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__test_warn_in_autograd_out =
+    foreign
+      "atg__test_warn_in_autograd_out"
+      (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__thnn_differentiable_gru_cell_backward =
+    foreign
+      "atg__thnn_differentiable_gru_cell_backward"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> returning void)
+  ;;
+
+  let stubs__thnn_differentiable_lstm_cell_backward =
+    foreign
+      "atg__thnn_differentiable_lstm_cell_backward"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> returning void)
+  ;;
+
+  let stubs__thnn_fused_gru_cell =
+    foreign
+      "atg__thnn_fused_gru_cell"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> returning void)
+  ;;
+
+  let stubs__thnn_fused_gru_cell_backward =
+    foreign
+      "atg__thnn_fused_gru_cell_backward"
+      (ptr raw_tensor @-> gc_tensor @-> gc_tensor @-> int @-> returning void)
+  ;;
+
+  let stubs__thnn_fused_gru_cell_backward_out =
+    foreign
+      "atg__thnn_fused_gru_cell_backward_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs__thnn_fused_gru_cell_out =
+    foreign
+      "atg__thnn_fused_gru_cell_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> returning void)
+  ;;
+
+  let stubs__thnn_fused_lstm_cell =
+    foreign
+      "atg__thnn_fused_lstm_cell"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> returning void)
+  ;;
+
+  let stubs__thnn_fused_lstm_cell_backward =
+    foreign
+      "atg__thnn_fused_lstm_cell_backward"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs__thnn_fused_lstm_cell_backward_impl =
+    foreign
+      "atg__thnn_fused_lstm_cell_backward_impl"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs__thnn_fused_lstm_cell_backward_impl_out =
+    foreign
+      "atg__thnn_fused_lstm_cell_backward_impl_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs__thnn_fused_lstm_cell_out =
+    foreign
+      "atg__thnn_fused_lstm_cell_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> returning void)
+  ;;
+
+  let stubs__to_copy =
+    foreign "atg__to_copy" (gc_tensor @-> int @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__to_copy_out =
+    foreign "atg__to_copy_out" (gc_tensor @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__to_cpu =
+    foreign "atg__to_cpu" (ptr gc_tensor @-> int @-> returning (ptr raw_tensor))
+  ;;
+
+  let stubs__to_dense =
+    foreign "atg__to_dense" (gc_tensor @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__to_dense_out =
+    foreign
+      "atg__to_dense_out"
+      (gc_tensor @-> gc_tensor @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__to_sparse_bsc =
+    foreign
+      "atg__to_sparse_bsc"
+      (gc_tensor @-> ptr int64_t @-> int @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__to_sparse_bsc_out =
+    foreign
+      "atg__to_sparse_bsc_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+end
+
+module C4 (F : Cstubs.FOREIGN) = struct
+  open F
+  open Type_defs
+
+  let stubs__to_sparse_bsr =
+    foreign
+      "atg__to_sparse_bsr"
+      (gc_tensor @-> ptr int64_t @-> int @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__to_sparse_bsr_out =
+    foreign
+      "atg__to_sparse_bsr_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__to_sparse_csc =
+    foreign "atg__to_sparse_csc" (gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__to_sparse_csc_out =
+    foreign
+      "atg__to_sparse_csc_out"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__to_sparse_csr =
+    foreign "atg__to_sparse_csr" (gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__to_sparse_csr_out =
+    foreign
+      "atg__to_sparse_csr_out"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__to_sparse_semi_structured =
+    foreign
+      "atg__to_sparse_semi_structured"
+      (ptr raw_tensor @-> gc_tensor @-> returning void)
+  ;;
+
+  let stubs__transform_bias_rescale_qkv =
+    foreign
+      "atg__transform_bias_rescale_qkv"
+      (ptr raw_tensor @-> gc_tensor @-> gc_tensor @-> int64_t @-> returning void)
+  ;;
+
+  let stubs__transform_bias_rescale_qkv_out =
+    foreign
+      "atg__transform_bias_rescale_qkv_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> returning void)
+  ;;
+
+  let stubs__transformer_encoder_layer_fwd =
+    foreign
+      "atg__transformer_encoder_layer_fwd"
+      (gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> int
+       @-> double
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__transformer_encoder_layer_fwd_out =
+    foreign
+      "atg__transformer_encoder_layer_fwd_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> int
+       @-> double
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__trilinear =
+    foreign
+      "atg__trilinear"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__trilinear_out =
+    foreign
+      "atg__trilinear_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__triton_multi_head_attention =
+    foreign
+      "atg__triton_multi_head_attention"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__triton_multi_head_attention_out =
+    foreign
+      "atg__triton_multi_head_attention_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__triton_scaled_dot_attention =
+    foreign
+      "atg__triton_scaled_dot_attention"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> double @-> returning raw_tensor)
+  ;;
+
+  let stubs__triton_scaled_dot_attention_out =
+    foreign
+      "atg__triton_scaled_dot_attention_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__unique =
+    foreign "atg__unique" (ptr raw_tensor @-> gc_tensor @-> int @-> int @-> returning void)
+  ;;
+
+  let stubs__unique2 =
+    foreign
+      "atg__unique2"
+      (ptr raw_tensor @-> gc_tensor @-> int @-> int @-> int @-> returning void)
+  ;;
+
+  let stubs__unique2_out =
+    foreign
+      "atg__unique2_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> int
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs__unique_out =
+    foreign
+      "atg__unique_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs__unpack_dual =
+    foreign
+      "atg__unpack_dual"
+      (ptr raw_tensor @-> gc_tensor @-> int64_t @-> returning void)
+  ;;
+
+  let stubs__unsafe_index =
+    foreign
+      "atg__unsafe_index"
+      (gc_tensor @-> ptr gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__unsafe_index_put =
+    foreign
+      "atg__unsafe_index_put"
+      (gc_tensor @-> ptr gc_tensor @-> int @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__unsafe_view =
+    foreign "atg__unsafe_view" (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__unsafe_view_out =
+    foreign
+      "atg__unsafe_view_out"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__upsample_bicubic2d_aa =
+    foreign
+      "atg__upsample_bicubic2d_aa"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__upsample_bicubic2d_aa_backward =
+    foreign
+      "atg__upsample_bicubic2d_aa_backward"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__upsample_bicubic2d_aa_backward_grad_input =
+    foreign
+      "atg__upsample_bicubic2d_aa_backward_grad_input"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__upsample_bicubic2d_aa_out =
+    foreign
+      "atg__upsample_bicubic2d_aa_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__upsample_bicubic2d_aa_vec =
+    foreign
+      "atg__upsample_bicubic2d_aa_vec"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> ptr double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__upsample_bilinear2d_aa =
+    foreign
+      "atg__upsample_bilinear2d_aa"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__upsample_bilinear2d_aa_backward =
+    foreign
+      "atg__upsample_bilinear2d_aa_backward"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__upsample_bilinear2d_aa_backward_grad_input =
+    foreign
+      "atg__upsample_bilinear2d_aa_backward_grad_input"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__upsample_bilinear2d_aa_out =
+    foreign
+      "atg__upsample_bilinear2d_aa_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__upsample_bilinear2d_aa_vec =
+    foreign
+      "atg__upsample_bilinear2d_aa_vec"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> ptr double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__upsample_nearest_exact1d =
+    foreign
+      "atg__upsample_nearest_exact1d"
+      (gc_tensor @-> ptr int64_t @-> int @-> double @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__upsample_nearest_exact1d_backward =
+    foreign
+      "atg__upsample_nearest_exact1d_backward"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__upsample_nearest_exact1d_backward_grad_input =
+    foreign
+      "atg__upsample_nearest_exact1d_backward_grad_input"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__upsample_nearest_exact1d_out =
+    foreign
+      "atg__upsample_nearest_exact1d_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__upsample_nearest_exact1d_vec =
+    foreign
+      "atg__upsample_nearest_exact1d_vec"
+      (gc_tensor @-> ptr int64_t @-> int @-> ptr double @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__upsample_nearest_exact2d =
+    foreign
+      "atg__upsample_nearest_exact2d"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__upsample_nearest_exact2d_backward =
+    foreign
+      "atg__upsample_nearest_exact2d_backward"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__upsample_nearest_exact2d_backward_grad_input =
+    foreign
+      "atg__upsample_nearest_exact2d_backward_grad_input"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__upsample_nearest_exact2d_out =
+    foreign
+      "atg__upsample_nearest_exact2d_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__upsample_nearest_exact2d_vec =
+    foreign
+      "atg__upsample_nearest_exact2d_vec"
+      (gc_tensor @-> ptr int64_t @-> int @-> ptr double @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__upsample_nearest_exact3d =
+    foreign
+      "atg__upsample_nearest_exact3d"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__upsample_nearest_exact3d_backward =
+    foreign
+      "atg__upsample_nearest_exact3d_backward"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__upsample_nearest_exact3d_backward_grad_input =
+    foreign
+      "atg__upsample_nearest_exact3d_backward_grad_input"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__upsample_nearest_exact3d_out =
+    foreign
+      "atg__upsample_nearest_exact3d_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs__upsample_nearest_exact3d_vec =
+    foreign
+      "atg__upsample_nearest_exact3d_vec"
+      (gc_tensor @-> ptr int64_t @-> int @-> ptr double @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs__use_cudnn_ctc_loss =
+    foreign
+      "atg__use_cudnn_ctc_loss"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> returning bool)
+  ;;
+
+  let stubs__use_cudnn_ctc_loss_tensor =
+    foreign
+      "atg__use_cudnn_ctc_loss_tensor"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> gc_tensor @-> int64_t @-> returning bool)
+  ;;
+
+  let stubs__use_cudnn_rnn_flatten_weight =
+    foreign "atg__use_cudnn_rnn_flatten_weight" (void @-> returning bool)
+  ;;
+
+  let stubs__validate_compressed_sparse_indices =
+    foreign
+      "atg__validate_compressed_sparse_indices"
+      (int
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> returning void)
+  ;;
+
+  let stubs__validate_sparse_bsc_tensor_args =
+    foreign
+      "atg__validate_sparse_bsc_tensor_args"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> returning void)
+  ;;
+
+  let stubs__validate_sparse_bsr_tensor_args =
+    foreign
+      "atg__validate_sparse_bsr_tensor_args"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> returning void)
+  ;;
+
+  let stubs__validate_sparse_csc_tensor_args =
+    foreign
+      "atg__validate_sparse_csc_tensor_args"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> returning void)
+  ;;
+
+  let stubs__values = foreign "atg__values" (gc_tensor @-> returning raw_tensor)
+  let stubs__values_copy = foreign "atg__values_copy" (gc_tensor @-> returning raw_tensor)
+
+  let stubs__values_copy_out =
+    foreign "atg__values_copy_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs__version = foreign "atg__version" (gc_tensor @-> returning int64_t)
+
+  let stubs__weight_norm =
+    foreign
+      "atg__weight_norm"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs__weight_norm_differentiable_backward =
+    foreign
+      "atg__weight_norm_differentiable_backward"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> returning void)
+  ;;
+
+  let stubs__weight_norm_interface =
+    foreign
+      "atg__weight_norm_interface"
+      (ptr raw_tensor @-> gc_tensor @-> gc_tensor @-> int64_t @-> returning void)
+  ;;
+
+  let stubs__weight_norm_interface_backward =
+    foreign
+      "atg__weight_norm_interface_backward"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> returning void)
+  ;;
+
+  let stubs__weight_norm_interface_backward_out =
+    foreign
+      "atg__weight_norm_interface_backward_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> returning void)
+  ;;
+
+  let stubs__weight_norm_interface_out =
+    foreign
+      "atg__weight_norm_interface_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> returning void)
+  ;;
+
+  let stubs_abs = foreign "atg_abs" (gc_tensor @-> returning raw_tensor)
+  let stubs_abs_ = foreign "atg_abs_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_abs_out =
+    foreign "atg_abs_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_absolute = foreign "atg_absolute" (gc_tensor @-> returning raw_tensor)
+  let stubs_absolute_ = foreign "atg_absolute_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_absolute_out =
+    foreign "atg_absolute_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_acos = foreign "atg_acos" (gc_tensor @-> returning raw_tensor)
+  let stubs_acos_ = foreign "atg_acos_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_acos_out =
+    foreign "atg_acos_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_acosh = foreign "atg_acosh" (gc_tensor @-> returning raw_tensor)
+  let stubs_acosh_ = foreign "atg_acosh_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_acosh_out =
+    foreign "atg_acosh_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_adaptive_avg_pool1d =
+    foreign
+      "atg_adaptive_avg_pool1d"
+      (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_adaptive_avg_pool2d =
+    foreign
+      "atg_adaptive_avg_pool2d"
+      (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_adaptive_avg_pool2d_out =
+    foreign
+      "atg_adaptive_avg_pool2d_out"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_adaptive_avg_pool3d =
+    foreign
+      "atg_adaptive_avg_pool3d"
+      (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_adaptive_avg_pool3d_backward =
+    foreign
+      "atg_adaptive_avg_pool3d_backward"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_adaptive_avg_pool3d_out =
+    foreign
+      "atg_adaptive_avg_pool3d_out"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_adaptive_max_pool1d =
+    foreign
+      "atg_adaptive_max_pool1d"
+      (ptr raw_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> returning void)
+  ;;
+
+  let stubs_adaptive_max_pool2d =
+    foreign
+      "atg_adaptive_max_pool2d"
+      (ptr raw_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> returning void)
+  ;;
+
+  let stubs_adaptive_max_pool2d_backward =
+    foreign
+      "atg_adaptive_max_pool2d_backward"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_adaptive_max_pool2d_backward_grad_input =
+    foreign
+      "atg_adaptive_max_pool2d_backward_grad_input"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_adaptive_max_pool2d_out =
+    foreign
+      "atg_adaptive_max_pool2d_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_adaptive_max_pool3d =
+    foreign
+      "atg_adaptive_max_pool3d"
+      (ptr raw_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> returning void)
+  ;;
+
+  let stubs_adaptive_max_pool3d_backward =
+    foreign
+      "atg_adaptive_max_pool3d_backward"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_adaptive_max_pool3d_backward_grad_input =
+    foreign
+      "atg_adaptive_max_pool3d_backward_grad_input"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_adaptive_max_pool3d_out =
+    foreign
+      "atg_adaptive_max_pool3d_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_add = foreign "atg_add" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  let stubs_add_ = foreign "atg_add_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+
+  let stubs_add_out =
+    foreign "atg_add_out" (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_add_scalar =
+    foreign "atg_add_scalar" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_add_scalar_ =
+    foreign "atg_add_scalar_" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+end
+
+module C5 (F : Cstubs.FOREIGN) = struct
+  open F
+  open Type_defs
+
+  let stubs_add_scalar_out =
+    foreign
+      "atg_add_scalar_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_addbmm =
+    foreign "atg_addbmm" (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_addbmm_ =
+    foreign "atg_addbmm_" (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_addbmm_out =
+    foreign
+      "atg_addbmm_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_addcdiv =
+    foreign "atg_addcdiv" (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_addcdiv_ =
+    foreign "atg_addcdiv_" (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_addcdiv_out =
+    foreign
+      "atg_addcdiv_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_addcmul =
+    foreign "atg_addcmul" (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_addcmul_ =
+    foreign "atg_addcmul_" (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_addcmul_out =
+    foreign
+      "atg_addcmul_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_addmm =
+    foreign "atg_addmm" (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_addmm_ =
+    foreign "atg_addmm_" (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_addmm_out =
+    foreign
+      "atg_addmm_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_addmv =
+    foreign "atg_addmv" (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_addmv_ =
+    foreign "atg_addmv_" (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_addmv_out =
+    foreign
+      "atg_addmv_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_addr =
+    foreign "atg_addr" (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_addr_ =
+    foreign "atg_addr_" (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_addr_out =
+    foreign
+      "atg_addr_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_adjoint = foreign "atg_adjoint" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_affine_grid_generator =
+    foreign
+      "atg_affine_grid_generator"
+      (gc_tensor @-> ptr int64_t @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_affine_grid_generator_backward =
+    foreign
+      "atg_affine_grid_generator_backward"
+      (gc_tensor @-> ptr int64_t @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_affine_grid_generator_out =
+    foreign
+      "atg_affine_grid_generator_out"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_alias = foreign "atg_alias" (gc_tensor @-> returning raw_tensor)
+  let stubs_alias_copy = foreign "atg_alias_copy" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_alias_copy_out =
+    foreign "atg_alias_copy_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_align_as =
+    foreign "atg_align_as" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_align_tensors =
+    foreign "atg_align_tensors" (ptr gc_tensor @-> int @-> returning (ptr raw_tensor))
+  ;;
+
+  let stubs_all = foreign "atg_all" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_all_all_out =
+    foreign "atg_all_all_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_all_dim =
+    foreign "atg_all_dim" (gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_all_out =
+    foreign
+      "atg_all_out"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_allclose =
+    foreign
+      "atg_allclose"
+      (gc_tensor @-> gc_tensor @-> double @-> double @-> int @-> returning bool)
+  ;;
+
+  let stubs_alpha_dropout =
+    foreign "atg_alpha_dropout" (gc_tensor @-> double @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_alpha_dropout_ =
+    foreign "atg_alpha_dropout_" (gc_tensor @-> double @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_amax =
+    foreign "atg_amax" (gc_tensor @-> ptr int64_t @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_amax_out =
+    foreign
+      "atg_amax_out"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_amin =
+    foreign "atg_amin" (gc_tensor @-> ptr int64_t @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_amin_out =
+    foreign
+      "atg_amin_out"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_aminmax =
+    foreign
+      "atg_aminmax"
+      (ptr raw_tensor @-> gc_tensor @-> int64_t @-> int @-> int @-> returning void)
+  ;;
+
+  let stubs_aminmax_out =
+    foreign
+      "atg_aminmax_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_angle = foreign "atg_angle" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_angle_out =
+    foreign "atg_angle_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_any = foreign "atg_any" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_any_all_out =
+    foreign "atg_any_all_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_any_dim =
+    foreign "atg_any_dim" (gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_any_out =
+    foreign
+      "atg_any_out"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_arange = foreign "atg_arange" (scalar @-> int @-> int @-> returning raw_tensor)
+
+  let stubs_arange_start =
+    foreign "atg_arange_start" (scalar @-> scalar @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_arange_start_step =
+    foreign
+      "atg_arange_start_step"
+      (scalar @-> scalar @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_arccos = foreign "atg_arccos" (gc_tensor @-> returning raw_tensor)
+  let stubs_arccos_ = foreign "atg_arccos_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_arccos_out =
+    foreign "atg_arccos_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_arccosh = foreign "atg_arccosh" (gc_tensor @-> returning raw_tensor)
+  let stubs_arccosh_ = foreign "atg_arccosh_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_arccosh_out =
+    foreign "atg_arccosh_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_arcsin = foreign "atg_arcsin" (gc_tensor @-> returning raw_tensor)
+  let stubs_arcsin_ = foreign "atg_arcsin_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_arcsin_out =
+    foreign "atg_arcsin_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_arcsinh = foreign "atg_arcsinh" (gc_tensor @-> returning raw_tensor)
+  let stubs_arcsinh_ = foreign "atg_arcsinh_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_arcsinh_out =
+    foreign "atg_arcsinh_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_arctan = foreign "atg_arctan" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_arctan2 =
+    foreign "atg_arctan2" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_arctan2_ =
+    foreign "atg_arctan2_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_arctan2_out =
+    foreign
+      "atg_arctan2_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_arctan_ = foreign "atg_arctan_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_arctan_out =
+    foreign "atg_arctan_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_arctanh = foreign "atg_arctanh" (gc_tensor @-> returning raw_tensor)
+  let stubs_arctanh_ = foreign "atg_arctanh_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_arctanh_out =
+    foreign "atg_arctanh_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_argmax =
+    foreign "atg_argmax" (gc_tensor @-> int64_t @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_argmax_out =
+    foreign
+      "atg_argmax_out"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_argmin =
+    foreign "atg_argmin" (gc_tensor @-> int64_t @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_argmin_out =
+    foreign
+      "atg_argmin_out"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_argsort =
+    foreign "atg_argsort" (gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_argsort_stable =
+    foreign
+      "atg_argsort_stable"
+      (gc_tensor @-> int @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_argsort_stable_out =
+    foreign
+      "atg_argsort_stable_out"
+      (gc_tensor @-> gc_tensor @-> int @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_argwhere = foreign "atg_argwhere" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_as_strided =
+    foreign
+      "atg_as_strided"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_as_strided_ =
+    foreign
+      "atg_as_strided_"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_as_strided_copy =
+    foreign
+      "atg_as_strided_copy"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_as_strided_copy_out =
+    foreign
+      "atg_as_strided_copy_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_as_strided_scatter =
+    foreign
+      "atg_as_strided_scatter"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_as_strided_scatter_out =
+    foreign
+      "atg_as_strided_scatter_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_asin = foreign "atg_asin" (gc_tensor @-> returning raw_tensor)
+  let stubs_asin_ = foreign "atg_asin_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_asin_out =
+    foreign "atg_asin_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_asinh = foreign "atg_asinh" (gc_tensor @-> returning raw_tensor)
+  let stubs_asinh_ = foreign "atg_asinh_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_asinh_out =
+    foreign "atg_asinh_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_atan = foreign "atg_atan" (gc_tensor @-> returning raw_tensor)
+  let stubs_atan2 = foreign "atg_atan2" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+
+  let stubs_atan2_ =
+    foreign "atg_atan2_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_atan2_out =
+    foreign
+      "atg_atan2_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_atan_ = foreign "atg_atan_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_atan_out =
+    foreign "atg_atan_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_atanh = foreign "atg_atanh" (gc_tensor @-> returning raw_tensor)
+  let stubs_atanh_ = foreign "atg_atanh_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_atanh_out =
+    foreign "atg_atanh_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+end
+
+module C6 (F : Cstubs.FOREIGN) = struct
+  open F
+  open Type_defs
+
+  let stubs_atleast_1d = foreign "atg_atleast_1d" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_atleast_1d_sequence =
+    foreign
+      "atg_atleast_1d_sequence"
+      (ptr gc_tensor @-> int @-> returning (ptr raw_tensor))
+  ;;
+
+  let stubs_atleast_2d = foreign "atg_atleast_2d" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_atleast_2d_sequence =
+    foreign
+      "atg_atleast_2d_sequence"
+      (ptr gc_tensor @-> int @-> returning (ptr raw_tensor))
+  ;;
+
+  let stubs_atleast_3d = foreign "atg_atleast_3d" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_atleast_3d_sequence =
+    foreign
+      "atg_atleast_3d_sequence"
+      (ptr gc_tensor @-> int @-> returning (ptr raw_tensor))
+  ;;
+
+  let stubs_avg_pool1d =
+    foreign
+      "atg_avg_pool1d"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_avg_pool2d =
+    foreign
+      "atg_avg_pool2d"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_avg_pool2d_backward =
+    foreign
+      "atg_avg_pool2d_backward"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_avg_pool2d_backward_grad_input =
+    foreign
+      "atg_avg_pool2d_backward_grad_input"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_avg_pool2d_out =
+    foreign
+      "atg_avg_pool2d_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_avg_pool3d =
+    foreign
+      "atg_avg_pool3d"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_avg_pool3d_backward =
+    foreign
+      "atg_avg_pool3d_backward"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_avg_pool3d_backward_grad_input =
+    foreign
+      "atg_avg_pool3d_backward_grad_input"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_avg_pool3d_out =
+    foreign
+      "atg_avg_pool3d_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_baddbmm =
+    foreign "atg_baddbmm" (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_baddbmm_ =
+    foreign "atg_baddbmm_" (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_baddbmm_out =
+    foreign
+      "atg_baddbmm_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_bartlett_window =
+    foreign "atg_bartlett_window" (int64_t @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_bartlett_window_out =
+    foreign "atg_bartlett_window_out" (gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_bartlett_window_periodic =
+    foreign
+      "atg_bartlett_window_periodic"
+      (int64_t @-> int @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_bartlett_window_periodic_out =
+    foreign
+      "atg_bartlett_window_periodic_out"
+      (gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_batch_norm =
+    foreign
+      "atg_batch_norm"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> double
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_batch_norm_backward_elemt =
+    foreign
+      "atg_batch_norm_backward_elemt"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_batch_norm_backward_elemt_out =
+    foreign
+      "atg_batch_norm_backward_elemt_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_batch_norm_backward_reduce =
+    foreign
+      "atg_batch_norm_backward_reduce"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> int
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_batch_norm_backward_reduce_out =
+    foreign
+      "atg_batch_norm_backward_reduce_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> int
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_batch_norm_elemt =
+    foreign
+      "atg_batch_norm_elemt"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_batch_norm_elemt_out =
+    foreign
+      "atg_batch_norm_elemt_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_batch_norm_gather_stats =
+    foreign
+      "atg_batch_norm_gather_stats"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> double
+       @-> int64_t
+       @-> returning void)
+  ;;
+
+  let stubs_batch_norm_gather_stats_out =
+    foreign
+      "atg_batch_norm_gather_stats_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> double
+       @-> int64_t
+       @-> returning void)
+  ;;
+
+  let stubs_batch_norm_gather_stats_with_counts =
+    foreign
+      "atg_batch_norm_gather_stats_with_counts"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> double
+       @-> gc_tensor
+       @-> returning void)
+  ;;
+
+  let stubs_batch_norm_gather_stats_with_counts_out =
+    foreign
+      "atg_batch_norm_gather_stats_with_counts_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> double
+       @-> gc_tensor
+       @-> returning void)
+  ;;
+
+  let stubs_batch_norm_stats =
+    foreign
+      "atg_batch_norm_stats"
+      (ptr raw_tensor @-> gc_tensor @-> double @-> returning void)
+  ;;
+
+  let stubs_batch_norm_stats_out =
+    foreign
+      "atg_batch_norm_stats_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> returning void)
+  ;;
+
+  let stubs_batch_norm_update_stats =
+    foreign
+      "atg_batch_norm_update_stats"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> returning void)
+  ;;
+
+  let stubs_batch_norm_update_stats_out =
+    foreign
+      "atg_batch_norm_update_stats_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> returning void)
+  ;;
+
+  let stubs_bernoulli = foreign "atg_bernoulli" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_bernoulli_ =
+    foreign "atg_bernoulli_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_bernoulli_float_ =
+    foreign "atg_bernoulli_float_" (gc_tensor @-> double @-> returning raw_tensor)
+  ;;
+
+  let stubs_bernoulli_p =
+    foreign "atg_bernoulli_p" (gc_tensor @-> double @-> returning raw_tensor)
+  ;;
+
+  let stubs_bernoulli_tensor =
+    foreign "atg_bernoulli_tensor" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_bilinear =
+    foreign
+      "atg_bilinear"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_binary_cross_entropy =
+    foreign
+      "atg_binary_cross_entropy"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_binary_cross_entropy_backward =
+    foreign
+      "atg_binary_cross_entropy_backward"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_binary_cross_entropy_backward_grad_input =
+    foreign
+      "atg_binary_cross_entropy_backward_grad_input"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_binary_cross_entropy_out =
+    foreign
+      "atg_binary_cross_entropy_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_binary_cross_entropy_with_logits =
+    foreign
+      "atg_binary_cross_entropy_with_logits"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_binary_cross_entropy_with_logits_out =
+    foreign
+      "atg_binary_cross_entropy_with_logits_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_bincount =
+    foreign "atg_bincount" (gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_bincount_out =
+    foreign
+      "atg_bincount_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_binomial =
+    foreign "atg_binomial" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_binomial_out =
+    foreign
+      "atg_binomial_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_bitwise_and =
+    foreign "atg_bitwise_and" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_bitwise_and_ =
+    foreign "atg_bitwise_and_" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_bitwise_and_scalar_out =
+    foreign
+      "atg_bitwise_and_scalar_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_bitwise_and_scalar_tensor =
+    foreign "atg_bitwise_and_scalar_tensor" (scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_bitwise_and_scalar_tensor_out =
+    foreign
+      "atg_bitwise_and_scalar_tensor_out"
+      (gc_tensor @-> scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_bitwise_and_tensor =
+    foreign "atg_bitwise_and_tensor" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_bitwise_and_tensor_ =
+    foreign "atg_bitwise_and_tensor_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_bitwise_and_tensor_out =
+    foreign
+      "atg_bitwise_and_tensor_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_bitwise_left_shift =
+    foreign "atg_bitwise_left_shift" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_bitwise_left_shift_ =
+    foreign "atg_bitwise_left_shift_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_bitwise_left_shift_scalar_tensor =
+    foreign
+      "atg_bitwise_left_shift_scalar_tensor"
+      (scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_bitwise_left_shift_scalar_tensor_out =
+    foreign
+      "atg_bitwise_left_shift_scalar_tensor_out"
+      (gc_tensor @-> scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_bitwise_left_shift_tensor_out =
+    foreign
+      "atg_bitwise_left_shift_tensor_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_bitwise_left_shift_tensor_scalar =
+    foreign
+      "atg_bitwise_left_shift_tensor_scalar"
+      (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_bitwise_left_shift_tensor_scalar_ =
+    foreign
+      "atg_bitwise_left_shift_tensor_scalar_"
+      (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_bitwise_left_shift_tensor_scalar_out =
+    foreign
+      "atg_bitwise_left_shift_tensor_scalar_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_bitwise_not = foreign "atg_bitwise_not" (gc_tensor @-> returning raw_tensor)
+  let stubs_bitwise_not_ = foreign "atg_bitwise_not_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_bitwise_not_out =
+    foreign "atg_bitwise_not_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_bitwise_or =
+    foreign "atg_bitwise_or" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_bitwise_or_ =
+    foreign "atg_bitwise_or_" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_bitwise_or_scalar_out =
+    foreign
+      "atg_bitwise_or_scalar_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_bitwise_or_scalar_tensor =
+    foreign "atg_bitwise_or_scalar_tensor" (scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_bitwise_or_scalar_tensor_out =
+    foreign
+      "atg_bitwise_or_scalar_tensor_out"
+      (gc_tensor @-> scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_bitwise_or_tensor =
+    foreign "atg_bitwise_or_tensor" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_bitwise_or_tensor_ =
+    foreign "atg_bitwise_or_tensor_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_bitwise_or_tensor_out =
+    foreign
+      "atg_bitwise_or_tensor_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_bitwise_right_shift =
+    foreign "atg_bitwise_right_shift" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_bitwise_right_shift_ =
+    foreign "atg_bitwise_right_shift_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_bitwise_right_shift_scalar_tensor =
+    foreign
+      "atg_bitwise_right_shift_scalar_tensor"
+      (scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_bitwise_right_shift_scalar_tensor_out =
+    foreign
+      "atg_bitwise_right_shift_scalar_tensor_out"
+      (gc_tensor @-> scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_bitwise_right_shift_tensor_out =
+    foreign
+      "atg_bitwise_right_shift_tensor_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_bitwise_right_shift_tensor_scalar =
+    foreign
+      "atg_bitwise_right_shift_tensor_scalar"
+      (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_bitwise_right_shift_tensor_scalar_ =
+    foreign
+      "atg_bitwise_right_shift_tensor_scalar_"
+      (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_bitwise_right_shift_tensor_scalar_out =
+    foreign
+      "atg_bitwise_right_shift_tensor_scalar_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_bitwise_xor =
+    foreign "atg_bitwise_xor" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_bitwise_xor_ =
+    foreign "atg_bitwise_xor_" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_bitwise_xor_scalar_out =
+    foreign
+      "atg_bitwise_xor_scalar_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_bitwise_xor_scalar_tensor =
+    foreign "atg_bitwise_xor_scalar_tensor" (scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_bitwise_xor_scalar_tensor_out =
+    foreign
+      "atg_bitwise_xor_scalar_tensor_out"
+      (gc_tensor @-> scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_bitwise_xor_tensor =
+    foreign "atg_bitwise_xor_tensor" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_bitwise_xor_tensor_ =
+    foreign "atg_bitwise_xor_tensor_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_bitwise_xor_tensor_out =
+    foreign
+      "atg_bitwise_xor_tensor_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_blackman_window =
+    foreign "atg_blackman_window" (int64_t @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_blackman_window_out =
+    foreign "atg_blackman_window_out" (gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_blackman_window_periodic =
+    foreign
+      "atg_blackman_window_periodic"
+      (int64_t @-> int @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_blackman_window_periodic_out =
+    foreign
+      "atg_blackman_window_periodic_out"
+      (gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+end
+
+module C7 (F : Cstubs.FOREIGN) = struct
+  open F
+  open Type_defs
+
+  let stubs_block_diag =
+    foreign "atg_block_diag" (ptr gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_block_diag_out =
+    foreign
+      "atg_block_diag_out"
+      (gc_tensor @-> ptr gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_bmm = foreign "atg_bmm" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+
+  let stubs_bmm_out =
+    foreign "atg_bmm_out" (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_broadcast_tensors =
+    foreign "atg_broadcast_tensors" (ptr gc_tensor @-> int @-> returning (ptr raw_tensor))
+  ;;
+
+  let stubs_broadcast_to =
+    foreign "atg_broadcast_to" (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_bucketize =
+    foreign
+      "atg_bucketize"
+      (gc_tensor @-> gc_tensor @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_bucketize_scalar =
+    foreign
+      "atg_bucketize_scalar"
+      (scalar @-> gc_tensor @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_bucketize_scalar_out =
+    foreign
+      "atg_bucketize_scalar_out"
+      (gc_tensor @-> scalar @-> gc_tensor @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_bucketize_tensor_out =
+    foreign
+      "atg_bucketize_tensor_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_can_cast = foreign "atg_can_cast" (int @-> int @-> returning bool)
+
+  let stubs_cartesian_prod =
+    foreign "atg_cartesian_prod" (ptr gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_cat =
+    foreign "atg_cat" (ptr gc_tensor @-> int @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_cat_out =
+    foreign
+      "atg_cat_out"
+      (gc_tensor @-> ptr gc_tensor @-> int @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_cauchy =
+    foreign "atg_cauchy" (gc_tensor @-> double @-> double @-> returning raw_tensor)
+  ;;
+
+  let stubs_cauchy_ =
+    foreign "atg_cauchy_" (gc_tensor @-> double @-> double @-> returning raw_tensor)
+  ;;
+
+  let stubs_cauchy_out =
+    foreign
+      "atg_cauchy_out"
+      (gc_tensor @-> gc_tensor @-> double @-> double @-> returning raw_tensor)
+  ;;
+
+  let stubs_ccol_indices = foreign "atg_ccol_indices" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_ccol_indices_copy =
+    foreign "atg_ccol_indices_copy" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_ccol_indices_copy_out =
+    foreign "atg_ccol_indices_copy_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_cdist =
+    foreign
+      "atg_cdist"
+      (gc_tensor @-> gc_tensor @-> double @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_ceil = foreign "atg_ceil" (gc_tensor @-> returning raw_tensor)
+  let stubs_ceil_ = foreign "atg_ceil_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_ceil_out =
+    foreign "atg_ceil_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_celu = foreign "atg_celu" (gc_tensor @-> returning raw_tensor)
+  let stubs_celu_ = foreign "atg_celu_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_celu_out =
+    foreign "atg_celu_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_chain_matmul =
+    foreign "atg_chain_matmul" (ptr gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_chain_matmul_out =
+    foreign
+      "atg_chain_matmul_out"
+      (gc_tensor @-> ptr gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_chalf = foreign "atg_chalf" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_channel_shuffle =
+    foreign "atg_channel_shuffle" (gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_channel_shuffle_out =
+    foreign
+      "atg_channel_shuffle_out"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_cholesky = foreign "atg_cholesky" (gc_tensor @-> int @-> returning raw_tensor)
+
+  let stubs_cholesky_inverse =
+    foreign "atg_cholesky_inverse" (gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_cholesky_inverse_out =
+    foreign
+      "atg_cholesky_inverse_out"
+      (gc_tensor @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_cholesky_out =
+    foreign "atg_cholesky_out" (gc_tensor @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_cholesky_solve =
+    foreign "atg_cholesky_solve" (gc_tensor @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_cholesky_solve_out =
+    foreign
+      "atg_cholesky_solve_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_choose_qparams_optimized =
+    foreign
+      "atg_choose_qparams_optimized"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> double
+       @-> int64_t
+       @-> returning void)
+  ;;
+
+  let stubs_chunk =
+    foreign "atg_chunk" (gc_tensor @-> int64_t @-> int64_t @-> returning (ptr raw_tensor))
+  ;;
+
+  let stubs_clamp =
+    foreign "atg_clamp" (gc_tensor @-> scalar @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_clamp_ =
+    foreign "atg_clamp_" (gc_tensor @-> scalar @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_clamp_max =
+    foreign "atg_clamp_max" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_clamp_max_ =
+    foreign "atg_clamp_max_" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_clamp_max_out =
+    foreign
+      "atg_clamp_max_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_clamp_max_tensor =
+    foreign "atg_clamp_max_tensor" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_clamp_max_tensor_ =
+    foreign "atg_clamp_max_tensor_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_clamp_max_tensor_out =
+    foreign
+      "atg_clamp_max_tensor_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_clamp_min =
+    foreign "atg_clamp_min" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_clamp_min_ =
+    foreign "atg_clamp_min_" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_clamp_min_out =
+    foreign
+      "atg_clamp_min_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_clamp_min_tensor =
+    foreign "atg_clamp_min_tensor" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_clamp_min_tensor_ =
+    foreign "atg_clamp_min_tensor_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_clamp_min_tensor_out =
+    foreign
+      "atg_clamp_min_tensor_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_clamp_out =
+    foreign
+      "atg_clamp_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_clamp_tensor =
+    foreign
+      "atg_clamp_tensor"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_clamp_tensor_ =
+    foreign
+      "atg_clamp_tensor_"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_clamp_tensor_out =
+    foreign
+      "atg_clamp_tensor_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_clip =
+    foreign "atg_clip" (gc_tensor @-> scalar @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_clip_ =
+    foreign "atg_clip_" (gc_tensor @-> scalar @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_clip_out =
+    foreign
+      "atg_clip_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_clip_tensor =
+    foreign
+      "atg_clip_tensor"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_clip_tensor_ =
+    foreign
+      "atg_clip_tensor_"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_clip_tensor_out =
+    foreign
+      "atg_clip_tensor_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_clone = foreign "atg_clone" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_clone_out =
+    foreign "atg_clone_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_coalesce = foreign "atg_coalesce" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_col2im =
+    foreign
+      "atg_col2im"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_col2im_out =
+    foreign
+      "atg_col2im_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_col_indices = foreign "atg_col_indices" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_col_indices_copy =
+    foreign "atg_col_indices_copy" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_col_indices_copy_out =
+    foreign "atg_col_indices_copy_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_column_stack =
+    foreign "atg_column_stack" (ptr gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_column_stack_out =
+    foreign
+      "atg_column_stack_out"
+      (gc_tensor @-> ptr gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_combinations =
+    foreign "atg_combinations" (gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_complex =
+    foreign "atg_complex" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_complex_out =
+    foreign
+      "atg_complex_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_concat =
+    foreign "atg_concat" (ptr gc_tensor @-> int @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_concat_out =
+    foreign
+      "atg_concat_out"
+      (gc_tensor @-> ptr gc_tensor @-> int @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_concatenate =
+    foreign "atg_concatenate" (ptr gc_tensor @-> int @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_concatenate_out =
+    foreign
+      "atg_concatenate_out"
+      (gc_tensor @-> ptr gc_tensor @-> int @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_conj = foreign "atg_conj" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_conj_physical =
+    foreign "atg_conj_physical" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_conj_physical_ =
+    foreign "atg_conj_physical_" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_conj_physical_out =
+    foreign "atg_conj_physical_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_constant_pad_nd =
+    foreign
+      "atg_constant_pad_nd"
+      (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_constant_pad_nd_out =
+    foreign
+      "atg_constant_pad_nd_out"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_contiguous = foreign "atg_contiguous" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_conv1d =
+    foreign
+      "atg_conv1d"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_conv1d_padding =
+    foreign
+      "atg_conv1d_padding"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> string
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_conv2d =
+    foreign
+      "atg_conv2d"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_conv2d_padding =
+    foreign
+      "atg_conv2d_padding"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> string
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_conv3d =
+    foreign
+      "atg_conv3d"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_conv3d_padding =
+    foreign
+      "atg_conv3d_padding"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> string
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_conv_depthwise3d =
+    foreign
+      "atg_conv_depthwise3d"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_conv_depthwise3d_out =
+    foreign
+      "atg_conv_depthwise3d_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_conv_tbc =
+    foreign
+      "atg_conv_tbc"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_conv_tbc_backward =
+    foreign
+      "atg_conv_tbc_backward"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> returning void)
+  ;;
+
+  let stubs_conv_tbc_out =
+    foreign
+      "atg_conv_tbc_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_conv_transpose1d =
+    foreign
+      "atg_conv_transpose1d"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> ptr int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+end
+
+module C8 (F : Cstubs.FOREIGN) = struct
+  open F
+  open Type_defs
+
+  let stubs_conv_transpose2d =
+    foreign
+      "atg_conv_transpose2d"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> ptr int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_conv_transpose3d =
+    foreign
+      "atg_conv_transpose3d"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> ptr int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_convolution =
+    foreign
+      "atg_convolution"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_convolution_out =
+    foreign
+      "atg_convolution_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_convolution_overrideable =
+    foreign
+      "atg_convolution_overrideable"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_convolution_overrideable_out =
+    foreign
+      "atg_convolution_overrideable_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_copy =
+    foreign "atg_copy" (gc_tensor @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_copy_out =
+    foreign
+      "atg_copy_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_copy_sparse_to_sparse =
+    foreign
+      "atg_copy_sparse_to_sparse"
+      (gc_tensor @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_copy_sparse_to_sparse_ =
+    foreign
+      "atg_copy_sparse_to_sparse_"
+      (gc_tensor @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_copy_sparse_to_sparse_out =
+    foreign
+      "atg_copy_sparse_to_sparse_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_copysign =
+    foreign "atg_copysign" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_copysign_ =
+    foreign "atg_copysign_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_copysign_out =
+    foreign
+      "atg_copysign_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_copysign_scalar =
+    foreign "atg_copysign_scalar" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_copysign_scalar_ =
+    foreign "atg_copysign_scalar_" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_copysign_scalar_out =
+    foreign
+      "atg_copysign_scalar_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_corrcoef = foreign "atg_corrcoef" (gc_tensor @-> returning raw_tensor)
+  let stubs_cos = foreign "atg_cos" (gc_tensor @-> returning raw_tensor)
+  let stubs_cos_ = foreign "atg_cos_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_cos_out =
+    foreign "atg_cos_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_cosh = foreign "atg_cosh" (gc_tensor @-> returning raw_tensor)
+  let stubs_cosh_ = foreign "atg_cosh_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_cosh_out =
+    foreign "atg_cosh_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_cosine_embedding_loss =
+    foreign
+      "atg_cosine_embedding_loss"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_cosine_similarity =
+    foreign
+      "atg_cosine_similarity"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> double @-> returning raw_tensor)
+  ;;
+
+  let stubs_count_nonzero =
+    foreign
+      "atg_count_nonzero"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_count_nonzero_out =
+    foreign
+      "atg_count_nonzero_out"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_cov =
+    foreign
+      "atg_cov"
+      (gc_tensor @-> int64_t @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_cross =
+    foreign
+      "atg_cross"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_cross_entropy_loss =
+    foreign
+      "atg_cross_entropy_loss"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> double
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_cross_out =
+    foreign
+      "atg_cross_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_crow_indices = foreign "atg_crow_indices" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_crow_indices_copy =
+    foreign "atg_crow_indices_copy" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_crow_indices_copy_out =
+    foreign "atg_crow_indices_copy_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_ctc_loss =
+    foreign
+      "atg_ctc_loss"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_ctc_loss_tensor =
+    foreign
+      "atg_ctc_loss_tensor"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_cudnn_affine_grid_generator =
+    foreign
+      "atg_cudnn_affine_grid_generator"
+      (gc_tensor @-> int64_t @-> int64_t @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_cudnn_affine_grid_generator_backward =
+    foreign
+      "atg_cudnn_affine_grid_generator_backward"
+      (gc_tensor @-> int64_t @-> int64_t @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_cudnn_affine_grid_generator_backward_out =
+    foreign
+      "atg_cudnn_affine_grid_generator_backward_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_cudnn_affine_grid_generator_out =
+    foreign
+      "atg_cudnn_affine_grid_generator_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_cudnn_batch_norm =
+    foreign
+      "atg_cudnn_batch_norm"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> double
+       @-> double
+       @-> returning void)
+  ;;
+
+  let stubs_cudnn_batch_norm_backward =
+    foreign
+      "atg_cudnn_batch_norm_backward"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> gc_tensor
+       @-> returning void)
+  ;;
+
+  let stubs_cudnn_batch_norm_backward_out =
+    foreign
+      "atg_cudnn_batch_norm_backward_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> gc_tensor
+       @-> returning void)
+  ;;
+
+  let stubs_cudnn_batch_norm_out =
+    foreign
+      "atg_cudnn_batch_norm_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> double
+       @-> double
+       @-> returning void)
+  ;;
+
+  let stubs_cudnn_convolution =
+    foreign
+      "atg_cudnn_convolution"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_cudnn_convolution_add_relu =
+    foreign
+      "atg_cudnn_convolution_add_relu"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> scalar
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_cudnn_convolution_add_relu_out =
+    foreign
+      "atg_cudnn_convolution_add_relu_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> scalar
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_cudnn_convolution_out =
+    foreign
+      "atg_cudnn_convolution_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_cudnn_convolution_relu =
+    foreign
+      "atg_cudnn_convolution_relu"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_cudnn_convolution_relu_out =
+    foreign
+      "atg_cudnn_convolution_relu_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_cudnn_convolution_transpose =
+    foreign
+      "atg_cudnn_convolution_transpose"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_cudnn_convolution_transpose_out =
+    foreign
+      "atg_cudnn_convolution_transpose_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_cudnn_grid_sampler =
+    foreign "atg_cudnn_grid_sampler" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_cudnn_grid_sampler_backward =
+    foreign
+      "atg_cudnn_grid_sampler_backward"
+      (ptr raw_tensor @-> gc_tensor @-> gc_tensor @-> gc_tensor @-> returning void)
+  ;;
+
+  let stubs_cudnn_grid_sampler_backward_out =
+    foreign
+      "atg_cudnn_grid_sampler_backward_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> returning void)
+  ;;
+
+  let stubs_cudnn_grid_sampler_out =
+    foreign
+      "atg_cudnn_grid_sampler_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_cudnn_is_acceptable =
+    foreign "atg_cudnn_is_acceptable" (gc_tensor @-> returning bool)
+  ;;
+
+  let stubs_cummax =
+    foreign "atg_cummax" (ptr raw_tensor @-> gc_tensor @-> int64_t @-> returning void)
+  ;;
+
+  let stubs_cummax_out =
+    foreign
+      "atg_cummax_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> returning void)
+  ;;
+
+  let stubs_cummaxmin_backward =
+    foreign
+      "atg_cummaxmin_backward"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_cummin =
+    foreign "atg_cummin" (ptr raw_tensor @-> gc_tensor @-> int64_t @-> returning void)
+  ;;
+
+  let stubs_cummin_out =
+    foreign
+      "atg_cummin_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> returning void)
+  ;;
+
+  let stubs_cumprod =
+    foreign "atg_cumprod" (gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_cumprod_ =
+    foreign "atg_cumprod_" (gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_cumprod_backward =
+    foreign
+      "atg_cumprod_backward"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_cumprod_out =
+    foreign
+      "atg_cumprod_out"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_cumsum =
+    foreign "atg_cumsum" (gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_cumsum_ =
+    foreign "atg_cumsum_" (gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_cumsum_out =
+    foreign
+      "atg_cumsum_out"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_cumulative_trapezoid =
+    foreign "atg_cumulative_trapezoid" (gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_cumulative_trapezoid_x =
+    foreign
+      "atg_cumulative_trapezoid_x"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_data = foreign "atg_data" (gc_tensor @-> returning raw_tensor)
+  let stubs_deg2rad = foreign "atg_deg2rad" (gc_tensor @-> returning raw_tensor)
+  let stubs_deg2rad_ = foreign "atg_deg2rad_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_deg2rad_out =
+    foreign "atg_deg2rad_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_dense_dim = foreign "atg_dense_dim" (gc_tensor @-> returning int64_t)
+  let stubs_dequantize = foreign "atg_dequantize" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_dequantize_self_out =
+    foreign "atg_dequantize_self_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_dequantize_tensors =
+    foreign "atg_dequantize_tensors" (ptr gc_tensor @-> int @-> returning (ptr raw_tensor))
+  ;;
+
+  let stubs_dequantize_tensors_out =
+    foreign
+      "atg_dequantize_tensors_out"
+      (ptr gc_tensor @-> int @-> ptr gc_tensor @-> int @-> returning void)
+  ;;
+
+  let stubs_det = foreign "atg_det" (gc_tensor @-> returning raw_tensor)
+  let stubs_detach = foreign "atg_detach" (gc_tensor @-> returning raw_tensor)
+  let stubs_detach_ = foreign "atg_detach_" (gc_tensor @-> returning raw_tensor)
+  let stubs_detach_copy = foreign "atg_detach_copy" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_detach_copy_out =
+    foreign "atg_detach_copy_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_diag = foreign "atg_diag" (gc_tensor @-> int64_t @-> returning raw_tensor)
+
+  let stubs_diag_embed =
+    foreign
+      "atg_diag_embed"
+      (gc_tensor @-> int64_t @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_diag_embed_out =
+    foreign
+      "atg_diag_embed_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_diag_out =
+    foreign "atg_diag_out" (gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_diagflat =
+    foreign "atg_diagflat" (gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_diagonal =
+    foreign
+      "atg_diagonal"
+      (gc_tensor @-> int64_t @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_diagonal_backward =
+    foreign
+      "atg_diagonal_backward"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_diagonal_backward_out =
+    foreign
+      "atg_diagonal_backward_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_diagonal_copy =
+    foreign
+      "atg_diagonal_copy"
+      (gc_tensor @-> int64_t @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_diagonal_copy_out =
+    foreign
+      "atg_diagonal_copy_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_diagonal_scatter =
+    foreign
+      "atg_diagonal_scatter"
+      (gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_diagonal_scatter_out =
+    foreign
+      "atg_diagonal_scatter_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_diff =
+    foreign
+      "atg_diff"
+      (gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_diff_out =
+    foreign
+      "atg_diff_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+end
+
+module C9 (F : Cstubs.FOREIGN) = struct
+  open F
+  open Type_defs
+
+  let stubs_digamma = foreign "atg_digamma" (gc_tensor @-> returning raw_tensor)
+  let stubs_digamma_ = foreign "atg_digamma_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_digamma_out =
+    foreign "atg_digamma_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_dist = foreign "atg_dist" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+
+  let stubs_dist_out =
+    foreign "atg_dist_out" (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_div = foreign "atg_div" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  let stubs_div_ = foreign "atg_div_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+
+  let stubs_div_out =
+    foreign "atg_div_out" (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_div_out_mode =
+    foreign
+      "atg_div_out_mode"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> string @-> returning raw_tensor)
+  ;;
+
+  let stubs_div_scalar =
+    foreign "atg_div_scalar" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_div_scalar_ =
+    foreign "atg_div_scalar_" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_div_scalar_mode =
+    foreign
+      "atg_div_scalar_mode"
+      (gc_tensor @-> scalar @-> string @-> returning raw_tensor)
+  ;;
+
+  let stubs_div_scalar_mode_ =
+    foreign
+      "atg_div_scalar_mode_"
+      (gc_tensor @-> scalar @-> string @-> returning raw_tensor)
+  ;;
+
+  let stubs_div_scalar_mode_out =
+    foreign
+      "atg_div_scalar_mode_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> string @-> returning raw_tensor)
+  ;;
+
+  let stubs_div_scalar_out =
+    foreign
+      "atg_div_scalar_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_div_tensor_mode =
+    foreign
+      "atg_div_tensor_mode"
+      (gc_tensor @-> gc_tensor @-> string @-> returning raw_tensor)
+  ;;
+
+  let stubs_div_tensor_mode_ =
+    foreign
+      "atg_div_tensor_mode_"
+      (gc_tensor @-> gc_tensor @-> string @-> returning raw_tensor)
+  ;;
+
+  let stubs_divide =
+    foreign "atg_divide" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_divide_ =
+    foreign "atg_divide_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_divide_out =
+    foreign
+      "atg_divide_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_divide_out_mode =
+    foreign
+      "atg_divide_out_mode"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> string @-> returning raw_tensor)
+  ;;
+
+  let stubs_divide_scalar =
+    foreign "atg_divide_scalar" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_divide_scalar_ =
+    foreign "atg_divide_scalar_" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_divide_scalar_mode =
+    foreign
+      "atg_divide_scalar_mode"
+      (gc_tensor @-> scalar @-> string @-> returning raw_tensor)
+  ;;
+
+  let stubs_divide_scalar_mode_ =
+    foreign
+      "atg_divide_scalar_mode_"
+      (gc_tensor @-> scalar @-> string @-> returning raw_tensor)
+  ;;
+
+  let stubs_divide_tensor_mode =
+    foreign
+      "atg_divide_tensor_mode"
+      (gc_tensor @-> gc_tensor @-> string @-> returning raw_tensor)
+  ;;
+
+  let stubs_divide_tensor_mode_ =
+    foreign
+      "atg_divide_tensor_mode_"
+      (gc_tensor @-> gc_tensor @-> string @-> returning raw_tensor)
+  ;;
+
+  let stubs_dot = foreign "atg_dot" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+
+  let stubs_dot_out =
+    foreign "atg_dot_out" (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_dropout =
+    foreign "atg_dropout" (gc_tensor @-> double @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_dropout_ =
+    foreign "atg_dropout_" (gc_tensor @-> double @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_dsplit =
+    foreign "atg_dsplit" (gc_tensor @-> int64_t @-> returning (ptr raw_tensor))
+  ;;
+
+  let stubs_dsplit_array =
+    foreign
+      "atg_dsplit_array"
+      (gc_tensor @-> ptr int64_t @-> int @-> returning (ptr raw_tensor))
+  ;;
+
+  let stubs_dstack = foreign "atg_dstack" (ptr gc_tensor @-> int @-> returning raw_tensor)
+
+  let stubs_dstack_out =
+    foreign "atg_dstack_out" (gc_tensor @-> ptr gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_einsum =
+    foreign
+      "atg_einsum"
+      (string @-> ptr gc_tensor @-> int @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_elu = foreign "atg_elu" (gc_tensor @-> returning raw_tensor)
+  let stubs_elu_ = foreign "atg_elu_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_elu_backward =
+    foreign
+      "atg_elu_backward"
+      (gc_tensor
+       @-> scalar
+       @-> scalar
+       @-> scalar
+       @-> int
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_elu_backward_grad_input =
+    foreign
+      "atg_elu_backward_grad_input"
+      (gc_tensor
+       @-> gc_tensor
+       @-> scalar
+       @-> scalar
+       @-> scalar
+       @-> int
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_elu_out =
+    foreign "atg_elu_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_embedding =
+    foreign
+      "atg_embedding"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_embedding_backward =
+    foreign
+      "atg_embedding_backward"
+      (gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_embedding_bag =
+    foreign
+      "atg_embedding_bag"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> gc_tensor
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_embedding_bag_padding_idx =
+    foreign
+      "atg_embedding_bag_padding_idx"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> gc_tensor
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_embedding_dense_backward =
+    foreign
+      "atg_embedding_dense_backward"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_embedding_dense_backward_out =
+    foreign
+      "atg_embedding_dense_backward_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_embedding_out =
+    foreign
+      "atg_embedding_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_embedding_renorm =
+    foreign
+      "atg_embedding_renorm"
+      (gc_tensor @-> gc_tensor @-> double @-> double @-> returning raw_tensor)
+  ;;
+
+  let stubs_embedding_renorm_ =
+    foreign
+      "atg_embedding_renorm_"
+      (gc_tensor @-> gc_tensor @-> double @-> double @-> returning raw_tensor)
+  ;;
+
+  let stubs_embedding_renorm_out =
+    foreign
+      "atg_embedding_renorm_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> double
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_embedding_sparse_backward =
+    foreign
+      "atg_embedding_sparse_backward"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_empty =
+    foreign "atg_empty" (ptr int64_t @-> int @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_empty_like = foreign "atg_empty_like" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_empty_like_out =
+    foreign "atg_empty_like_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_empty_out =
+    foreign "atg_empty_out" (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_empty_permuted =
+    foreign
+      "atg_empty_permuted"
+      (ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_empty_permuted_out =
+    foreign
+      "atg_empty_permuted_out"
+      (gc_tensor @-> ptr int64_t @-> int @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_empty_quantized =
+    foreign
+      "atg_empty_quantized"
+      (ptr int64_t @-> int @-> gc_tensor @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_empty_quantized_out =
+    foreign
+      "atg_empty_quantized_out"
+      (gc_tensor @-> ptr int64_t @-> int @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_empty_strided =
+    foreign
+      "atg_empty_strided"
+      (ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_empty_strided_out =
+    foreign
+      "atg_empty_strided_out"
+      (gc_tensor @-> ptr int64_t @-> int @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_eq = foreign "atg_eq" (gc_tensor @-> scalar @-> returning raw_tensor)
+  let stubs_eq_ = foreign "atg_eq_" (gc_tensor @-> scalar @-> returning raw_tensor)
+
+  let stubs_eq_scalar_out =
+    foreign
+      "atg_eq_scalar_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_eq_tensor =
+    foreign "atg_eq_tensor" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_eq_tensor_ =
+    foreign "atg_eq_tensor_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_eq_tensor_out =
+    foreign
+      "atg_eq_tensor_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_equal = foreign "atg_equal" (gc_tensor @-> gc_tensor @-> returning bool)
+  let stubs_erf = foreign "atg_erf" (gc_tensor @-> returning raw_tensor)
+  let stubs_erf_ = foreign "atg_erf_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_erf_out =
+    foreign "atg_erf_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_erfc = foreign "atg_erfc" (gc_tensor @-> returning raw_tensor)
+  let stubs_erfc_ = foreign "atg_erfc_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_erfc_out =
+    foreign "atg_erfc_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_erfinv = foreign "atg_erfinv" (gc_tensor @-> returning raw_tensor)
+  let stubs_erfinv_ = foreign "atg_erfinv_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_erfinv_out =
+    foreign "atg_erfinv_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_exp = foreign "atg_exp" (gc_tensor @-> returning raw_tensor)
+  let stubs_exp2 = foreign "atg_exp2" (gc_tensor @-> returning raw_tensor)
+  let stubs_exp2_ = foreign "atg_exp2_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_exp2_out =
+    foreign "atg_exp2_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_exp_ = foreign "atg_exp_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_exp_out =
+    foreign "atg_exp_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_expand =
+    foreign
+      "atg_expand"
+      (gc_tensor @-> ptr int64_t @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_expand_as =
+    foreign "atg_expand_as" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_expand_copy =
+    foreign
+      "atg_expand_copy"
+      (gc_tensor @-> ptr int64_t @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_expand_copy_out =
+    foreign
+      "atg_expand_copy_out"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_expm1 = foreign "atg_expm1" (gc_tensor @-> returning raw_tensor)
+  let stubs_expm1_ = foreign "atg_expm1_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_expm1_out =
+    foreign "atg_expm1_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_exponential =
+    foreign "atg_exponential" (gc_tensor @-> double @-> returning raw_tensor)
+  ;;
+
+  let stubs_exponential_ =
+    foreign "atg_exponential_" (gc_tensor @-> double @-> returning raw_tensor)
+  ;;
+
+  let stubs_exponential_out =
+    foreign
+      "atg_exponential_out"
+      (gc_tensor @-> gc_tensor @-> double @-> returning raw_tensor)
+  ;;
+
+  let stubs_eye = foreign "atg_eye" (int64_t @-> int @-> int @-> returning raw_tensor)
+
+  let stubs_eye_m =
+    foreign "atg_eye_m" (int64_t @-> int64_t @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_eye_m_out =
+    foreign "atg_eye_m_out" (gc_tensor @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_eye_out =
+    foreign "atg_eye_out" (gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_fake_quantize_per_channel_affine =
+    foreign
+      "atg_fake_quantize_per_channel_affine"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_fake_quantize_per_channel_affine_cachemask =
+    foreign
+      "atg_fake_quantize_per_channel_affine_cachemask"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> returning void)
+  ;;
+end
+
+module C10 (F : Cstubs.FOREIGN) = struct
+  open F
+  open Type_defs
+
+  let stubs_fake_quantize_per_channel_affine_cachemask_backward =
+    foreign
+      "atg_fake_quantize_per_channel_affine_cachemask_backward"
+      (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_fake_quantize_per_channel_affine_cachemask_out =
+    foreign
+      "atg_fake_quantize_per_channel_affine_cachemask_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> returning void)
+  ;;
+
+  let stubs_fake_quantize_per_tensor_affine =
+    foreign
+      "atg_fake_quantize_per_tensor_affine"
+      (gc_tensor @-> double @-> int64_t @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_fake_quantize_per_tensor_affine_cachemask =
+    foreign
+      "atg_fake_quantize_per_tensor_affine_cachemask"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> double
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> returning void)
+  ;;
+
+  let stubs_fake_quantize_per_tensor_affine_cachemask_backward =
+    foreign
+      "atg_fake_quantize_per_tensor_affine_cachemask_backward"
+      (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_fake_quantize_per_tensor_affine_cachemask_out =
+    foreign
+      "atg_fake_quantize_per_tensor_affine_cachemask_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> returning void)
+  ;;
+
+  let stubs_fake_quantize_per_tensor_affine_tensor_qparams =
+    foreign
+      "atg_fake_quantize_per_tensor_affine_tensor_qparams"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_fbgemm_linear_fp16_weight =
+    foreign
+      "atg_fbgemm_linear_fp16_weight"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_fbgemm_linear_fp16_weight_fp32_activation =
+    foreign
+      "atg_fbgemm_linear_fp16_weight_fp32_activation"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_fbgemm_linear_int8_weight =
+    foreign
+      "atg_fbgemm_linear_int8_weight"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> scalar
+       @-> scalar
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_fbgemm_linear_int8_weight_fp32_activation =
+    foreign
+      "atg_fbgemm_linear_int8_weight_fp32_activation"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> scalar
+       @-> scalar
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_fbgemm_pack_gemm_matrix_fp16 =
+    foreign "atg_fbgemm_pack_gemm_matrix_fp16" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_fbgemm_pack_quantized_matrix =
+    foreign "atg_fbgemm_pack_quantized_matrix" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_fbgemm_pack_quantized_matrix_kn =
+    foreign
+      "atg_fbgemm_pack_quantized_matrix_kn"
+      (gc_tensor @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_feature_alpha_dropout =
+    foreign
+      "atg_feature_alpha_dropout"
+      (gc_tensor @-> double @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_feature_alpha_dropout_ =
+    foreign
+      "atg_feature_alpha_dropout_"
+      (gc_tensor @-> double @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_feature_dropout =
+    foreign "atg_feature_dropout" (gc_tensor @-> double @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_feature_dropout_ =
+    foreign "atg_feature_dropout_" (gc_tensor @-> double @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_fft_fft =
+    foreign
+      "atg_fft_fft"
+      (gc_tensor @-> int64_t @-> int @-> int64_t @-> string @-> returning raw_tensor)
+  ;;
+
+  let stubs_fft_fft2 =
+    foreign
+      "atg_fft_fft2"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> string
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_fft_fft2_out =
+    foreign
+      "atg_fft_fft2_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> string
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_fft_fft_out =
+    foreign
+      "atg_fft_fft_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int
+       @-> int64_t
+       @-> string
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_fft_fftfreq =
+    foreign "atg_fft_fftfreq" (int64_t @-> double @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_fft_fftfreq_out =
+    foreign
+      "atg_fft_fftfreq_out"
+      (gc_tensor @-> int64_t @-> double @-> returning raw_tensor)
+  ;;
+
+  let stubs_fft_fftn =
+    foreign
+      "atg_fft_fftn"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> string
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_fft_fftn_out =
+    foreign
+      "atg_fft_fftn_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> string
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_fft_fftshift =
+    foreign "atg_fft_fftshift" (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_fft_hfft =
+    foreign
+      "atg_fft_hfft"
+      (gc_tensor @-> int64_t @-> int @-> int64_t @-> string @-> returning raw_tensor)
+  ;;
+
+  let stubs_fft_hfft2 =
+    foreign
+      "atg_fft_hfft2"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> string
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_fft_hfft2_out =
+    foreign
+      "atg_fft_hfft2_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> string
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_fft_hfft_out =
+    foreign
+      "atg_fft_hfft_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int
+       @-> int64_t
+       @-> string
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_fft_hfftn =
+    foreign
+      "atg_fft_hfftn"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> string
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_fft_hfftn_out =
+    foreign
+      "atg_fft_hfftn_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> string
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_fft_ifft =
+    foreign
+      "atg_fft_ifft"
+      (gc_tensor @-> int64_t @-> int @-> int64_t @-> string @-> returning raw_tensor)
+  ;;
+
+  let stubs_fft_ifft2 =
+    foreign
+      "atg_fft_ifft2"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> string
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_fft_ifft2_out =
+    foreign
+      "atg_fft_ifft2_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> string
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_fft_ifft_out =
+    foreign
+      "atg_fft_ifft_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int
+       @-> int64_t
+       @-> string
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_fft_ifftn =
+    foreign
+      "atg_fft_ifftn"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> string
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_fft_ifftn_out =
+    foreign
+      "atg_fft_ifftn_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> string
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_fft_ifftshift =
+    foreign
+      "atg_fft_ifftshift"
+      (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_fft_ihfft =
+    foreign
+      "atg_fft_ihfft"
+      (gc_tensor @-> int64_t @-> int @-> int64_t @-> string @-> returning raw_tensor)
+  ;;
+
+  let stubs_fft_ihfft2 =
+    foreign
+      "atg_fft_ihfft2"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> string
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_fft_ihfft2_out =
+    foreign
+      "atg_fft_ihfft2_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> string
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_fft_ihfft_out =
+    foreign
+      "atg_fft_ihfft_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int
+       @-> int64_t
+       @-> string
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_fft_ihfftn =
+    foreign
+      "atg_fft_ihfftn"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> string
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_fft_ihfftn_out =
+    foreign
+      "atg_fft_ihfftn_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> string
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_fft_irfft =
+    foreign
+      "atg_fft_irfft"
+      (gc_tensor @-> int64_t @-> int @-> int64_t @-> string @-> returning raw_tensor)
+  ;;
+
+  let stubs_fft_irfft2 =
+    foreign
+      "atg_fft_irfft2"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> string
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_fft_irfft2_out =
+    foreign
+      "atg_fft_irfft2_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> string
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_fft_irfft_out =
+    foreign
+      "atg_fft_irfft_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int
+       @-> int64_t
+       @-> string
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_fft_irfftn =
+    foreign
+      "atg_fft_irfftn"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> string
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_fft_irfftn_out =
+    foreign
+      "atg_fft_irfftn_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> string
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_fft_rfft =
+    foreign
+      "atg_fft_rfft"
+      (gc_tensor @-> int64_t @-> int @-> int64_t @-> string @-> returning raw_tensor)
+  ;;
+
+  let stubs_fft_rfft2 =
+    foreign
+      "atg_fft_rfft2"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> string
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_fft_rfft2_out =
+    foreign
+      "atg_fft_rfft2_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> string
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_fft_rfft_out =
+    foreign
+      "atg_fft_rfft_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int
+       @-> int64_t
+       @-> string
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_fft_rfftfreq =
+    foreign
+      "atg_fft_rfftfreq"
+      (int64_t @-> double @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_fft_rfftfreq_out =
+    foreign
+      "atg_fft_rfftfreq_out"
+      (gc_tensor @-> int64_t @-> double @-> returning raw_tensor)
+  ;;
+
+  let stubs_fft_rfftn =
+    foreign
+      "atg_fft_rfftn"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> string
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_fft_rfftn_out =
+    foreign
+      "atg_fft_rfftn_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> string
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_fill = foreign "atg_fill" (gc_tensor @-> scalar @-> returning raw_tensor)
+  let stubs_fill_ = foreign "atg_fill_" (gc_tensor @-> scalar @-> returning raw_tensor)
+
+  let stubs_fill_diagonal_ =
+    foreign "atg_fill_diagonal_" (gc_tensor @-> scalar @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_fill_scalar_out =
+    foreign
+      "atg_fill_scalar_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_fill_tensor =
+    foreign "atg_fill_tensor" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_fill_tensor_ =
+    foreign "atg_fill_tensor_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_fill_tensor_out =
+    foreign
+      "atg_fill_tensor_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_fix = foreign "atg_fix" (gc_tensor @-> returning raw_tensor)
+  let stubs_fix_ = foreign "atg_fix_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_fix_out =
+    foreign "atg_fix_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_flatten =
+    foreign "atg_flatten" (gc_tensor @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_flatten_dense_tensors =
+    foreign "atg_flatten_dense_tensors" (ptr gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_flip =
+    foreign "atg_flip" (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_flip_out =
+    foreign
+      "atg_flip_out"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_fliplr = foreign "atg_fliplr" (gc_tensor @-> returning raw_tensor)
+  let stubs_flipud = foreign "atg_flipud" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_float_power =
+    foreign "atg_float_power" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_float_power_ =
+    foreign "atg_float_power_" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_float_power_scalar =
+    foreign "atg_float_power_scalar" (scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_float_power_scalar_out =
+    foreign
+      "atg_float_power_scalar_out"
+      (gc_tensor @-> scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_float_power_tensor_ =
+    foreign "atg_float_power_tensor_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_float_power_tensor_scalar =
+    foreign "atg_float_power_tensor_scalar" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_float_power_tensor_scalar_out =
+    foreign
+      "atg_float_power_tensor_scalar_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_float_power_tensor_tensor_out =
+    foreign
+      "atg_float_power_tensor_tensor_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_floor = foreign "atg_floor" (gc_tensor @-> returning raw_tensor)
+  let stubs_floor_ = foreign "atg_floor_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_floor_divide =
+    foreign "atg_floor_divide" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_floor_divide_ =
+    foreign "atg_floor_divide_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_floor_divide_out =
+    foreign
+      "atg_floor_divide_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_floor_divide_scalar =
+    foreign "atg_floor_divide_scalar" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_floor_divide_scalar_ =
+    foreign "atg_floor_divide_scalar_" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_floor_out =
+    foreign "atg_floor_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_fmax = foreign "atg_fmax" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+
+  let stubs_fmax_out =
+    foreign "atg_fmax_out" (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_fmin = foreign "atg_fmin" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+
+  let stubs_fmin_out =
+    foreign "atg_fmin_out" (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_fmod = foreign "atg_fmod" (gc_tensor @-> scalar @-> returning raw_tensor)
+  let stubs_fmod_ = foreign "atg_fmod_" (gc_tensor @-> scalar @-> returning raw_tensor)
+
+  let stubs_fmod_scalar_out =
+    foreign
+      "atg_fmod_scalar_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_fmod_tensor =
+    foreign "atg_fmod_tensor" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+end
+
+module C11 (F : Cstubs.FOREIGN) = struct
+  open F
+  open Type_defs
+
+  let stubs_fmod_tensor_ =
+    foreign "atg_fmod_tensor_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_fmod_tensor_out =
+    foreign
+      "atg_fmod_tensor_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_frac = foreign "atg_frac" (gc_tensor @-> returning raw_tensor)
+  let stubs_frac_ = foreign "atg_frac_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_frac_out =
+    foreign "atg_frac_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_fractional_max_pool2d =
+    foreign
+      "atg_fractional_max_pool2d"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> gc_tensor
+       @-> returning void)
+  ;;
+
+  let stubs_fractional_max_pool2d_backward =
+    foreign
+      "atg_fractional_max_pool2d_backward"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_fractional_max_pool2d_backward_grad_input =
+    foreign
+      "atg_fractional_max_pool2d_backward_grad_input"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_fractional_max_pool2d_output =
+    foreign
+      "atg_fractional_max_pool2d_output"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> gc_tensor
+       @-> returning void)
+  ;;
+
+  let stubs_fractional_max_pool3d =
+    foreign
+      "atg_fractional_max_pool3d"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> gc_tensor
+       @-> returning void)
+  ;;
+
+  let stubs_fractional_max_pool3d_backward =
+    foreign
+      "atg_fractional_max_pool3d_backward"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_fractional_max_pool3d_backward_grad_input =
+    foreign
+      "atg_fractional_max_pool3d_backward_grad_input"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_fractional_max_pool3d_output =
+    foreign
+      "atg_fractional_max_pool3d_output"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> gc_tensor
+       @-> returning void)
+  ;;
+
+  let stubs_frexp = foreign "atg_frexp" (ptr raw_tensor @-> gc_tensor @-> returning void)
+
+  let stubs_frexp_tensor_out =
+    foreign
+      "atg_frexp_tensor_out"
+      (ptr raw_tensor @-> gc_tensor @-> gc_tensor @-> gc_tensor @-> returning void)
+  ;;
+
+  let stubs_frobenius_norm =
+    foreign
+      "atg_frobenius_norm"
+      (gc_tensor @-> ptr int64_t @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_frobenius_norm_out =
+    foreign
+      "atg_frobenius_norm_out"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_from_file =
+    foreign
+      "atg_from_file"
+      (string @-> int @-> int64_t @-> int @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_from_file_out =
+    foreign
+      "atg_from_file_out"
+      (gc_tensor @-> string @-> int @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_full =
+    foreign
+      "atg_full"
+      (ptr int64_t @-> int @-> scalar @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_full_like =
+    foreign "atg_full_like" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_full_like_out =
+    foreign
+      "atg_full_like_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_full_out =
+    foreign
+      "atg_full_out"
+      (gc_tensor @-> ptr int64_t @-> int @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_fused_moving_avg_obs_fake_quant =
+    foreign
+      "atg_fused_moving_avg_obs_fake_quant"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_gather =
+    foreign
+      "atg_gather"
+      (gc_tensor @-> int64_t @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_gather_backward =
+    foreign
+      "atg_gather_backward"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_gather_out =
+    foreign
+      "atg_gather_out"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_gcd = foreign "atg_gcd" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  let stubs_gcd_ = foreign "atg_gcd_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+
+  let stubs_gcd_out =
+    foreign "atg_gcd_out" (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_ge = foreign "atg_ge" (gc_tensor @-> scalar @-> returning raw_tensor)
+  let stubs_ge_ = foreign "atg_ge_" (gc_tensor @-> scalar @-> returning raw_tensor)
+
+  let stubs_ge_scalar_out =
+    foreign
+      "atg_ge_scalar_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_ge_tensor =
+    foreign "atg_ge_tensor" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_ge_tensor_ =
+    foreign "atg_ge_tensor_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_ge_tensor_out =
+    foreign
+      "atg_ge_tensor_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_gelu = foreign "atg_gelu" (gc_tensor @-> string @-> returning raw_tensor)
+  let stubs_gelu_ = foreign "atg_gelu_" (gc_tensor @-> string @-> returning raw_tensor)
+
+  let stubs_gelu_backward =
+    foreign
+      "atg_gelu_backward"
+      (gc_tensor @-> gc_tensor @-> string @-> returning raw_tensor)
+  ;;
+
+  let stubs_gelu_backward_grad_input =
+    foreign
+      "atg_gelu_backward_grad_input"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> string @-> returning raw_tensor)
+  ;;
+
+  let stubs_gelu_out =
+    foreign "atg_gelu_out" (gc_tensor @-> gc_tensor @-> string @-> returning raw_tensor)
+  ;;
+
+  let stubs_geometric =
+    foreign "atg_geometric" (gc_tensor @-> double @-> returning raw_tensor)
+  ;;
+
+  let stubs_geometric_ =
+    foreign "atg_geometric_" (gc_tensor @-> double @-> returning raw_tensor)
+  ;;
+
+  let stubs_geometric_out =
+    foreign
+      "atg_geometric_out"
+      (gc_tensor @-> gc_tensor @-> double @-> returning raw_tensor)
+  ;;
+
+  let stubs_geqrf = foreign "atg_geqrf" (ptr raw_tensor @-> gc_tensor @-> returning void)
+
+  let stubs_geqrf_a =
+    foreign
+      "atg_geqrf_a"
+      (ptr raw_tensor @-> gc_tensor @-> gc_tensor @-> gc_tensor @-> returning void)
+  ;;
+
+  let stubs_ger = foreign "atg_ger" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+
+  let stubs_ger_out =
+    foreign "atg_ger_out" (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_glu = foreign "atg_glu" (gc_tensor @-> int64_t @-> returning raw_tensor)
+
+  let stubs_glu_backward =
+    foreign
+      "atg_glu_backward"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_glu_backward_grad_input =
+    foreign
+      "atg_glu_backward_grad_input"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_glu_backward_jvp =
+    foreign
+      "atg_glu_backward_jvp"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_glu_backward_jvp_out =
+    foreign
+      "atg_glu_backward_jvp_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_glu_jvp =
+    foreign
+      "atg_glu_jvp"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_glu_jvp_out =
+    foreign
+      "atg_glu_jvp_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_glu_out =
+    foreign "atg_glu_out" (gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_grad = foreign "atg_grad" (gc_tensor @-> returning raw_tensor)
+  let stubs_greater = foreign "atg_greater" (gc_tensor @-> scalar @-> returning raw_tensor)
+
+  let stubs_greater_ =
+    foreign "atg_greater_" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_greater_equal =
+    foreign "atg_greater_equal" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_greater_equal_ =
+    foreign "atg_greater_equal_" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_greater_equal_scalar_out =
+    foreign
+      "atg_greater_equal_scalar_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_greater_equal_tensor =
+    foreign "atg_greater_equal_tensor" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_greater_equal_tensor_ =
+    foreign "atg_greater_equal_tensor_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_greater_equal_tensor_out =
+    foreign
+      "atg_greater_equal_tensor_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_greater_scalar_out =
+    foreign
+      "atg_greater_scalar_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_greater_tensor =
+    foreign "atg_greater_tensor" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_greater_tensor_ =
+    foreign "atg_greater_tensor_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_greater_tensor_out =
+    foreign
+      "atg_greater_tensor_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_grid_sampler =
+    foreign
+      "atg_grid_sampler"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_grid_sampler_2d =
+    foreign
+      "atg_grid_sampler_2d"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_grid_sampler_2d_out =
+    foreign
+      "atg_grid_sampler_2d_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_grid_sampler_3d =
+    foreign
+      "atg_grid_sampler_3d"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_grid_sampler_3d_out =
+    foreign
+      "atg_grid_sampler_3d_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_group_norm =
+    foreign
+      "atg_group_norm"
+      (gc_tensor
+       @-> int64_t
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_gru =
+    foreign
+      "atg_gru"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr gc_tensor
+       @-> int
+       @-> int
+       @-> int64_t
+       @-> double
+       @-> int
+       @-> int
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_gru_cell =
+    foreign
+      "atg_gru_cell"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_gru_data =
+    foreign
+      "atg_gru_data"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr gc_tensor
+       @-> int
+       @-> int
+       @-> int64_t
+       @-> double
+       @-> int
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_gt = foreign "atg_gt" (gc_tensor @-> scalar @-> returning raw_tensor)
+  let stubs_gt_ = foreign "atg_gt_" (gc_tensor @-> scalar @-> returning raw_tensor)
+
+  let stubs_gt_scalar_out =
+    foreign
+      "atg_gt_scalar_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_gt_tensor =
+    foreign "atg_gt_tensor" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_gt_tensor_ =
+    foreign "atg_gt_tensor_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_gt_tensor_out =
+    foreign
+      "atg_gt_tensor_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_hamming_window =
+    foreign "atg_hamming_window" (int64_t @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_hamming_window_out =
+    foreign "atg_hamming_window_out" (gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_hamming_window_periodic =
+    foreign
+      "atg_hamming_window_periodic"
+      (int64_t @-> int @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_hamming_window_periodic_alpha =
+    foreign
+      "atg_hamming_window_periodic_alpha"
+      (int64_t @-> int @-> double @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_hamming_window_periodic_alpha_beta =
+    foreign
+      "atg_hamming_window_periodic_alpha_beta"
+      (int64_t @-> int @-> double @-> double @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_hamming_window_periodic_alpha_beta_out =
+    foreign
+      "atg_hamming_window_periodic_alpha_beta_out"
+      (gc_tensor @-> int64_t @-> int @-> double @-> double @-> returning raw_tensor)
+  ;;
+
+  let stubs_hamming_window_periodic_alpha_out =
+    foreign
+      "atg_hamming_window_periodic_alpha_out"
+      (gc_tensor @-> int64_t @-> int @-> double @-> returning raw_tensor)
+  ;;
+
+  let stubs_hamming_window_periodic_out =
+    foreign
+      "atg_hamming_window_periodic_out"
+      (gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_hann_window =
+    foreign "atg_hann_window" (int64_t @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_hann_window_out =
+    foreign "atg_hann_window_out" (gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_hann_window_periodic =
+    foreign
+      "atg_hann_window_periodic"
+      (int64_t @-> int @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_hann_window_periodic_out =
+    foreign
+      "atg_hann_window_periodic_out"
+      (gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_hardshrink = foreign "atg_hardshrink" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_hardshrink_backward =
+    foreign
+      "atg_hardshrink_backward"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_hardshrink_backward_grad_input =
+    foreign
+      "atg_hardshrink_backward_grad_input"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_hardshrink_out =
+    foreign "atg_hardshrink_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+end
+
+module C12 (F : Cstubs.FOREIGN) = struct
+  open F
+  open Type_defs
+
+  let stubs_hardsigmoid = foreign "atg_hardsigmoid" (gc_tensor @-> returning raw_tensor)
+  let stubs_hardsigmoid_ = foreign "atg_hardsigmoid_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_hardsigmoid_backward =
+    foreign "atg_hardsigmoid_backward" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_hardsigmoid_backward_grad_input =
+    foreign
+      "atg_hardsigmoid_backward_grad_input"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_hardsigmoid_out =
+    foreign "atg_hardsigmoid_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_hardswish = foreign "atg_hardswish" (gc_tensor @-> returning raw_tensor)
+  let stubs_hardswish_ = foreign "atg_hardswish_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_hardswish_backward =
+    foreign "atg_hardswish_backward" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_hardswish_backward_out =
+    foreign
+      "atg_hardswish_backward_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_hardswish_out =
+    foreign "atg_hardswish_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_hardtanh = foreign "atg_hardtanh" (gc_tensor @-> returning raw_tensor)
+  let stubs_hardtanh_ = foreign "atg_hardtanh_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_hardtanh_backward =
+    foreign
+      "atg_hardtanh_backward"
+      (gc_tensor @-> gc_tensor @-> scalar @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_hardtanh_backward_grad_input =
+    foreign
+      "atg_hardtanh_backward_grad_input"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> scalar
+       @-> scalar
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_hardtanh_out =
+    foreign "atg_hardtanh_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_heaviside =
+    foreign "atg_heaviside" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_heaviside_ =
+    foreign "atg_heaviside_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_heaviside_out =
+    foreign
+      "atg_heaviside_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_hinge_embedding_loss =
+    foreign
+      "atg_hinge_embedding_loss"
+      (gc_tensor @-> gc_tensor @-> double @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_histc = foreign "atg_histc" (gc_tensor @-> int64_t @-> returning raw_tensor)
+
+  let stubs_histc_out =
+    foreign "atg_histc_out" (gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_hsplit =
+    foreign "atg_hsplit" (gc_tensor @-> int64_t @-> returning (ptr raw_tensor))
+  ;;
+
+  let stubs_hsplit_array =
+    foreign
+      "atg_hsplit_array"
+      (gc_tensor @-> ptr int64_t @-> int @-> returning (ptr raw_tensor))
+  ;;
+
+  let stubs_hspmm = foreign "atg_hspmm" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+
+  let stubs_hspmm_out =
+    foreign
+      "atg_hspmm_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_hstack = foreign "atg_hstack" (ptr gc_tensor @-> int @-> returning raw_tensor)
+
+  let stubs_hstack_out =
+    foreign "atg_hstack_out" (gc_tensor @-> ptr gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_huber_loss =
+    foreign
+      "atg_huber_loss"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> double @-> returning raw_tensor)
+  ;;
+
+  let stubs_huber_loss_backward =
+    foreign
+      "atg_huber_loss_backward"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> double
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_huber_loss_backward_out =
+    foreign
+      "atg_huber_loss_backward_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> double
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_huber_loss_out =
+    foreign
+      "atg_huber_loss_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> double
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_hypot = foreign "atg_hypot" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+
+  let stubs_hypot_ =
+    foreign "atg_hypot_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_hypot_out =
+    foreign
+      "atg_hypot_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_i0 = foreign "atg_i0" (gc_tensor @-> returning raw_tensor)
+  let stubs_i0_ = foreign "atg_i0_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_i0_out =
+    foreign "atg_i0_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_igamma =
+    foreign "atg_igamma" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_igamma_ =
+    foreign "atg_igamma_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_igamma_out =
+    foreign
+      "atg_igamma_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_igammac =
+    foreign "atg_igammac" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_igammac_ =
+    foreign "atg_igammac_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_igammac_out =
+    foreign
+      "atg_igammac_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_im2col =
+    foreign
+      "atg_im2col"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_im2col_out =
+    foreign
+      "atg_im2col_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_imag = foreign "atg_imag" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_index =
+    foreign "atg_index" (gc_tensor @-> ptr gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_index_add =
+    foreign
+      "atg_index_add"
+      (gc_tensor @-> int64_t @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_index_add_ =
+    foreign
+      "atg_index_add_"
+      (gc_tensor @-> int64_t @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_index_add_out =
+    foreign
+      "atg_index_add_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_index_copy =
+    foreign
+      "atg_index_copy"
+      (gc_tensor @-> int64_t @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_index_copy_ =
+    foreign
+      "atg_index_copy_"
+      (gc_tensor @-> int64_t @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_index_copy_out =
+    foreign
+      "atg_index_copy_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_index_fill =
+    foreign
+      "atg_index_fill"
+      (gc_tensor @-> int64_t @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_index_fill_ =
+    foreign
+      "atg_index_fill_"
+      (gc_tensor @-> int64_t @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_index_fill_int_scalar_out =
+    foreign
+      "atg_index_fill_int_scalar_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> gc_tensor
+       @-> scalar
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_index_fill_int_tensor =
+    foreign
+      "atg_index_fill_int_tensor"
+      (gc_tensor @-> int64_t @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_index_fill_int_tensor_ =
+    foreign
+      "atg_index_fill_int_tensor_"
+      (gc_tensor @-> int64_t @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_index_fill_int_tensor_out =
+    foreign
+      "atg_index_fill_int_tensor_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_index_put =
+    foreign
+      "atg_index_put"
+      (gc_tensor @-> ptr gc_tensor @-> int @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_index_put_ =
+    foreign
+      "atg_index_put_"
+      (gc_tensor @-> ptr gc_tensor @-> int @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_index_put_out =
+    foreign
+      "atg_index_put_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr gc_tensor
+       @-> int
+       @-> gc_tensor
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_index_reduce =
+    foreign
+      "atg_index_reduce"
+      (gc_tensor
+       @-> int64_t
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> string
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_index_reduce_ =
+    foreign
+      "atg_index_reduce_"
+      (gc_tensor
+       @-> int64_t
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> string
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_index_reduce_out =
+    foreign
+      "atg_index_reduce_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> string
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_index_select =
+    foreign
+      "atg_index_select"
+      (gc_tensor @-> int64_t @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_index_select_backward =
+    foreign
+      "atg_index_select_backward"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_index_select_out =
+    foreign
+      "atg_index_select_out"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_index_tensor_out =
+    foreign
+      "atg_index_tensor_out"
+      (gc_tensor @-> gc_tensor @-> ptr gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_indices = foreign "atg_indices" (gc_tensor @-> returning raw_tensor)
+  let stubs_indices_copy = foreign "atg_indices_copy" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_indices_copy_out =
+    foreign "atg_indices_copy_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_infinitely_differentiable_gelu_backward =
+    foreign
+      "atg_infinitely_differentiable_gelu_backward"
+      (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_inner = foreign "atg_inner" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+
+  let stubs_inner_out =
+    foreign
+      "atg_inner_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_instance_norm =
+    foreign
+      "atg_instance_norm"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> double
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_int_repr = foreign "atg_int_repr" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_int_repr_out =
+    foreign "atg_int_repr_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_inverse = foreign "atg_inverse" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_inverse_out =
+    foreign "atg_inverse_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_is_coalesced = foreign "atg_is_coalesced" (gc_tensor @-> returning bool)
+  let stubs_is_complex = foreign "atg_is_complex" (gc_tensor @-> returning bool)
+  let stubs_is_conj = foreign "atg_is_conj" (gc_tensor @-> returning bool)
+  let stubs_is_distributed = foreign "atg_is_distributed" (gc_tensor @-> returning bool)
+
+  let stubs_is_floating_point =
+    foreign "atg_is_floating_point" (gc_tensor @-> returning bool)
+  ;;
+
+  let stubs_is_inference = foreign "atg_is_inference" (gc_tensor @-> returning bool)
+  let stubs_is_leaf = foreign "atg_is_leaf" (gc_tensor @-> returning bool)
+  let stubs_is_neg = foreign "atg_is_neg" (gc_tensor @-> returning bool)
+  let stubs_is_nonzero = foreign "atg_is_nonzero" (gc_tensor @-> returning bool)
+  let stubs_is_pinned = foreign "atg_is_pinned" (gc_tensor @-> int @-> returning bool)
+
+  let stubs_is_same_size =
+    foreign "atg_is_same_size" (gc_tensor @-> gc_tensor @-> returning bool)
+  ;;
+
+  let stubs_is_set_to =
+    foreign "atg_is_set_to" (gc_tensor @-> gc_tensor @-> returning bool)
+  ;;
+
+  let stubs_is_signed = foreign "atg_is_signed" (gc_tensor @-> returning bool)
+
+  let stubs_is_vulkan_available =
+    foreign "atg_is_vulkan_available" (void @-> returning bool)
+  ;;
+
+  let stubs_isclose =
+    foreign
+      "atg_isclose"
+      (gc_tensor @-> gc_tensor @-> double @-> double @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_isfinite = foreign "atg_isfinite" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_isin =
+    foreign "atg_isin" (gc_tensor @-> gc_tensor @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_isin_scalar_tensor =
+    foreign
+      "atg_isin_scalar_tensor"
+      (scalar @-> gc_tensor @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_isin_scalar_tensor_out =
+    foreign
+      "atg_isin_scalar_tensor_out"
+      (gc_tensor @-> scalar @-> gc_tensor @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_isin_tensor_scalar =
+    foreign
+      "atg_isin_tensor_scalar"
+      (gc_tensor @-> scalar @-> int @-> int @-> returning raw_tensor)
+  ;;
+end
+
+module C13 (F : Cstubs.FOREIGN) = struct
+  open F
+  open Type_defs
+
+  let stubs_isin_tensor_scalar_out =
+    foreign
+      "atg_isin_tensor_scalar_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_isin_tensor_tensor_out =
+    foreign
+      "atg_isin_tensor_tensor_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_isinf = foreign "atg_isinf" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_isinf_out =
+    foreign "atg_isinf_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_isnan = foreign "atg_isnan" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_isnan_out =
+    foreign "atg_isnan_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_isneginf = foreign "atg_isneginf" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_isneginf_out =
+    foreign "atg_isneginf_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_isposinf = foreign "atg_isposinf" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_isposinf_out =
+    foreign "atg_isposinf_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_isreal = foreign "atg_isreal" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_istft =
+    foreign
+      "atg_istft"
+      (gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> gc_tensor
+       @-> int
+       @-> int
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_kaiser_window =
+    foreign "atg_kaiser_window" (int64_t @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_kaiser_window_beta =
+    foreign
+      "atg_kaiser_window_beta"
+      (int64_t @-> int @-> double @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_kaiser_window_beta_out =
+    foreign
+      "atg_kaiser_window_beta_out"
+      (gc_tensor @-> int64_t @-> int @-> double @-> returning raw_tensor)
+  ;;
+
+  let stubs_kaiser_window_out =
+    foreign "atg_kaiser_window_out" (gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_kaiser_window_periodic =
+    foreign
+      "atg_kaiser_window_periodic"
+      (int64_t @-> int @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_kaiser_window_periodic_out =
+    foreign
+      "atg_kaiser_window_periodic_out"
+      (gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_kl_div =
+    foreign
+      "atg_kl_div"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_kron = foreign "atg_kron" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+
+  let stubs_kron_out =
+    foreign "atg_kron_out" (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_kthvalue =
+    foreign
+      "atg_kthvalue"
+      (ptr raw_tensor @-> gc_tensor @-> int64_t @-> int64_t @-> int @-> returning void)
+  ;;
+
+  let stubs_kthvalue_values =
+    foreign
+      "atg_kthvalue_values"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_l1_loss =
+    foreign "atg_l1_loss" (gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_layer_norm =
+    foreign
+      "atg_layer_norm"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_lcm = foreign "atg_lcm" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  let stubs_lcm_ = foreign "atg_lcm_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+
+  let stubs_lcm_out =
+    foreign "atg_lcm_out" (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_ldexp = foreign "atg_ldexp" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+
+  let stubs_ldexp_ =
+    foreign "atg_ldexp_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_ldexp_out =
+    foreign
+      "atg_ldexp_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_le = foreign "atg_le" (gc_tensor @-> scalar @-> returning raw_tensor)
+  let stubs_le_ = foreign "atg_le_" (gc_tensor @-> scalar @-> returning raw_tensor)
+
+  let stubs_le_scalar_out =
+    foreign
+      "atg_le_scalar_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_le_tensor =
+    foreign "atg_le_tensor" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_le_tensor_ =
+    foreign "atg_le_tensor_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_le_tensor_out =
+    foreign
+      "atg_le_tensor_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_leaky_relu = foreign "atg_leaky_relu" (gc_tensor @-> returning raw_tensor)
+  let stubs_leaky_relu_ = foreign "atg_leaky_relu_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_leaky_relu_backward =
+    foreign
+      "atg_leaky_relu_backward"
+      (gc_tensor @-> gc_tensor @-> scalar @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_leaky_relu_backward_grad_input =
+    foreign
+      "atg_leaky_relu_backward_grad_input"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> scalar @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_leaky_relu_out =
+    foreign "atg_leaky_relu_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_lerp =
+    foreign "atg_lerp" (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_lerp_ =
+    foreign "atg_lerp_" (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_lerp_scalar_out =
+    foreign
+      "atg_lerp_scalar_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_lerp_tensor =
+    foreign
+      "atg_lerp_tensor"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_lerp_tensor_ =
+    foreign
+      "atg_lerp_tensor_"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_lerp_tensor_out =
+    foreign
+      "atg_lerp_tensor_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_less = foreign "atg_less" (gc_tensor @-> scalar @-> returning raw_tensor)
+  let stubs_less_ = foreign "atg_less_" (gc_tensor @-> scalar @-> returning raw_tensor)
+
+  let stubs_less_equal =
+    foreign "atg_less_equal" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_less_equal_ =
+    foreign "atg_less_equal_" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_less_equal_scalar_out =
+    foreign
+      "atg_less_equal_scalar_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_less_equal_tensor =
+    foreign "atg_less_equal_tensor" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_less_equal_tensor_ =
+    foreign "atg_less_equal_tensor_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_less_equal_tensor_out =
+    foreign
+      "atg_less_equal_tensor_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_less_scalar_out =
+    foreign
+      "atg_less_scalar_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_less_tensor =
+    foreign "atg_less_tensor" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_less_tensor_ =
+    foreign "atg_less_tensor_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_less_tensor_out =
+    foreign
+      "atg_less_tensor_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_lgamma = foreign "atg_lgamma" (gc_tensor @-> returning raw_tensor)
+  let stubs_lgamma_ = foreign "atg_lgamma_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_lgamma_out =
+    foreign "atg_lgamma_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_lift = foreign "atg_lift" (gc_tensor @-> returning raw_tensor)
+  let stubs_lift_fresh = foreign "atg_lift_fresh" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_lift_fresh_copy =
+    foreign "atg_lift_fresh_copy" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_lift_fresh_copy_out =
+    foreign "atg_lift_fresh_copy_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_lift_out =
+    foreign "atg_lift_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_cholesky =
+    foreign "atg_linalg_cholesky" (gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_cholesky_ex =
+    foreign
+      "atg_linalg_cholesky_ex"
+      (ptr raw_tensor @-> gc_tensor @-> int @-> int @-> returning void)
+  ;;
+
+  let stubs_linalg_cholesky_ex_l =
+    foreign
+      "atg_linalg_cholesky_ex_l"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_linalg_cholesky_out =
+    foreign
+      "atg_linalg_cholesky_out"
+      (gc_tensor @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_cond =
+    foreign "atg_linalg_cond" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_cond_out =
+    foreign
+      "atg_linalg_cond_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_cond_p_str =
+    foreign "atg_linalg_cond_p_str" (gc_tensor @-> string @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_cond_p_str_out =
+    foreign
+      "atg_linalg_cond_p_str_out"
+      (gc_tensor @-> gc_tensor @-> string @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_cross =
+    foreign
+      "atg_linalg_cross"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_cross_out =
+    foreign
+      "atg_linalg_cross_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_det = foreign "atg_linalg_det" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_linalg_det_out =
+    foreign "atg_linalg_det_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_diagonal =
+    foreign
+      "atg_linalg_diagonal"
+      (gc_tensor @-> int64_t @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_eig =
+    foreign "atg_linalg_eig" (ptr raw_tensor @-> gc_tensor @-> returning void)
+  ;;
+
+  let stubs_linalg_eig_out =
+    foreign
+      "atg_linalg_eig_out"
+      (ptr raw_tensor @-> gc_tensor @-> gc_tensor @-> gc_tensor @-> returning void)
+  ;;
+
+  let stubs_linalg_eigh =
+    foreign "atg_linalg_eigh" (ptr raw_tensor @-> gc_tensor @-> string @-> returning void)
+  ;;
+
+  let stubs_linalg_eigh_eigvals =
+    foreign
+      "atg_linalg_eigh_eigvals"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> string
+       @-> returning void)
+  ;;
+
+  let stubs_linalg_eigvals =
+    foreign "atg_linalg_eigvals" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_eigvals_out =
+    foreign "atg_linalg_eigvals_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_eigvalsh =
+    foreign "atg_linalg_eigvalsh" (gc_tensor @-> string @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_eigvalsh_out =
+    foreign
+      "atg_linalg_eigvalsh_out"
+      (gc_tensor @-> gc_tensor @-> string @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_householder_product =
+    foreign
+      "atg_linalg_householder_product"
+      (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_householder_product_out =
+    foreign
+      "atg_linalg_householder_product_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_inv = foreign "atg_linalg_inv" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_linalg_inv_ex =
+    foreign "atg_linalg_inv_ex" (ptr raw_tensor @-> gc_tensor @-> int @-> returning void)
+  ;;
+
+  let stubs_linalg_inv_ex_inverse =
+    foreign
+      "atg_linalg_inv_ex_inverse"
+      (ptr raw_tensor @-> gc_tensor @-> gc_tensor @-> gc_tensor @-> int @-> returning void)
+  ;;
+
+  let stubs_linalg_inv_out =
+    foreign "atg_linalg_inv_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_ldl_factor =
+    foreign
+      "atg_linalg_ldl_factor"
+      (ptr raw_tensor @-> gc_tensor @-> int @-> returning void)
+  ;;
+
+  let stubs_linalg_ldl_factor_ex =
+    foreign
+      "atg_linalg_ldl_factor_ex"
+      (ptr raw_tensor @-> gc_tensor @-> int @-> int @-> returning void)
+  ;;
+
+  let stubs_linalg_ldl_factor_ex_out =
+    foreign
+      "atg_linalg_ldl_factor_ex_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_linalg_ldl_factor_out =
+    foreign
+      "atg_linalg_ldl_factor_out"
+      (ptr raw_tensor @-> gc_tensor @-> gc_tensor @-> gc_tensor @-> int @-> returning void)
+  ;;
+
+  let stubs_linalg_ldl_solve =
+    foreign
+      "atg_linalg_ldl_solve"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+end
+
+module C14 (F : Cstubs.FOREIGN) = struct
+  open F
+  open Type_defs
+
+  let stubs_linalg_ldl_solve_out =
+    foreign
+      "atg_linalg_ldl_solve_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_lstsq =
+    foreign
+      "atg_linalg_lstsq"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> int
+       @-> string
+       @-> returning void)
+  ;;
+
+  let stubs_linalg_lstsq_out =
+    foreign
+      "atg_linalg_lstsq_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> int
+       @-> string
+       @-> returning void)
+  ;;
+
+  let stubs_linalg_lu =
+    foreign "atg_linalg_lu" (ptr raw_tensor @-> gc_tensor @-> int @-> returning void)
+  ;;
+
+  let stubs_linalg_lu_factor =
+    foreign
+      "atg_linalg_lu_factor"
+      (ptr raw_tensor @-> gc_tensor @-> int @-> returning void)
+  ;;
+
+  let stubs_linalg_lu_factor_ex =
+    foreign
+      "atg_linalg_lu_factor_ex"
+      (ptr raw_tensor @-> gc_tensor @-> int @-> int @-> returning void)
+  ;;
+
+  let stubs_linalg_lu_factor_ex_out =
+    foreign
+      "atg_linalg_lu_factor_ex_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_linalg_lu_factor_out =
+    foreign
+      "atg_linalg_lu_factor_out"
+      (ptr raw_tensor @-> gc_tensor @-> gc_tensor @-> gc_tensor @-> int @-> returning void)
+  ;;
+
+  let stubs_linalg_lu_out =
+    foreign
+      "atg_linalg_lu_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_linalg_lu_solve =
+    foreign
+      "atg_linalg_lu_solve"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_lu_solve_out =
+    foreign
+      "atg_linalg_lu_solve_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_matmul =
+    foreign "atg_linalg_matmul" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_matmul_out =
+    foreign
+      "atg_linalg_matmul_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_matrix_exp =
+    foreign "atg_linalg_matrix_exp" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_matrix_exp_out =
+    foreign "atg_linalg_matrix_exp_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_matrix_power =
+    foreign "atg_linalg_matrix_power" (gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_matrix_power_out =
+    foreign
+      "atg_linalg_matrix_power_out"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_matrix_rank =
+    foreign
+      "atg_linalg_matrix_rank"
+      (gc_tensor @-> double @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_matrix_rank_atol_rtol_float =
+    foreign
+      "atg_linalg_matrix_rank_atol_rtol_float"
+      (gc_tensor @-> double @-> int @-> double @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_matrix_rank_atol_rtol_float_out =
+    foreign
+      "atg_linalg_matrix_rank_atol_rtol_float_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_matrix_rank_atol_rtol_tensor =
+    foreign
+      "atg_linalg_matrix_rank_atol_rtol_tensor"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_matrix_rank_atol_rtol_tensor_out =
+    foreign
+      "atg_linalg_matrix_rank_atol_rtol_tensor_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_matrix_rank_out =
+    foreign
+      "atg_linalg_matrix_rank_out"
+      (gc_tensor @-> gc_tensor @-> double @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_matrix_rank_out_tol_tensor =
+    foreign
+      "atg_linalg_matrix_rank_out_tol_tensor"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_matrix_rank_tol_tensor =
+    foreign
+      "atg_linalg_matrix_rank_tol_tensor"
+      (gc_tensor @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_multi_dot =
+    foreign "atg_linalg_multi_dot" (ptr gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_multi_dot_out =
+    foreign
+      "atg_linalg_multi_dot_out"
+      (gc_tensor @-> ptr gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_pinv =
+    foreign "atg_linalg_pinv" (gc_tensor @-> double @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_pinv_atol_rtol_float =
+    foreign
+      "atg_linalg_pinv_atol_rtol_float"
+      (gc_tensor @-> double @-> int @-> double @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_pinv_atol_rtol_float_out =
+    foreign
+      "atg_linalg_pinv_atol_rtol_float_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_pinv_atol_rtol_tensor =
+    foreign
+      "atg_linalg_pinv_atol_rtol_tensor"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_pinv_atol_rtol_tensor_out =
+    foreign
+      "atg_linalg_pinv_atol_rtol_tensor_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_pinv_out =
+    foreign
+      "atg_linalg_pinv_out"
+      (gc_tensor @-> gc_tensor @-> double @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_pinv_out_rcond_tensor =
+    foreign
+      "atg_linalg_pinv_out_rcond_tensor"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_pinv_rcond_tensor =
+    foreign
+      "atg_linalg_pinv_rcond_tensor"
+      (gc_tensor @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_qr =
+    foreign "atg_linalg_qr" (ptr raw_tensor @-> gc_tensor @-> string @-> returning void)
+  ;;
+
+  let stubs_linalg_qr_out =
+    foreign
+      "atg_linalg_qr_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> string
+       @-> returning void)
+  ;;
+
+  let stubs_linalg_slogdet =
+    foreign "atg_linalg_slogdet" (ptr raw_tensor @-> gc_tensor @-> returning void)
+  ;;
+
+  let stubs_linalg_slogdet_out =
+    foreign
+      "atg_linalg_slogdet_out"
+      (ptr raw_tensor @-> gc_tensor @-> gc_tensor @-> gc_tensor @-> returning void)
+  ;;
+
+  let stubs_linalg_solve =
+    foreign "atg_linalg_solve" (gc_tensor @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_solve_ex =
+    foreign
+      "atg_linalg_solve_ex"
+      (ptr raw_tensor @-> gc_tensor @-> gc_tensor @-> int @-> int @-> returning void)
+  ;;
+
+  let stubs_linalg_solve_ex_out =
+    foreign
+      "atg_linalg_solve_ex_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_linalg_solve_out =
+    foreign
+      "atg_linalg_solve_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_solve_triangular =
+    foreign
+      "atg_linalg_solve_triangular"
+      (gc_tensor @-> gc_tensor @-> int @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_solve_triangular_out =
+    foreign
+      "atg_linalg_solve_triangular_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_svd =
+    foreign
+      "atg_linalg_svd"
+      (ptr raw_tensor @-> gc_tensor @-> int @-> string @-> returning void)
+  ;;
+
+  let stubs_linalg_svd_u =
+    foreign
+      "atg_linalg_svd_u"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> string
+       @-> returning void)
+  ;;
+
+  let stubs_linalg_svdvals =
+    foreign "atg_linalg_svdvals" (gc_tensor @-> string @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_svdvals_out =
+    foreign
+      "atg_linalg_svdvals_out"
+      (gc_tensor @-> gc_tensor @-> string @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_tensorinv =
+    foreign "atg_linalg_tensorinv" (gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_tensorinv_out =
+    foreign
+      "atg_linalg_tensorinv_out"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_tensorsolve =
+    foreign
+      "atg_linalg_tensorsolve"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_tensorsolve_out =
+    foreign
+      "atg_linalg_tensorsolve_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_vander =
+    foreign "atg_linalg_vander" (gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_vecdot =
+    foreign
+      "atg_linalg_vecdot"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_linalg_vecdot_out =
+    foreign
+      "atg_linalg_vecdot_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_linear =
+    foreign "atg_linear" (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_linear_out =
+    foreign
+      "atg_linear_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_linspace =
+    foreign
+      "atg_linspace"
+      (scalar @-> scalar @-> int64_t @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_linspace_out =
+    foreign
+      "atg_linspace_out"
+      (gc_tensor @-> scalar @-> scalar @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_log = foreign "atg_log" (gc_tensor @-> returning raw_tensor)
+  let stubs_log10 = foreign "atg_log10" (gc_tensor @-> returning raw_tensor)
+  let stubs_log10_ = foreign "atg_log10_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_log10_out =
+    foreign "atg_log10_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_log1p = foreign "atg_log1p" (gc_tensor @-> returning raw_tensor)
+  let stubs_log1p_ = foreign "atg_log1p_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_log1p_out =
+    foreign "atg_log1p_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_log2 = foreign "atg_log2" (gc_tensor @-> returning raw_tensor)
+  let stubs_log2_ = foreign "atg_log2_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_log2_out =
+    foreign "atg_log2_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_log_ = foreign "atg_log_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_log_normal =
+    foreign "atg_log_normal" (gc_tensor @-> double @-> double @-> returning raw_tensor)
+  ;;
+
+  let stubs_log_normal_ =
+    foreign "atg_log_normal_" (gc_tensor @-> double @-> double @-> returning raw_tensor)
+  ;;
+
+  let stubs_log_normal_out =
+    foreign
+      "atg_log_normal_out"
+      (gc_tensor @-> gc_tensor @-> double @-> double @-> returning raw_tensor)
+  ;;
+
+  let stubs_log_out =
+    foreign "atg_log_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_log_sigmoid = foreign "atg_log_sigmoid" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_log_sigmoid_backward =
+    foreign
+      "atg_log_sigmoid_backward"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_log_sigmoid_backward_grad_input =
+    foreign
+      "atg_log_sigmoid_backward_grad_input"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_log_sigmoid_out =
+    foreign "atg_log_sigmoid_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_log_softmax =
+    foreign "atg_log_softmax" (gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_log_softmax_int_out =
+    foreign
+      "atg_log_softmax_int_out"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_logaddexp =
+    foreign "atg_logaddexp" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_logaddexp2 =
+    foreign "atg_logaddexp2" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_logaddexp2_out =
+    foreign
+      "atg_logaddexp2_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_logaddexp_out =
+    foreign
+      "atg_logaddexp_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_logcumsumexp =
+    foreign "atg_logcumsumexp" (gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_logcumsumexp_out =
+    foreign
+      "atg_logcumsumexp_out"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_logdet = foreign "atg_logdet" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_logical_and =
+    foreign "atg_logical_and" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_logical_and_ =
+    foreign "atg_logical_and_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_logical_and_out =
+    foreign
+      "atg_logical_and_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_logical_not = foreign "atg_logical_not" (gc_tensor @-> returning raw_tensor)
+  let stubs_logical_not_ = foreign "atg_logical_not_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_logical_not_out =
+    foreign "atg_logical_not_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_logical_or =
+    foreign "atg_logical_or" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_logical_or_ =
+    foreign "atg_logical_or_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_logical_or_out =
+    foreign
+      "atg_logical_or_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_logical_xor =
+    foreign "atg_logical_xor" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_logical_xor_ =
+    foreign "atg_logical_xor_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_logical_xor_out =
+    foreign
+      "atg_logical_xor_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+end
+
+module C15 (F : Cstubs.FOREIGN) = struct
+  open F
+  open Type_defs
+
+  let stubs_logit =
+    foreign "atg_logit" (gc_tensor @-> double @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_logit_ =
+    foreign "atg_logit_" (gc_tensor @-> double @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_logit_backward =
+    foreign
+      "atg_logit_backward"
+      (gc_tensor @-> gc_tensor @-> double @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_logit_backward_grad_input =
+    foreign
+      "atg_logit_backward_grad_input"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> double @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_logit_out =
+    foreign
+      "atg_logit_out"
+      (gc_tensor @-> gc_tensor @-> double @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_logspace =
+    foreign
+      "atg_logspace"
+      (scalar @-> scalar @-> int64_t @-> double @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_logspace_out =
+    foreign
+      "atg_logspace_out"
+      (gc_tensor @-> scalar @-> scalar @-> int64_t @-> double @-> returning raw_tensor)
+  ;;
+
+  let stubs_logsumexp =
+    foreign
+      "atg_logsumexp"
+      (gc_tensor @-> ptr int64_t @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_logsumexp_out =
+    foreign
+      "atg_logsumexp_out"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_lstm =
+    foreign
+      "atg_lstm"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> int
+       @-> int64_t
+       @-> double
+       @-> int
+       @-> int
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_lstm_cell =
+    foreign
+      "atg_lstm_cell"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> ptr gc_tensor
+       @-> int
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> returning void)
+  ;;
+
+  let stubs_lstm_data =
+    foreign
+      "atg_lstm_data"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> int
+       @-> int64_t
+       @-> double
+       @-> int
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_lstm_mps_backward =
+    foreign
+      "atg_lstm_mps_backward"
+      (gc_tensor
+       @-> ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> int
+       @-> int64_t
+       @-> double
+       @-> int
+       @-> int
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_lt = foreign "atg_lt" (gc_tensor @-> scalar @-> returning raw_tensor)
+  let stubs_lt_ = foreign "atg_lt_" (gc_tensor @-> scalar @-> returning raw_tensor)
+
+  let stubs_lt_scalar_out =
+    foreign
+      "atg_lt_scalar_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_lt_tensor =
+    foreign "atg_lt_tensor" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_lt_tensor_ =
+    foreign "atg_lt_tensor_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_lt_tensor_out =
+    foreign
+      "atg_lt_tensor_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_lu_solve =
+    foreign "atg_lu_solve" (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_lu_solve_out =
+    foreign
+      "atg_lu_solve_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_lu_unpack =
+    foreign
+      "atg_lu_unpack"
+      (ptr raw_tensor @-> gc_tensor @-> gc_tensor @-> int @-> int @-> returning void)
+  ;;
+
+  let stubs_lu_unpack_out =
+    foreign
+      "atg_lu_unpack_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_margin_ranking_loss =
+    foreign
+      "atg_margin_ranking_loss"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_masked_fill =
+    foreign "atg_masked_fill" (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_masked_fill_ =
+    foreign
+      "atg_masked_fill_"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_masked_fill_scalar_out =
+    foreign
+      "atg_masked_fill_scalar_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_masked_fill_tensor =
+    foreign
+      "atg_masked_fill_tensor"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_masked_fill_tensor_ =
+    foreign
+      "atg_masked_fill_tensor_"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_masked_fill_tensor_out =
+    foreign
+      "atg_masked_fill_tensor_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_masked_scatter =
+    foreign
+      "atg_masked_scatter"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_masked_scatter_ =
+    foreign
+      "atg_masked_scatter_"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_masked_scatter_out =
+    foreign
+      "atg_masked_scatter_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_masked_select =
+    foreign "atg_masked_select" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_masked_select_backward =
+    foreign
+      "atg_masked_select_backward"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_masked_select_out =
+    foreign
+      "atg_masked_select_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_matmul =
+    foreign "atg_matmul" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_matmul_out =
+    foreign
+      "atg_matmul_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_matrix_exp = foreign "atg_matrix_exp" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_matrix_exp_backward =
+    foreign "atg_matrix_exp_backward" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_matrix_h = foreign "atg_matrix_h" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_matrix_power =
+    foreign "atg_matrix_power" (gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_matrix_power_out =
+    foreign
+      "atg_matrix_power_out"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_max = foreign "atg_max" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_max_dim =
+    foreign
+      "atg_max_dim"
+      (ptr raw_tensor @-> gc_tensor @-> int64_t @-> int @-> returning void)
+  ;;
+
+  let stubs_max_dim_max =
+    foreign
+      "atg_max_dim_max"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_max_other =
+    foreign "atg_max_other" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_max_out =
+    foreign "atg_max_out" (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_max_pool1d =
+    foreign
+      "atg_max_pool1d"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_max_pool1d_with_indices =
+    foreign
+      "atg_max_pool1d_with_indices"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_max_pool2d =
+    foreign
+      "atg_max_pool2d"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_max_pool2d_backward =
+    foreign
+      "atg_max_pool2d_backward"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_max_pool2d_backward_out =
+    foreign
+      "atg_max_pool2d_backward_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_max_pool2d_with_indices =
+    foreign
+      "atg_max_pool2d_with_indices"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_max_pool2d_with_indices_backward =
+    foreign
+      "atg_max_pool2d_with_indices_backward"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_max_pool2d_with_indices_backward_grad_input =
+    foreign
+      "atg_max_pool2d_with_indices_backward_grad_input"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_max_pool2d_with_indices_out =
+    foreign
+      "atg_max_pool2d_with_indices_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_max_pool3d =
+    foreign
+      "atg_max_pool3d"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_max_pool3d_with_indices =
+    foreign
+      "atg_max_pool3d_with_indices"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_max_pool3d_with_indices_backward =
+    foreign
+      "atg_max_pool3d_with_indices_backward"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_max_pool3d_with_indices_backward_grad_input =
+    foreign
+      "atg_max_pool3d_with_indices_backward_grad_input"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_max_pool3d_with_indices_out =
+    foreign
+      "atg_max_pool3d_with_indices_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_max_unary_out =
+    foreign "atg_max_unary_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_max_unpool2d =
+    foreign
+      "atg_max_unpool2d"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_max_unpool2d_out =
+    foreign
+      "atg_max_unpool2d_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_max_unpool3d =
+    foreign
+      "atg_max_unpool3d"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_max_unpool3d_out =
+    foreign
+      "atg_max_unpool3d_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_maximum =
+    foreign "atg_maximum" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_maximum_out =
+    foreign
+      "atg_maximum_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_mean = foreign "atg_mean" (gc_tensor @-> int @-> returning raw_tensor)
+
+  let stubs_mean_dim =
+    foreign
+      "atg_mean_dim"
+      (gc_tensor @-> ptr int64_t @-> int @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_mean_out =
+    foreign
+      "atg_mean_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_median = foreign "atg_median" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_median_dim =
+    foreign
+      "atg_median_dim"
+      (ptr raw_tensor @-> gc_tensor @-> int64_t @-> int @-> returning void)
+  ;;
+
+  let stubs_median_dim_values =
+    foreign
+      "atg_median_dim_values"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_median_out =
+    foreign "atg_median_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_meshgrid =
+    foreign "atg_meshgrid" (ptr gc_tensor @-> int @-> returning (ptr raw_tensor))
+  ;;
+
+  let stubs_meshgrid_indexing =
+    foreign
+      "atg_meshgrid_indexing"
+      (ptr gc_tensor @-> int @-> string @-> returning (ptr raw_tensor))
+  ;;
+
+  let stubs_mh = foreign "atg_mh" (gc_tensor @-> returning raw_tensor)
+  let stubs_min = foreign "atg_min" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_min_dim =
+    foreign
+      "atg_min_dim"
+      (ptr raw_tensor @-> gc_tensor @-> int64_t @-> int @-> returning void)
+  ;;
+
+  let stubs_min_dim_min =
+    foreign
+      "atg_min_dim_min"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_min_other =
+    foreign "atg_min_other" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_min_out =
+    foreign "atg_min_out" (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_min_unary_out =
+    foreign "atg_min_unary_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_minimum =
+    foreign "atg_minimum" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_minimum_out =
+    foreign
+      "atg_minimum_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_miopen_batch_norm =
+    foreign
+      "atg_miopen_batch_norm"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> double
+       @-> double
+       @-> returning void)
+  ;;
+
+  let stubs_miopen_batch_norm_backward =
+    foreign
+      "atg_miopen_batch_norm_backward"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> returning void)
+  ;;
+
+  let stubs_miopen_batch_norm_backward_out =
+    foreign
+      "atg_miopen_batch_norm_backward_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> returning void)
+  ;;
+
+  let stubs_miopen_batch_norm_out =
+    foreign
+      "atg_miopen_batch_norm_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> double
+       @-> double
+       @-> returning void)
+  ;;
+
+  let stubs_miopen_convolution =
+    foreign
+      "atg_miopen_convolution"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_miopen_convolution_add_relu =
+    foreign
+      "atg_miopen_convolution_add_relu"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> scalar
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_miopen_convolution_out =
+    foreign
+      "atg_miopen_convolution_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_miopen_convolution_relu =
+    foreign
+      "atg_miopen_convolution_relu"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_miopen_convolution_transpose =
+    foreign
+      "atg_miopen_convolution_transpose"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_miopen_convolution_transpose_out =
+    foreign
+      "atg_miopen_convolution_transpose_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_miopen_depthwise_convolution =
+    foreign
+      "atg_miopen_depthwise_convolution"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_miopen_depthwise_convolution_out =
+    foreign
+      "atg_miopen_depthwise_convolution_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_miopen_rnn =
+    foreign
+      "atg_miopen_rnn"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> ptr gc_tensor
+       @-> int
+       @-> int64_t
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> int
+       @-> double
+       @-> int
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> gc_tensor
+       @-> returning void)
+  ;;
+end
+
+module C16 (F : Cstubs.FOREIGN) = struct
+  open F
+  open Type_defs
+
+  let stubs_miopen_rnn_out =
+    foreign
+      "atg_miopen_rnn_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr gc_tensor
+       @-> int
+       @-> int64_t
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> int
+       @-> double
+       @-> int
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> gc_tensor
+       @-> returning void)
+  ;;
+
+  let stubs_mish = foreign "atg_mish" (gc_tensor @-> returning raw_tensor)
+  let stubs_mish_ = foreign "atg_mish_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_mish_backward =
+    foreign "atg_mish_backward" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_mish_out =
+    foreign "atg_mish_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_mkldnn_adaptive_avg_pool2d =
+    foreign
+      "atg_mkldnn_adaptive_avg_pool2d"
+      (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_mkldnn_adaptive_avg_pool2d_backward =
+    foreign
+      "atg_mkldnn_adaptive_avg_pool2d_backward"
+      (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_mkldnn_adaptive_avg_pool2d_backward_out =
+    foreign
+      "atg_mkldnn_adaptive_avg_pool2d_backward_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_mkldnn_adaptive_avg_pool2d_out =
+    foreign
+      "atg_mkldnn_adaptive_avg_pool2d_out"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_mkldnn_convolution =
+    foreign
+      "atg_mkldnn_convolution"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_mkldnn_convolution_out =
+    foreign
+      "atg_mkldnn_convolution_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_mkldnn_linear =
+    foreign
+      "atg_mkldnn_linear"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_mkldnn_linear_backward_input =
+    foreign
+      "atg_mkldnn_linear_backward_input"
+      (ptr int64_t @-> int @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_mkldnn_linear_backward_input_out =
+    foreign
+      "atg_mkldnn_linear_backward_input_out"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_mkldnn_linear_backward_weights =
+    foreign
+      "atg_mkldnn_linear_backward_weights"
+      (ptr raw_tensor @-> gc_tensor @-> gc_tensor @-> gc_tensor @-> int @-> returning void)
+  ;;
+
+  let stubs_mkldnn_linear_backward_weights_out =
+    foreign
+      "atg_mkldnn_linear_backward_weights_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_mkldnn_linear_out =
+    foreign
+      "atg_mkldnn_linear_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_mkldnn_max_pool2d =
+    foreign
+      "atg_mkldnn_max_pool2d"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_mkldnn_max_pool2d_backward =
+    foreign
+      "atg_mkldnn_max_pool2d_backward"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_mkldnn_max_pool2d_backward_out =
+    foreign
+      "atg_mkldnn_max_pool2d_backward_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_mkldnn_max_pool2d_out =
+    foreign
+      "atg_mkldnn_max_pool2d_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_mkldnn_max_pool3d =
+    foreign
+      "atg_mkldnn_max_pool3d"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_mkldnn_max_pool3d_backward =
+    foreign
+      "atg_mkldnn_max_pool3d_backward"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_mkldnn_max_pool3d_backward_out =
+    foreign
+      "atg_mkldnn_max_pool3d_backward_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_mkldnn_max_pool3d_out =
+    foreign
+      "atg_mkldnn_max_pool3d_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_mkldnn_reorder_conv2d_weight =
+    foreign
+      "atg_mkldnn_reorder_conv2d_weight"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> ptr int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_mkldnn_reorder_conv2d_weight_out =
+    foreign
+      "atg_mkldnn_reorder_conv2d_weight_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> ptr int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_mkldnn_reorder_conv3d_weight =
+    foreign
+      "atg_mkldnn_reorder_conv3d_weight"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_mkldnn_reorder_conv3d_weight_out =
+    foreign
+      "atg_mkldnn_reorder_conv3d_weight_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_mkldnn_rnn_layer =
+    foreign
+      "atg_mkldnn_rnn_layer"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_mkldnn_rnn_layer_backward =
+    foreign
+      "atg_mkldnn_rnn_layer_backward"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> gc_tensor
+       @-> returning void)
+  ;;
+
+  let stubs_mkldnn_rnn_layer_backward_out =
+    foreign
+      "atg_mkldnn_rnn_layer_backward_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> gc_tensor
+       @-> returning void)
+  ;;
+
+  let stubs_mkldnn_rnn_layer_out =
+    foreign
+      "atg_mkldnn_rnn_layer_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_mm = foreign "atg_mm" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+
+  let stubs_mm_out =
+    foreign "atg_mm_out" (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_mode =
+    foreign
+      "atg_mode"
+      (ptr raw_tensor @-> gc_tensor @-> int64_t @-> int @-> returning void)
+  ;;
+
+  let stubs_mode_values =
+    foreign
+      "atg_mode_values"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_moveaxis =
+    foreign
+      "atg_moveaxis"
+      (gc_tensor @-> ptr int64_t @-> int @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_moveaxis_int =
+    foreign "atg_moveaxis_int" (gc_tensor @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_movedim =
+    foreign
+      "atg_movedim"
+      (gc_tensor @-> ptr int64_t @-> int @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_movedim_int =
+    foreign "atg_movedim_int" (gc_tensor @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_mse_loss =
+    foreign "atg_mse_loss" (gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_mse_loss_backward =
+    foreign
+      "atg_mse_loss_backward"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_mse_loss_backward_grad_input =
+    foreign
+      "atg_mse_loss_backward_grad_input"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_mse_loss_out =
+    foreign
+      "atg_mse_loss_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_msort = foreign "atg_msort" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_msort_out =
+    foreign "atg_msort_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_mt = foreign "atg_mt" (gc_tensor @-> returning raw_tensor)
+  let stubs_mul = foreign "atg_mul" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  let stubs_mul_ = foreign "atg_mul_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+
+  let stubs_mul_out =
+    foreign "atg_mul_out" (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_mul_scalar =
+    foreign "atg_mul_scalar" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_mul_scalar_ =
+    foreign "atg_mul_scalar_" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_mul_scalar_out =
+    foreign
+      "atg_mul_scalar_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_multi_margin_loss_backward =
+    foreign
+      "atg_multi_margin_loss_backward"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> scalar
+       @-> scalar
+       @-> gc_tensor
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_multi_margin_loss_backward_grad_input =
+    foreign
+      "atg_multi_margin_loss_backward_grad_input"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> scalar
+       @-> scalar
+       @-> gc_tensor
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_multilabel_margin_loss =
+    foreign
+      "atg_multilabel_margin_loss"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_multilabel_margin_loss_backward =
+    foreign
+      "atg_multilabel_margin_loss_backward"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_multilabel_margin_loss_backward_grad_input =
+    foreign
+      "atg_multilabel_margin_loss_backward_grad_input"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_multilabel_margin_loss_out =
+    foreign
+      "atg_multilabel_margin_loss_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_multinomial =
+    foreign "atg_multinomial" (gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_multinomial_out =
+    foreign
+      "atg_multinomial_out"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_multiply =
+    foreign "atg_multiply" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_multiply_ =
+    foreign "atg_multiply_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_multiply_out =
+    foreign
+      "atg_multiply_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_multiply_scalar =
+    foreign "atg_multiply_scalar" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_multiply_scalar_ =
+    foreign "atg_multiply_scalar_" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_mv = foreign "atg_mv" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+
+  let stubs_mv_out =
+    foreign "atg_mv_out" (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_mvlgamma =
+    foreign "atg_mvlgamma" (gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_mvlgamma_ =
+    foreign "atg_mvlgamma_" (gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_mvlgamma_out =
+    foreign
+      "atg_mvlgamma_out"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_nan_to_num =
+    foreign
+      "atg_nan_to_num"
+      (gc_tensor
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_nan_to_num_ =
+    foreign
+      "atg_nan_to_num_"
+      (gc_tensor
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_nan_to_num_out =
+    foreign
+      "atg_nan_to_num_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_nanmean =
+    foreign
+      "atg_nanmean"
+      (gc_tensor @-> ptr int64_t @-> int @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_nanmean_out =
+    foreign
+      "atg_nanmean_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_nanmedian = foreign "atg_nanmedian" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_nanmedian_dim =
+    foreign
+      "atg_nanmedian_dim"
+      (ptr raw_tensor @-> gc_tensor @-> int64_t @-> int @-> returning void)
+  ;;
+
+  let stubs_nanmedian_dim_values =
+    foreign
+      "atg_nanmedian_dim_values"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_nanmedian_out =
+    foreign "atg_nanmedian_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_nanquantile =
+    foreign
+      "atg_nanquantile"
+      (gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int
+       @-> int
+       @-> string
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_nanquantile_out =
+    foreign
+      "atg_nanquantile_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int
+       @-> int
+       @-> string
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_nanquantile_scalar =
+    foreign
+      "atg_nanquantile_scalar"
+      (gc_tensor
+       @-> double
+       @-> int64_t
+       @-> int
+       @-> int
+       @-> string
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_nanquantile_scalar_out =
+    foreign
+      "atg_nanquantile_scalar_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> int64_t
+       @-> int
+       @-> int
+       @-> string
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_nansum =
+    foreign
+      "atg_nansum"
+      (gc_tensor @-> ptr int64_t @-> int @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_nansum_out =
+    foreign
+      "atg_nansum_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_narrow =
+    foreign
+      "atg_narrow"
+      (gc_tensor @-> int64_t @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_narrow_copy =
+    foreign
+      "atg_narrow_copy"
+      (gc_tensor @-> int64_t @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_narrow_copy_out =
+    foreign
+      "atg_narrow_copy_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_narrow_tensor =
+    foreign
+      "atg_narrow_tensor"
+      (gc_tensor @-> int64_t @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_native_batch_norm =
+    foreign
+      "atg_native_batch_norm"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> double
+       @-> double
+       @-> returning void)
+  ;;
+
+  let stubs_native_batch_norm_out =
+    foreign
+      "atg_native_batch_norm_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> double
+       @-> double
+       @-> returning void)
+  ;;
+
+  let stubs_native_channel_shuffle =
+    foreign "atg_native_channel_shuffle" (gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_native_dropout =
+    foreign
+      "atg_native_dropout"
+      (ptr raw_tensor @-> gc_tensor @-> double @-> int @-> returning void)
+  ;;
+
+  let stubs_native_dropout_backward =
+    foreign
+      "atg_native_dropout_backward"
+      (gc_tensor @-> gc_tensor @-> double @-> returning raw_tensor)
+  ;;
+
+  let stubs_native_dropout_backward_out =
+    foreign
+      "atg_native_dropout_backward_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> double @-> returning raw_tensor)
+  ;;
+
+  let stubs_native_dropout_out =
+    foreign
+      "atg_native_dropout_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_native_group_norm =
+    foreign
+      "atg_native_group_norm"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> double
+       @-> returning void)
+  ;;
+
+  let stubs_native_group_norm_out =
+    foreign
+      "atg_native_group_norm_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> double
+       @-> returning void)
+  ;;
+end
+
+module C17 (F : Cstubs.FOREIGN) = struct
+  open F
+  open Type_defs
+
+  let stubs_native_layer_norm =
+    foreign
+      "atg_native_layer_norm"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> returning void)
+  ;;
+
+  let stubs_native_layer_norm_out =
+    foreign
+      "atg_native_layer_norm_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> returning void)
+  ;;
+
+  let stubs_native_norm = foreign "atg_native_norm" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_native_norm_out =
+    foreign "atg_native_norm_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_native_norm_scalaropt_dim_dtype =
+    foreign
+      "atg_native_norm_scalaropt_dim_dtype"
+      (gc_tensor
+       @-> scalar
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_native_norm_scalaropt_dim_dtype_out =
+    foreign
+      "atg_native_norm_scalaropt_dim_dtype_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> scalar
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_ne = foreign "atg_ne" (gc_tensor @-> scalar @-> returning raw_tensor)
+  let stubs_ne_ = foreign "atg_ne_" (gc_tensor @-> scalar @-> returning raw_tensor)
+
+  let stubs_ne_scalar_out =
+    foreign
+      "atg_ne_scalar_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_ne_tensor =
+    foreign "atg_ne_tensor" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_ne_tensor_ =
+    foreign "atg_ne_tensor_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_ne_tensor_out =
+    foreign
+      "atg_ne_tensor_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_neg = foreign "atg_neg" (gc_tensor @-> returning raw_tensor)
+  let stubs_neg_ = foreign "atg_neg_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_neg_out =
+    foreign "atg_neg_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_negative = foreign "atg_negative" (gc_tensor @-> returning raw_tensor)
+  let stubs_negative_ = foreign "atg_negative_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_negative_out =
+    foreign "atg_negative_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_nested_to_padded_tensor =
+    foreign
+      "atg_nested_to_padded_tensor"
+      (gc_tensor @-> double @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_new_empty =
+    foreign
+      "atg_new_empty"
+      (gc_tensor @-> ptr int64_t @-> int @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_new_empty_out =
+    foreign
+      "atg_new_empty_out"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_new_empty_strided =
+    foreign
+      "atg_new_empty_strided"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_new_empty_strided_out =
+    foreign
+      "atg_new_empty_strided_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_new_full =
+    foreign
+      "atg_new_full"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> scalar
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_new_full_out =
+    foreign
+      "atg_new_full_out"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_new_ones =
+    foreign
+      "atg_new_ones"
+      (gc_tensor @-> ptr int64_t @-> int @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_new_ones_out =
+    foreign
+      "atg_new_ones_out"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_new_zeros =
+    foreign
+      "atg_new_zeros"
+      (gc_tensor @-> ptr int64_t @-> int @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_new_zeros_out =
+    foreign
+      "atg_new_zeros_out"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_nextafter =
+    foreign "atg_nextafter" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_nextafter_ =
+    foreign "atg_nextafter_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_nextafter_out =
+    foreign
+      "atg_nextafter_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_nll_loss =
+    foreign
+      "atg_nll_loss"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_nll_loss2d =
+    foreign
+      "atg_nll_loss2d"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_nll_loss2d_backward =
+    foreign
+      "atg_nll_loss2d_backward"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_nll_loss2d_backward_grad_input =
+    foreign
+      "atg_nll_loss2d_backward_grad_input"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_nll_loss2d_out =
+    foreign
+      "atg_nll_loss2d_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_nll_loss_backward =
+    foreign
+      "atg_nll_loss_backward"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_nll_loss_backward_grad_input =
+    foreign
+      "atg_nll_loss_backward_grad_input"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_nll_loss_nd =
+    foreign
+      "atg_nll_loss_nd"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_nll_loss_out =
+    foreign
+      "atg_nll_loss_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_nonzero = foreign "atg_nonzero" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_nonzero_numpy =
+    foreign "atg_nonzero_numpy" (gc_tensor @-> returning (ptr raw_tensor))
+  ;;
+
+  let stubs_nonzero_out =
+    foreign "atg_nonzero_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_nonzero_static =
+    foreign
+      "atg_nonzero_static"
+      (gc_tensor @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_nonzero_static_out =
+    foreign
+      "atg_nonzero_static_out"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_norm = foreign "atg_norm" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_norm_dtype_out =
+    foreign
+      "atg_norm_dtype_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> scalar
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_norm_except_dim =
+    foreign
+      "atg_norm_except_dim"
+      (gc_tensor @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_norm_out =
+    foreign
+      "atg_norm_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> scalar
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_norm_scalar_out =
+    foreign "atg_norm_scalar_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_norm_scalaropt_dim =
+    foreign
+      "atg_norm_scalaropt_dim"
+      (gc_tensor @-> scalar @-> ptr int64_t @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_norm_scalaropt_dim_dtype =
+    foreign
+      "atg_norm_scalaropt_dim_dtype"
+      (gc_tensor
+       @-> scalar
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_norm_scalaropt_dtype =
+    foreign
+      "atg_norm_scalaropt_dtype"
+      (gc_tensor @-> scalar @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_norm_scalaropt_dtype_out =
+    foreign
+      "atg_norm_scalaropt_dtype_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_normal_ =
+    foreign "atg_normal_" (gc_tensor @-> double @-> double @-> returning raw_tensor)
+  ;;
+
+  let stubs_normal_functional =
+    foreign
+      "atg_normal_functional"
+      (gc_tensor @-> double @-> double @-> returning raw_tensor)
+  ;;
+
+  let stubs_not_equal =
+    foreign "atg_not_equal" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_not_equal_ =
+    foreign "atg_not_equal_" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_not_equal_scalar_out =
+    foreign
+      "atg_not_equal_scalar_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_not_equal_tensor =
+    foreign "atg_not_equal_tensor" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_not_equal_tensor_ =
+    foreign "atg_not_equal_tensor_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_not_equal_tensor_out =
+    foreign
+      "atg_not_equal_tensor_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_nuclear_norm =
+    foreign "atg_nuclear_norm" (gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_nuclear_norm_dim =
+    foreign
+      "atg_nuclear_norm_dim"
+      (gc_tensor @-> ptr int64_t @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_nuclear_norm_dim_out =
+    foreign
+      "atg_nuclear_norm_dim_out"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_nuclear_norm_out =
+    foreign
+      "atg_nuclear_norm_out"
+      (gc_tensor @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_numpy_t = foreign "atg_numpy_t" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_one_hot =
+    foreign "atg_one_hot" (gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_ones =
+    foreign "atg_ones" (ptr int64_t @-> int @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_ones_like = foreign "atg_ones_like" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_ones_like_out =
+    foreign "atg_ones_like_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_ones_out =
+    foreign "atg_ones_out" (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_orgqr = foreign "atg_orgqr" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+
+  let stubs_orgqr_out =
+    foreign
+      "atg_orgqr_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_ormqr =
+    foreign
+      "atg_ormqr"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_ormqr_out =
+    foreign
+      "atg_ormqr_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_outer = foreign "atg_outer" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+
+  let stubs_outer_out =
+    foreign
+      "atg_outer_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_output_nr = foreign "atg_output_nr" (gc_tensor @-> returning int64_t)
+
+  let stubs_pad =
+    foreign
+      "atg_pad"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> string
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_pad_sequence =
+    foreign
+      "atg_pad_sequence"
+      (ptr gc_tensor @-> int @-> int @-> double @-> returning raw_tensor)
+  ;;
+
+  let stubs_pairwise_distance =
+    foreign
+      "atg_pairwise_distance"
+      (gc_tensor @-> gc_tensor @-> double @-> double @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_pdist = foreign "atg_pdist" (gc_tensor @-> double @-> returning raw_tensor)
+
+  let stubs_permute =
+    foreign "atg_permute" (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_permute_copy =
+    foreign "atg_permute_copy" (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_permute_copy_out =
+    foreign
+      "atg_permute_copy_out"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_pin_memory =
+    foreign "atg_pin_memory" (gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_pinverse =
+    foreign "atg_pinverse" (gc_tensor @-> double @-> returning raw_tensor)
+  ;;
+
+  let stubs_pixel_shuffle =
+    foreign "atg_pixel_shuffle" (gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_pixel_shuffle_out =
+    foreign
+      "atg_pixel_shuffle_out"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_pixel_unshuffle =
+    foreign "atg_pixel_unshuffle" (gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_pixel_unshuffle_out =
+    foreign
+      "atg_pixel_unshuffle_out"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_poisson = foreign "atg_poisson" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_poisson_nll_loss =
+    foreign
+      "atg_poisson_nll_loss"
+      (gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> int
+       @-> double
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_poisson_out =
+    foreign "atg_poisson_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_polar = foreign "atg_polar" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+
+  let stubs_polar_out =
+    foreign
+      "atg_polar_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_polygamma =
+    foreign "atg_polygamma" (int64_t @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_polygamma_ =
+    foreign "atg_polygamma_" (gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+end
+
+module C18 (F : Cstubs.FOREIGN) = struct
+  open F
+  open Type_defs
+
+  let stubs_polygamma_out =
+    foreign
+      "atg_polygamma_out"
+      (gc_tensor @-> int64_t @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_positive = foreign "atg_positive" (gc_tensor @-> returning raw_tensor)
+  let stubs_pow = foreign "atg_pow" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  let stubs_pow_ = foreign "atg_pow_" (gc_tensor @-> scalar @-> returning raw_tensor)
+
+  let stubs_pow_scalar =
+    foreign "atg_pow_scalar" (scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_pow_scalar_out =
+    foreign
+      "atg_pow_scalar_out"
+      (gc_tensor @-> scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_pow_tensor_ =
+    foreign "atg_pow_tensor_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_pow_tensor_scalar =
+    foreign "atg_pow_tensor_scalar" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_pow_tensor_scalar_out =
+    foreign
+      "atg_pow_tensor_scalar_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_pow_tensor_tensor_out =
+    foreign
+      "atg_pow_tensor_tensor_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_prelu = foreign "atg_prelu" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  let stubs_prod = foreign "atg_prod" (gc_tensor @-> int @-> returning raw_tensor)
+
+  let stubs_prod_dim_int =
+    foreign
+      "atg_prod_dim_int"
+      (gc_tensor @-> int64_t @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_prod_int_out =
+    foreign
+      "atg_prod_int_out"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_prod_out =
+    foreign "atg_prod_out" (gc_tensor @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_put =
+    foreign
+      "atg_put"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_put_ =
+    foreign
+      "atg_put_"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_put_out =
+    foreign
+      "atg_put_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_q_per_channel_axis =
+    foreign "atg_q_per_channel_axis" (gc_tensor @-> returning int64_t)
+  ;;
+
+  let stubs_q_per_channel_scales =
+    foreign "atg_q_per_channel_scales" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_q_per_channel_scales_out =
+    foreign
+      "atg_q_per_channel_scales_out"
+      (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_q_per_channel_zero_points =
+    foreign "atg_q_per_channel_zero_points" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_q_per_channel_zero_points_out =
+    foreign
+      "atg_q_per_channel_zero_points_out"
+      (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_q_scale = foreign "atg_q_scale" (gc_tensor @-> returning double)
+  let stubs_q_zero_point = foreign "atg_q_zero_point" (gc_tensor @-> returning int64_t)
+  let stubs_qr = foreign "atg_qr" (ptr raw_tensor @-> gc_tensor @-> int @-> returning void)
+
+  let stubs_qr_q =
+    foreign
+      "atg_qr_q"
+      (ptr raw_tensor @-> gc_tensor @-> gc_tensor @-> gc_tensor @-> int @-> returning void)
+  ;;
+
+  let stubs_quantile =
+    foreign
+      "atg_quantile"
+      (gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int
+       @-> int
+       @-> string
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_quantile_out =
+    foreign
+      "atg_quantile_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int
+       @-> int
+       @-> string
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_quantile_scalar =
+    foreign
+      "atg_quantile_scalar"
+      (gc_tensor
+       @-> double
+       @-> int64_t
+       @-> int
+       @-> int
+       @-> string
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_quantile_scalar_out =
+    foreign
+      "atg_quantile_scalar_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> int64_t
+       @-> int
+       @-> int
+       @-> string
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_quantize_per_channel =
+    foreign
+      "atg_quantize_per_channel"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_quantize_per_channel_out =
+    foreign
+      "atg_quantize_per_channel_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_quantize_per_tensor =
+    foreign
+      "atg_quantize_per_tensor"
+      (gc_tensor @-> double @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_quantize_per_tensor_dynamic =
+    foreign
+      "atg_quantize_per_tensor_dynamic"
+      (gc_tensor @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_quantize_per_tensor_dynamic_out =
+    foreign
+      "atg_quantize_per_tensor_dynamic_out"
+      (gc_tensor @-> gc_tensor @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_quantize_per_tensor_out =
+    foreign
+      "atg_quantize_per_tensor_out"
+      (gc_tensor @-> gc_tensor @-> double @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_quantize_per_tensor_tensor_qparams =
+    foreign
+      "atg_quantize_per_tensor_tensor_qparams"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_quantize_per_tensor_tensor_qparams_out =
+    foreign
+      "atg_quantize_per_tensor_tensor_qparams_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_quantize_per_tensor_tensors =
+    foreign
+      "atg_quantize_per_tensor_tensors"
+      (ptr gc_tensor
+       @-> int
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> returning (ptr raw_tensor))
+  ;;
+
+  let stubs_quantize_per_tensor_tensors_out =
+    foreign
+      "atg_quantize_per_tensor_tensors_out"
+      (ptr gc_tensor
+       @-> int
+       @-> ptr gc_tensor
+       @-> int
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_quantized_batch_norm =
+    foreign
+      "atg_quantized_batch_norm"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> double
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_quantized_batch_norm_out =
+    foreign
+      "atg_quantized_batch_norm_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> double
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_quantized_gru_cell =
+    foreign
+      "atg_quantized_gru_cell"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> scalar
+       @-> scalar
+       @-> scalar
+       @-> scalar
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_quantized_lstm_cell =
+    foreign
+      "atg_quantized_lstm_cell"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> ptr gc_tensor
+       @-> int
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> scalar
+       @-> scalar
+       @-> scalar
+       @-> scalar
+       @-> returning void)
+  ;;
+
+  let stubs_quantized_max_pool1d =
+    foreign
+      "atg_quantized_max_pool1d"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_quantized_max_pool1d_out =
+    foreign
+      "atg_quantized_max_pool1d_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_quantized_max_pool2d =
+    foreign
+      "atg_quantized_max_pool2d"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_quantized_max_pool2d_out =
+    foreign
+      "atg_quantized_max_pool2d_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_quantized_max_pool3d =
+    foreign
+      "atg_quantized_max_pool3d"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_quantized_max_pool3d_out =
+    foreign
+      "atg_quantized_max_pool3d_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_quantized_rnn_relu_cell =
+    foreign
+      "atg_quantized_rnn_relu_cell"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> scalar
+       @-> scalar
+       @-> scalar
+       @-> scalar
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_quantized_rnn_tanh_cell =
+    foreign
+      "atg_quantized_rnn_tanh_cell"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> scalar
+       @-> scalar
+       @-> scalar
+       @-> scalar
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_rad2deg = foreign "atg_rad2deg" (gc_tensor @-> returning raw_tensor)
+  let stubs_rad2deg_ = foreign "atg_rad2deg_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_rad2deg_out =
+    foreign "atg_rad2deg_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_rand =
+    foreign "atg_rand" (ptr int64_t @-> int @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_rand_like = foreign "atg_rand_like" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_rand_like_out =
+    foreign "atg_rand_like_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_rand_out =
+    foreign "atg_rand_out" (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_randint =
+    foreign
+      "atg_randint"
+      (int64_t @-> ptr int64_t @-> int @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_randint_like =
+    foreign "atg_randint_like" (gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_randint_like_low_dtype =
+    foreign
+      "atg_randint_like_low_dtype"
+      (gc_tensor @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_randint_like_low_dtype_out =
+    foreign
+      "atg_randint_like_low_dtype_out"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_randint_like_out =
+    foreign
+      "atg_randint_like_out"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_randint_low =
+    foreign
+      "atg_randint_low"
+      (int64_t
+       @-> int64_t
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_randint_low_out =
+    foreign
+      "atg_randint_low_out"
+      (gc_tensor @-> int64_t @-> int64_t @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_randint_out =
+    foreign
+      "atg_randint_out"
+      (gc_tensor @-> int64_t @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_randn =
+    foreign "atg_randn" (ptr int64_t @-> int @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_randn_like = foreign "atg_randn_like" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_randn_like_out =
+    foreign "atg_randn_like_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_randn_out =
+    foreign "atg_randn_out" (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_random = foreign "atg_random" (gc_tensor @-> returning raw_tensor)
+  let stubs_random_ = foreign "atg_random_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_random_from =
+    foreign
+      "atg_random_from"
+      (gc_tensor @-> int64_t @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_random_from_ =
+    foreign
+      "atg_random_from_"
+      (gc_tensor @-> int64_t @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_random_from_out =
+    foreign
+      "atg_random_from_out"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_random_out =
+    foreign "atg_random_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_random_to =
+    foreign "atg_random_to" (gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_random_to_ =
+    foreign "atg_random_to_" (gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_random_to_out =
+    foreign
+      "atg_random_to_out"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_randperm =
+    foreign "atg_randperm" (int64_t @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_randperm_out =
+    foreign "atg_randperm_out" (gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_range =
+    foreign "atg_range" (scalar @-> scalar @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_range_out =
+    foreign "atg_range_out" (gc_tensor @-> scalar @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_range_out_ =
+    foreign "atg_range_out_" (gc_tensor @-> scalar @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_range_step =
+    foreign "atg_range_step" (scalar @-> scalar @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_ravel = foreign "atg_ravel" (gc_tensor @-> returning raw_tensor)
+  let stubs_real = foreign "atg_real" (gc_tensor @-> returning raw_tensor)
+  let stubs_reciprocal = foreign "atg_reciprocal" (gc_tensor @-> returning raw_tensor)
+  let stubs_reciprocal_ = foreign "atg_reciprocal_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_reciprocal_out =
+    foreign "atg_reciprocal_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_reflection_pad1d =
+    foreign
+      "atg_reflection_pad1d"
+      (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_reflection_pad1d_backward =
+    foreign
+      "atg_reflection_pad1d_backward"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_reflection_pad1d_backward_grad_input =
+    foreign
+      "atg_reflection_pad1d_backward_grad_input"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_reflection_pad1d_out =
+    foreign
+      "atg_reflection_pad1d_out"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_reflection_pad2d =
+    foreign
+      "atg_reflection_pad2d"
+      (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_reflection_pad2d_backward =
+    foreign
+      "atg_reflection_pad2d_backward"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_reflection_pad2d_backward_grad_input =
+    foreign
+      "atg_reflection_pad2d_backward_grad_input"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_reflection_pad2d_out =
+    foreign
+      "atg_reflection_pad2d_out"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+end
+
+module C19 (F : Cstubs.FOREIGN) = struct
+  open F
+  open Type_defs
+
+  let stubs_reflection_pad3d =
+    foreign
+      "atg_reflection_pad3d"
+      (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_reflection_pad3d_backward =
+    foreign
+      "atg_reflection_pad3d_backward"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_reflection_pad3d_backward_grad_input =
+    foreign
+      "atg_reflection_pad3d_backward_grad_input"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_reflection_pad3d_out =
+    foreign
+      "atg_reflection_pad3d_out"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_relu = foreign "atg_relu" (gc_tensor @-> returning raw_tensor)
+  let stubs_relu6 = foreign "atg_relu6" (gc_tensor @-> returning raw_tensor)
+  let stubs_relu6_ = foreign "atg_relu6_" (gc_tensor @-> returning raw_tensor)
+  let stubs_relu_ = foreign "atg_relu_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_relu_out =
+    foreign "atg_relu_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_remainder =
+    foreign "atg_remainder" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_remainder_ =
+    foreign "atg_remainder_" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_remainder_scalar_out =
+    foreign
+      "atg_remainder_scalar_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_remainder_scalar_tensor =
+    foreign "atg_remainder_scalar_tensor" (scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_remainder_scalar_tensor_out =
+    foreign
+      "atg_remainder_scalar_tensor_out"
+      (gc_tensor @-> scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_remainder_tensor =
+    foreign "atg_remainder_tensor" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_remainder_tensor_ =
+    foreign "atg_remainder_tensor_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_remainder_tensor_out =
+    foreign
+      "atg_remainder_tensor_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_renorm =
+    foreign
+      "atg_renorm"
+      (gc_tensor @-> scalar @-> int64_t @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_renorm_ =
+    foreign
+      "atg_renorm_"
+      (gc_tensor @-> scalar @-> int64_t @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_renorm_out =
+    foreign
+      "atg_renorm_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> int64_t @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_repeat =
+    foreign "atg_repeat" (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_repeat_interleave =
+    foreign
+      "atg_repeat_interleave"
+      (gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_repeat_interleave_self_int =
+    foreign
+      "atg_repeat_interleave_self_int"
+      (gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_repeat_interleave_self_tensor =
+    foreign
+      "atg_repeat_interleave_self_tensor"
+      (gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_repeat_interleave_tensor_out =
+    foreign
+      "atg_repeat_interleave_tensor_out"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_repeat_out =
+    foreign
+      "atg_repeat_out"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_replication_pad1d =
+    foreign
+      "atg_replication_pad1d"
+      (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_replication_pad1d_backward =
+    foreign
+      "atg_replication_pad1d_backward"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_replication_pad1d_backward_grad_input =
+    foreign
+      "atg_replication_pad1d_backward_grad_input"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_replication_pad1d_out =
+    foreign
+      "atg_replication_pad1d_out"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_replication_pad2d =
+    foreign
+      "atg_replication_pad2d"
+      (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_replication_pad2d_backward =
+    foreign
+      "atg_replication_pad2d_backward"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_replication_pad2d_backward_grad_input =
+    foreign
+      "atg_replication_pad2d_backward_grad_input"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_replication_pad2d_out =
+    foreign
+      "atg_replication_pad2d_out"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_replication_pad3d =
+    foreign
+      "atg_replication_pad3d"
+      (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_replication_pad3d_backward =
+    foreign
+      "atg_replication_pad3d_backward"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_replication_pad3d_backward_grad_input =
+    foreign
+      "atg_replication_pad3d_backward_grad_input"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_replication_pad3d_out =
+    foreign
+      "atg_replication_pad3d_out"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_requires_grad_ =
+    foreign "atg_requires_grad_" (gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_reshape =
+    foreign "atg_reshape" (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_reshape_as =
+    foreign "atg_reshape_as" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_resize =
+    foreign "atg_resize" (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_resize_ =
+    foreign "atg_resize_" (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_resize_as =
+    foreign "atg_resize_as" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_resize_as_ =
+    foreign "atg_resize_as_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_resize_as_out =
+    foreign
+      "atg_resize_as_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_resize_as_sparse =
+    foreign "atg_resize_as_sparse" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_resize_as_sparse_ =
+    foreign "atg_resize_as_sparse_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_resize_as_sparse_out =
+    foreign
+      "atg_resize_as_sparse_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_resize_out =
+    foreign
+      "atg_resize_out"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_resolve_conj = foreign "atg_resolve_conj" (gc_tensor @-> returning raw_tensor)
+  let stubs_resolve_neg = foreign "atg_resolve_neg" (gc_tensor @-> returning raw_tensor)
+  let stubs_retains_grad = foreign "atg_retains_grad" (gc_tensor @-> returning bool)
+
+  let stubs_rnn_relu =
+    foreign
+      "atg_rnn_relu"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr gc_tensor
+       @-> int
+       @-> int
+       @-> int64_t
+       @-> double
+       @-> int
+       @-> int
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_rnn_relu_cell =
+    foreign
+      "atg_rnn_relu_cell"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_rnn_relu_data =
+    foreign
+      "atg_rnn_relu_data"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr gc_tensor
+       @-> int
+       @-> int
+       @-> int64_t
+       @-> double
+       @-> int
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_rnn_tanh =
+    foreign
+      "atg_rnn_tanh"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr gc_tensor
+       @-> int
+       @-> int
+       @-> int64_t
+       @-> double
+       @-> int
+       @-> int
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_rnn_tanh_cell =
+    foreign
+      "atg_rnn_tanh_cell"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_rnn_tanh_data =
+    foreign
+      "atg_rnn_tanh_data"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr gc_tensor
+       @-> int
+       @-> int
+       @-> int64_t
+       @-> double
+       @-> int
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_roll =
+    foreign
+      "atg_roll"
+      (gc_tensor @-> ptr int64_t @-> int @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_roll_out =
+    foreign
+      "atg_roll_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_rot90 =
+    foreign
+      "atg_rot90"
+      (gc_tensor @-> int64_t @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_rot90_out =
+    foreign
+      "atg_rot90_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> ptr int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_round = foreign "atg_round" (gc_tensor @-> returning raw_tensor)
+  let stubs_round_ = foreign "atg_round_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_round_decimals =
+    foreign "atg_round_decimals" (gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_round_decimals_ =
+    foreign "atg_round_decimals_" (gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_round_decimals_out =
+    foreign
+      "atg_round_decimals_out"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_round_out =
+    foreign "atg_round_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_row_indices = foreign "atg_row_indices" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_row_indices_copy =
+    foreign "atg_row_indices_copy" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_row_indices_copy_out =
+    foreign "atg_row_indices_copy_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_row_stack =
+    foreign "atg_row_stack" (ptr gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_row_stack_out =
+    foreign
+      "atg_row_stack_out"
+      (gc_tensor @-> ptr gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_rrelu = foreign "atg_rrelu" (gc_tensor @-> int @-> returning raw_tensor)
+  let stubs_rrelu_ = foreign "atg_rrelu_" (gc_tensor @-> int @-> returning raw_tensor)
+
+  let stubs_rrelu_with_noise =
+    foreign
+      "atg_rrelu_with_noise"
+      (gc_tensor @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_rrelu_with_noise_ =
+    foreign
+      "atg_rrelu_with_noise_"
+      (gc_tensor @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_rrelu_with_noise_backward =
+    foreign
+      "atg_rrelu_with_noise_backward"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> scalar
+       @-> scalar
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_rrelu_with_noise_backward_out =
+    foreign
+      "atg_rrelu_with_noise_backward_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> scalar
+       @-> scalar
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_rrelu_with_noise_out =
+    foreign
+      "atg_rrelu_with_noise_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_rsqrt = foreign "atg_rsqrt" (gc_tensor @-> returning raw_tensor)
+  let stubs_rsqrt_ = foreign "atg_rsqrt_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_rsqrt_out =
+    foreign "atg_rsqrt_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_rsub = foreign "atg_rsub" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+
+  let stubs_rsub_scalar =
+    foreign "atg_rsub_scalar" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_rsub_scalar_out =
+    foreign
+      "atg_rsub_scalar_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_rsub_tensor_out =
+    foreign
+      "atg_rsub_tensor_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_scalar_tensor =
+    foreign "atg_scalar_tensor" (scalar @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_scalar_tensor_out =
+    foreign "atg_scalar_tensor_out" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_scaled_dot_product_attention =
+    foreign
+      "atg_scaled_dot_product_attention"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_scatter =
+    foreign
+      "atg_scatter"
+      (gc_tensor @-> int64_t @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_scatter_ =
+    foreign
+      "atg_scatter_"
+      (gc_tensor @-> int64_t @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_scatter_add =
+    foreign
+      "atg_scatter_add"
+      (gc_tensor @-> int64_t @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_scatter_add_ =
+    foreign
+      "atg_scatter_add_"
+      (gc_tensor @-> int64_t @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_scatter_add_out =
+    foreign
+      "atg_scatter_add_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_scatter_reduce =
+    foreign
+      "atg_scatter_reduce"
+      (gc_tensor
+       @-> int64_t
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> string
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_scatter_reduce_ =
+    foreign
+      "atg_scatter_reduce_"
+      (gc_tensor
+       @-> int64_t
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> string
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_scatter_reduce_out =
+    foreign
+      "atg_scatter_reduce_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> string
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_scatter_src_out =
+    foreign
+      "atg_scatter_src_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+end
+
+module C20 (F : Cstubs.FOREIGN) = struct
+  open F
+  open Type_defs
+
+  let stubs_scatter_value =
+    foreign
+      "atg_scatter_value"
+      (gc_tensor @-> int64_t @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_scatter_value_ =
+    foreign
+      "atg_scatter_value_"
+      (gc_tensor @-> int64_t @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_scatter_value_out =
+    foreign
+      "atg_scatter_value_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> gc_tensor
+       @-> scalar
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_scatter_value_reduce =
+    foreign
+      "atg_scatter_value_reduce"
+      (gc_tensor @-> int64_t @-> gc_tensor @-> scalar @-> string @-> returning raw_tensor)
+  ;;
+
+  let stubs_scatter_value_reduce_ =
+    foreign
+      "atg_scatter_value_reduce_"
+      (gc_tensor @-> int64_t @-> gc_tensor @-> scalar @-> string @-> returning raw_tensor)
+  ;;
+
+  let stubs_scatter_value_reduce_out =
+    foreign
+      "atg_scatter_value_reduce_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> gc_tensor
+       @-> scalar
+       @-> string
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_searchsorted =
+    foreign
+      "atg_searchsorted"
+      (gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> int
+       @-> string
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_searchsorted_scalar =
+    foreign
+      "atg_searchsorted_scalar"
+      (gc_tensor
+       @-> scalar
+       @-> int
+       @-> int
+       @-> string
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_searchsorted_scalar_out =
+    foreign
+      "atg_searchsorted_scalar_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> scalar
+       @-> int
+       @-> int
+       @-> string
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_searchsorted_tensor_out =
+    foreign
+      "atg_searchsorted_tensor_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> int
+       @-> string
+       @-> gc_tensor
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_segment_reduce =
+    foreign
+      "atg_segment_reduce"
+      (gc_tensor
+       @-> string
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int
+       @-> scalar
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_segment_reduce_out =
+    foreign
+      "atg_segment_reduce_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> string
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int
+       @-> scalar
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_select =
+    foreign "atg_select" (gc_tensor @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_select_backward =
+    foreign
+      "atg_select_backward"
+      (gc_tensor @-> ptr int64_t @-> int @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_select_backward_out =
+    foreign
+      "atg_select_backward_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_select_copy =
+    foreign "atg_select_copy" (gc_tensor @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_select_copy_int_out =
+    foreign
+      "atg_select_copy_int_out"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_select_scatter =
+    foreign
+      "atg_select_scatter"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_select_scatter_out =
+    foreign
+      "atg_select_scatter_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_selu = foreign "atg_selu" (gc_tensor @-> returning raw_tensor)
+  let stubs_selu_ = foreign "atg_selu_" (gc_tensor @-> returning raw_tensor)
+  let stubs_set = foreign "atg_set" (gc_tensor @-> returning raw_tensor)
+  let stubs_set_ = foreign "atg_set_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_set_out =
+    foreign "atg_set_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_set_requires_grad =
+    foreign "atg_set_requires_grad" (gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_set_source_tensor =
+    foreign "atg_set_source_tensor" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_set_source_tensor_ =
+    foreign "atg_set_source_tensor_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_set_source_tensor_out =
+    foreign
+      "atg_set_source_tensor_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_set_source_tensor_storage_offset_ =
+    foreign
+      "atg_set_source_tensor_storage_offset_"
+      (gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_sgn = foreign "atg_sgn" (gc_tensor @-> returning raw_tensor)
+  let stubs_sgn_ = foreign "atg_sgn_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_sgn_out =
+    foreign "atg_sgn_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_sigmoid = foreign "atg_sigmoid" (gc_tensor @-> returning raw_tensor)
+  let stubs_sigmoid_ = foreign "atg_sigmoid_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_sigmoid_backward =
+    foreign "atg_sigmoid_backward" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_sigmoid_backward_grad_input =
+    foreign
+      "atg_sigmoid_backward_grad_input"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_sigmoid_out =
+    foreign "atg_sigmoid_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_sign = foreign "atg_sign" (gc_tensor @-> returning raw_tensor)
+  let stubs_sign_ = foreign "atg_sign_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_sign_out =
+    foreign "atg_sign_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_signbit = foreign "atg_signbit" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_signbit_out =
+    foreign "atg_signbit_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_silu = foreign "atg_silu" (gc_tensor @-> returning raw_tensor)
+  let stubs_silu_ = foreign "atg_silu_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_silu_backward =
+    foreign "atg_silu_backward" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_silu_backward_grad_input =
+    foreign
+      "atg_silu_backward_grad_input"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_silu_out =
+    foreign "atg_silu_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_sin = foreign "atg_sin" (gc_tensor @-> returning raw_tensor)
+  let stubs_sin_ = foreign "atg_sin_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_sin_out =
+    foreign "atg_sin_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_sinc = foreign "atg_sinc" (gc_tensor @-> returning raw_tensor)
+  let stubs_sinc_ = foreign "atg_sinc_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_sinc_out =
+    foreign "atg_sinc_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_sinh = foreign "atg_sinh" (gc_tensor @-> returning raw_tensor)
+  let stubs_sinh_ = foreign "atg_sinh_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_sinh_out =
+    foreign "atg_sinh_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_size = foreign "atg_size" (gc_tensor @-> int64_t @-> returning int64_t)
+
+  let stubs_slice =
+    foreign
+      "atg_slice"
+      (gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_slice_backward =
+    foreign
+      "atg_slice_backward"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_slice_backward_out =
+    foreign
+      "atg_slice_backward_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_slice_copy =
+    foreign
+      "atg_slice_copy"
+      (gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_slice_copy_tensor_out =
+    foreign
+      "atg_slice_copy_tensor_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_slice_scatter =
+    foreign
+      "atg_slice_scatter"
+      (gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_slice_scatter_out =
+    foreign
+      "atg_slice_scatter_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_slogdet =
+    foreign "atg_slogdet" (ptr raw_tensor @-> gc_tensor @-> returning void)
+  ;;
+
+  let stubs_slogdet_out =
+    foreign
+      "atg_slogdet_out"
+      (ptr raw_tensor @-> gc_tensor @-> gc_tensor @-> gc_tensor @-> returning void)
+  ;;
+
+  let stubs_slow_conv3d =
+    foreign
+      "atg_slow_conv3d"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_slow_conv3d_out =
+    foreign
+      "atg_slow_conv3d_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_slow_conv_dilated2d =
+    foreign
+      "atg_slow_conv_dilated2d"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_slow_conv_dilated2d_out =
+    foreign
+      "atg_slow_conv_dilated2d_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_slow_conv_dilated3d =
+    foreign
+      "atg_slow_conv_dilated3d"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_slow_conv_dilated3d_out =
+    foreign
+      "atg_slow_conv_dilated3d_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_slow_conv_transpose2d =
+    foreign
+      "atg_slow_conv_transpose2d"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_slow_conv_transpose2d_out =
+    foreign
+      "atg_slow_conv_transpose2d_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_slow_conv_transpose3d =
+    foreign
+      "atg_slow_conv_transpose3d"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_slow_conv_transpose3d_out =
+    foreign
+      "atg_slow_conv_transpose3d_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_smm = foreign "atg_smm" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+
+  let stubs_smooth_l1_loss =
+    foreign
+      "atg_smooth_l1_loss"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> double @-> returning raw_tensor)
+  ;;
+
+  let stubs_smooth_l1_loss_backward =
+    foreign
+      "atg_smooth_l1_loss_backward"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> double
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_smooth_l1_loss_backward_grad_input =
+    foreign
+      "atg_smooth_l1_loss_backward_grad_input"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> double
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_smooth_l1_loss_out =
+    foreign
+      "atg_smooth_l1_loss_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> double
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_soft_margin_loss =
+    foreign
+      "atg_soft_margin_loss"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_soft_margin_loss_backward =
+    foreign
+      "atg_soft_margin_loss_backward"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_soft_margin_loss_backward_grad_input =
+    foreign
+      "atg_soft_margin_loss_backward_grad_input"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_soft_margin_loss_out =
+    foreign
+      "atg_soft_margin_loss_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_softmax =
+    foreign "atg_softmax" (gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_softmax_int_out =
+    foreign
+      "atg_softmax_int_out"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_softplus = foreign "atg_softplus" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_softplus_backward =
+    foreign
+      "atg_softplus_backward"
+      (gc_tensor @-> gc_tensor @-> scalar @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_softplus_backward_grad_input =
+    foreign
+      "atg_softplus_backward_grad_input"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> scalar
+       @-> scalar
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_softplus_out =
+    foreign "atg_softplus_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_softshrink = foreign "atg_softshrink" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_softshrink_backward =
+    foreign
+      "atg_softshrink_backward"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_softshrink_backward_grad_input =
+    foreign
+      "atg_softshrink_backward_grad_input"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_softshrink_out =
+    foreign "atg_softshrink_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_sort =
+    foreign
+      "atg_sort"
+      (ptr raw_tensor @-> gc_tensor @-> int64_t @-> int @-> returning void)
+  ;;
+
+  let stubs_sort_stable =
+    foreign
+      "atg_sort_stable"
+      (ptr raw_tensor @-> gc_tensor @-> int @-> int64_t @-> int @-> returning void)
+  ;;
+
+  let stubs_sort_values =
+    foreign
+      "atg_sort_values"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_sort_values_stable =
+    foreign
+      "atg_sort_values_stable"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_sparse_bsc_tensor =
+    foreign
+      "atg_sparse_bsc_tensor"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int @-> int @-> returning raw_tensor)
+  ;;
+end
+
+module C21 (F : Cstubs.FOREIGN) = struct
+  open F
+  open Type_defs
+
+  let stubs_sparse_bsc_tensor_ccol_row_value_size =
+    foreign
+      "atg_sparse_bsc_tensor_ccol_row_value_size"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_sparse_bsr_tensor =
+    foreign
+      "atg_sparse_bsr_tensor"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_sparse_bsr_tensor_crow_col_value_size =
+    foreign
+      "atg_sparse_bsr_tensor_crow_col_value_size"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_sparse_compressed_tensor =
+    foreign
+      "atg_sparse_compressed_tensor"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_sparse_compressed_tensor_comp_plain_value_size =
+    foreign
+      "atg_sparse_compressed_tensor_comp_plain_value_size"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_sparse_coo_tensor =
+    foreign
+      "atg_sparse_coo_tensor"
+      (ptr int64_t @-> int @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_sparse_coo_tensor_indices =
+    foreign
+      "atg_sparse_coo_tensor_indices"
+      (gc_tensor @-> gc_tensor @-> int @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_sparse_coo_tensor_indices_size =
+    foreign
+      "atg_sparse_coo_tensor_indices_size"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_sparse_coo_tensor_size_out =
+    foreign
+      "atg_sparse_coo_tensor_size_out"
+      (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_sparse_csc_tensor =
+    foreign
+      "atg_sparse_csc_tensor"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_sparse_csc_tensor_ccol_row_value_size =
+    foreign
+      "atg_sparse_csc_tensor_ccol_row_value_size"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_sparse_csr_tensor =
+    foreign
+      "atg_sparse_csr_tensor"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_sparse_csr_tensor_crow_col_value_size =
+    foreign
+      "atg_sparse_csr_tensor_crow_col_value_size"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_sparse_dim = foreign "atg_sparse_dim" (gc_tensor @-> returning int64_t)
+
+  let stubs_sparse_mask =
+    foreign "atg_sparse_mask" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_sparse_mask_out =
+    foreign
+      "atg_sparse_mask_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_sparse_resize =
+    foreign
+      "atg_sparse_resize"
+      (gc_tensor @-> ptr int64_t @-> int @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_sparse_resize_ =
+    foreign
+      "atg_sparse_resize_"
+      (gc_tensor @-> ptr int64_t @-> int @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_sparse_resize_and_clear =
+    foreign
+      "atg_sparse_resize_and_clear"
+      (gc_tensor @-> ptr int64_t @-> int @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_sparse_resize_and_clear_ =
+    foreign
+      "atg_sparse_resize_and_clear_"
+      (gc_tensor @-> ptr int64_t @-> int @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_sparse_resize_and_clear_out =
+    foreign
+      "atg_sparse_resize_and_clear_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_sparse_resize_out =
+    foreign
+      "atg_sparse_resize_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_sparse_sampled_addmm =
+    foreign
+      "atg_sparse_sampled_addmm"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_sparse_sampled_addmm_out =
+    foreign
+      "atg_sparse_sampled_addmm_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_airy_ai =
+    foreign "atg_special_airy_ai" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_airy_ai_out =
+    foreign "atg_special_airy_ai_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_bessel_j0 =
+    foreign "atg_special_bessel_j0" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_bessel_j0_out =
+    foreign "atg_special_bessel_j0_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_bessel_j1 =
+    foreign "atg_special_bessel_j1" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_bessel_j1_out =
+    foreign "atg_special_bessel_j1_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_bessel_y0 =
+    foreign "atg_special_bessel_y0" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_bessel_y0_out =
+    foreign "atg_special_bessel_y0_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_bessel_y1 =
+    foreign "atg_special_bessel_y1" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_bessel_y1_out =
+    foreign "atg_special_bessel_y1_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_chebyshev_polynomial_t =
+    foreign
+      "atg_special_chebyshev_polynomial_t"
+      (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_chebyshev_polynomial_t_n_scalar =
+    foreign
+      "atg_special_chebyshev_polynomial_t_n_scalar"
+      (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_chebyshev_polynomial_t_n_scalar_out =
+    foreign
+      "atg_special_chebyshev_polynomial_t_n_scalar_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_chebyshev_polynomial_t_out =
+    foreign
+      "atg_special_chebyshev_polynomial_t_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_chebyshev_polynomial_t_x_scalar =
+    foreign
+      "atg_special_chebyshev_polynomial_t_x_scalar"
+      (scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_chebyshev_polynomial_t_x_scalar_out =
+    foreign
+      "atg_special_chebyshev_polynomial_t_x_scalar_out"
+      (gc_tensor @-> scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_chebyshev_polynomial_u =
+    foreign
+      "atg_special_chebyshev_polynomial_u"
+      (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_chebyshev_polynomial_u_n_scalar =
+    foreign
+      "atg_special_chebyshev_polynomial_u_n_scalar"
+      (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_chebyshev_polynomial_u_n_scalar_out =
+    foreign
+      "atg_special_chebyshev_polynomial_u_n_scalar_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_chebyshev_polynomial_u_out =
+    foreign
+      "atg_special_chebyshev_polynomial_u_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_chebyshev_polynomial_u_x_scalar =
+    foreign
+      "atg_special_chebyshev_polynomial_u_x_scalar"
+      (scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_chebyshev_polynomial_u_x_scalar_out =
+    foreign
+      "atg_special_chebyshev_polynomial_u_x_scalar_out"
+      (gc_tensor @-> scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_chebyshev_polynomial_v =
+    foreign
+      "atg_special_chebyshev_polynomial_v"
+      (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_chebyshev_polynomial_v_n_scalar =
+    foreign
+      "atg_special_chebyshev_polynomial_v_n_scalar"
+      (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_chebyshev_polynomial_v_n_scalar_out =
+    foreign
+      "atg_special_chebyshev_polynomial_v_n_scalar_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_chebyshev_polynomial_v_out =
+    foreign
+      "atg_special_chebyshev_polynomial_v_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_chebyshev_polynomial_v_x_scalar =
+    foreign
+      "atg_special_chebyshev_polynomial_v_x_scalar"
+      (scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_chebyshev_polynomial_v_x_scalar_out =
+    foreign
+      "atg_special_chebyshev_polynomial_v_x_scalar_out"
+      (gc_tensor @-> scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_chebyshev_polynomial_w =
+    foreign
+      "atg_special_chebyshev_polynomial_w"
+      (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_chebyshev_polynomial_w_n_scalar =
+    foreign
+      "atg_special_chebyshev_polynomial_w_n_scalar"
+      (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_chebyshev_polynomial_w_n_scalar_out =
+    foreign
+      "atg_special_chebyshev_polynomial_w_n_scalar_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_chebyshev_polynomial_w_out =
+    foreign
+      "atg_special_chebyshev_polynomial_w_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_chebyshev_polynomial_w_x_scalar =
+    foreign
+      "atg_special_chebyshev_polynomial_w_x_scalar"
+      (scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_chebyshev_polynomial_w_x_scalar_out =
+    foreign
+      "atg_special_chebyshev_polynomial_w_x_scalar_out"
+      (gc_tensor @-> scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_digamma =
+    foreign "atg_special_digamma" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_digamma_out =
+    foreign "atg_special_digamma_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_entr = foreign "atg_special_entr" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_special_entr_out =
+    foreign "atg_special_entr_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_erf = foreign "atg_special_erf" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_special_erf_out =
+    foreign "atg_special_erf_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_erfc = foreign "atg_special_erfc" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_special_erfc_out =
+    foreign "atg_special_erfc_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_erfcx =
+    foreign "atg_special_erfcx" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_erfcx_out =
+    foreign "atg_special_erfcx_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_erfinv =
+    foreign "atg_special_erfinv" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_erfinv_out =
+    foreign "atg_special_erfinv_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_exp2 = foreign "atg_special_exp2" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_special_exp2_out =
+    foreign "atg_special_exp2_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_expit =
+    foreign "atg_special_expit" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_expit_out =
+    foreign "atg_special_expit_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_expm1 =
+    foreign "atg_special_expm1" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_expm1_out =
+    foreign "atg_special_expm1_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_gammainc =
+    foreign "atg_special_gammainc" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_gammainc_out =
+    foreign
+      "atg_special_gammainc_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_gammaincc =
+    foreign "atg_special_gammaincc" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_gammaincc_out =
+    foreign
+      "atg_special_gammaincc_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_gammaln =
+    foreign "atg_special_gammaln" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_gammaln_out =
+    foreign "atg_special_gammaln_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_hermite_polynomial_h =
+    foreign
+      "atg_special_hermite_polynomial_h"
+      (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_hermite_polynomial_h_n_scalar =
+    foreign
+      "atg_special_hermite_polynomial_h_n_scalar"
+      (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_hermite_polynomial_h_n_scalar_out =
+    foreign
+      "atg_special_hermite_polynomial_h_n_scalar_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_hermite_polynomial_h_out =
+    foreign
+      "atg_special_hermite_polynomial_h_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_hermite_polynomial_h_x_scalar =
+    foreign
+      "atg_special_hermite_polynomial_h_x_scalar"
+      (scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_hermite_polynomial_h_x_scalar_out =
+    foreign
+      "atg_special_hermite_polynomial_h_x_scalar_out"
+      (gc_tensor @-> scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_hermite_polynomial_he =
+    foreign
+      "atg_special_hermite_polynomial_he"
+      (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_hermite_polynomial_he_n_scalar =
+    foreign
+      "atg_special_hermite_polynomial_he_n_scalar"
+      (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_hermite_polynomial_he_n_scalar_out =
+    foreign
+      "atg_special_hermite_polynomial_he_n_scalar_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_hermite_polynomial_he_out =
+    foreign
+      "atg_special_hermite_polynomial_he_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_hermite_polynomial_he_x_scalar =
+    foreign
+      "atg_special_hermite_polynomial_he_x_scalar"
+      (scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_hermite_polynomial_he_x_scalar_out =
+    foreign
+      "atg_special_hermite_polynomial_he_x_scalar_out"
+      (gc_tensor @-> scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_i0 = foreign "atg_special_i0" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_special_i0_out =
+    foreign "atg_special_i0_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_i0e = foreign "atg_special_i0e" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_special_i0e_out =
+    foreign "atg_special_i0e_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_i1 = foreign "atg_special_i1" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_special_i1_out =
+    foreign "atg_special_i1_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+end
+
+module C22 (F : Cstubs.FOREIGN) = struct
+  open F
+  open Type_defs
+
+  let stubs_special_i1e = foreign "atg_special_i1e" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_special_i1e_out =
+    foreign "atg_special_i1e_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_laguerre_polynomial_l =
+    foreign
+      "atg_special_laguerre_polynomial_l"
+      (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_laguerre_polynomial_l_n_scalar =
+    foreign
+      "atg_special_laguerre_polynomial_l_n_scalar"
+      (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_laguerre_polynomial_l_n_scalar_out =
+    foreign
+      "atg_special_laguerre_polynomial_l_n_scalar_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_laguerre_polynomial_l_out =
+    foreign
+      "atg_special_laguerre_polynomial_l_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_laguerre_polynomial_l_x_scalar =
+    foreign
+      "atg_special_laguerre_polynomial_l_x_scalar"
+      (scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_laguerre_polynomial_l_x_scalar_out =
+    foreign
+      "atg_special_laguerre_polynomial_l_x_scalar_out"
+      (gc_tensor @-> scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_legendre_polynomial_p =
+    foreign
+      "atg_special_legendre_polynomial_p"
+      (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_legendre_polynomial_p_n_scalar =
+    foreign
+      "atg_special_legendre_polynomial_p_n_scalar"
+      (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_legendre_polynomial_p_n_scalar_out =
+    foreign
+      "atg_special_legendre_polynomial_p_n_scalar_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_legendre_polynomial_p_out =
+    foreign
+      "atg_special_legendre_polynomial_p_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_legendre_polynomial_p_x_scalar =
+    foreign
+      "atg_special_legendre_polynomial_p_x_scalar"
+      (scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_legendre_polynomial_p_x_scalar_out =
+    foreign
+      "atg_special_legendre_polynomial_p_x_scalar_out"
+      (gc_tensor @-> scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_log1p =
+    foreign "atg_special_log1p" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_log1p_out =
+    foreign "atg_special_log1p_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_log_ndtr =
+    foreign "atg_special_log_ndtr" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_log_ndtr_out =
+    foreign "atg_special_log_ndtr_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_log_softmax =
+    foreign
+      "atg_special_log_softmax"
+      (gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_logit =
+    foreign "atg_special_logit" (gc_tensor @-> double @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_logit_out =
+    foreign
+      "atg_special_logit_out"
+      (gc_tensor @-> gc_tensor @-> double @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_logsumexp =
+    foreign
+      "atg_special_logsumexp"
+      (gc_tensor @-> ptr int64_t @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_logsumexp_out =
+    foreign
+      "atg_special_logsumexp_out"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_modified_bessel_i0 =
+    foreign "atg_special_modified_bessel_i0" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_modified_bessel_i0_out =
+    foreign
+      "atg_special_modified_bessel_i0_out"
+      (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_modified_bessel_i1 =
+    foreign "atg_special_modified_bessel_i1" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_modified_bessel_i1_out =
+    foreign
+      "atg_special_modified_bessel_i1_out"
+      (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_modified_bessel_k0 =
+    foreign "atg_special_modified_bessel_k0" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_modified_bessel_k0_out =
+    foreign
+      "atg_special_modified_bessel_k0_out"
+      (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_modified_bessel_k1 =
+    foreign "atg_special_modified_bessel_k1" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_modified_bessel_k1_out =
+    foreign
+      "atg_special_modified_bessel_k1_out"
+      (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_multigammaln =
+    foreign "atg_special_multigammaln" (gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_multigammaln_out =
+    foreign
+      "atg_special_multigammaln_out"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_ndtr = foreign "atg_special_ndtr" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_special_ndtr_out =
+    foreign "atg_special_ndtr_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_ndtri =
+    foreign "atg_special_ndtri" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_ndtri_out =
+    foreign "atg_special_ndtri_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_polygamma =
+    foreign "atg_special_polygamma" (int64_t @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_polygamma_out =
+    foreign
+      "atg_special_polygamma_out"
+      (gc_tensor @-> int64_t @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_psi = foreign "atg_special_psi" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_special_psi_out =
+    foreign "atg_special_psi_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_round =
+    foreign "atg_special_round" (gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_round_out =
+    foreign
+      "atg_special_round_out"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_scaled_modified_bessel_k0 =
+    foreign "atg_special_scaled_modified_bessel_k0" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_scaled_modified_bessel_k0_out =
+    foreign
+      "atg_special_scaled_modified_bessel_k0_out"
+      (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_scaled_modified_bessel_k1 =
+    foreign "atg_special_scaled_modified_bessel_k1" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_scaled_modified_bessel_k1_out =
+    foreign
+      "atg_special_scaled_modified_bessel_k1_out"
+      (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_shifted_chebyshev_polynomial_t =
+    foreign
+      "atg_special_shifted_chebyshev_polynomial_t"
+      (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_shifted_chebyshev_polynomial_t_n_scalar =
+    foreign
+      "atg_special_shifted_chebyshev_polynomial_t_n_scalar"
+      (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_shifted_chebyshev_polynomial_t_n_scalar_out =
+    foreign
+      "atg_special_shifted_chebyshev_polynomial_t_n_scalar_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_shifted_chebyshev_polynomial_t_out =
+    foreign
+      "atg_special_shifted_chebyshev_polynomial_t_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_shifted_chebyshev_polynomial_t_x_scalar =
+    foreign
+      "atg_special_shifted_chebyshev_polynomial_t_x_scalar"
+      (scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_shifted_chebyshev_polynomial_t_x_scalar_out =
+    foreign
+      "atg_special_shifted_chebyshev_polynomial_t_x_scalar_out"
+      (gc_tensor @-> scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_shifted_chebyshev_polynomial_u =
+    foreign
+      "atg_special_shifted_chebyshev_polynomial_u"
+      (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_shifted_chebyshev_polynomial_u_n_scalar =
+    foreign
+      "atg_special_shifted_chebyshev_polynomial_u_n_scalar"
+      (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_shifted_chebyshev_polynomial_u_n_scalar_out =
+    foreign
+      "atg_special_shifted_chebyshev_polynomial_u_n_scalar_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_shifted_chebyshev_polynomial_u_out =
+    foreign
+      "atg_special_shifted_chebyshev_polynomial_u_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_shifted_chebyshev_polynomial_u_x_scalar =
+    foreign
+      "atg_special_shifted_chebyshev_polynomial_u_x_scalar"
+      (scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_shifted_chebyshev_polynomial_u_x_scalar_out =
+    foreign
+      "atg_special_shifted_chebyshev_polynomial_u_x_scalar_out"
+      (gc_tensor @-> scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_shifted_chebyshev_polynomial_v =
+    foreign
+      "atg_special_shifted_chebyshev_polynomial_v"
+      (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_shifted_chebyshev_polynomial_v_n_scalar =
+    foreign
+      "atg_special_shifted_chebyshev_polynomial_v_n_scalar"
+      (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_shifted_chebyshev_polynomial_v_n_scalar_out =
+    foreign
+      "atg_special_shifted_chebyshev_polynomial_v_n_scalar_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_shifted_chebyshev_polynomial_v_out =
+    foreign
+      "atg_special_shifted_chebyshev_polynomial_v_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_shifted_chebyshev_polynomial_v_x_scalar =
+    foreign
+      "atg_special_shifted_chebyshev_polynomial_v_x_scalar"
+      (scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_shifted_chebyshev_polynomial_v_x_scalar_out =
+    foreign
+      "atg_special_shifted_chebyshev_polynomial_v_x_scalar_out"
+      (gc_tensor @-> scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_shifted_chebyshev_polynomial_w =
+    foreign
+      "atg_special_shifted_chebyshev_polynomial_w"
+      (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_shifted_chebyshev_polynomial_w_n_scalar =
+    foreign
+      "atg_special_shifted_chebyshev_polynomial_w_n_scalar"
+      (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_shifted_chebyshev_polynomial_w_n_scalar_out =
+    foreign
+      "atg_special_shifted_chebyshev_polynomial_w_n_scalar_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_shifted_chebyshev_polynomial_w_out =
+    foreign
+      "atg_special_shifted_chebyshev_polynomial_w_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_shifted_chebyshev_polynomial_w_x_scalar =
+    foreign
+      "atg_special_shifted_chebyshev_polynomial_w_x_scalar"
+      (scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_shifted_chebyshev_polynomial_w_x_scalar_out =
+    foreign
+      "atg_special_shifted_chebyshev_polynomial_w_x_scalar_out"
+      (gc_tensor @-> scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_sinc = foreign "atg_special_sinc" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_special_sinc_out =
+    foreign "atg_special_sinc_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_softmax =
+    foreign "atg_special_softmax" (gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_spherical_bessel_j0 =
+    foreign "atg_special_spherical_bessel_j0" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_spherical_bessel_j0_out =
+    foreign
+      "atg_special_spherical_bessel_j0_out"
+      (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_xlog1py =
+    foreign "atg_special_xlog1py" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_xlog1py_other_scalar =
+    foreign
+      "atg_special_xlog1py_other_scalar"
+      (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_xlog1py_other_scalar_out =
+    foreign
+      "atg_special_xlog1py_other_scalar_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_xlog1py_out =
+    foreign
+      "atg_special_xlog1py_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_xlog1py_self_scalar =
+    foreign
+      "atg_special_xlog1py_self_scalar"
+      (scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_xlog1py_self_scalar_out =
+    foreign
+      "atg_special_xlog1py_self_scalar_out"
+      (gc_tensor @-> scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_xlogy =
+    foreign "atg_special_xlogy" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_xlogy_other_scalar =
+    foreign
+      "atg_special_xlogy_other_scalar"
+      (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_xlogy_other_scalar_out =
+    foreign
+      "atg_special_xlogy_other_scalar_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_xlogy_out =
+    foreign
+      "atg_special_xlogy_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_xlogy_self_scalar =
+    foreign "atg_special_xlogy_self_scalar" (scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_xlogy_self_scalar_out =
+    foreign
+      "atg_special_xlogy_self_scalar_out"
+      (gc_tensor @-> scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_zeta =
+    foreign "atg_special_zeta" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_zeta_other_scalar =
+    foreign "atg_special_zeta_other_scalar" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_zeta_other_scalar_out =
+    foreign
+      "atg_special_zeta_other_scalar_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_zeta_out =
+    foreign
+      "atg_special_zeta_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_zeta_self_scalar =
+    foreign "atg_special_zeta_self_scalar" (scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_special_zeta_self_scalar_out =
+    foreign
+      "atg_special_zeta_self_scalar_out"
+      (gc_tensor @-> scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_split =
+    foreign "atg_split" (gc_tensor @-> int64_t @-> int64_t @-> returning (ptr raw_tensor))
+  ;;
+
+  let stubs_split_copy =
+    foreign
+      "atg_split_copy"
+      (gc_tensor @-> int64_t @-> int64_t @-> returning (ptr raw_tensor))
+  ;;
+
+  let stubs_split_copy_tensor_out =
+    foreign
+      "atg_split_copy_tensor_out"
+      (ptr gc_tensor @-> int @-> gc_tensor @-> int64_t @-> int64_t @-> returning void)
+  ;;
+
+  let stubs_split_sizes =
+    foreign
+      "atg_split_sizes"
+      (gc_tensor @-> ptr int64_t @-> int @-> int64_t @-> returning (ptr raw_tensor))
+  ;;
+
+  let stubs_split_with_sizes =
+    foreign
+      "atg_split_with_sizes"
+      (gc_tensor @-> ptr int64_t @-> int @-> int64_t @-> returning (ptr raw_tensor))
+  ;;
+
+  let stubs_split_with_sizes_copy =
+    foreign
+      "atg_split_with_sizes_copy"
+      (gc_tensor @-> ptr int64_t @-> int @-> int64_t @-> returning (ptr raw_tensor))
+  ;;
+end
+
+module C23 (F : Cstubs.FOREIGN) = struct
+  open F
+  open Type_defs
+
+  let stubs_split_with_sizes_copy_out =
+    foreign
+      "atg_split_with_sizes_copy_out"
+      (ptr gc_tensor
+       @-> int
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> returning void)
+  ;;
+
+  let stubs_sqrt = foreign "atg_sqrt" (gc_tensor @-> returning raw_tensor)
+  let stubs_sqrt_ = foreign "atg_sqrt_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_sqrt_out =
+    foreign "atg_sqrt_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_square = foreign "atg_square" (gc_tensor @-> returning raw_tensor)
+  let stubs_square_ = foreign "atg_square_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_square_out =
+    foreign "atg_square_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_squeeze = foreign "atg_squeeze" (gc_tensor @-> returning raw_tensor)
+  let stubs_squeeze_ = foreign "atg_squeeze_" (gc_tensor @-> returning raw_tensor)
+  let stubs_squeeze_copy = foreign "atg_squeeze_copy" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_squeeze_copy_dim =
+    foreign "atg_squeeze_copy_dim" (gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_squeeze_copy_dim_out =
+    foreign
+      "atg_squeeze_copy_dim_out"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_squeeze_copy_dims =
+    foreign
+      "atg_squeeze_copy_dims"
+      (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_squeeze_copy_dims_out =
+    foreign
+      "atg_squeeze_copy_dims_out"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_squeeze_copy_out =
+    foreign "atg_squeeze_copy_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_squeeze_dim =
+    foreign "atg_squeeze_dim" (gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_squeeze_dim_ =
+    foreign "atg_squeeze_dim_" (gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_squeeze_dims =
+    foreign "atg_squeeze_dims" (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_squeeze_dims_ =
+    foreign
+      "atg_squeeze_dims_"
+      (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_sspaddmm =
+    foreign "atg_sspaddmm" (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_sspaddmm_out =
+    foreign
+      "atg_sspaddmm_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_stack =
+    foreign "atg_stack" (ptr gc_tensor @-> int @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_stack_out =
+    foreign
+      "atg_stack_out"
+      (gc_tensor @-> ptr gc_tensor @-> int @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_std = foreign "atg_std" (gc_tensor @-> int @-> returning raw_tensor)
+
+  let stubs_std_correction =
+    foreign
+      "atg_std_correction"
+      (gc_tensor @-> ptr int64_t @-> int @-> scalar @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_std_correction_out =
+    foreign
+      "atg_std_correction_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> scalar
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_std_dim =
+    foreign
+      "atg_std_dim"
+      (gc_tensor @-> ptr int64_t @-> int @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_std_mean =
+    foreign "atg_std_mean" (ptr raw_tensor @-> gc_tensor @-> int @-> returning void)
+  ;;
+
+  let stubs_std_mean_correction =
+    foreign
+      "atg_std_mean_correction"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> scalar
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_std_mean_correction_out =
+    foreign
+      "atg_std_mean_correction_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> scalar
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_std_mean_dim =
+    foreign
+      "atg_std_mean_dim"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_std_out =
+    foreign
+      "atg_std_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_stft =
+    foreign
+      "atg_stft"
+      (gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> gc_tensor
+       @-> int
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_stft_center =
+    foreign
+      "atg_stft_center"
+      (gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> gc_tensor
+       @-> int
+       @-> string
+       @-> int
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_stride = foreign "atg_stride" (gc_tensor @-> int64_t @-> returning int64_t)
+  let stubs_sub = foreign "atg_sub" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  let stubs_sub_ = foreign "atg_sub_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+
+  let stubs_sub_out =
+    foreign "atg_sub_out" (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_sub_scalar =
+    foreign "atg_sub_scalar" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_sub_scalar_ =
+    foreign "atg_sub_scalar_" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_sub_scalar_out =
+    foreign
+      "atg_sub_scalar_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_subtract =
+    foreign "atg_subtract" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_subtract_ =
+    foreign "atg_subtract_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_subtract_out =
+    foreign
+      "atg_subtract_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_subtract_scalar =
+    foreign "atg_subtract_scalar" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_subtract_scalar_ =
+    foreign "atg_subtract_scalar_" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_sum = foreign "atg_sum" (gc_tensor @-> int @-> returning raw_tensor)
+
+  let stubs_sum_dim_intlist =
+    foreign
+      "atg_sum_dim_intlist"
+      (gc_tensor @-> ptr int64_t @-> int @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_sum_intlist_out =
+    foreign
+      "atg_sum_intlist_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_sum_out =
+    foreign "atg_sum_out" (gc_tensor @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_sum_to_size =
+    foreign "atg_sum_to_size" (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_svd =
+    foreign "atg_svd" (ptr raw_tensor @-> gc_tensor @-> int @-> int @-> returning void)
+  ;;
+
+  let stubs_svd_u =
+    foreign
+      "atg_svd_u"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_swapaxes =
+    foreign "atg_swapaxes" (gc_tensor @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_swapaxes_ =
+    foreign "atg_swapaxes_" (gc_tensor @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_swapdims =
+    foreign "atg_swapdims" (gc_tensor @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_swapdims_ =
+    foreign "atg_swapdims_" (gc_tensor @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_tr = foreign "atg_t" (gc_tensor @-> returning raw_tensor)
+  let stubs_t_ = foreign "atg_t_" (gc_tensor @-> returning raw_tensor)
+  let stubs_t_copy = foreign "atg_t_copy" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_t_copy_out =
+    foreign "atg_t_copy_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_take = foreign "atg_take" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+
+  let stubs_take_along_dim =
+    foreign
+      "atg_take_along_dim"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_take_along_dim_out =
+    foreign
+      "atg_take_along_dim_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_take_out =
+    foreign "atg_take_out" (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_tan = foreign "atg_tan" (gc_tensor @-> returning raw_tensor)
+  let stubs_tan_ = foreign "atg_tan_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_tan_out =
+    foreign "atg_tan_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_tanh = foreign "atg_tanh" (gc_tensor @-> returning raw_tensor)
+  let stubs_tanh_ = foreign "atg_tanh_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_tanh_backward =
+    foreign "atg_tanh_backward" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_tanh_backward_grad_input =
+    foreign
+      "atg_tanh_backward_grad_input"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_tanh_out =
+    foreign "atg_tanh_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_tensor_split =
+    foreign
+      "atg_tensor_split"
+      (gc_tensor @-> int64_t @-> int64_t @-> returning (ptr raw_tensor))
+  ;;
+
+  let stubs_tensor_split_indices =
+    foreign
+      "atg_tensor_split_indices"
+      (gc_tensor @-> ptr int64_t @-> int @-> int64_t @-> returning (ptr raw_tensor))
+  ;;
+
+  let stubs_tensor_split_tensor_indices_or_sections =
+    foreign
+      "atg_tensor_split_tensor_indices_or_sections"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> returning (ptr raw_tensor))
+  ;;
+
+  let stubs_tensordot =
+    foreign
+      "atg_tensordot"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_tensordot_out =
+    foreign
+      "atg_tensordot_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_threshold =
+    foreign "atg_threshold" (gc_tensor @-> scalar @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_threshold_ =
+    foreign "atg_threshold_" (gc_tensor @-> scalar @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_threshold_backward =
+    foreign
+      "atg_threshold_backward"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_threshold_backward_grad_input =
+    foreign
+      "atg_threshold_backward_grad_input"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_threshold_out =
+    foreign
+      "atg_threshold_out"
+      (gc_tensor @-> gc_tensor @-> scalar @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_tile =
+    foreign "atg_tile" (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_to_ = foreign "atg_to" (gc_tensor @-> int @-> returning raw_tensor)
+
+  let stubs_to_dense =
+    foreign "atg_to_dense" (gc_tensor @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_to_dense_backward =
+    foreign
+      "atg_to_dense_backward"
+      (gc_tensor @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_to_device =
+    foreign
+      "atg_to_device"
+      (gc_tensor @-> int @-> int @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_to_dtype =
+    foreign "atg_to_dtype" (gc_tensor @-> int @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_to_dtype_layout =
+    foreign
+      "atg_to_dtype_layout"
+      (gc_tensor @-> int @-> int @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_to_mkldnn =
+    foreign "atg_to_mkldnn" (gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_to_mkldnn_backward =
+    foreign "atg_to_mkldnn_backward" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_to_mkldnn_out =
+    foreign "atg_to_mkldnn_out" (gc_tensor @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_to_other =
+    foreign
+      "atg_to_other"
+      (gc_tensor @-> gc_tensor @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_to_padded_tensor =
+    foreign
+      "atg_to_padded_tensor"
+      (gc_tensor @-> double @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_to_padded_tensor_out =
+    foreign
+      "atg_to_padded_tensor_out"
+      (gc_tensor @-> gc_tensor @-> double @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_topk =
+    foreign
+      "atg_topk"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> int
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_topk_values =
+    foreign
+      "atg_topk_values"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> int
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_totype = foreign "atg_totype" (gc_tensor @-> int @-> returning raw_tensor)
+  let stubs_trace = foreign "atg_trace" (gc_tensor @-> returning raw_tensor)
+end
+
+module C24 (F : Cstubs.FOREIGN) = struct
+  open F
+  open Type_defs
+
+  let stubs_trace_backward =
+    foreign
+      "atg_trace_backward"
+      (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_trace_out =
+    foreign "atg_trace_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_transpose =
+    foreign "atg_transpose" (gc_tensor @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_transpose_ =
+    foreign "atg_transpose_" (gc_tensor @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_transpose_copy =
+    foreign
+      "atg_transpose_copy"
+      (gc_tensor @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_transpose_copy_int_out =
+    foreign
+      "atg_transpose_copy_int_out"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_trapezoid =
+    foreign "atg_trapezoid" (gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_trapezoid_x =
+    foreign
+      "atg_trapezoid_x"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_trapz =
+    foreign "atg_trapz" (gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_trapz_dx =
+    foreign "atg_trapz_dx" (gc_tensor @-> double @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_triangular_solve =
+    foreign
+      "atg_triangular_solve"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> int
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_triangular_solve_x =
+    foreign
+      "atg_triangular_solve_x"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> int
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_tril = foreign "atg_tril" (gc_tensor @-> int64_t @-> returning raw_tensor)
+  let stubs_tril_ = foreign "atg_tril_" (gc_tensor @-> int64_t @-> returning raw_tensor)
+
+  let stubs_tril_indices =
+    foreign
+      "atg_tril_indices"
+      (int64_t @-> int64_t @-> int64_t @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_tril_indices_out =
+    foreign
+      "atg_tril_indices_out"
+      (gc_tensor @-> int64_t @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_tril_out =
+    foreign "atg_tril_out" (gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_triplet_margin_loss =
+    foreign
+      "atg_triplet_margin_loss"
+      (gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> double
+       @-> double
+       @-> double
+       @-> int
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_triu = foreign "atg_triu" (gc_tensor @-> int64_t @-> returning raw_tensor)
+  let stubs_triu_ = foreign "atg_triu_" (gc_tensor @-> int64_t @-> returning raw_tensor)
+
+  let stubs_triu_indices =
+    foreign
+      "atg_triu_indices"
+      (int64_t @-> int64_t @-> int64_t @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_triu_indices_out =
+    foreign
+      "atg_triu_indices_out"
+      (gc_tensor @-> int64_t @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_triu_out =
+    foreign "atg_triu_out" (gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_true_divide =
+    foreign "atg_true_divide" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_true_divide_ =
+    foreign "atg_true_divide_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_true_divide_out =
+    foreign
+      "atg_true_divide_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_true_divide_scalar =
+    foreign "atg_true_divide_scalar" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_true_divide_scalar_ =
+    foreign "atg_true_divide_scalar_" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_trunc = foreign "atg_trunc" (gc_tensor @-> returning raw_tensor)
+  let stubs_trunc_ = foreign "atg_trunc_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_trunc_out =
+    foreign "atg_trunc_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_type_as =
+    foreign "atg_type_as" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_unbind =
+    foreign "atg_unbind" (gc_tensor @-> int64_t @-> returning (ptr raw_tensor))
+  ;;
+
+  let stubs_unbind_copy =
+    foreign "atg_unbind_copy" (gc_tensor @-> int64_t @-> returning (ptr raw_tensor))
+  ;;
+
+  let stubs_unbind_copy_int_out =
+    foreign
+      "atg_unbind_copy_int_out"
+      (ptr gc_tensor @-> int @-> gc_tensor @-> int64_t @-> returning void)
+  ;;
+
+  let stubs_unflatten =
+    foreign
+      "atg_unflatten"
+      (gc_tensor @-> int64_t @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_unflatten_dense_tensors =
+    foreign
+      "atg_unflatten_dense_tensors"
+      (gc_tensor @-> ptr gc_tensor @-> int @-> returning (ptr raw_tensor))
+  ;;
+
+  let stubs_unfold =
+    foreign
+      "atg_unfold"
+      (gc_tensor @-> int64_t @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_unfold_backward =
+    foreign
+      "atg_unfold_backward"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_unfold_backward_out =
+    foreign
+      "atg_unfold_backward_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_unfold_copy =
+    foreign
+      "atg_unfold_copy"
+      (gc_tensor @-> int64_t @-> int64_t @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_unfold_copy_out =
+    foreign
+      "atg_unfold_copy_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int64_t
+       @-> int64_t
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_uniform =
+    foreign "atg_uniform" (gc_tensor @-> double @-> double @-> returning raw_tensor)
+  ;;
+
+  let stubs_uniform_ =
+    foreign "atg_uniform_" (gc_tensor @-> double @-> double @-> returning raw_tensor)
+  ;;
+
+  let stubs_uniform_out =
+    foreign
+      "atg_uniform_out"
+      (gc_tensor @-> gc_tensor @-> double @-> double @-> returning raw_tensor)
+  ;;
+
+  let stubs_unique_consecutive =
+    foreign
+      "atg_unique_consecutive"
+      (ptr raw_tensor @-> gc_tensor @-> int @-> int @-> int64_t @-> int @-> returning void)
+  ;;
+
+  let stubs_unique_consecutive_out =
+    foreign
+      "atg_unique_consecutive_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int
+       @-> int
+       @-> int64_t
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_unique_dim =
+    foreign
+      "atg_unique_dim"
+      (ptr raw_tensor @-> gc_tensor @-> int64_t @-> int @-> int @-> int @-> returning void)
+  ;;
+
+  let stubs_unique_dim_consecutive =
+    foreign
+      "atg_unique_dim_consecutive"
+      (ptr raw_tensor @-> gc_tensor @-> int64_t @-> int @-> int @-> returning void)
+  ;;
+
+  let stubs_unique_dim_consecutive_out =
+    foreign
+      "atg_unique_dim_consecutive_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_unique_dim_out =
+    foreign
+      "atg_unique_dim_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_unsafe_chunk =
+    foreign
+      "atg_unsafe_chunk"
+      (gc_tensor @-> int64_t @-> int64_t @-> returning (ptr raw_tensor))
+  ;;
+
+  let stubs_unsafe_split =
+    foreign
+      "atg_unsafe_split"
+      (gc_tensor @-> int64_t @-> int64_t @-> returning (ptr raw_tensor))
+  ;;
+
+  let stubs_unsafe_split_tensor_out =
+    foreign
+      "atg_unsafe_split_tensor_out"
+      (ptr gc_tensor @-> int @-> gc_tensor @-> int64_t @-> int64_t @-> returning void)
+  ;;
+
+  let stubs_unsafe_split_with_sizes =
+    foreign
+      "atg_unsafe_split_with_sizes"
+      (gc_tensor @-> ptr int64_t @-> int @-> int64_t @-> returning (ptr raw_tensor))
+  ;;
+
+  let stubs_unsafe_split_with_sizes_out =
+    foreign
+      "atg_unsafe_split_with_sizes_out"
+      (ptr gc_tensor
+       @-> int
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int64_t
+       @-> returning void)
+  ;;
+
+  let stubs_unsqueeze =
+    foreign "atg_unsqueeze" (gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_unsqueeze_ =
+    foreign "atg_unsqueeze_" (gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_unsqueeze_copy =
+    foreign "atg_unsqueeze_copy" (gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_unsqueeze_copy_out =
+    foreign
+      "atg_unsqueeze_copy_out"
+      (gc_tensor @-> gc_tensor @-> int64_t @-> returning raw_tensor)
+  ;;
+
+  let stubs_upsample_bicubic2d =
+    foreign
+      "atg_upsample_bicubic2d"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_upsample_bicubic2d_backward =
+    foreign
+      "atg_upsample_bicubic2d_backward"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_upsample_bicubic2d_backward_grad_input =
+    foreign
+      "atg_upsample_bicubic2d_backward_grad_input"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_upsample_bicubic2d_out =
+    foreign
+      "atg_upsample_bicubic2d_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_upsample_bicubic2d_vec =
+    foreign
+      "atg_upsample_bicubic2d_vec"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> ptr double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_upsample_bilinear2d =
+    foreign
+      "atg_upsample_bilinear2d"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_upsample_bilinear2d_backward =
+    foreign
+      "atg_upsample_bilinear2d_backward"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_upsample_bilinear2d_backward_grad_input =
+    foreign
+      "atg_upsample_bilinear2d_backward_grad_input"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_upsample_bilinear2d_out =
+    foreign
+      "atg_upsample_bilinear2d_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_upsample_bilinear2d_vec =
+    foreign
+      "atg_upsample_bilinear2d_vec"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> ptr double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_upsample_linear1d =
+    foreign
+      "atg_upsample_linear1d"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_upsample_linear1d_backward =
+    foreign
+      "atg_upsample_linear1d_backward"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_upsample_linear1d_backward_grad_input =
+    foreign
+      "atg_upsample_linear1d_backward_grad_input"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_upsample_linear1d_out =
+    foreign
+      "atg_upsample_linear1d_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_upsample_linear1d_vec =
+    foreign
+      "atg_upsample_linear1d_vec"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> ptr double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_upsample_nearest1d =
+    foreign
+      "atg_upsample_nearest1d"
+      (gc_tensor @-> ptr int64_t @-> int @-> double @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_upsample_nearest1d_backward =
+    foreign
+      "atg_upsample_nearest1d_backward"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_upsample_nearest1d_backward_grad_input =
+    foreign
+      "atg_upsample_nearest1d_backward_grad_input"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_upsample_nearest1d_out =
+    foreign
+      "atg_upsample_nearest1d_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_upsample_nearest1d_vec =
+    foreign
+      "atg_upsample_nearest1d_vec"
+      (gc_tensor @-> ptr int64_t @-> int @-> ptr double @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_upsample_nearest2d =
+    foreign
+      "atg_upsample_nearest2d"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_upsample_nearest2d_backward =
+    foreign
+      "atg_upsample_nearest2d_backward"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_upsample_nearest2d_backward_grad_input =
+    foreign
+      "atg_upsample_nearest2d_backward_grad_input"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_upsample_nearest2d_out =
+    foreign
+      "atg_upsample_nearest2d_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_upsample_nearest2d_vec =
+    foreign
+      "atg_upsample_nearest2d_vec"
+      (gc_tensor @-> ptr int64_t @-> int @-> ptr double @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_upsample_nearest3d =
+    foreign
+      "atg_upsample_nearest3d"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_upsample_nearest3d_backward =
+    foreign
+      "atg_upsample_nearest3d_backward"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_upsample_nearest3d_backward_grad_input =
+    foreign
+      "atg_upsample_nearest3d_backward_grad_input"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_upsample_nearest3d_out =
+    foreign
+      "atg_upsample_nearest3d_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_upsample_nearest3d_vec =
+    foreign
+      "atg_upsample_nearest3d_vec"
+      (gc_tensor @-> ptr int64_t @-> int @-> ptr double @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_upsample_trilinear3d =
+    foreign
+      "atg_upsample_trilinear3d"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_upsample_trilinear3d_backward =
+    foreign
+      "atg_upsample_trilinear3d_backward"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_upsample_trilinear3d_backward_grad_input =
+    foreign
+      "atg_upsample_trilinear3d_backward_grad_input"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_upsample_trilinear3d_out =
+    foreign
+      "atg_upsample_trilinear3d_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_upsample_trilinear3d_vec =
+    foreign
+      "atg_upsample_trilinear3d_vec"
+      (gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> ptr double
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_value_selecting_reduction_backward =
+    foreign
+      "atg_value_selecting_reduction_backward"
+      (gc_tensor
+       @-> int64_t
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_values = foreign "atg_values" (gc_tensor @-> returning raw_tensor)
+  let stubs_values_copy = foreign "atg_values_copy" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_values_copy_out =
+    foreign "atg_values_copy_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_vander =
+    foreign "atg_vander" (gc_tensor @-> int64_t @-> int @-> int @-> returning raw_tensor)
+  ;;
+end
+
+module C25 (F : Cstubs.FOREIGN) = struct
+  open F
+  open Type_defs
+
+  let stubs_var = foreign "atg_var" (gc_tensor @-> int @-> returning raw_tensor)
+
+  let stubs_var_correction =
+    foreign
+      "atg_var_correction"
+      (gc_tensor @-> ptr int64_t @-> int @-> scalar @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_var_correction_out =
+    foreign
+      "atg_var_correction_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> scalar
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_var_dim =
+    foreign
+      "atg_var_dim"
+      (gc_tensor @-> ptr int64_t @-> int @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_var_mean =
+    foreign "atg_var_mean" (ptr raw_tensor @-> gc_tensor @-> int @-> returning void)
+  ;;
+
+  let stubs_var_mean_correction =
+    foreign
+      "atg_var_mean_correction"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> scalar
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_var_mean_correction_out =
+    foreign
+      "atg_var_mean_correction_out"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> scalar
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_var_mean_dim =
+    foreign
+      "atg_var_mean_dim"
+      (ptr raw_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> returning void)
+  ;;
+
+  let stubs_var_out =
+    foreign
+      "atg_var_out"
+      (gc_tensor
+       @-> gc_tensor
+       @-> ptr int64_t
+       @-> int
+       @-> int
+       @-> int
+       @-> returning raw_tensor)
+  ;;
+
+  let stubs_vdot = foreign "atg_vdot" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+
+  let stubs_vdot_out =
+    foreign "atg_vdot_out" (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_view =
+    foreign "atg_view" (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_view_as =
+    foreign "atg_view_as" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_view_as_complex =
+    foreign "atg_view_as_complex" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_view_as_complex_copy =
+    foreign "atg_view_as_complex_copy" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_view_as_complex_copy_out =
+    foreign
+      "atg_view_as_complex_copy_out"
+      (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_view_as_real = foreign "atg_view_as_real" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_view_as_real_copy =
+    foreign "atg_view_as_real_copy" (gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_view_as_real_copy_out =
+    foreign "atg_view_as_real_copy_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_view_copy =
+    foreign "atg_view_copy" (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_view_copy_dtype =
+    foreign "atg_view_copy_dtype" (gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_view_copy_dtype_out =
+    foreign
+      "atg_view_copy_dtype_out"
+      (gc_tensor @-> gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_view_copy_out =
+    foreign
+      "atg_view_copy_out"
+      (gc_tensor @-> gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_view_dtype =
+    foreign "atg_view_dtype" (gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_vsplit =
+    foreign "atg_vsplit" (gc_tensor @-> int64_t @-> returning (ptr raw_tensor))
+  ;;
+
+  let stubs_vsplit_array =
+    foreign
+      "atg_vsplit_array"
+      (gc_tensor @-> ptr int64_t @-> int @-> returning (ptr raw_tensor))
+  ;;
+
+  let stubs_vstack = foreign "atg_vstack" (ptr gc_tensor @-> int @-> returning raw_tensor)
+
+  let stubs_vstack_out =
+    foreign "atg_vstack_out" (gc_tensor @-> ptr gc_tensor @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_where = foreign "atg_where" (gc_tensor @-> returning (ptr raw_tensor))
+
+  let stubs_where_scalar =
+    foreign "atg_where_scalar" (gc_tensor @-> scalar @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_where_scalarother =
+    foreign
+      "atg_where_scalarother"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_where_scalarself =
+    foreign
+      "atg_where_scalarself"
+      (gc_tensor @-> scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_where_self =
+    foreign
+      "atg_where_self"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_where_self_out =
+    foreign
+      "atg_where_self_out"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_xlogy = foreign "atg_xlogy" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+
+  let stubs_xlogy_ =
+    foreign "atg_xlogy_" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_xlogy_outscalar_other =
+    foreign
+      "atg_xlogy_outscalar_other"
+      (gc_tensor @-> gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_xlogy_outscalar_self =
+    foreign
+      "atg_xlogy_outscalar_self"
+      (gc_tensor @-> scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_xlogy_outtensor =
+    foreign
+      "atg_xlogy_outtensor"
+      (gc_tensor @-> gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_xlogy_scalar_other =
+    foreign "atg_xlogy_scalar_other" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_xlogy_scalar_other_ =
+    foreign "atg_xlogy_scalar_other_" (gc_tensor @-> scalar @-> returning raw_tensor)
+  ;;
+
+  let stubs_xlogy_scalar_self =
+    foreign "atg_xlogy_scalar_self" (scalar @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_zero = foreign "atg_zero" (gc_tensor @-> returning raw_tensor)
+  let stubs_zero_ = foreign "atg_zero_" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_zero_out =
+    foreign "atg_zero_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_zeros =
+    foreign "atg_zeros" (ptr int64_t @-> int @-> int @-> int @-> returning raw_tensor)
+  ;;
+
+  let stubs_zeros_like = foreign "atg_zeros_like" (gc_tensor @-> returning raw_tensor)
+
+  let stubs_zeros_like_out =
+    foreign "atg_zeros_like_out" (gc_tensor @-> gc_tensor @-> returning raw_tensor)
+  ;;
+
+  let stubs_zeros_out =
+    foreign "atg_zeros_out" (gc_tensor @-> ptr int64_t @-> int @-> returning raw_tensor)
+  ;;
+end
+
+module C (F : Cstubs.FOREIGN) = struct
+  include C0 (F)
+  include C1 (F)
+  include C2 (F)
+  include C3 (F)
+  include C4 (F)
+  include C5 (F)
+  include C6 (F)
+  include C7 (F)
+  include C8 (F)
+  include C9 (F)
+  include C10 (F)
+  include C11 (F)
+  include C12 (F)
+  include C13 (F)
+  include C14 (F)
+  include C15 (F)
+  include C16 (F)
+  include C17 (F)
+  include C18 (F)
+  include C19 (F)
+  include C20 (F)
+  include C21 (F)
+  include C22 (F)
+  include C23 (F)
+  include C24 (F)
+  include C25 (F)
+end
diff --git a/src/bindings/type_defs.ml b/src/bindings/type_defs.ml
new file mode 100644
index 0000000..0fdcb48
--- /dev/null
+++ b/src/bindings/type_defs.ml
@@ -0,0 +1,23 @@
+open Ctypes
+
+type raw_tensor = unit ptr
+type gc_tensor = unit ptr
+type ivalue = unit ptr
+type module_ = unit ptr
+type optimizer = unit ptr
+type scalar = unit ptr
+
+let raw_tensor : raw_tensor typ = ptr void
+let gc_tensor : gc_tensor typ = ptr void
+let ivalue : ivalue typ = ptr void
+let module_ : module_ typ = ptr void
+let optimizer : optimizer typ = ptr void
+let scalar : scalar typ = ptr void
+let none_gc_tensor = null
+let gc_tensor_of_voidp t = Cstubs_internals.make_ptr void t
+let is_none_raw_tensor t = is_null t
+
+let fatptr_of_raw_tensor (t : raw_tensor) =
+  let (CPointer fatptr) = t in
+  fatptr
+;;
diff --git a/src/bindings/type_defs.mli b/src/bindings/type_defs.mli
new file mode 100644
index 0000000..e95e576
--- /dev/null
+++ b/src/bindings/type_defs.mli
@@ -0,0 +1,19 @@
+open Ctypes
+
+type raw_tensor
+type gc_tensor
+type ivalue
+type module_
+type optimizer
+type scalar
+
+val raw_tensor : raw_tensor typ
+val gc_tensor : gc_tensor typ
+val ivalue : ivalue typ
+val module_ : module_ typ
+val optimizer : optimizer typ
+val scalar : scalar typ
+val none_gc_tensor : gc_tensor
+val gc_tensor_of_voidp : Ctypes_ptr.voidp -> gc_tensor
+val is_none_raw_tensor : raw_tensor -> bool
+val fatptr_of_raw_tensor : raw_tensor -> (Obj.t option, unit) Cstubs_internals.fatptr
diff --git a/src/gen_bindings/dune b/src/gen_bindings/dune
new file mode 100644
index 0000000..ff8bc33
--- /dev/null
+++ b/src/gen_bindings/dune
@@ -0,0 +1,6 @@
+(executables
+ (modes byte exe)
+ (names gen)
+ (libraries core.command core_unix.command_unix stdio yaml)
+ (preprocess
+  (pps ppx_let ppx_string)))
diff --git a/src/gen/gen.ml b/src/gen_bindings/gen.ml
similarity index 83%
rename from src/gen/gen.ml
rename to src/gen_bindings/gen.ml
index ac86e84..7f3412b 100644
--- a/src/gen/gen.ml
+++ b/src/gen_bindings/gen.ml
@@ -4,6 +4,10 @@
 open Base
 open Stdio
 
+let cpp_filename = "torch_api_generated"
+let bindings_filename = "torch_bindings_generated.ml"
+let wrapper_filename = "wrapper_generated"
+
 let excluded_functions =
   Set.of_list
     (module String)
@@ -193,7 +197,7 @@ module Func = struct
         Printf.sprintf "int64_t *%s_data, int %s_len" arg_name arg_name
       | DoubleList -> Printf.sprintf "double *%s_data, int %s_len" arg_name arg_name
       | TensorOptList | TensorList ->
-        Printf.sprintf "tensor *%s_data, int %s_len" arg_name arg_name
+        Printf.sprintf "gc_tensor *%s_data, int %s_len" arg_name arg_name
       | TensorOptions -> Printf.sprintf "int %s_kind, int %s_device" arg_name arg_name
       | Int64Option -> Printf.sprintf "int64_t %s_v, int %s_null" arg_name arg_name
       | DoubleOption -> Printf.sprintf "double %s_v, int %s_null" arg_name arg_name
@@ -203,8 +207,7 @@ module Func = struct
           | Bool -> "int"
           | Int64 -> "int64_t"
           | Double -> "double"
-          | Tensor -> "tensor"
-          | TensorOption -> "tensor"
+          | Tensor | TensorOption -> "gc_tensor"
           | ScalarType -> "int"
           | Device -> "int"
           | Scalar -> "scalar"
@@ -225,8 +228,10 @@ module Func = struct
   let c_args_list args =
     List.map args ~f:(fun { arg_name; arg_type; _ } ->
       match arg_type with
-      | Scalar | Tensor -> "*" ^ arg_name
-      | TensorOption -> [%string "(%{arg_name} ? *%{arg_name} : torch::Tensor())"]
+      | Scalar -> "*" ^ arg_name
+      | Tensor -> [%string "*tensor_ptr_from_ocaml(%{arg_name})"]
+      | TensorOption ->
+        [%string "(%{arg_name} ? tensor_from_ocaml(%{arg_name}) : torch::Tensor())"]
       | Bool -> "(bool)" ^ arg_name
       | IntList -> [%string "torch::IntArrayRef(%{arg_name}_data, %{arg_name}_len)"]
       | IntListOption ->
@@ -258,7 +263,13 @@ module Func = struct
     | `function_ -> [%string "torch::%{t.name}(%{c_args_list t.args})"]
     | `method_ ->
       (match t.args with
-       | head :: tail -> [%string "%{head.arg_name}->%{t.name}(%{c_args_list tail})"]
+       | head :: tail ->
+         let obj =
+           match head.arg_type with
+           | Tensor -> [%string "tensor_ptr_from_ocaml(%{head.arg_name})"]
+           | _ -> head.arg_name
+         in
+         [%string "%{obj}->%{t.name}(%{c_args_list tail})"]
        | [] ->
          failwith [%string "Method calls should have at least one argument %{t.name}"])
   ;;
@@ -273,7 +284,7 @@ module Func = struct
     | other -> other
   ;;
 
-  let stubs_signature t =
+  let bindings_signature t =
     let args =
       List.concat_map t.args ~f:(fun arg ->
         match arg.arg_type with
@@ -282,30 +293,30 @@ module Func = struct
         | Int64Option -> [ "int64_t"; "int" ]
         | Double -> [ "double" ]
         | DoubleOption -> [ "double"; "int" ]
-        | Tensor -> [ "t" ]
-        | TensorOption -> [ "t" ]
+        | Tensor | TensorOption -> [ "gc_tensor" ]
         | TensorOptions -> [ "int"; "int" ]
         | ScalarType -> [ "int" ]
         | Device -> [ "int" ]
         | IntList | IntListOption -> [ "ptr int64_t"; "int" ]
         | DoubleList -> [ "ptr double"; "int" ]
-        | TensorOptList | TensorList -> [ "ptr t"; "int" ]
+        | TensorOptList | TensorList -> [ "ptr gc_tensor"; "int" ]
         | String -> [ "string" ]
         | Scalar -> [ "scalar" ])
       |> String.concat ~sep:" @-> "
     in
-    let simple_stub args return_type =
+    let simple_binding args return_type =
       if String.length args > 0
       then [%string "%{args} @-> returning %{return_type}"]
       else [%string "void @-> returning %{return_type}"]
     in
     match t.returns with
-    | `fixed _ -> [%string "ptr t @-> %{args} @-> returning void"]
-    | `dynamic -> [%string "%{args} @-> returning (ptr t)"]
-    | `nothing -> simple_stub args "void"
-    | `bool -> simple_stub args "bool"
-    | `int64_t -> simple_stub args "int64_t"
-    | `double -> simple_stub args "double"
+    | `fixed 1 -> [%string "%{args} @-> returning raw_tensor"]
+    | `fixed _ -> [%string "ptr raw_tensor @-> %{args} @-> returning void"]
+    | `dynamic -> [%string "%{args} @-> returning (ptr raw_tensor)"]
+    | `nothing -> simple_binding args "void"
+    | `bool -> simple_binding args "bool"
+    | `int64_t -> simple_binding args "int64_t"
+    | `double -> simple_binding args "double"
   ;;
 
   let replace_map =
@@ -343,11 +354,12 @@ module Func = struct
         [%string
           {|(%{name} |> CArray.of_list double |> CArray.start) (List.length %{name})|}]
       | TensorList ->
-        [%string "(CArray.of_list t %{name} |> CArray.start) (List.length %{name})"]
+        [%string
+          "(CArray.of_list gc_tensor %{name} |> CArray.start) (List.length %{name})"]
       | TensorOptList ->
         [%string
-          "(List.map (function Some x -> x | None -> null) %{name} |> CArray.of_list t \
-           |> CArray.start) (List.length %{name})"]
+          "(List.map (function Some x -> x | None -> none_gc_tensor) %{name} |> \
+           CArray.of_list gc_tensor |> CArray.start) (List.length %{name})"]
       | Bool -> [%string "(if %{name} then 1 else 0)"]
       | ScalarType -> [%string "(Kind.packed_to_int %{name})"]
       | TensorOptions ->
@@ -357,7 +369,8 @@ module Func = struct
         if String.( = ) name "reduction"
         then "(Reduction.to_int reduction |> Int64.of_int)"
         else [%string "(Int64.of_int %{name})"]
-      | TensorOption -> [%string "(match %{name} with | Some v -> v | None -> null)"]
+      | TensorOption ->
+        [%string "(match %{name} with | Some v -> v | None -> none_gc_tensor)"]
       | Double | String | Scalar | Tensor -> name)
     |> String.concat ~sep:" "
   ;;
@@ -466,7 +479,7 @@ let p out_channel s =
 ;;
 
 let write_cpp funcs filename =
-  Out_channel.with_file (filename ^ ".cpp.h") ~f:(fun out_cpp ->
+  Out_channel.with_file (filename ^ ".cpp") ~f:(fun out_cpp ->
     Out_channel.with_file (filename ^ ".h") ~f:(fun out_h ->
       let pc s = p out_cpp s in
       let ph s = p out_h s in
@@ -478,22 +491,20 @@ let write_cpp funcs filename =
         let c_typed_args_list = Func.c_typed_args_list func in
         match func.returns with
         | `dynamic ->
-          pc "tensor *atg_%s(%s) {" exported_name c_typed_args_list;
+          pc "raw_tensor *atg_%s(%s) {" exported_name c_typed_args_list;
           pc "  PROTECT(";
-          pc "    auto outputs__ = %s;" (Func.c_call func);
+          pc "    auto results__ = %s;" (Func.c_call func);
           (* the returned type is a C++ vector of tensors *)
-          pc "    int sz = outputs__.size();";
-          pc
-            "    torch::Tensor **out__ = (torch::Tensor**)malloc((sz + 1) * \
-             sizeof(torch::Tensor*));";
+          pc "    int sz = results__.size();";
+          pc "    raw_tensor *out__ = (raw_tensor*)malloc((sz + 1) * sizeof(raw_tensor));";
           pc "    for (int i = 0; i < sz; ++i)";
-          pc "      out__[i] = new torch::Tensor(outputs__[i]);";
+          pc "      out__[i] = tensor_to_ocaml(results__[i]);";
           pc "    out__[sz] = nullptr;";
           pc "    return out__;";
           pc "  )";
           pc "}";
           pc "";
-          ph "tensor *atg_%s(%s);" exported_name c_typed_args_list
+          ph "raw_tensor *atg_%s(%s);" exported_name c_typed_args_list
         | `nothing ->
           pc "void atg_%s(%s) {" exported_name c_typed_args_list;
           pc "  PROTECT(";
@@ -502,20 +513,26 @@ let write_cpp funcs filename =
           pc "}";
           pc "";
           ph "void atg_%s(%s);" exported_name c_typed_args_list
+        | `fixed 1 ->
+          pc "raw_tensor atg_%s(%s) {" exported_name c_typed_args_list;
+          pc "  PROTECT(";
+          pc "    torch::Tensor results__ = %s;" (Func.c_call func);
+          pc "    return tensor_to_ocaml(results__);";
+          pc "  )";
+          pc "}";
+          pc "";
+          ph "raw_tensor atg_%s(%s);" exported_name c_typed_args_list
         | `fixed ntensors ->
-          pc "void atg_%s(tensor *out__, %s) {" exported_name c_typed_args_list;
+          pc "void atg_%s(raw_tensor *out__, %s) {" exported_name c_typed_args_list;
           pc "  PROTECT(";
-          pc "    auto outputs__ = %s;" (Func.c_call func);
-          if ntensors = 1
-          then pc "    out__[0] = new torch::Tensor(outputs__);"
-          else
-            for i = 0 to ntensors - 1 do
-              pc "    out__[%d] = new torch::Tensor(std::get<%d>(outputs__));" i i
-            done;
+          pc "    auto results__ = %s;" (Func.c_call func);
+          for i = 0 to ntensors - 1 do
+            pc "    out__[%d] = tensor_to_ocaml(std::get<%d>(results__));" i i
+          done;
           pc "  )";
           pc "}";
           pc "";
-          ph "void atg_%s(tensor *, %s);" exported_name c_typed_args_list
+          ph "void atg_%s(raw_tensor *, %s);" exported_name c_typed_args_list
         | (`bool | `int64_t | `double) as returns ->
           let c_type =
             match returns with
@@ -533,7 +550,7 @@ let write_cpp funcs filename =
           ph "%s atg_%s(%s);" c_type exported_name c_typed_args_list)))
 ;;
 
-let write_stubs funcs filename =
+let write_bindings funcs filename =
   Out_channel.with_file filename ~f:(fun out_channel ->
     let p s = p out_channel s in
     p "(* THIS FILE IS AUTOMATICALLY GENERATED, DO NOT EDIT BY HAND! *)";
@@ -544,14 +561,11 @@ let write_stubs funcs filename =
     List.iteri funcs ~f:(fun idx funcs ->
       p "module C%d(F: Cstubs.FOREIGN) = struct" idx;
       p "  open F";
-      p "  type t = unit ptr";
-      p "  let t : t typ = ptr void";
-      p "  type scalar = unit ptr";
-      p "  let scalar : scalar typ = ptr void";
+      p "  open Type_defs";
       List.iter funcs ~f:(fun (exported_name, func) ->
         p "  let stubs_%s =" (Func.caml_name exported_name);
         p "    foreign \"atg_%s\"" exported_name;
-        p "    (%s)" (Func.stubs_signature func);
+        p "    (%s)" (Func.bindings_signature func);
         p "");
       p "end");
     p "module C(F: Cstubs.FOREIGN) = struct";
@@ -567,23 +581,9 @@ let write_wrapper funcs filename =
       pm "(* THIS FILE IS AUTOMATICALLY GENERATED, DO NOT EDIT BY HAND! *)";
       pm "";
       pm "open Ctypes";
-      pm "";
-      pm "module C = Torch_bindings.C(Torch_generated)";
-      pm "open C.TensorG";
-      pm "";
-      pm "let to_tensor_list ptr =";
-      pm "  let rec loop ptr acc =";
-      pm "    let tensor = !@ptr in";
-      pm "    if is_null tensor";
-      pm "    then acc";
-      pm "    else begin";
-      pm "      Gc.finalise C.Tensor.free tensor;";
-      pm "      loop (ptr +@ 1) (tensor :: acc)";
-      pm "    end";
-      pm "  in";
-      pm "  let result = loop ptr [] in";
-      pm "  C.free (to_voidp ptr);";
-      pm "  List.rev result";
+      pm "open Torch_bindings.Type_defs";
+      pm "open Torch_stubs";
+      pm "open C.Generated";
       pm "";
       pi "(* THIS FILE IS AUTOMATICALLY GENERATED, DO NOT EDIT BY HAND! *)";
       pi "";
@@ -597,15 +597,16 @@ let write_wrapper funcs filename =
         (match func.returns with
          | `nothing | `bool | `int64_t | `double ->
            pm "  stubs_%s %s" caml_name (Func.caml_binding_args func)
+         | `fixed 1 ->
+           pm "  stubs_%s %s |> with_tensor_gc" caml_name (Func.caml_binding_args func)
          | `fixed ntensors ->
-           pm "  let out__ = CArray.make t %d in" ntensors;
+           pm "  let out__ = CArray.make raw_tensor %d in" ntensors;
            pm
              "  stubs_%s (CArray.start out__) %s;"
              caml_name
              (Func.caml_binding_args func);
            for i = 0 to ntensors - 1 do
-             pm "  let t%d = CArray.get out__ %d in" i i;
-             pm "  Gc.finalise C.Tensor.free t%d;" i
+             pm "  let t%d = CArray.get out__ %d |> with_tensor_gc in" i i
            done;
            pm
              "  %s"
@@ -659,8 +660,8 @@ let methods =
   ]
 ;;
 
-let run ~yaml_filename ~cpp_filename ~stubs_filename ~wrapper_filename =
-  let funcs = read_yaml yaml_filename in
+let run ~declarations_filename ~gen_bindings ~gen_wrappers =
+  let funcs = read_yaml declarations_filename in
   let funcs = methods @ funcs in
   printf "Generating code for %d functions.\n%!" (List.length funcs);
   (* Generate some unique names for overloaded functions. *)
@@ -694,15 +695,24 @@ let run ~yaml_filename ~cpp_filename ~stubs_filename ~wrapper_filename =
                 name, func))
     |> Map.of_alist_exn (module String)
   in
-  write_cpp funcs cpp_filename;
-  write_stubs funcs stubs_filename;
-  write_wrapper funcs wrapper_filename
+  if gen_bindings then write_bindings funcs bindings_filename;
+  if gen_wrappers
+  then (
+    write_cpp funcs cpp_filename;
+    write_wrapper funcs wrapper_filename)
 ;;
 
-let () =
-  run
-    ~yaml_filename:"third_party/pytorch/Declarations-v2.1.0.yaml"
-    ~cpp_filename:"src/wrapper/torch_api_generated"
-    ~stubs_filename:"src/stubs/torch_bindings_generated.ml"
-    ~wrapper_filename:"src/wrapper/wrapper_generated"
+let command =
+  Command.basic
+    ~summary:"generate bindings or wrapper code for toch functions"
+    (let%map_open.Command declarations_filename =
+       flag "declarations" (required string) ~doc:"PATH path to Declarations.yaml"
+     and gen_bindings =
+       flag "bindings" no_arg ~doc:"if passed in, generate ctypes bindings OCaml code"
+     and gen_wrappers =
+       flag "wrappers" no_arg ~doc:"if passed in, generate wrapper C++ and OCaml code"
+     in
+     fun () -> run ~declarations_filename ~gen_bindings ~gen_wrappers)
 ;;
+
+let () = Command_unix.run command
diff --git a/src/gen/dune b/src/gen_stubs/dune
similarity index 52%
rename from src/gen/dune
rename to src/gen_stubs/dune
index 41d2ddf..7d0e753 100644
--- a/src/gen/dune
+++ b/src/gen_stubs/dune
@@ -1,6 +1,6 @@
 (executables
  (modes byte exe)
- (names gen)
- (libraries base stdio yaml)
+ (names gen_stubs)
+ (libraries ctypes.stubs torch_bindings)
  (preprocess
   (pps ppx_string)))
diff --git a/src/stubs/torch_gen.ml b/src/gen_stubs/gen_stubs.ml
similarity index 74%
rename from src/stubs/torch_gen.ml
rename to src/gen_stubs/gen_stubs.ml
index f8d4d17..2a3067e 100644
--- a/src/stubs/torch_gen.ml
+++ b/src/gen_stubs/gen_stubs.ml
@@ -1,9 +1,9 @@
 let () =
   let fmt file = Format.formatter_of_out_channel (open_out file) in
-  let fmt_c = fmt "torch_stubs.c" in
+  let fmt_c = fmt "torch_stubs_generated.c" in
   Format.fprintf fmt_c "#include \"torch_api.h\"@.";
   Cstubs.write_c fmt_c ~prefix:"caml_" (module Torch_bindings.C);
-  let fmt_ml = fmt "torch_generated.ml" in
+  let fmt_ml = fmt "torch_stubs_generated.ml" in
   Cstubs.write_ml fmt_ml ~prefix:"caml_" (module Torch_bindings.C);
   flush_all ()
 ;;
diff --git a/src/stubs/dune b/src/stubs/dune
deleted file mode 100644
index e3dc775..0000000
--- a/src/stubs/dune
+++ /dev/null
@@ -1,6 +0,0 @@
-(executables
- (modes byte exe)
- (names torch_gen)
- (libraries ctypes.stubs)
- (preprocess
-  (pps ppx_jane)))
diff --git a/src/stubs/torch_bindings.ml b/src/stubs/torch_bindings.ml
deleted file mode 100644
index b854d7b..0000000
--- a/src/stubs/torch_bindings.ml
+++ /dev/null
@@ -1,269 +0,0 @@
-open Ctypes
-
-module C (F : Cstubs.FOREIGN) = struct
-  open F
-
-  let manual_seed = foreign "at_manual_seed" (int64_t @-> returning void)
-  let free = foreign "free" (ptr void @-> returning void)
-  let get_num_threads = foreign "at_get_num_threads" (void @-> returning int)
-  let set_num_threads = foreign "at_set_num_threads" (int @-> returning void)
-
-  module Tensor = struct
-    type t = unit ptr
-
-    let t : t typ = ptr void
-    let new_tensor = foreign "at_new_tensor" (void @-> returning t)
-
-    let tensor_of_data =
-      foreign
-        "at_tensor_of_data"
-        (ptr void
-         (* data *)
-         @-> ptr int64_t
-         (* dims *)
-         @-> int
-         (* ndims *)
-         @-> int
-         (* element size in bytes *)
-         @-> int
-         (* kind *)
-         @-> returning t)
-    ;;
-
-    let copy_data =
-      foreign
-        "at_copy_data"
-        (t
-         (* tensor *)
-         @-> ptr void
-         (* data *)
-         @-> int64_t
-         (* numel *)
-         @-> int
-         (* element size in bytes *)
-         @-> returning void)
-    ;;
-
-    let copy_ = foreign "at_copy_" (t (* dst *) @-> t (* src *) @-> returning void)
-    let set_data = foreign "at_set_data" (t (* dst *) @-> t (* src *) @-> returning void)
-
-    let float_vec =
-      foreign
-        "at_float_vec"
-        (ptr double (* values *)
-         @-> int (* num values *)
-         @-> int (* kind *)
-         @-> returning t)
-    ;;
-
-    let int_vec =
-      foreign
-        "at_int_vec"
-        (ptr int64_t
-         (* values *)
-         @-> int
-         (* num values *)
-         @-> int
-         (* kind *)
-         @-> returning t)
-    ;;
-
-    let device = foreign "at_device" (t @-> returning int)
-    let defined = foreign "at_defined" (t @-> returning bool)
-    let num_dims = foreign "at_dim" (t @-> returning int)
-    let shape = foreign "at_shape" (t @-> ptr int (* dims *) @-> returning void)
-    let scalar_type = foreign "at_scalar_type" (t @-> returning int)
-    let backward = foreign "at_backward" (t @-> int @-> int @-> returning void)
-    let requires_grad = foreign "at_requires_grad" (t @-> returning int)
-    let grad_set_enabled = foreign "at_grad_set_enabled" (int @-> returning int)
-    let get = foreign "at_get" (t @-> int @-> returning t)
-
-    let double_value =
-      foreign "at_double_value_at_indexes" (t @-> ptr int @-> int @-> returning float)
-    ;;
-
-    let int64_value =
-      foreign "at_int64_value_at_indexes" (t @-> ptr int @-> int @-> returning int64_t)
-    ;;
-
-    let double_value_set =
-      foreign
-        "at_set_double_value_at_indexes"
-        (t @-> ptr int @-> int @-> float @-> returning void)
-    ;;
-
-    let int64_value_set =
-      foreign
-        "at_set_int64_value_at_indexes"
-        (t @-> ptr int @-> int @-> int64_t @-> returning void)
-    ;;
-
-    let fill_double = foreign "at_fill_double" (t @-> float @-> returning void)
-    let fill_int64 = foreign "at_fill_int64" (t @-> int64_t @-> returning void)
-    let print = foreign "at_print" (t @-> returning void)
-    let to_string = foreign "at_to_string" (t @-> int @-> returning string)
-    let free = foreign "at_free" (t @-> returning void)
-
-    let run_backward =
-      foreign
-        "at_run_backward"
-        (ptr t @-> int @-> ptr t @-> int @-> ptr t @-> int @-> int @-> returning void)
-    ;;
-  end
-
-  module Scalar = struct
-    type t = unit ptr
-
-    let t : t typ = ptr void
-    let int = foreign "ats_int" (int64_t @-> returning t)
-    let float = foreign "ats_float" (float @-> returning t)
-    let free = foreign "ats_free" (t @-> returning void)
-  end
-
-  module Serialize = struct
-    let t = Tensor.t
-    let save = foreign "at_save" (t @-> string @-> returning void)
-    let load = foreign "at_load" (string @-> returning t)
-
-    let save_multi =
-      foreign
-        "at_save_multi"
-        (ptr t @-> ptr (ptr char) @-> int @-> string @-> returning void)
-    ;;
-
-    let load_multi =
-      foreign
-        "at_load_multi"
-        (ptr t @-> ptr (ptr char) @-> int @-> string @-> returning void)
-    ;;
-
-    let load_multi_ =
-      foreign
-        "at_load_multi_"
-        (ptr t @-> ptr (ptr char) @-> int @-> string @-> returning void)
-    ;;
-
-    let load_callback =
-      foreign
-        "at_load_callback"
-        (string
-         @-> static_funptr Ctypes.(string @-> t @-> returning void)
-         @-> returning void)
-    ;;
-  end
-
-  module Optimizer = struct
-    type t = unit ptr
-
-    let t : t typ = ptr void
-
-    let adam =
-      foreign "ato_adam" (float @-> float @-> float @-> float @-> float @-> returning t)
-    ;;
-
-    let rmsprop =
-      foreign
-        "ato_rmsprop"
-        (float
-         (* learning rate *)
-         @-> float
-         (* alpha *)
-         @-> float
-         (* eps *)
-         @-> float
-         (* weight decay *)
-         @-> float
-         (* momentum *)
-         @-> int
-         (* centered *)
-         @-> returning t)
-    ;;
-
-    let sgd =
-      foreign
-        "ato_sgd"
-        (float
-         (* learning rate *)
-         @-> float
-         (* momentum *)
-         @-> float
-         (* dampening *)
-         @-> float
-         (* weight decay *)
-         @-> bool
-         (* nesterov *)
-         @-> returning t)
-    ;;
-
-    let add_parameters =
-      foreign "ato_add_parameters" (t @-> ptr Tensor.t @-> int @-> returning void)
-    ;;
-
-    let set_learning_rate =
-      foreign "ato_set_learning_rate" (t @-> float @-> returning void)
-    ;;
-
-    let set_momentum = foreign "ato_set_momentum" (t @-> float @-> returning void)
-    let zero_grad = foreign "ato_zero_grad" (t @-> returning void)
-    let step = foreign "ato_step" (t @-> returning void)
-    let free = foreign "ato_free" (t @-> returning void)
-  end
-
-  module Cuda = struct
-    let device_count = foreign "atc_cuda_device_count" (void @-> returning int)
-    let is_available = foreign "atc_cuda_is_available" (void @-> returning int)
-    let cudnn_is_available = foreign "atc_cudnn_is_available" (void @-> returning int)
-    let set_benchmark_cudnn = foreign "atc_set_benchmark_cudnn" (int @-> returning void)
-  end
-
-  module Ivalue = struct
-    type t = unit ptr
-
-    let t : t typ = ptr void
-    let to_int64 = foreign "ati_to_int" (t @-> returning int64_t)
-    let to_bool = foreign "ati_to_bool" (t @-> returning int)
-    let to_double = foreign "ati_to_double" (t @-> returning double)
-    let to_tensor = foreign "ati_to_tensor" (t @-> returning Tensor.t)
-    let tuple_length = foreign "ati_tuple_length" (t @-> returning int)
-    let list_length = foreign "ati_list_length" (t @-> returning int)
-    let to_tuple = foreign "ati_to_tuple" (t @-> ptr t @-> int @-> returning void)
-
-    let to_tensor_list =
-      foreign "ati_to_tensor_list" (t @-> ptr Tensor.t @-> int @-> returning void)
-    ;;
-
-    let to_generic_list =
-      foreign "ati_to_generic_list" (t @-> ptr t @-> int @-> returning void)
-    ;;
-
-    let to_string = foreign "ati_to_string" (t @-> returning string)
-    let none = foreign "ati_none" (void @-> returning t)
-    let bool = foreign "ati_bool" (int @-> returning t)
-    let tensor = foreign "ati_tensor" (Tensor.t @-> returning t)
-    let int64 = foreign "ati_int" (int64_t @-> returning t)
-    let double = foreign "ati_double" (float @-> returning t)
-    let tuple = foreign "ati_tuple" (ptr t @-> int @-> returning t)
-    let tensor_list = foreign "ati_tensor_list" (ptr Tensor.t @-> int @-> returning t)
-    let string = foreign "ati_string" (string @-> returning t)
-    let tag = foreign "ati_tag" (t @-> returning int)
-    let free = foreign "ati_free" (t @-> returning void)
-  end
-
-  module Module = struct
-    type t = unit ptr
-
-    let t : t typ = ptr void
-    let load = foreign "atm_load" (string @-> int @-> returning t)
-    let load_str = foreign "atm_load_str" (string @-> int @-> int @-> returning t)
-    let forward = foreign "atm_forward" (t @-> ptr Tensor.t @-> int @-> returning Tensor.t)
-
-    let forward_ =
-      foreign "atm_forward_" (t @-> ptr Ivalue.t @-> int @-> returning Ivalue.t)
-    ;;
-
-    let named_buffers = foreign "atm_named_buffers" (t @-> returning Ivalue.t)
-    let free = foreign "atm_free" (t @-> returning void)
-  end
-
-  module TensorG = Torch_bindings_generated.C (F)
-end
diff --git a/src/stubs/torch_bindings_generated.ml b/src/stubs/torch_bindings_generated.ml
deleted file mode 100644
index ecbf933..0000000
--- a/src/stubs/torch_bindings_generated.ml
+++ /dev/null
@@ -1,15190 +0,0 @@
-(* THIS FILE IS AUTOMATICALLY GENERATED, DO NOT EDIT BY HAND! *)
-
-open Ctypes
-
-module C0 (F : Cstubs.FOREIGN) = struct
-  open F
-
-  type t = unit ptr
-
-  let t : t typ = ptr void
-
-  type scalar = unit ptr
-
-  let scalar : scalar typ = ptr void
-  let stubs___and__ = foreign "atg___and__" (ptr t @-> t @-> scalar @-> returning void)
-
-  let stubs___and__tensor_ =
-    foreign "atg___and__tensor_" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs___iand__ = foreign "atg___iand__" (ptr t @-> t @-> scalar @-> returning void)
-
-  let stubs___iand__tensor_ =
-    foreign "atg___iand__tensor_" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs___ilshift__ =
-    foreign "atg___ilshift__" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs___ilshift__tensor_ =
-    foreign "atg___ilshift__tensor_" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs___ior__ = foreign "atg___ior__" (ptr t @-> t @-> scalar @-> returning void)
-
-  let stubs___ior__tensor_ =
-    foreign "atg___ior__tensor_" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs___irshift__ =
-    foreign "atg___irshift__" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs___irshift__tensor_ =
-    foreign "atg___irshift__tensor_" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs___ixor__ = foreign "atg___ixor__" (ptr t @-> t @-> scalar @-> returning void)
-
-  let stubs___ixor__tensor_ =
-    foreign "atg___ixor__tensor_" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs___lshift__ =
-    foreign "atg___lshift__" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs___lshift__scalar_out_ =
-    foreign "atg___lshift__scalar_out_" (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs___lshift__tensor_ =
-    foreign "atg___lshift__tensor_" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs___lshift__tensor_out_ =
-    foreign "atg___lshift__tensor_out_" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs___or__ = foreign "atg___or__" (ptr t @-> t @-> scalar @-> returning void)
-
-  let stubs___or__tensor_ =
-    foreign "atg___or__tensor_" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs___rshift__ =
-    foreign "atg___rshift__" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs___rshift__scalar_out_ =
-    foreign "atg___rshift__scalar_out_" (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs___rshift__tensor_ =
-    foreign "atg___rshift__tensor_" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs___rshift__tensor_out_ =
-    foreign "atg___rshift__tensor_out_" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs___xor__ = foreign "atg___xor__" (ptr t @-> t @-> scalar @-> returning void)
-
-  let stubs___xor__tensor_ =
-    foreign "atg___xor__tensor_" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__adaptive_avg_pool2d =
-    foreign
-      "atg__adaptive_avg_pool2d"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__adaptive_avg_pool2d_backward =
-    foreign "atg__adaptive_avg_pool2d_backward" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__adaptive_avg_pool2d_backward_out =
-    foreign
-      "atg__adaptive_avg_pool2d_backward_out"
-      (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__adaptive_avg_pool2d_out =
-    foreign
-      "atg__adaptive_avg_pool2d_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__adaptive_avg_pool3d =
-    foreign
-      "atg__adaptive_avg_pool3d"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__adaptive_avg_pool3d_backward =
-    foreign "atg__adaptive_avg_pool3d_backward" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__adaptive_avg_pool3d_backward_out =
-    foreign
-      "atg__adaptive_avg_pool3d_backward_out"
-      (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__adaptive_avg_pool3d_out =
-    foreign
-      "atg__adaptive_avg_pool3d_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__add_batch_dim =
-    foreign "atg__add_batch_dim" (ptr t @-> t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__add_relu = foreign "atg__add_relu" (ptr t @-> t @-> t @-> returning void)
-  let stubs__add_relu_ = foreign "atg__add_relu_" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs__add_relu_out =
-    foreign "atg__add_relu_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__add_relu_scalar =
-    foreign "atg__add_relu_scalar" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs__add_relu_scalar_ =
-    foreign "atg__add_relu_scalar_" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs__add_relu_scalar_out =
-    foreign "atg__add_relu_scalar_out" (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs__addmm_activation =
-    foreign "atg__addmm_activation" (ptr t @-> t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs__addmm_activation_out =
-    foreign
-      "atg__addmm_activation_out"
-      (ptr t @-> t @-> t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs__aminmax = foreign "atg__aminmax" (ptr t @-> t @-> returning void)
-
-  let stubs__aminmax_dim =
-    foreign "atg__aminmax_dim" (ptr t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__aminmax_dim_out =
-    foreign
-      "atg__aminmax_dim_out"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__aminmax_out =
-    foreign "atg__aminmax_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__amp_update_scale =
-    foreign
-      "atg__amp_update_scale"
-      (ptr t @-> t @-> t @-> t @-> double @-> double @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__amp_update_scale_ =
-    foreign
-      "atg__amp_update_scale_"
-      (ptr t @-> t @-> t @-> t @-> double @-> double @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__amp_update_scale_out =
-    foreign
-      "atg__amp_update_scale_out"
-      (ptr t @-> t @-> t @-> t @-> t @-> double @-> double @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__assert_tensor_metadata =
-    foreign
-      "atg__assert_tensor_metadata"
-      (t @-> ptr int64_t @-> int @-> ptr int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs__autocast_to_full_precision =
-    foreign
-      "atg__autocast_to_full_precision"
-      (ptr t @-> t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs__autocast_to_reduced_precision =
-    foreign
-      "atg__autocast_to_reduced_precision"
-      (ptr t @-> t @-> int @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs__cast_byte = foreign "atg__cast_byte" (ptr t @-> t @-> int @-> returning void)
-  let stubs__cast_char = foreign "atg__cast_char" (ptr t @-> t @-> int @-> returning void)
-
-  let stubs__cast_double =
-    foreign "atg__cast_double" (ptr t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs__cast_float =
-    foreign "atg__cast_float" (ptr t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs__cast_half = foreign "atg__cast_half" (ptr t @-> t @-> int @-> returning void)
-  let stubs__cast_int = foreign "atg__cast_int" (ptr t @-> t @-> int @-> returning void)
-  let stubs__cast_long = foreign "atg__cast_long" (ptr t @-> t @-> int @-> returning void)
-
-  let stubs__cast_short =
-    foreign "atg__cast_short" (ptr t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs__cdist_backward =
-    foreign
-      "atg__cdist_backward"
-      (ptr t @-> t @-> t @-> t @-> double @-> t @-> returning void)
-  ;;
-
-  let stubs__cdist_backward_out =
-    foreign
-      "atg__cdist_backward_out"
-      (ptr t @-> t @-> t @-> t @-> t @-> double @-> t @-> returning void)
-  ;;
-
-  let stubs__cholesky_solve_helper =
-    foreign "atg__cholesky_solve_helper" (ptr t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs__cholesky_solve_helper_out =
-    foreign
-      "atg__cholesky_solve_helper_out"
-      (ptr t @-> t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs__coalesce = foreign "atg__coalesce" (ptr t @-> t @-> returning void)
-
-  let stubs__coalesce_out =
-    foreign "atg__coalesce_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__coalesced = foreign "atg__coalesced" (ptr t @-> t @-> int @-> returning void)
-
-  let stubs__coalesced_ =
-    foreign "atg__coalesced_" (ptr t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs__coalesced_out =
-    foreign "atg__coalesced_out" (ptr t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs__compute_linear_combination =
-    foreign "atg__compute_linear_combination" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__compute_linear_combination_out =
-    foreign
-      "atg__compute_linear_combination_out"
-      (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__conj = foreign "atg__conj" (ptr t @-> t @-> returning void)
-  let stubs__conj_copy = foreign "atg__conj_copy" (ptr t @-> t @-> returning void)
-
-  let stubs__conj_copy_out =
-    foreign "atg__conj_copy_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__conj_physical = foreign "atg__conj_physical" (ptr t @-> t @-> returning void)
-
-  let stubs__conj_physical_out =
-    foreign "atg__conj_physical_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__conv_depthwise2d =
-    foreign
-      "atg__conv_depthwise2d"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__conv_depthwise2d_out =
-    foreign
-      "atg__conv_depthwise2d_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__convert_indices_from_coo_to_csr =
-    foreign
-      "atg__convert_indices_from_coo_to_csr"
-      (ptr t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__convert_indices_from_coo_to_csr_out =
-    foreign
-      "atg__convert_indices_from_coo_to_csr_out"
-      (ptr t @-> t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__convert_indices_from_csr_to_coo =
-    foreign
-      "atg__convert_indices_from_csr_to_coo"
-      (ptr t @-> t @-> t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs__convert_indices_from_csr_to_coo_out =
-    foreign
-      "atg__convert_indices_from_csr_to_coo_out"
-      (ptr t @-> t @-> t @-> t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs__convolution =
-    foreign
-      "atg__convolution"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> int
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__convolution_deprecated =
-    foreign
-      "atg__convolution_deprecated"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__convolution_mode =
-    foreign
-      "atg__convolution_mode"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> string
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs__convolution_out =
-    foreign
-      "atg__convolution_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> int
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__copy_from =
-    foreign "atg__copy_from" (ptr t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs__copy_from_and_resize =
-    foreign "atg__copy_from_and_resize" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__copy_from_and_resize_out =
-    foreign "atg__copy_from_and_resize_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__copy_from_out =
-    foreign "atg__copy_from_out" (ptr t @-> t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs__cslt_compress = foreign "atg__cslt_compress" (ptr t @-> t @-> returning void)
-
-  let stubs__cslt_sparse_mm =
-    foreign "atg__cslt_sparse_mm" (ptr t @-> t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs__ctc_loss =
-    foreign
-      "atg__ctc_loss"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__ctc_loss_backward =
-    foreign
-      "atg__ctc_loss_backward"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> t
-       @-> t
-       @-> int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__ctc_loss_backward_out =
-    foreign
-      "atg__ctc_loss_backward_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> t
-       @-> t
-       @-> int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__ctc_loss_backward_tensor =
-    foreign
-      "atg__ctc_loss_backward_tensor"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__ctc_loss_out =
-    foreign
-      "atg__ctc_loss_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__ctc_loss_tensor =
-    foreign
-      "atg__ctc_loss_tensor"
-      (ptr t @-> t @-> t @-> t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__ctc_loss_tensor_out =
-    foreign
-      "atg__ctc_loss_tensor_out"
-      (ptr t @-> t @-> t @-> t @-> t @-> t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__cudnn_ctc_loss =
-    foreign
-      "atg__cudnn_ctc_loss"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__cudnn_ctc_loss_out =
-    foreign
-      "atg__cudnn_ctc_loss_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-end
-
-module C1 (F : Cstubs.FOREIGN) = struct
-  open F
-
-  type t = unit ptr
-
-  let t : t typ = ptr void
-
-  type scalar = unit ptr
-
-  let scalar : scalar typ = ptr void
-
-  let stubs__cudnn_ctc_loss_tensor =
-    foreign
-      "atg__cudnn_ctc_loss_tensor"
-      (ptr t @-> t @-> t @-> t @-> t @-> int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs__cudnn_init_dropout_state =
-    foreign
-      "atg__cudnn_init_dropout_state"
-      (ptr t @-> double @-> int @-> int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs__cudnn_init_dropout_state_out =
-    foreign
-      "atg__cudnn_init_dropout_state_out"
-      (ptr t @-> t @-> double @-> int @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__cudnn_rnn =
-    foreign
-      "atg__cudnn_rnn"
-      (ptr t
-       @-> t
-       @-> ptr t
-       @-> int
-       @-> int64_t
-       @-> t
-       @-> t
-       @-> t
-       @-> int64_t
-       @-> int64_t
-       @-> int64_t
-       @-> int64_t
-       @-> int
-       @-> double
-       @-> int
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> t
-       @-> returning void)
-  ;;
-
-  let stubs__cudnn_rnn_flatten_weight =
-    foreign
-      "atg__cudnn_rnn_flatten_weight"
-      (ptr t
-       @-> ptr t
-       @-> int
-       @-> int64_t
-       @-> int64_t
-       @-> int64_t
-       @-> int64_t
-       @-> int64_t
-       @-> int64_t
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__cudnn_rnn_flatten_weight_out =
-    foreign
-      "atg__cudnn_rnn_flatten_weight_out"
-      (ptr t
-       @-> t
-       @-> ptr t
-       @-> int
-       @-> int64_t
-       @-> int64_t
-       @-> int64_t
-       @-> int64_t
-       @-> int64_t
-       @-> int64_t
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__cudnn_rnn_out =
-    foreign
-      "atg__cudnn_rnn_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr t
-       @-> int
-       @-> int64_t
-       @-> t
-       @-> t
-       @-> t
-       @-> int64_t
-       @-> int64_t
-       @-> int64_t
-       @-> int64_t
-       @-> int
-       @-> double
-       @-> int
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> t
-       @-> returning void)
-  ;;
-
-  let stubs__debug_has_internal_overlap =
-    foreign "atg__debug_has_internal_overlap" (t @-> returning int64_t)
-  ;;
-
-  let stubs__dim_arange =
-    foreign "atg__dim_arange" (ptr t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__dimi = foreign "atg__dimi" (t @-> returning int64_t)
-  let stubs__dimv = foreign "atg__dimv" (t @-> returning int64_t)
-
-  let stubs__dirichlet_grad =
-    foreign "atg__dirichlet_grad" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__dirichlet_grad_out =
-    foreign "atg__dirichlet_grad_out" (ptr t @-> t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__efficient_attention_backward =
-    foreign
-      "atg__efficient_attention_backward"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int64_t
-       @-> int64_t
-       @-> t
-       @-> double
-       @-> t
-       @-> t
-       @-> int64_t
-       @-> int
-       @-> double
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__efficientzerotensor =
-    foreign
-      "atg__efficientzerotensor"
-      (ptr t @-> ptr int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs__efficientzerotensor_out =
-    foreign
-      "atg__efficientzerotensor_out"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__embedding_bag =
-    foreign
-      "atg__embedding_bag"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> t
-       @-> int
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs__embedding_bag_backward =
-    foreign
-      "atg__embedding_bag_backward"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int64_t
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> t
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs__embedding_bag_dense_backward =
-    foreign
-      "atg__embedding_bag_dense_backward"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int64_t
-       @-> int
-       @-> int64_t
-       @-> t
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs__embedding_bag_dense_backward_out =
-    foreign
-      "atg__embedding_bag_dense_backward_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int64_t
-       @-> int
-       @-> int64_t
-       @-> t
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs__embedding_bag_forward_only =
-    foreign
-      "atg__embedding_bag_forward_only"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> t
-       @-> int
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs__embedding_bag_forward_only_out =
-    foreign
-      "atg__embedding_bag_forward_only_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> t
-       @-> int
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs__embedding_bag_out =
-    foreign
-      "atg__embedding_bag_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> t
-       @-> int
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs__embedding_bag_per_sample_weights_backward =
-    foreign
-      "atg__embedding_bag_per_sample_weights_backward"
-      (ptr t @-> t @-> t @-> t @-> t @-> t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__embedding_bag_per_sample_weights_backward_out =
-    foreign
-      "atg__embedding_bag_per_sample_weights_backward_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int64_t
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs__embedding_bag_sparse_backward =
-    foreign
-      "atg__embedding_bag_sparse_backward"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int64_t
-       @-> int
-       @-> int64_t
-       @-> t
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs__empty_affine_quantized =
-    foreign
-      "atg__empty_affine_quantized"
-      (ptr t
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> int
-       @-> double
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs__empty_affine_quantized_out =
-    foreign
-      "atg__empty_affine_quantized_out"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> double @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__empty_per_channel_affine_quantized =
-    foreign
-      "atg__empty_per_channel_affine_quantized"
-      (ptr t
-       @-> ptr int64_t
-       @-> int
-       @-> t
-       @-> t
-       @-> int64_t
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__empty_per_channel_affine_quantized_out =
-    foreign
-      "atg__empty_per_channel_affine_quantized_out"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__euclidean_dist =
-    foreign "atg__euclidean_dist" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__euclidean_dist_out =
-    foreign "atg__euclidean_dist_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__fake_quantize_learnable_per_channel_affine =
-    foreign
-      "atg__fake_quantize_learnable_per_channel_affine"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> int64_t
-       @-> int64_t
-       @-> int64_t
-       @-> double
-       @-> returning void)
-  ;;
-
-  let stubs__fake_quantize_learnable_per_channel_affine_backward =
-    foreign
-      "atg__fake_quantize_learnable_per_channel_affine_backward"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int64_t
-       @-> int64_t
-       @-> int64_t
-       @-> double
-       @-> returning void)
-  ;;
-
-  let stubs__fake_quantize_learnable_per_channel_affine_out =
-    foreign
-      "atg__fake_quantize_learnable_per_channel_affine_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int64_t
-       @-> int64_t
-       @-> int64_t
-       @-> double
-       @-> returning void)
-  ;;
-
-  let stubs__fake_quantize_learnable_per_tensor_affine =
-    foreign
-      "atg__fake_quantize_learnable_per_tensor_affine"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> int64_t @-> double @-> returning void)
-  ;;
-
-  let stubs__fake_quantize_learnable_per_tensor_affine_backward =
-    foreign
-      "atg__fake_quantize_learnable_per_tensor_affine_backward"
-      (ptr t @-> t @-> t @-> t @-> t @-> int64_t @-> int64_t @-> double @-> returning void)
-  ;;
-
-  let stubs__fake_quantize_learnable_per_tensor_affine_out =
-    foreign
-      "atg__fake_quantize_learnable_per_tensor_affine_out"
-      (ptr t @-> t @-> t @-> t @-> t @-> int64_t @-> int64_t @-> double @-> returning void)
-  ;;
-
-  let stubs__fake_quantize_per_tensor_affine_cachemask_tensor_qparams =
-    foreign
-      "atg__fake_quantize_per_tensor_affine_cachemask_tensor_qparams"
-      (ptr t @-> t @-> t @-> t @-> t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__fake_quantize_per_tensor_affine_cachemask_tensor_qparams_out =
-    foreign
-      "atg__fake_quantize_per_tensor_affine_cachemask_tensor_qparams_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int64_t
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs__fft_c2c =
-    foreign
-      "atg__fft_c2c"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__fft_c2c_out =
-    foreign
-      "atg__fft_c2c_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__fft_c2r =
-    foreign
-      "atg__fft_c2r"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__fft_c2r_out =
-    foreign
-      "atg__fft_c2r_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs__fft_r2c =
-    foreign
-      "atg__fft_r2c"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__fft_r2c_out =
-    foreign
-      "atg__fft_r2c_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__fill_mem_eff_dropout_mask_ =
-    foreign
-      "atg__fill_mem_eff_dropout_mask_"
-      (ptr t @-> t @-> double @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__flash_attention_backward =
-    foreign
-      "atg__flash_attention_backward"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int64_t
-       @-> int64_t
-       @-> double
-       @-> int
-       @-> t
-       @-> t
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__foobar =
-    foreign "atg__foobar" (ptr t @-> t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs__foobar_out =
-    foreign
-      "atg__foobar_out"
-      (ptr t @-> t @-> t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs__functional_assert_async =
-    foreign
-      "atg__functional_assert_async"
-      (ptr t @-> t @-> string @-> t @-> returning void)
-  ;;
-
-  let stubs__functional_sym_constrain_range =
-    foreign
-      "atg__functional_sym_constrain_range"
-      (ptr t @-> scalar @-> int64_t @-> int @-> int64_t @-> int @-> t @-> returning void)
-  ;;
-
-  let stubs__functional_sym_constrain_range_for_size =
-    foreign
-      "atg__functional_sym_constrain_range_for_size"
-      (ptr t @-> scalar @-> int64_t @-> int @-> int64_t @-> int @-> t @-> returning void)
-  ;;
-
-  let stubs__fused_adam =
-    foreign
-      "atg__fused_adam"
-      (ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> double
-       @-> double
-       @-> double
-       @-> double
-       @-> double
-       @-> int
-       @-> int
-       @-> t
-       @-> t
-       @-> returning void)
-  ;;
-
-  let stubs__fused_adam_ =
-    foreign
-      "atg__fused_adam_"
-      (ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> double
-       @-> double
-       @-> double
-       @-> double
-       @-> double
-       @-> int
-       @-> int
-       @-> t
-       @-> t
-       @-> returning void)
-  ;;
-
-  let stubs__fused_adam_tensor_lr_ =
-    foreign
-      "atg__fused_adam_tensor_lr_"
-      (ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> t
-       @-> double
-       @-> double
-       @-> double
-       @-> double
-       @-> int
-       @-> int
-       @-> t
-       @-> t
-       @-> returning void)
-  ;;
-
-  let stubs__fused_adam_tensor_lr_out =
-    foreign
-      "atg__fused_adam_tensor_lr_out"
-      (ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> t
-       @-> double
-       @-> double
-       @-> double
-       @-> double
-       @-> int
-       @-> int
-       @-> t
-       @-> t
-       @-> returning void)
-  ;;
-
-  let stubs__fused_adamw =
-    foreign
-      "atg__fused_adamw"
-      (ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> double
-       @-> double
-       @-> double
-       @-> double
-       @-> double
-       @-> int
-       @-> int
-       @-> t
-       @-> t
-       @-> returning void)
-  ;;
-
-  let stubs__fused_adamw_ =
-    foreign
-      "atg__fused_adamw_"
-      (ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> double
-       @-> double
-       @-> double
-       @-> double
-       @-> double
-       @-> int
-       @-> int
-       @-> t
-       @-> t
-       @-> returning void)
-  ;;
-
-  let stubs__fused_adamw_tensor_lr_ =
-    foreign
-      "atg__fused_adamw_tensor_lr_"
-      (ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> t
-       @-> double
-       @-> double
-       @-> double
-       @-> double
-       @-> int
-       @-> int
-       @-> t
-       @-> t
-       @-> returning void)
-  ;;
-
-  let stubs__fused_adamw_tensor_lr_out =
-    foreign
-      "atg__fused_adamw_tensor_lr_out"
-      (ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> t
-       @-> double
-       @-> double
-       @-> double
-       @-> double
-       @-> int
-       @-> int
-       @-> t
-       @-> t
-       @-> returning void)
-  ;;
-
-  let stubs__fused_dropout =
-    foreign "atg__fused_dropout" (ptr t @-> t @-> double @-> returning void)
-  ;;
-
-  let stubs__fused_dropout_out =
-    foreign
-      "atg__fused_dropout_out"
-      (ptr t @-> t @-> t @-> t @-> double @-> returning void)
-  ;;
-
-  let stubs__fused_moving_avg_obs_fq_helper =
-    foreign
-      "atg__fused_moving_avg_obs_fq_helper"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> double
-       @-> int64_t
-       @-> int64_t
-       @-> int64_t
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__fused_moving_avg_obs_fq_helper_functional =
-    foreign
-      "atg__fused_moving_avg_obs_fq_helper_functional"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> double
-       @-> int64_t
-       @-> int64_t
-       @-> int64_t
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__fused_moving_avg_obs_fq_helper_out =
-    foreign
-      "atg__fused_moving_avg_obs_fq_helper_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> double
-       @-> int64_t
-       @-> int64_t
-       @-> int64_t
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__fused_sdp_choice =
-    foreign
-      "atg__fused_sdp_choice"
-      (t @-> t @-> t @-> t @-> double @-> int @-> double @-> int @-> returning int64_t)
-  ;;
-
-  let stubs__fw_primal =
-    foreign "atg__fw_primal" (ptr t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__fw_primal_copy =
-    foreign "atg__fw_primal_copy" (ptr t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__fw_primal_copy_out =
-    foreign "atg__fw_primal_copy_out" (ptr t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__gather_sparse_backward =
-    foreign
-      "atg__gather_sparse_backward"
-      (ptr t @-> t @-> int64_t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__grid_sampler_2d_cpu_fallback =
-    foreign
-      "atg__grid_sampler_2d_cpu_fallback"
-      (ptr t @-> t @-> t @-> int64_t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__grid_sampler_2d_cpu_fallback_backward =
-    foreign
-      "atg__grid_sampler_2d_cpu_fallback_backward"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__grid_sampler_2d_cpu_fallback_out =
-    foreign
-      "atg__grid_sampler_2d_cpu_fallback_out"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__has_compatible_shallow_copy_type =
-    foreign "atg__has_compatible_shallow_copy_type" (t @-> t @-> returning bool)
-  ;;
-
-  let stubs__has_same_storage_numel =
-    foreign "atg__has_same_storage_numel" (t @-> t @-> returning bool)
-  ;;
-
-  let stubs__histogramdd_bin_edges =
-    foreign
-      "atg__histogramdd_bin_edges"
-      (t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr double
-       @-> int
-       @-> t
-       @-> int
-       @-> returning (ptr t))
-  ;;
-
-  let stubs__histogramdd_bin_edges_out =
-    foreign
-      "atg__histogramdd_bin_edges_out"
-      (ptr t
-       @-> int
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr double
-       @-> int
-       @-> t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__histogramdd_from_bin_cts =
-    foreign
-      "atg__histogramdd_from_bin_cts"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr double
-       @-> int
-       @-> t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__histogramdd_from_bin_cts_out =
-    foreign
-      "atg__histogramdd_from_bin_cts_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr double
-       @-> int
-       @-> t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__histogramdd_from_bin_tensors =
-    foreign
-      "atg__histogramdd_from_bin_tensors"
-      (ptr t @-> t @-> ptr t @-> int @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs__histogramdd_from_bin_tensors_out =
-    foreign
-      "atg__histogramdd_from_bin_tensors_out"
-      (ptr t @-> t @-> t @-> ptr t @-> int @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs__index_put_impl =
-    foreign
-      "atg__index_put_impl"
-      (ptr t @-> t @-> ptr t @-> int @-> t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs__index_put_impl_ =
-    foreign
-      "atg__index_put_impl_"
-      (ptr t @-> t @-> ptr t @-> int @-> t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs__index_put_impl_out =
-    foreign
-      "atg__index_put_impl_out"
-      (ptr t @-> t @-> t @-> ptr t @-> int @-> t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs__indices = foreign "atg__indices" (ptr t @-> t @-> returning void)
-  let stubs__indices_copy = foreign "atg__indices_copy" (ptr t @-> t @-> returning void)
-
-  let stubs__indices_copy_out =
-    foreign "atg__indices_copy_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__int_mm = foreign "atg__int_mm" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs__int_mm_out =
-    foreign "atg__int_mm_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__is_all_true = foreign "atg__is_all_true" (ptr t @-> t @-> returning void)
-  let stubs__is_any_true = foreign "atg__is_any_true" (ptr t @-> t @-> returning void)
-  let stubs__is_zerotensor = foreign "atg__is_zerotensor" (t @-> returning bool)
-
-  let stubs__linalg_check_errors =
-    foreign "atg__linalg_check_errors" (t @-> string @-> int @-> returning void)
-  ;;
-
-  let stubs__linalg_det = foreign "atg__linalg_det" (ptr t @-> t @-> returning void)
-
-  let stubs__linalg_det_result =
-    foreign "atg__linalg_det_result" (ptr t @-> t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__linalg_eigh =
-    foreign "atg__linalg_eigh" (ptr t @-> t @-> string @-> int @-> returning void)
-  ;;
-
-  let stubs__linalg_eigh_eigenvalues =
-    foreign
-      "atg__linalg_eigh_eigenvalues"
-      (ptr t @-> t @-> t @-> t @-> string @-> int @-> returning void)
-  ;;
-
-  let stubs__linalg_slogdet =
-    foreign "atg__linalg_slogdet" (ptr t @-> t @-> returning void)
-  ;;
-
-  let stubs__linalg_slogdet_sign =
-    foreign
-      "atg__linalg_slogdet_sign"
-      (ptr t @-> t @-> t @-> t @-> t @-> t @-> returning void)
-  ;;
-end
-
-module C2 (F : Cstubs.FOREIGN) = struct
-  open F
-
-  type t = unit ptr
-
-  let t : t typ = ptr void
-
-  type scalar = unit ptr
-
-  let scalar : scalar typ = ptr void
-
-  let stubs__linalg_solve_ex =
-    foreign "atg__linalg_solve_ex" (ptr t @-> t @-> t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs__linalg_solve_ex_result =
-    foreign
-      "atg__linalg_solve_ex_result"
-      (ptr t @-> t @-> t @-> t @-> t @-> t @-> t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs__linalg_svd =
-    foreign "atg__linalg_svd" (ptr t @-> t @-> int @-> int @-> string @-> returning void)
-  ;;
-
-  let stubs__linalg_svd_u =
-    foreign
-      "atg__linalg_svd_u"
-      (ptr t @-> t @-> t @-> t @-> t @-> int @-> int @-> string @-> returning void)
-  ;;
-
-  let stubs__log_softmax =
-    foreign "atg__log_softmax" (ptr t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__log_softmax_backward_data =
-    foreign
-      "atg__log_softmax_backward_data"
-      (ptr t @-> t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__log_softmax_backward_data_out =
-    foreign
-      "atg__log_softmax_backward_data_out"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__log_softmax_out =
-    foreign
-      "atg__log_softmax_out"
-      (ptr t @-> t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__logcumsumexp =
-    foreign "atg__logcumsumexp" (ptr t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__logcumsumexp_out =
-    foreign "atg__logcumsumexp_out" (ptr t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__lstm_mps =
-    foreign
-      "atg__lstm_mps"
-      (ptr t
-       @-> t
-       @-> ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> int
-       @-> int64_t
-       @-> double
-       @-> int
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__lstm_mps_out =
-    foreign
-      "atg__lstm_mps_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> int
-       @-> int64_t
-       @-> double
-       @-> int
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__lu_with_info =
-    foreign "atg__lu_with_info" (ptr t @-> t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs__make_dep_token =
-    foreign "atg__make_dep_token" (ptr t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs__make_dual =
-    foreign "atg__make_dual" (ptr t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__make_dual_copy =
-    foreign "atg__make_dual_copy" (ptr t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__make_dual_copy_out =
-    foreign
-      "atg__make_dual_copy_out"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__make_per_channel_quantized_tensor =
-    foreign
-      "atg__make_per_channel_quantized_tensor"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__make_per_channel_quantized_tensor_out =
-    foreign
-      "atg__make_per_channel_quantized_tensor_out"
-      (ptr t @-> t @-> t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__make_per_tensor_quantized_tensor =
-    foreign
-      "atg__make_per_tensor_quantized_tensor"
-      (ptr t @-> t @-> double @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__make_per_tensor_quantized_tensor_out =
-    foreign
-      "atg__make_per_tensor_quantized_tensor_out"
-      (ptr t @-> t @-> t @-> double @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__masked_scale =
-    foreign "atg__masked_scale" (ptr t @-> t @-> t @-> double @-> returning void)
-  ;;
-
-  let stubs__masked_scale_out =
-    foreign "atg__masked_scale_out" (ptr t @-> t @-> t @-> t @-> double @-> returning void)
-  ;;
-
-  let stubs__masked_softmax =
-    foreign
-      "atg__masked_softmax"
-      (ptr t @-> t @-> t @-> int64_t @-> int @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__masked_softmax_backward =
-    foreign
-      "atg__masked_softmax_backward"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__masked_softmax_backward_out =
-    foreign
-      "atg__masked_softmax_backward_out"
-      (ptr t @-> t @-> t @-> t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__masked_softmax_out =
-    foreign
-      "atg__masked_softmax_out"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> int @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__mkldnn_reshape =
-    foreign "atg__mkldnn_reshape" (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__mkldnn_reshape_out =
-    foreign
-      "atg__mkldnn_reshape_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__mkldnn_transpose =
-    foreign
-      "atg__mkldnn_transpose"
-      (ptr t @-> t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__mkldnn_transpose_ =
-    foreign
-      "atg__mkldnn_transpose_"
-      (ptr t @-> t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__mkldnn_transpose_out =
-    foreign
-      "atg__mkldnn_transpose_out"
-      (ptr t @-> t @-> t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__mps_convolution =
-    foreign
-      "atg__mps_convolution"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs__mps_convolution_out =
-    foreign
-      "atg__mps_convolution_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs__mps_convolution_transpose =
-    foreign
-      "atg__mps_convolution_transpose"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs__mps_convolution_transpose_out =
-    foreign
-      "atg__mps_convolution_transpose_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs__native_batch_norm_legit =
-    foreign
-      "atg__native_batch_norm_legit"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int
-       @-> double
-       @-> double
-       @-> returning void)
-  ;;
-
-  let stubs__native_batch_norm_legit_functional =
-    foreign
-      "atg__native_batch_norm_legit_functional"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int
-       @-> double
-       @-> double
-       @-> returning void)
-  ;;
-
-  let stubs__native_batch_norm_legit_no_stats =
-    foreign
-      "atg__native_batch_norm_legit_no_stats"
-      (ptr t @-> t @-> t @-> t @-> int @-> double @-> double @-> returning void)
-  ;;
-
-  let stubs__native_batch_norm_legit_no_stats_out =
-    foreign
-      "atg__native_batch_norm_legit_no_stats_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int
-       @-> double
-       @-> double
-       @-> returning void)
-  ;;
-
-  let stubs__native_batch_norm_legit_no_training =
-    foreign
-      "atg__native_batch_norm_legit_no_training"
-      (ptr t @-> t @-> t @-> t @-> t @-> t @-> double @-> double @-> returning void)
-  ;;
-
-  let stubs__native_batch_norm_legit_no_training_out =
-    foreign
-      "atg__native_batch_norm_legit_no_training_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> double
-       @-> double
-       @-> returning void)
-  ;;
-
-  let stubs__native_batch_norm_legit_out =
-    foreign
-      "atg__native_batch_norm_legit_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int
-       @-> double
-       @-> double
-       @-> returning void)
-  ;;
-
-  let stubs__native_multi_head_attention =
-    foreign
-      "atg__native_multi_head_attention"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> int64_t
-       @-> int64_t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__native_multi_head_attention_out =
-    foreign
-      "atg__native_multi_head_attention_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int64_t
-       @-> int64_t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__neg_view = foreign "atg__neg_view" (ptr t @-> t @-> returning void)
-  let stubs__neg_view_copy = foreign "atg__neg_view_copy" (ptr t @-> t @-> returning void)
-
-  let stubs__neg_view_copy_out =
-    foreign "atg__neg_view_copy_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__nested_from_padded =
-    foreign "atg__nested_from_padded" (ptr t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs__nested_from_padded_and_nested_example =
-    foreign
-      "atg__nested_from_padded_and_nested_example"
-      (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__nested_from_padded_and_nested_example_out =
-    foreign
-      "atg__nested_from_padded_and_nested_example_out"
-      (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__nested_from_padded_out =
-    foreign
-      "atg__nested_from_padded_out"
-      (ptr t @-> t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs__nested_select_backward =
-    foreign
-      "atg__nested_select_backward"
-      (ptr t @-> t @-> t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__nested_sum_backward =
-    foreign
-      "atg__nested_sum_backward"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs__nested_view_from_buffer =
-    foreign
-      "atg__nested_view_from_buffer"
-      (ptr t @-> t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__nested_view_from_buffer_copy =
-    foreign
-      "atg__nested_view_from_buffer_copy"
-      (ptr t @-> t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__nested_view_from_buffer_copy_out =
-    foreign
-      "atg__nested_view_from_buffer_copy_out"
-      (ptr t @-> t @-> t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__new_zeros_with_same_feature_meta =
-    foreign
-      "atg__new_zeros_with_same_feature_meta"
-      (ptr t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__new_zeros_with_same_feature_meta_out =
-    foreign
-      "atg__new_zeros_with_same_feature_meta_out"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__nnpack_available = foreign "atg__nnpack_available" (void @-> returning bool)
-
-  let stubs__nnpack_spatial_convolution =
-    foreign
-      "atg__nnpack_spatial_convolution"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__nnpack_spatial_convolution_out =
-    foreign
-      "atg__nnpack_spatial_convolution_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__nnz = foreign "atg__nnz" (t @-> returning int64_t)
-
-  let stubs__pack_padded_sequence =
-    foreign "atg__pack_padded_sequence" (ptr t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs__pack_padded_sequence_backward =
-    foreign
-      "atg__pack_padded_sequence_backward"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs__pack_padded_sequence_out =
-    foreign
-      "atg__pack_padded_sequence_out"
-      (ptr t @-> t @-> t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs__pad_circular =
-    foreign "atg__pad_circular" (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__pad_enum =
-    foreign
-      "atg__pad_enum"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__pad_packed_sequence =
-    foreign
-      "atg__pad_packed_sequence"
-      (ptr t @-> t @-> t @-> int @-> scalar @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__pdist_backward =
-    foreign "atg__pdist_backward" (ptr t @-> t @-> t @-> double @-> t @-> returning void)
-  ;;
-
-  let stubs__pdist_backward_out =
-    foreign
-      "atg__pdist_backward_out"
-      (ptr t @-> t @-> t @-> t @-> double @-> t @-> returning void)
-  ;;
-
-  let stubs__pin_memory =
-    foreign "atg__pin_memory" (ptr t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs__pin_memory_out =
-    foreign "atg__pin_memory_out" (ptr t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs__prelu_kernel =
-    foreign "atg__prelu_kernel" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__prelu_kernel_backward =
-    foreign "atg__prelu_kernel_backward" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__propagate_xla_data =
-    foreign "atg__propagate_xla_data" (t @-> t @-> returning void)
-  ;;
-
-  let stubs__remove_batch_dim =
-    foreign
-      "atg__remove_batch_dim"
-      (ptr t @-> t @-> int64_t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__reshape_alias =
-    foreign
-      "atg__reshape_alias"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__reshape_alias_copy =
-    foreign
-      "atg__reshape_alias_copy"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__reshape_alias_copy_out =
-    foreign
-      "atg__reshape_alias_copy_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__reshape_copy =
-    foreign "atg__reshape_copy" (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__reshape_from_tensor =
-    foreign "atg__reshape_from_tensor" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__resize_output =
-    foreign
-      "atg__resize_output"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs__resize_output_ =
-    foreign
-      "atg__resize_output_"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs__resize_output_out =
-    foreign
-      "atg__resize_output_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs__rowwise_prune =
-    foreign "atg__rowwise_prune" (ptr t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs__sample_dirichlet =
-    foreign "atg__sample_dirichlet" (ptr t @-> t @-> returning void)
-  ;;
-
-  let stubs__sample_dirichlet_out =
-    foreign "atg__sample_dirichlet_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__saturate_weight_to_fp16 =
-    foreign "atg__saturate_weight_to_fp16" (ptr t @-> t @-> returning void)
-  ;;
-
-  let stubs__scaled_dot_product_attention_math =
-    foreign
-      "atg__scaled_dot_product_attention_math"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> double
-       @-> int
-       @-> t
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__scaled_dot_product_efficient_attention =
-    foreign
-      "atg__scaled_dot_product_efficient_attention"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__scaled_dot_product_flash_attention_backward =
-    foreign
-      "atg__scaled_dot_product_flash_attention_backward"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int64_t
-       @-> int64_t
-       @-> double
-       @-> int
-       @-> t
-       @-> t
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__scaled_mm =
-    foreign
-      "atg__scaled_mm"
-      (ptr t @-> t @-> t @-> t @-> int @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__scaled_mm_out =
-    foreign
-      "atg__scaled_mm_out"
-      (ptr t @-> t @-> t @-> t @-> t @-> t @-> int @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__scatter_reduce =
-    foreign
-      "atg__scatter_reduce"
-      (ptr t @-> t @-> int64_t @-> t @-> t @-> string @-> int @-> returning void)
-  ;;
-
-  let stubs__scatter_reduce_ =
-    foreign
-      "atg__scatter_reduce_"
-      (ptr t @-> t @-> int64_t @-> t @-> t @-> string @-> int @-> returning void)
-  ;;
-
-  let stubs__scatter_reduce_two_out =
-    foreign
-      "atg__scatter_reduce_two_out"
-      (ptr t @-> t @-> t @-> int64_t @-> t @-> t @-> string @-> int @-> returning void)
-  ;;
-
-  let stubs__segment_reduce_backward =
-    foreign
-      "atg__segment_reduce_backward"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> string
-       @-> t
-       @-> t
-       @-> int64_t
-       @-> scalar
-       @-> returning void)
-  ;;
-
-  let stubs__segment_reduce_backward_out =
-    foreign
-      "atg__segment_reduce_backward_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> string
-       @-> t
-       @-> t
-       @-> int64_t
-       @-> scalar
-       @-> returning void)
-  ;;
-
-  let stubs__shape_as_tensor =
-    foreign "atg__shape_as_tensor" (ptr t @-> t @-> returning void)
-  ;;
-end
-
-module C3 (F : Cstubs.FOREIGN) = struct
-  open F
-
-  type t = unit ptr
-
-  let t : t typ = ptr void
-
-  type scalar = unit ptr
-
-  let scalar : scalar typ = ptr void
-
-  let stubs__slow_conv2d_backward =
-    foreign
-      "atg__slow_conv2d_backward"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__sobol_engine_draw =
-    foreign
-      "atg__sobol_engine_draw"
-      (ptr t @-> t @-> int64_t @-> t @-> int64_t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__sobol_engine_ff_ =
-    foreign
-      "atg__sobol_engine_ff_"
-      (ptr t @-> t @-> int64_t @-> t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__sobol_engine_initialize_state_ =
-    foreign
-      "atg__sobol_engine_initialize_state_"
-      (ptr t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__sobol_engine_scramble_ =
-    foreign
-      "atg__sobol_engine_scramble_"
-      (ptr t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__softmax =
-    foreign "atg__softmax" (ptr t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__softmax_backward_data =
-    foreign
-      "atg__softmax_backward_data"
-      (ptr t @-> t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__softmax_backward_data_out =
-    foreign
-      "atg__softmax_backward_data_out"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__softmax_out =
-    foreign "atg__softmax_out" (ptr t @-> t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__sparse_addmm =
-    foreign "atg__sparse_addmm" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__sparse_addmm_out =
-    foreign "atg__sparse_addmm_out" (ptr t @-> t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__sparse_broadcast_to =
-    foreign
-      "atg__sparse_broadcast_to"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__sparse_broadcast_to_copy =
-    foreign
-      "atg__sparse_broadcast_to_copy"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__sparse_broadcast_to_copy_out =
-    foreign
-      "atg__sparse_broadcast_to_copy_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__sparse_bsc_tensor_unsafe =
-    foreign
-      "atg__sparse_bsc_tensor_unsafe"
-      (ptr t @-> t @-> t @-> t @-> ptr int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs__sparse_bsr_tensor_unsafe =
-    foreign
-      "atg__sparse_bsr_tensor_unsafe"
-      (ptr t @-> t @-> t @-> t @-> ptr int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs__sparse_compressed_tensor_unsafe =
-    foreign
-      "atg__sparse_compressed_tensor_unsafe"
-      (ptr t @-> t @-> t @-> t @-> ptr int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs__sparse_coo_tensor_unsafe =
-    foreign
-      "atg__sparse_coo_tensor_unsafe"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__sparse_coo_tensor_with_dims =
-    foreign
-      "atg__sparse_coo_tensor_with_dims"
-      (ptr t
-       @-> int64_t
-       @-> int64_t
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__sparse_coo_tensor_with_dims_and_tensors =
-    foreign
-      "atg__sparse_coo_tensor_with_dims_and_tensors"
-      (ptr t
-       @-> int64_t
-       @-> int64_t
-       @-> ptr int64_t
-       @-> int
-       @-> t
-       @-> t
-       @-> int
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__sparse_coo_tensor_with_dims_and_tensors_out =
-    foreign
-      "atg__sparse_coo_tensor_with_dims_and_tensors_out"
-      (ptr t
-       @-> t
-       @-> int64_t
-       @-> int64_t
-       @-> ptr int64_t
-       @-> int
-       @-> t
-       @-> t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__sparse_coo_tensor_with_dims_out =
-    foreign
-      "atg__sparse_coo_tensor_with_dims_out"
-      (ptr t @-> t @-> int64_t @-> int64_t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__sparse_csc_tensor_unsafe =
-    foreign
-      "atg__sparse_csc_tensor_unsafe"
-      (ptr t @-> t @-> t @-> t @-> ptr int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs__sparse_csr_prod =
-    foreign
-      "atg__sparse_csr_prod"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs__sparse_csr_prod_dim_dtype_out =
-    foreign
-      "atg__sparse_csr_prod_dim_dtype_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs__sparse_csr_sum =
-    foreign
-      "atg__sparse_csr_sum"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs__sparse_csr_sum_dim_dtype_out =
-    foreign
-      "atg__sparse_csr_sum_dim_dtype_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs__sparse_csr_tensor_unsafe =
-    foreign
-      "atg__sparse_csr_tensor_unsafe"
-      (ptr t @-> t @-> t @-> t @-> ptr int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs__sparse_log_softmax =
-    foreign "atg__sparse_log_softmax" (ptr t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__sparse_log_softmax_backward_data =
-    foreign
-      "atg__sparse_log_softmax_backward_data"
-      (ptr t @-> t @-> t @-> int64_t @-> t @-> returning void)
-  ;;
-
-  let stubs__sparse_log_softmax_backward_data_out =
-    foreign
-      "atg__sparse_log_softmax_backward_data_out"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> t @-> returning void)
-  ;;
-
-  let stubs__sparse_log_softmax_int =
-    foreign
-      "atg__sparse_log_softmax_int"
-      (ptr t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__sparse_log_softmax_out =
-    foreign
-      "atg__sparse_log_softmax_out"
-      (ptr t @-> t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__sparse_mask_projection =
-    foreign "atg__sparse_mask_projection" (ptr t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs__sparse_mask_projection_out =
-    foreign
-      "atg__sparse_mask_projection_out"
-      (ptr t @-> t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs__sparse_mm = foreign "atg__sparse_mm" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs__sparse_mm_reduce =
-    foreign "atg__sparse_mm_reduce" (ptr t @-> t @-> t @-> string @-> returning void)
-  ;;
-
-  let stubs__sparse_mm_reduce_impl =
-    foreign "atg__sparse_mm_reduce_impl" (ptr t @-> t @-> t @-> string @-> returning void)
-  ;;
-
-  let stubs__sparse_semi_structured_linear =
-    foreign
-      "atg__sparse_semi_structured_linear"
-      (ptr t @-> t @-> t @-> t @-> t @-> string @-> returning void)
-  ;;
-
-  let stubs__sparse_softmax =
-    foreign "atg__sparse_softmax" (ptr t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__sparse_softmax_backward_data =
-    foreign
-      "atg__sparse_softmax_backward_data"
-      (ptr t @-> t @-> t @-> int64_t @-> t @-> returning void)
-  ;;
-
-  let stubs__sparse_softmax_backward_data_out =
-    foreign
-      "atg__sparse_softmax_backward_data_out"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> t @-> returning void)
-  ;;
-
-  let stubs__sparse_softmax_int =
-    foreign "atg__sparse_softmax_int" (ptr t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__sparse_softmax_out =
-    foreign
-      "atg__sparse_softmax_out"
-      (ptr t @-> t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__sparse_sparse_matmul =
-    foreign "atg__sparse_sparse_matmul" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__sparse_sparse_matmul_out =
-    foreign "atg__sparse_sparse_matmul_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__sparse_sum = foreign "atg__sparse_sum" (ptr t @-> t @-> returning void)
-
-  let stubs__sparse_sum_backward =
-    foreign
-      "atg__sparse_sum_backward"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__sparse_sum_backward_out =
-    foreign
-      "atg__sparse_sum_backward_out"
-      (ptr t @-> t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__sparse_sum_dim =
-    foreign "atg__sparse_sum_dim" (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__sparse_sum_dim_dtype =
-    foreign
-      "atg__sparse_sum_dim_dtype"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs__sparse_sum_dim_out =
-    foreign
-      "atg__sparse_sum_dim_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__sparse_sum_dtype =
-    foreign "atg__sparse_sum_dtype" (ptr t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs__spdiags =
-    foreign "atg__spdiags" (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__spdiags_out =
-    foreign
-      "atg__spdiags_out"
-      (ptr t @-> t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__stack =
-    foreign "atg__stack" (ptr t @-> ptr t @-> int @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__stack_out =
-    foreign "atg__stack_out" (ptr t @-> t @-> ptr t @-> int @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__standard_gamma =
-    foreign "atg__standard_gamma" (ptr t @-> t @-> returning void)
-  ;;
-
-  let stubs__standard_gamma_grad =
-    foreign "atg__standard_gamma_grad" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__standard_gamma_grad_out =
-    foreign "atg__standard_gamma_grad_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__standard_gamma_out =
-    foreign "atg__standard_gamma_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__test_ambiguous_defaults =
-    foreign
-      "atg__test_ambiguous_defaults"
-      (ptr t @-> t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__test_ambiguous_defaults_b =
-    foreign
-      "atg__test_ambiguous_defaults_b"
-      (ptr t @-> t @-> int64_t @-> string @-> returning void)
-  ;;
-
-  let stubs__test_autograd_multiple_dispatch =
-    foreign "atg__test_autograd_multiple_dispatch" (ptr t @-> t @-> returning void)
-  ;;
-
-  let stubs__test_autograd_multiple_dispatch_fullcoverage_out =
-    foreign
-      "atg__test_autograd_multiple_dispatch_fullcoverage_out"
-      (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__test_autograd_multiple_dispatch_ntonly =
-    foreign
-      "atg__test_autograd_multiple_dispatch_ntonly"
-      (ptr t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs__test_autograd_multiple_dispatch_view =
-    foreign "atg__test_autograd_multiple_dispatch_view" (ptr t @-> t @-> returning void)
-  ;;
-
-  let stubs__test_autograd_multiple_dispatch_view_copy =
-    foreign
-      "atg__test_autograd_multiple_dispatch_view_copy"
-      (ptr t @-> t @-> returning void)
-  ;;
-
-  let stubs__test_autograd_multiple_dispatch_view_copy_out =
-    foreign
-      "atg__test_autograd_multiple_dispatch_view_copy_out"
-      (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__test_check_tensor =
-    foreign "atg__test_check_tensor" (ptr t @-> t @-> returning void)
-  ;;
-
-  let stubs__test_functorch_fallback =
-    foreign "atg__test_functorch_fallback" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__test_functorch_fallback_out =
-    foreign "atg__test_functorch_fallback_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__test_optional_filled_intlist =
-    foreign
-      "atg__test_optional_filled_intlist"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__test_optional_filled_intlist_out =
-    foreign
-      "atg__test_optional_filled_intlist_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__test_optional_floatlist =
-    foreign
-      "atg__test_optional_floatlist"
-      (ptr t @-> t @-> ptr double @-> int @-> returning void)
-  ;;
-
-  let stubs__test_optional_floatlist_out =
-    foreign
-      "atg__test_optional_floatlist_out"
-      (ptr t @-> t @-> t @-> ptr double @-> int @-> returning void)
-  ;;
-
-  let stubs__test_optional_intlist =
-    foreign
-      "atg__test_optional_intlist"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__test_optional_intlist_out =
-    foreign
-      "atg__test_optional_intlist_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__test_serialization_subcmul =
-    foreign "atg__test_serialization_subcmul" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__test_string_default =
-    foreign
-      "atg__test_string_default"
-      (ptr t @-> t @-> string @-> string @-> returning void)
-  ;;
-
-  let stubs__test_warn_in_autograd =
-    foreign "atg__test_warn_in_autograd" (ptr t @-> t @-> returning void)
-  ;;
-
-  let stubs__test_warn_in_autograd_out =
-    foreign "atg__test_warn_in_autograd_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__thnn_differentiable_gru_cell_backward =
-    foreign
-      "atg__thnn_differentiable_gru_cell_backward"
-      (ptr t @-> t @-> t @-> t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__thnn_differentiable_lstm_cell_backward =
-    foreign
-      "atg__thnn_differentiable_lstm_cell_backward"
-      (ptr t @-> t @-> t @-> t @-> t @-> t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__thnn_fused_gru_cell =
-    foreign
-      "atg__thnn_fused_gru_cell"
-      (ptr t @-> t @-> t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__thnn_fused_gru_cell_backward =
-    foreign
-      "atg__thnn_fused_gru_cell_backward"
-      (ptr t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs__thnn_fused_gru_cell_backward_out =
-    foreign
-      "atg__thnn_fused_gru_cell_backward_out"
-      (ptr t @-> t @-> t @-> t @-> t @-> t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs__thnn_fused_gru_cell_out =
-    foreign
-      "atg__thnn_fused_gru_cell_out"
-      (ptr t @-> t @-> t @-> t @-> t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__thnn_fused_lstm_cell =
-    foreign
-      "atg__thnn_fused_lstm_cell"
-      (ptr t @-> t @-> t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__thnn_fused_lstm_cell_backward =
-    foreign
-      "atg__thnn_fused_lstm_cell_backward"
-      (ptr t @-> t @-> t @-> t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs__thnn_fused_lstm_cell_backward_impl =
-    foreign
-      "atg__thnn_fused_lstm_cell_backward_impl"
-      (ptr t @-> t @-> t @-> t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs__thnn_fused_lstm_cell_backward_impl_out =
-    foreign
-      "atg__thnn_fused_lstm_cell_backward_impl_out"
-      (ptr t @-> t @-> t @-> t @-> t @-> t @-> t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs__thnn_fused_lstm_cell_out =
-    foreign
-      "atg__thnn_fused_lstm_cell_out"
-      (ptr t @-> t @-> t @-> t @-> t @-> t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__to_copy =
-    foreign "atg__to_copy" (ptr t @-> t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs__to_copy_out =
-    foreign "atg__to_copy_out" (ptr t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs__to_cpu = foreign "atg__to_cpu" (ptr t @-> int @-> returning (ptr t))
-
-  let stubs__to_dense =
-    foreign "atg__to_dense" (ptr t @-> t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs__to_dense_out =
-    foreign "atg__to_dense_out" (ptr t @-> t @-> t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs__to_sparse_bsc =
-    foreign
-      "atg__to_sparse_bsc"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__to_sparse_bsc_out =
-    foreign
-      "atg__to_sparse_bsc_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> int64_t @-> int @-> returning void)
-  ;;
-end
-
-module C4 (F : Cstubs.FOREIGN) = struct
-  open F
-
-  type t = unit ptr
-
-  let t : t typ = ptr void
-
-  type scalar = unit ptr
-
-  let scalar : scalar typ = ptr void
-
-  let stubs__to_sparse_bsr =
-    foreign
-      "atg__to_sparse_bsr"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__to_sparse_bsr_out =
-    foreign
-      "atg__to_sparse_bsr_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__to_sparse_csc =
-    foreign "atg__to_sparse_csc" (ptr t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__to_sparse_csc_out =
-    foreign
-      "atg__to_sparse_csc_out"
-      (ptr t @-> t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__to_sparse_csr =
-    foreign "atg__to_sparse_csr" (ptr t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__to_sparse_csr_out =
-    foreign
-      "atg__to_sparse_csr_out"
-      (ptr t @-> t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__to_sparse_semi_structured =
-    foreign "atg__to_sparse_semi_structured" (ptr t @-> t @-> returning void)
-  ;;
-
-  let stubs__transform_bias_rescale_qkv =
-    foreign
-      "atg__transform_bias_rescale_qkv"
-      (ptr t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__transform_bias_rescale_qkv_out =
-    foreign
-      "atg__transform_bias_rescale_qkv_out"
-      (ptr t @-> t @-> t @-> t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__transformer_encoder_layer_fwd =
-    foreign
-      "atg__transformer_encoder_layer_fwd"
-      (ptr t
-       @-> t
-       @-> int64_t
-       @-> int64_t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int
-       @-> int
-       @-> double
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__transformer_encoder_layer_fwd_out =
-    foreign
-      "atg__transformer_encoder_layer_fwd_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> int64_t
-       @-> int64_t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int
-       @-> int
-       @-> double
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__trilinear =
-    foreign
-      "atg__trilinear"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs__trilinear_out =
-    foreign
-      "atg__trilinear_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs__triton_multi_head_attention =
-    foreign
-      "atg__triton_multi_head_attention"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> int64_t
-       @-> int64_t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> returning void)
-  ;;
-
-  let stubs__triton_multi_head_attention_out =
-    foreign
-      "atg__triton_multi_head_attention_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int64_t
-       @-> int64_t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> returning void)
-  ;;
-
-  let stubs__triton_scaled_dot_attention =
-    foreign
-      "atg__triton_scaled_dot_attention"
-      (ptr t @-> t @-> t @-> t @-> double @-> returning void)
-  ;;
-
-  let stubs__triton_scaled_dot_attention_out =
-    foreign
-      "atg__triton_scaled_dot_attention_out"
-      (ptr t @-> t @-> t @-> t @-> t @-> double @-> returning void)
-  ;;
-
-  let stubs__unique =
-    foreign "atg__unique" (ptr t @-> t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs__unique2 =
-    foreign "atg__unique2" (ptr t @-> t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs__unique2_out =
-    foreign
-      "atg__unique2_out"
-      (ptr t @-> t @-> t @-> t @-> t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs__unique_out =
-    foreign "atg__unique_out" (ptr t @-> t @-> t @-> t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs__unpack_dual =
-    foreign "atg__unpack_dual" (ptr t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__unsafe_index =
-    foreign "atg__unsafe_index" (ptr t @-> t @-> ptr t @-> int @-> returning void)
-  ;;
-
-  let stubs__unsafe_index_put =
-    foreign
-      "atg__unsafe_index_put"
-      (ptr t @-> t @-> ptr t @-> int @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs__unsafe_view =
-    foreign "atg__unsafe_view" (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__unsafe_view_out =
-    foreign
-      "atg__unsafe_view_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__upsample_bicubic2d_aa =
-    foreign
-      "atg__upsample_bicubic2d_aa"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__upsample_bicubic2d_aa_backward =
-    foreign
-      "atg__upsample_bicubic2d_aa_backward"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__upsample_bicubic2d_aa_backward_grad_input =
-    foreign
-      "atg__upsample_bicubic2d_aa_backward_grad_input"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__upsample_bicubic2d_aa_out =
-    foreign
-      "atg__upsample_bicubic2d_aa_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__upsample_bicubic2d_aa_vec =
-    foreign
-      "atg__upsample_bicubic2d_aa_vec"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> ptr double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__upsample_bilinear2d_aa =
-    foreign
-      "atg__upsample_bilinear2d_aa"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__upsample_bilinear2d_aa_backward =
-    foreign
-      "atg__upsample_bilinear2d_aa_backward"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__upsample_bilinear2d_aa_backward_grad_input =
-    foreign
-      "atg__upsample_bilinear2d_aa_backward_grad_input"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__upsample_bilinear2d_aa_out =
-    foreign
-      "atg__upsample_bilinear2d_aa_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__upsample_bilinear2d_aa_vec =
-    foreign
-      "atg__upsample_bilinear2d_aa_vec"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> ptr double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__upsample_nearest_exact1d =
-    foreign
-      "atg__upsample_nearest_exact1d"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> double @-> int @-> returning void)
-  ;;
-
-  let stubs__upsample_nearest_exact1d_backward =
-    foreign
-      "atg__upsample_nearest_exact1d_backward"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__upsample_nearest_exact1d_backward_grad_input =
-    foreign
-      "atg__upsample_nearest_exact1d_backward_grad_input"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__upsample_nearest_exact1d_out =
-    foreign
-      "atg__upsample_nearest_exact1d_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> double @-> int @-> returning void)
-  ;;
-
-  let stubs__upsample_nearest_exact1d_vec =
-    foreign
-      "atg__upsample_nearest_exact1d_vec"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> ptr double @-> int @-> returning void)
-  ;;
-
-  let stubs__upsample_nearest_exact2d =
-    foreign
-      "atg__upsample_nearest_exact2d"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__upsample_nearest_exact2d_backward =
-    foreign
-      "atg__upsample_nearest_exact2d_backward"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__upsample_nearest_exact2d_backward_grad_input =
-    foreign
-      "atg__upsample_nearest_exact2d_backward_grad_input"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__upsample_nearest_exact2d_out =
-    foreign
-      "atg__upsample_nearest_exact2d_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__upsample_nearest_exact2d_vec =
-    foreign
-      "atg__upsample_nearest_exact2d_vec"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> ptr double @-> int @-> returning void)
-  ;;
-
-  let stubs__upsample_nearest_exact3d =
-    foreign
-      "atg__upsample_nearest_exact3d"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__upsample_nearest_exact3d_backward =
-    foreign
-      "atg__upsample_nearest_exact3d_backward"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__upsample_nearest_exact3d_backward_grad_input =
-    foreign
-      "atg__upsample_nearest_exact3d_backward_grad_input"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__upsample_nearest_exact3d_out =
-    foreign
-      "atg__upsample_nearest_exact3d_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs__upsample_nearest_exact3d_vec =
-    foreign
-      "atg__upsample_nearest_exact3d_vec"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> ptr double @-> int @-> returning void)
-  ;;
-
-  let stubs__use_cudnn_ctc_loss =
-    foreign
-      "atg__use_cudnn_ctc_loss"
-      (t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> returning bool)
-  ;;
-
-  let stubs__use_cudnn_ctc_loss_tensor =
-    foreign
-      "atg__use_cudnn_ctc_loss_tensor"
-      (t @-> t @-> t @-> t @-> int64_t @-> returning bool)
-  ;;
-
-  let stubs__use_cudnn_rnn_flatten_weight =
-    foreign "atg__use_cudnn_rnn_flatten_weight" (void @-> returning bool)
-  ;;
-
-  let stubs__validate_compressed_sparse_indices =
-    foreign
-      "atg__validate_compressed_sparse_indices"
-      (int @-> t @-> t @-> int64_t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__validate_sparse_bsc_tensor_args =
-    foreign
-      "atg__validate_sparse_bsc_tensor_args"
-      (t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__validate_sparse_bsr_tensor_args =
-    foreign
-      "atg__validate_sparse_bsr_tensor_args"
-      (t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__validate_sparse_csc_tensor_args =
-    foreign
-      "atg__validate_sparse_csc_tensor_args"
-      (t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs__values = foreign "atg__values" (ptr t @-> t @-> returning void)
-  let stubs__values_copy = foreign "atg__values_copy" (ptr t @-> t @-> returning void)
-
-  let stubs__values_copy_out =
-    foreign "atg__values_copy_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs__version = foreign "atg__version" (t @-> returning int64_t)
-
-  let stubs__weight_norm =
-    foreign "atg__weight_norm" (ptr t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__weight_norm_differentiable_backward =
-    foreign
-      "atg__weight_norm_differentiable_backward"
-      (ptr t @-> t @-> t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__weight_norm_interface =
-    foreign "atg__weight_norm_interface" (ptr t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__weight_norm_interface_backward =
-    foreign
-      "atg__weight_norm_interface_backward"
-      (ptr t @-> t @-> t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__weight_norm_interface_backward_out =
-    foreign
-      "atg__weight_norm_interface_backward_out"
-      (ptr t @-> t @-> t @-> t @-> t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs__weight_norm_interface_out =
-    foreign
-      "atg__weight_norm_interface_out"
-      (ptr t @-> t @-> t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_abs = foreign "atg_abs" (ptr t @-> t @-> returning void)
-  let stubs_abs_ = foreign "atg_abs_" (ptr t @-> t @-> returning void)
-  let stubs_abs_out = foreign "atg_abs_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_absolute = foreign "atg_absolute" (ptr t @-> t @-> returning void)
-  let stubs_absolute_ = foreign "atg_absolute_" (ptr t @-> t @-> returning void)
-
-  let stubs_absolute_out =
-    foreign "atg_absolute_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_acos = foreign "atg_acos" (ptr t @-> t @-> returning void)
-  let stubs_acos_ = foreign "atg_acos_" (ptr t @-> t @-> returning void)
-  let stubs_acos_out = foreign "atg_acos_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_acosh = foreign "atg_acosh" (ptr t @-> t @-> returning void)
-  let stubs_acosh_ = foreign "atg_acosh_" (ptr t @-> t @-> returning void)
-  let stubs_acosh_out = foreign "atg_acosh_out" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_adaptive_avg_pool1d =
-    foreign
-      "atg_adaptive_avg_pool1d"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_adaptive_avg_pool2d =
-    foreign
-      "atg_adaptive_avg_pool2d"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_adaptive_avg_pool2d_out =
-    foreign
-      "atg_adaptive_avg_pool2d_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_adaptive_avg_pool3d =
-    foreign
-      "atg_adaptive_avg_pool3d"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_adaptive_avg_pool3d_backward =
-    foreign "atg_adaptive_avg_pool3d_backward" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_adaptive_avg_pool3d_out =
-    foreign
-      "atg_adaptive_avg_pool3d_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_adaptive_max_pool1d =
-    foreign
-      "atg_adaptive_max_pool1d"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_adaptive_max_pool2d =
-    foreign
-      "atg_adaptive_max_pool2d"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_adaptive_max_pool2d_backward =
-    foreign "atg_adaptive_max_pool2d_backward" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_adaptive_max_pool2d_backward_grad_input =
-    foreign
-      "atg_adaptive_max_pool2d_backward_grad_input"
-      (ptr t @-> t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_adaptive_max_pool2d_out =
-    foreign
-      "atg_adaptive_max_pool2d_out"
-      (ptr t @-> t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_adaptive_max_pool3d =
-    foreign
-      "atg_adaptive_max_pool3d"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_adaptive_max_pool3d_backward =
-    foreign "atg_adaptive_max_pool3d_backward" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_adaptive_max_pool3d_backward_grad_input =
-    foreign
-      "atg_adaptive_max_pool3d_backward_grad_input"
-      (ptr t @-> t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_adaptive_max_pool3d_out =
-    foreign
-      "atg_adaptive_max_pool3d_out"
-      (ptr t @-> t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_add = foreign "atg_add" (ptr t @-> t @-> t @-> returning void)
-  let stubs_add_ = foreign "atg_add_" (ptr t @-> t @-> t @-> returning void)
-  let stubs_add_out = foreign "atg_add_out" (ptr t @-> t @-> t @-> t @-> returning void)
-
-  let stubs_add_scalar =
-    foreign "atg_add_scalar" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_add_scalar_ =
-    foreign "atg_add_scalar_" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-end
-
-module C5 (F : Cstubs.FOREIGN) = struct
-  open F
-
-  type t = unit ptr
-
-  let t : t typ = ptr void
-
-  type scalar = unit ptr
-
-  let scalar : scalar typ = ptr void
-
-  let stubs_add_scalar_out =
-    foreign "atg_add_scalar_out" (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_addbmm = foreign "atg_addbmm" (ptr t @-> t @-> t @-> t @-> returning void)
-  let stubs_addbmm_ = foreign "atg_addbmm_" (ptr t @-> t @-> t @-> t @-> returning void)
-
-  let stubs_addbmm_out =
-    foreign "atg_addbmm_out" (ptr t @-> t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_addcdiv = foreign "atg_addcdiv" (ptr t @-> t @-> t @-> t @-> returning void)
-  let stubs_addcdiv_ = foreign "atg_addcdiv_" (ptr t @-> t @-> t @-> t @-> returning void)
-
-  let stubs_addcdiv_out =
-    foreign "atg_addcdiv_out" (ptr t @-> t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_addcmul = foreign "atg_addcmul" (ptr t @-> t @-> t @-> t @-> returning void)
-  let stubs_addcmul_ = foreign "atg_addcmul_" (ptr t @-> t @-> t @-> t @-> returning void)
-
-  let stubs_addcmul_out =
-    foreign "atg_addcmul_out" (ptr t @-> t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_addmm = foreign "atg_addmm" (ptr t @-> t @-> t @-> t @-> returning void)
-  let stubs_addmm_ = foreign "atg_addmm_" (ptr t @-> t @-> t @-> t @-> returning void)
-
-  let stubs_addmm_out =
-    foreign "atg_addmm_out" (ptr t @-> t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_addmv = foreign "atg_addmv" (ptr t @-> t @-> t @-> t @-> returning void)
-  let stubs_addmv_ = foreign "atg_addmv_" (ptr t @-> t @-> t @-> t @-> returning void)
-
-  let stubs_addmv_out =
-    foreign "atg_addmv_out" (ptr t @-> t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_addr = foreign "atg_addr" (ptr t @-> t @-> t @-> t @-> returning void)
-  let stubs_addr_ = foreign "atg_addr_" (ptr t @-> t @-> t @-> t @-> returning void)
-
-  let stubs_addr_out =
-    foreign "atg_addr_out" (ptr t @-> t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_adjoint = foreign "atg_adjoint" (ptr t @-> t @-> returning void)
-
-  let stubs_affine_grid_generator =
-    foreign
-      "atg_affine_grid_generator"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_affine_grid_generator_backward =
-    foreign
-      "atg_affine_grid_generator_backward"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_affine_grid_generator_out =
-    foreign
-      "atg_affine_grid_generator_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_alias = foreign "atg_alias" (ptr t @-> t @-> returning void)
-  let stubs_alias_copy = foreign "atg_alias_copy" (ptr t @-> t @-> returning void)
-
-  let stubs_alias_copy_out =
-    foreign "atg_alias_copy_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_align_as = foreign "atg_align_as" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_align_tensors =
-    foreign "atg_align_tensors" (ptr t @-> int @-> returning (ptr t))
-  ;;
-
-  let stubs_all = foreign "atg_all" (ptr t @-> t @-> returning void)
-  let stubs_all_all_out = foreign "atg_all_all_out" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_all_dim =
-    foreign "atg_all_dim" (ptr t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_all_out =
-    foreign "atg_all_out" (ptr t @-> t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_allclose =
-    foreign "atg_allclose" (t @-> t @-> double @-> double @-> int @-> returning bool)
-  ;;
-
-  let stubs_alpha_dropout =
-    foreign "atg_alpha_dropout" (ptr t @-> t @-> double @-> int @-> returning void)
-  ;;
-
-  let stubs_alpha_dropout_ =
-    foreign "atg_alpha_dropout_" (ptr t @-> t @-> double @-> int @-> returning void)
-  ;;
-
-  let stubs_amax =
-    foreign "atg_amax" (ptr t @-> t @-> ptr int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_amax_out =
-    foreign
-      "atg_amax_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_amin =
-    foreign "atg_amin" (ptr t @-> t @-> ptr int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_amin_out =
-    foreign
-      "atg_amin_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_aminmax =
-    foreign "atg_aminmax" (ptr t @-> t @-> int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_aminmax_out =
-    foreign
-      "atg_aminmax_out"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_angle = foreign "atg_angle" (ptr t @-> t @-> returning void)
-  let stubs_angle_out = foreign "atg_angle_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_any = foreign "atg_any" (ptr t @-> t @-> returning void)
-  let stubs_any_all_out = foreign "atg_any_all_out" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_any_dim =
-    foreign "atg_any_dim" (ptr t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_any_out =
-    foreign "atg_any_out" (ptr t @-> t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_arange =
-    foreign "atg_arange" (ptr t @-> scalar @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_arange_start =
-    foreign
-      "atg_arange_start"
-      (ptr t @-> scalar @-> scalar @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_arange_start_step =
-    foreign
-      "atg_arange_start_step"
-      (ptr t @-> scalar @-> scalar @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_arccos = foreign "atg_arccos" (ptr t @-> t @-> returning void)
-  let stubs_arccos_ = foreign "atg_arccos_" (ptr t @-> t @-> returning void)
-  let stubs_arccos_out = foreign "atg_arccos_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_arccosh = foreign "atg_arccosh" (ptr t @-> t @-> returning void)
-  let stubs_arccosh_ = foreign "atg_arccosh_" (ptr t @-> t @-> returning void)
-  let stubs_arccosh_out = foreign "atg_arccosh_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_arcsin = foreign "atg_arcsin" (ptr t @-> t @-> returning void)
-  let stubs_arcsin_ = foreign "atg_arcsin_" (ptr t @-> t @-> returning void)
-  let stubs_arcsin_out = foreign "atg_arcsin_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_arcsinh = foreign "atg_arcsinh" (ptr t @-> t @-> returning void)
-  let stubs_arcsinh_ = foreign "atg_arcsinh_" (ptr t @-> t @-> returning void)
-  let stubs_arcsinh_out = foreign "atg_arcsinh_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_arctan = foreign "atg_arctan" (ptr t @-> t @-> returning void)
-  let stubs_arctan2 = foreign "atg_arctan2" (ptr t @-> t @-> t @-> returning void)
-  let stubs_arctan2_ = foreign "atg_arctan2_" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_arctan2_out =
-    foreign "atg_arctan2_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_arctan_ = foreign "atg_arctan_" (ptr t @-> t @-> returning void)
-  let stubs_arctan_out = foreign "atg_arctan_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_arctanh = foreign "atg_arctanh" (ptr t @-> t @-> returning void)
-  let stubs_arctanh_ = foreign "atg_arctanh_" (ptr t @-> t @-> returning void)
-  let stubs_arctanh_out = foreign "atg_arctanh_out" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_argmax =
-    foreign "atg_argmax" (ptr t @-> t @-> int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_argmax_out =
-    foreign
-      "atg_argmax_out"
-      (ptr t @-> t @-> t @-> int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_argmin =
-    foreign "atg_argmin" (ptr t @-> t @-> int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_argmin_out =
-    foreign
-      "atg_argmin_out"
-      (ptr t @-> t @-> t @-> int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_argsort =
-    foreign "atg_argsort" (ptr t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_argsort_stable =
-    foreign
-      "atg_argsort_stable"
-      (ptr t @-> t @-> int @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_argsort_stable_out =
-    foreign
-      "atg_argsort_stable_out"
-      (ptr t @-> t @-> t @-> int @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_argwhere = foreign "atg_argwhere" (ptr t @-> t @-> returning void)
-
-  let stubs_as_strided =
-    foreign
-      "atg_as_strided"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_as_strided_ =
-    foreign
-      "atg_as_strided_"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_as_strided_copy =
-    foreign
-      "atg_as_strided_copy"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_as_strided_copy_out =
-    foreign
-      "atg_as_strided_copy_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_as_strided_scatter =
-    foreign
-      "atg_as_strided_scatter"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_as_strided_scatter_out =
-    foreign
-      "atg_as_strided_scatter_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_asin = foreign "atg_asin" (ptr t @-> t @-> returning void)
-  let stubs_asin_ = foreign "atg_asin_" (ptr t @-> t @-> returning void)
-  let stubs_asin_out = foreign "atg_asin_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_asinh = foreign "atg_asinh" (ptr t @-> t @-> returning void)
-  let stubs_asinh_ = foreign "atg_asinh_" (ptr t @-> t @-> returning void)
-  let stubs_asinh_out = foreign "atg_asinh_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_atan = foreign "atg_atan" (ptr t @-> t @-> returning void)
-  let stubs_atan2 = foreign "atg_atan2" (ptr t @-> t @-> t @-> returning void)
-  let stubs_atan2_ = foreign "atg_atan2_" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_atan2_out =
-    foreign "atg_atan2_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_atan_ = foreign "atg_atan_" (ptr t @-> t @-> returning void)
-  let stubs_atan_out = foreign "atg_atan_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_atanh = foreign "atg_atanh" (ptr t @-> t @-> returning void)
-  let stubs_atanh_ = foreign "atg_atanh_" (ptr t @-> t @-> returning void)
-  let stubs_atanh_out = foreign "atg_atanh_out" (ptr t @-> t @-> t @-> returning void)
-end
-
-module C6 (F : Cstubs.FOREIGN) = struct
-  open F
-
-  type t = unit ptr
-
-  let t : t typ = ptr void
-
-  type scalar = unit ptr
-
-  let scalar : scalar typ = ptr void
-  let stubs_atleast_1d = foreign "atg_atleast_1d" (ptr t @-> t @-> returning void)
-
-  let stubs_atleast_1d_sequence =
-    foreign "atg_atleast_1d_sequence" (ptr t @-> int @-> returning (ptr t))
-  ;;
-
-  let stubs_atleast_2d = foreign "atg_atleast_2d" (ptr t @-> t @-> returning void)
-
-  let stubs_atleast_2d_sequence =
-    foreign "atg_atleast_2d_sequence" (ptr t @-> int @-> returning (ptr t))
-  ;;
-
-  let stubs_atleast_3d = foreign "atg_atleast_3d" (ptr t @-> t @-> returning void)
-
-  let stubs_atleast_3d_sequence =
-    foreign "atg_atleast_3d_sequence" (ptr t @-> int @-> returning (ptr t))
-  ;;
-
-  let stubs_avg_pool1d =
-    foreign
-      "atg_avg_pool1d"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_avg_pool2d =
-    foreign
-      "atg_avg_pool2d"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_avg_pool2d_backward =
-    foreign
-      "atg_avg_pool2d_backward"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_avg_pool2d_backward_grad_input =
-    foreign
-      "atg_avg_pool2d_backward_grad_input"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_avg_pool2d_out =
-    foreign
-      "atg_avg_pool2d_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_avg_pool3d =
-    foreign
-      "atg_avg_pool3d"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_avg_pool3d_backward =
-    foreign
-      "atg_avg_pool3d_backward"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_avg_pool3d_backward_grad_input =
-    foreign
-      "atg_avg_pool3d_backward_grad_input"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_avg_pool3d_out =
-    foreign
-      "atg_avg_pool3d_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_baddbmm = foreign "atg_baddbmm" (ptr t @-> t @-> t @-> t @-> returning void)
-  let stubs_baddbmm_ = foreign "atg_baddbmm_" (ptr t @-> t @-> t @-> t @-> returning void)
-
-  let stubs_baddbmm_out =
-    foreign "atg_baddbmm_out" (ptr t @-> t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_bartlett_window =
-    foreign "atg_bartlett_window" (ptr t @-> int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_bartlett_window_out =
-    foreign "atg_bartlett_window_out" (ptr t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_bartlett_window_periodic =
-    foreign
-      "atg_bartlett_window_periodic"
-      (ptr t @-> int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_bartlett_window_periodic_out =
-    foreign
-      "atg_bartlett_window_periodic_out"
-      (ptr t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_batch_norm =
-    foreign
-      "atg_batch_norm"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int
-       @-> double
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_batch_norm_backward_elemt =
-    foreign
-      "atg_batch_norm_backward_elemt"
-      (ptr t @-> t @-> t @-> t @-> t @-> t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_batch_norm_backward_elemt_out =
-    foreign
-      "atg_batch_norm_backward_elemt_out"
-      (ptr t @-> t @-> t @-> t @-> t @-> t @-> t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_batch_norm_backward_reduce =
-    foreign
-      "atg_batch_norm_backward_reduce"
-      (ptr t @-> t @-> t @-> t @-> t @-> t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_batch_norm_backward_reduce_out =
-    foreign
-      "atg_batch_norm_backward_reduce_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_batch_norm_elemt =
-    foreign
-      "atg_batch_norm_elemt"
-      (ptr t @-> t @-> t @-> t @-> t @-> t @-> double @-> returning void)
-  ;;
-
-  let stubs_batch_norm_elemt_out =
-    foreign
-      "atg_batch_norm_elemt_out"
-      (ptr t @-> t @-> t @-> t @-> t @-> t @-> t @-> double @-> returning void)
-  ;;
-
-  let stubs_batch_norm_gather_stats =
-    foreign
-      "atg_batch_norm_gather_stats"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> double
-       @-> double
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs_batch_norm_gather_stats_out =
-    foreign
-      "atg_batch_norm_gather_stats_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> double
-       @-> double
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs_batch_norm_gather_stats_with_counts =
-    foreign
-      "atg_batch_norm_gather_stats_with_counts"
-      (ptr t @-> t @-> t @-> t @-> t @-> t @-> double @-> double @-> t @-> returning void)
-  ;;
-
-  let stubs_batch_norm_gather_stats_with_counts_out =
-    foreign
-      "atg_batch_norm_gather_stats_with_counts_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> double
-       @-> double
-       @-> t
-       @-> returning void)
-  ;;
-
-  let stubs_batch_norm_stats =
-    foreign "atg_batch_norm_stats" (ptr t @-> t @-> double @-> returning void)
-  ;;
-
-  let stubs_batch_norm_stats_out =
-    foreign
-      "atg_batch_norm_stats_out"
-      (ptr t @-> t @-> t @-> t @-> double @-> returning void)
-  ;;
-
-  let stubs_batch_norm_update_stats =
-    foreign
-      "atg_batch_norm_update_stats"
-      (ptr t @-> t @-> t @-> t @-> double @-> returning void)
-  ;;
-
-  let stubs_batch_norm_update_stats_out =
-    foreign
-      "atg_batch_norm_update_stats_out"
-      (ptr t @-> t @-> t @-> t @-> t @-> t @-> double @-> returning void)
-  ;;
-
-  let stubs_bernoulli = foreign "atg_bernoulli" (ptr t @-> t @-> returning void)
-  let stubs_bernoulli_ = foreign "atg_bernoulli_" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_bernoulli_float_ =
-    foreign "atg_bernoulli_float_" (ptr t @-> t @-> double @-> returning void)
-  ;;
-
-  let stubs_bernoulli_p =
-    foreign "atg_bernoulli_p" (ptr t @-> t @-> double @-> returning void)
-  ;;
-
-  let stubs_bernoulli_tensor =
-    foreign "atg_bernoulli_tensor" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_bilinear =
-    foreign "atg_bilinear" (ptr t @-> t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_binary_cross_entropy =
-    foreign
-      "atg_binary_cross_entropy"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_binary_cross_entropy_backward =
-    foreign
-      "atg_binary_cross_entropy_backward"
-      (ptr t @-> t @-> t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_binary_cross_entropy_backward_grad_input =
-    foreign
-      "atg_binary_cross_entropy_backward_grad_input"
-      (ptr t @-> t @-> t @-> t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_binary_cross_entropy_out =
-    foreign
-      "atg_binary_cross_entropy_out"
-      (ptr t @-> t @-> t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_binary_cross_entropy_with_logits =
-    foreign
-      "atg_binary_cross_entropy_with_logits"
-      (ptr t @-> t @-> t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_binary_cross_entropy_with_logits_out =
-    foreign
-      "atg_binary_cross_entropy_with_logits_out"
-      (ptr t @-> t @-> t @-> t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_bincount =
-    foreign "atg_bincount" (ptr t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_bincount_out =
-    foreign "atg_bincount_out" (ptr t @-> t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_binomial = foreign "atg_binomial" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_binomial_out =
-    foreign "atg_binomial_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_bitwise_and =
-    foreign "atg_bitwise_and" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_bitwise_and_ =
-    foreign "atg_bitwise_and_" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_bitwise_and_scalar_out =
-    foreign "atg_bitwise_and_scalar_out" (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_bitwise_and_scalar_tensor =
-    foreign "atg_bitwise_and_scalar_tensor" (ptr t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_bitwise_and_scalar_tensor_out =
-    foreign
-      "atg_bitwise_and_scalar_tensor_out"
-      (ptr t @-> t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_bitwise_and_tensor =
-    foreign "atg_bitwise_and_tensor" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_bitwise_and_tensor_ =
-    foreign "atg_bitwise_and_tensor_" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_bitwise_and_tensor_out =
-    foreign "atg_bitwise_and_tensor_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_bitwise_left_shift =
-    foreign "atg_bitwise_left_shift" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_bitwise_left_shift_ =
-    foreign "atg_bitwise_left_shift_" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_bitwise_left_shift_scalar_tensor =
-    foreign
-      "atg_bitwise_left_shift_scalar_tensor"
-      (ptr t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_bitwise_left_shift_scalar_tensor_out =
-    foreign
-      "atg_bitwise_left_shift_scalar_tensor_out"
-      (ptr t @-> t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_bitwise_left_shift_tensor_out =
-    foreign
-      "atg_bitwise_left_shift_tensor_out"
-      (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_bitwise_left_shift_tensor_scalar =
-    foreign
-      "atg_bitwise_left_shift_tensor_scalar"
-      (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_bitwise_left_shift_tensor_scalar_ =
-    foreign
-      "atg_bitwise_left_shift_tensor_scalar_"
-      (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_bitwise_left_shift_tensor_scalar_out =
-    foreign
-      "atg_bitwise_left_shift_tensor_scalar_out"
-      (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_bitwise_not = foreign "atg_bitwise_not" (ptr t @-> t @-> returning void)
-  let stubs_bitwise_not_ = foreign "atg_bitwise_not_" (ptr t @-> t @-> returning void)
-
-  let stubs_bitwise_not_out =
-    foreign "atg_bitwise_not_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_bitwise_or =
-    foreign "atg_bitwise_or" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_bitwise_or_ =
-    foreign "atg_bitwise_or_" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_bitwise_or_scalar_out =
-    foreign "atg_bitwise_or_scalar_out" (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_bitwise_or_scalar_tensor =
-    foreign "atg_bitwise_or_scalar_tensor" (ptr t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_bitwise_or_scalar_tensor_out =
-    foreign
-      "atg_bitwise_or_scalar_tensor_out"
-      (ptr t @-> t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_bitwise_or_tensor =
-    foreign "atg_bitwise_or_tensor" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_bitwise_or_tensor_ =
-    foreign "atg_bitwise_or_tensor_" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_bitwise_or_tensor_out =
-    foreign "atg_bitwise_or_tensor_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_bitwise_right_shift =
-    foreign "atg_bitwise_right_shift" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_bitwise_right_shift_ =
-    foreign "atg_bitwise_right_shift_" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_bitwise_right_shift_scalar_tensor =
-    foreign
-      "atg_bitwise_right_shift_scalar_tensor"
-      (ptr t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_bitwise_right_shift_scalar_tensor_out =
-    foreign
-      "atg_bitwise_right_shift_scalar_tensor_out"
-      (ptr t @-> t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_bitwise_right_shift_tensor_out =
-    foreign
-      "atg_bitwise_right_shift_tensor_out"
-      (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_bitwise_right_shift_tensor_scalar =
-    foreign
-      "atg_bitwise_right_shift_tensor_scalar"
-      (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_bitwise_right_shift_tensor_scalar_ =
-    foreign
-      "atg_bitwise_right_shift_tensor_scalar_"
-      (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_bitwise_right_shift_tensor_scalar_out =
-    foreign
-      "atg_bitwise_right_shift_tensor_scalar_out"
-      (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_bitwise_xor =
-    foreign "atg_bitwise_xor" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_bitwise_xor_ =
-    foreign "atg_bitwise_xor_" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_bitwise_xor_scalar_out =
-    foreign "atg_bitwise_xor_scalar_out" (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_bitwise_xor_scalar_tensor =
-    foreign "atg_bitwise_xor_scalar_tensor" (ptr t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_bitwise_xor_scalar_tensor_out =
-    foreign
-      "atg_bitwise_xor_scalar_tensor_out"
-      (ptr t @-> t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_bitwise_xor_tensor =
-    foreign "atg_bitwise_xor_tensor" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_bitwise_xor_tensor_ =
-    foreign "atg_bitwise_xor_tensor_" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_bitwise_xor_tensor_out =
-    foreign "atg_bitwise_xor_tensor_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_blackman_window =
-    foreign "atg_blackman_window" (ptr t @-> int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_blackman_window_out =
-    foreign "atg_blackman_window_out" (ptr t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_blackman_window_periodic =
-    foreign
-      "atg_blackman_window_periodic"
-      (ptr t @-> int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_blackman_window_periodic_out =
-    foreign
-      "atg_blackman_window_periodic_out"
-      (ptr t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-end
-
-module C7 (F : Cstubs.FOREIGN) = struct
-  open F
-
-  type t = unit ptr
-
-  let t : t typ = ptr void
-
-  type scalar = unit ptr
-
-  let scalar : scalar typ = ptr void
-
-  let stubs_block_diag =
-    foreign "atg_block_diag" (ptr t @-> ptr t @-> int @-> returning void)
-  ;;
-
-  let stubs_block_diag_out =
-    foreign "atg_block_diag_out" (ptr t @-> t @-> ptr t @-> int @-> returning void)
-  ;;
-
-  let stubs_bmm = foreign "atg_bmm" (ptr t @-> t @-> t @-> returning void)
-  let stubs_bmm_out = foreign "atg_bmm_out" (ptr t @-> t @-> t @-> t @-> returning void)
-
-  let stubs_broadcast_tensors =
-    foreign "atg_broadcast_tensors" (ptr t @-> int @-> returning (ptr t))
-  ;;
-
-  let stubs_broadcast_to =
-    foreign "atg_broadcast_to" (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_bucketize =
-    foreign "atg_bucketize" (ptr t @-> t @-> t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_bucketize_scalar =
-    foreign
-      "atg_bucketize_scalar"
-      (ptr t @-> scalar @-> t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_bucketize_scalar_out =
-    foreign
-      "atg_bucketize_scalar_out"
-      (ptr t @-> t @-> scalar @-> t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_bucketize_tensor_out =
-    foreign
-      "atg_bucketize_tensor_out"
-      (ptr t @-> t @-> t @-> t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_can_cast = foreign "atg_can_cast" (int @-> int @-> returning bool)
-
-  let stubs_cartesian_prod =
-    foreign "atg_cartesian_prod" (ptr t @-> ptr t @-> int @-> returning void)
-  ;;
-
-  let stubs_cat =
-    foreign "atg_cat" (ptr t @-> ptr t @-> int @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_cat_out =
-    foreign "atg_cat_out" (ptr t @-> t @-> ptr t @-> int @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_cauchy =
-    foreign "atg_cauchy" (ptr t @-> t @-> double @-> double @-> returning void)
-  ;;
-
-  let stubs_cauchy_ =
-    foreign "atg_cauchy_" (ptr t @-> t @-> double @-> double @-> returning void)
-  ;;
-
-  let stubs_cauchy_out =
-    foreign "atg_cauchy_out" (ptr t @-> t @-> t @-> double @-> double @-> returning void)
-  ;;
-
-  let stubs_ccol_indices = foreign "atg_ccol_indices" (ptr t @-> t @-> returning void)
-
-  let stubs_ccol_indices_copy =
-    foreign "atg_ccol_indices_copy" (ptr t @-> t @-> returning void)
-  ;;
-
-  let stubs_ccol_indices_copy_out =
-    foreign "atg_ccol_indices_copy_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_cdist =
-    foreign
-      "atg_cdist"
-      (ptr t @-> t @-> t @-> double @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_ceil = foreign "atg_ceil" (ptr t @-> t @-> returning void)
-  let stubs_ceil_ = foreign "atg_ceil_" (ptr t @-> t @-> returning void)
-  let stubs_ceil_out = foreign "atg_ceil_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_celu = foreign "atg_celu" (ptr t @-> t @-> returning void)
-  let stubs_celu_ = foreign "atg_celu_" (ptr t @-> t @-> returning void)
-  let stubs_celu_out = foreign "atg_celu_out" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_chain_matmul =
-    foreign "atg_chain_matmul" (ptr t @-> ptr t @-> int @-> returning void)
-  ;;
-
-  let stubs_chain_matmul_out =
-    foreign "atg_chain_matmul_out" (ptr t @-> t @-> ptr t @-> int @-> returning void)
-  ;;
-
-  let stubs_chalf = foreign "atg_chalf" (ptr t @-> t @-> returning void)
-
-  let stubs_channel_shuffle =
-    foreign "atg_channel_shuffle" (ptr t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_channel_shuffle_out =
-    foreign "atg_channel_shuffle_out" (ptr t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_cholesky = foreign "atg_cholesky" (ptr t @-> t @-> int @-> returning void)
-
-  let stubs_cholesky_inverse =
-    foreign "atg_cholesky_inverse" (ptr t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_cholesky_inverse_out =
-    foreign "atg_cholesky_inverse_out" (ptr t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_cholesky_out =
-    foreign "atg_cholesky_out" (ptr t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_cholesky_solve =
-    foreign "atg_cholesky_solve" (ptr t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_cholesky_solve_out =
-    foreign "atg_cholesky_solve_out" (ptr t @-> t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_choose_qparams_optimized =
-    foreign
-      "atg_choose_qparams_optimized"
-      (ptr t @-> t @-> int64_t @-> int64_t @-> double @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_chunk = foreign "atg_chunk" (t @-> int64_t @-> int64_t @-> returning (ptr t))
-
-  let stubs_clamp =
-    foreign "atg_clamp" (ptr t @-> t @-> scalar @-> scalar @-> returning void)
-  ;;
-
-  let stubs_clamp_ =
-    foreign "atg_clamp_" (ptr t @-> t @-> scalar @-> scalar @-> returning void)
-  ;;
-
-  let stubs_clamp_max = foreign "atg_clamp_max" (ptr t @-> t @-> scalar @-> returning void)
-
-  let stubs_clamp_max_ =
-    foreign "atg_clamp_max_" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_clamp_max_out =
-    foreign "atg_clamp_max_out" (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_clamp_max_tensor =
-    foreign "atg_clamp_max_tensor" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_clamp_max_tensor_ =
-    foreign "atg_clamp_max_tensor_" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_clamp_max_tensor_out =
-    foreign "atg_clamp_max_tensor_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_clamp_min = foreign "atg_clamp_min" (ptr t @-> t @-> scalar @-> returning void)
-
-  let stubs_clamp_min_ =
-    foreign "atg_clamp_min_" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_clamp_min_out =
-    foreign "atg_clamp_min_out" (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_clamp_min_tensor =
-    foreign "atg_clamp_min_tensor" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_clamp_min_tensor_ =
-    foreign "atg_clamp_min_tensor_" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_clamp_min_tensor_out =
-    foreign "atg_clamp_min_tensor_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_clamp_out =
-    foreign "atg_clamp_out" (ptr t @-> t @-> t @-> scalar @-> scalar @-> returning void)
-  ;;
-
-  let stubs_clamp_tensor =
-    foreign "atg_clamp_tensor" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_clamp_tensor_ =
-    foreign "atg_clamp_tensor_" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_clamp_tensor_out =
-    foreign "atg_clamp_tensor_out" (ptr t @-> t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_clip =
-    foreign "atg_clip" (ptr t @-> t @-> scalar @-> scalar @-> returning void)
-  ;;
-
-  let stubs_clip_ =
-    foreign "atg_clip_" (ptr t @-> t @-> scalar @-> scalar @-> returning void)
-  ;;
-
-  let stubs_clip_out =
-    foreign "atg_clip_out" (ptr t @-> t @-> t @-> scalar @-> scalar @-> returning void)
-  ;;
-
-  let stubs_clip_tensor =
-    foreign "atg_clip_tensor" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_clip_tensor_ =
-    foreign "atg_clip_tensor_" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_clip_tensor_out =
-    foreign "atg_clip_tensor_out" (ptr t @-> t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_clone = foreign "atg_clone" (ptr t @-> t @-> returning void)
-  let stubs_clone_out = foreign "atg_clone_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_coalesce = foreign "atg_coalesce" (ptr t @-> t @-> returning void)
-
-  let stubs_col2im =
-    foreign
-      "atg_col2im"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_col2im_out =
-    foreign
-      "atg_col2im_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_col_indices = foreign "atg_col_indices" (ptr t @-> t @-> returning void)
-
-  let stubs_col_indices_copy =
-    foreign "atg_col_indices_copy" (ptr t @-> t @-> returning void)
-  ;;
-
-  let stubs_col_indices_copy_out =
-    foreign "atg_col_indices_copy_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_column_stack =
-    foreign "atg_column_stack" (ptr t @-> ptr t @-> int @-> returning void)
-  ;;
-
-  let stubs_column_stack_out =
-    foreign "atg_column_stack_out" (ptr t @-> t @-> ptr t @-> int @-> returning void)
-  ;;
-
-  let stubs_combinations =
-    foreign "atg_combinations" (ptr t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_complex = foreign "atg_complex" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_complex_out =
-    foreign "atg_complex_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_concat =
-    foreign "atg_concat" (ptr t @-> ptr t @-> int @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_concat_out =
-    foreign "atg_concat_out" (ptr t @-> t @-> ptr t @-> int @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_concatenate =
-    foreign "atg_concatenate" (ptr t @-> ptr t @-> int @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_concatenate_out =
-    foreign
-      "atg_concatenate_out"
-      (ptr t @-> t @-> ptr t @-> int @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_conj = foreign "atg_conj" (ptr t @-> t @-> returning void)
-  let stubs_conj_physical = foreign "atg_conj_physical" (ptr t @-> t @-> returning void)
-  let stubs_conj_physical_ = foreign "atg_conj_physical_" (ptr t @-> t @-> returning void)
-
-  let stubs_conj_physical_out =
-    foreign "atg_conj_physical_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_constant_pad_nd =
-    foreign "atg_constant_pad_nd" (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_constant_pad_nd_out =
-    foreign
-      "atg_constant_pad_nd_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_contiguous = foreign "atg_contiguous" (ptr t @-> t @-> returning void)
-
-  let stubs_conv1d =
-    foreign
-      "atg_conv1d"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs_conv1d_padding =
-    foreign
-      "atg_conv1d_padding"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> string
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs_conv2d =
-    foreign
-      "atg_conv2d"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs_conv2d_padding =
-    foreign
-      "atg_conv2d_padding"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> string
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs_conv3d =
-    foreign
-      "atg_conv3d"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs_conv3d_padding =
-    foreign
-      "atg_conv3d_padding"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> string
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs_conv_depthwise3d =
-    foreign
-      "atg_conv_depthwise3d"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_conv_depthwise3d_out =
-    foreign
-      "atg_conv_depthwise3d_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_conv_tbc =
-    foreign "atg_conv_tbc" (ptr t @-> t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_conv_tbc_backward =
-    foreign
-      "atg_conv_tbc_backward"
-      (ptr t @-> t @-> t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_conv_tbc_out =
-    foreign
-      "atg_conv_tbc_out"
-      (ptr t @-> t @-> t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_conv_transpose1d =
-    foreign
-      "atg_conv_transpose1d"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> ptr int64_t
-       @-> int
-       @-> returning void)
-  ;;
-end
-
-module C8 (F : Cstubs.FOREIGN) = struct
-  open F
-
-  type t = unit ptr
-
-  let t : t typ = ptr void
-
-  type scalar = unit ptr
-
-  let scalar : scalar typ = ptr void
-
-  let stubs_conv_transpose2d =
-    foreign
-      "atg_conv_transpose2d"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> ptr int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_conv_transpose3d =
-    foreign
-      "atg_conv_transpose3d"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> ptr int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_convolution =
-    foreign
-      "atg_convolution"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs_convolution_out =
-    foreign
-      "atg_convolution_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs_convolution_overrideable =
-    foreign
-      "atg_convolution_overrideable"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs_convolution_overrideable_out =
-    foreign
-      "atg_convolution_overrideable_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs_copy = foreign "atg_copy" (ptr t @-> t @-> t @-> int @-> returning void)
-
-  let stubs_copy_out =
-    foreign "atg_copy_out" (ptr t @-> t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_copy_sparse_to_sparse =
-    foreign "atg_copy_sparse_to_sparse" (ptr t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_copy_sparse_to_sparse_ =
-    foreign "atg_copy_sparse_to_sparse_" (ptr t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_copy_sparse_to_sparse_out =
-    foreign
-      "atg_copy_sparse_to_sparse_out"
-      (ptr t @-> t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_copysign = foreign "atg_copysign" (ptr t @-> t @-> t @-> returning void)
-  let stubs_copysign_ = foreign "atg_copysign_" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_copysign_out =
-    foreign "atg_copysign_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_copysign_scalar =
-    foreign "atg_copysign_scalar" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_copysign_scalar_ =
-    foreign "atg_copysign_scalar_" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_copysign_scalar_out =
-    foreign "atg_copysign_scalar_out" (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_corrcoef = foreign "atg_corrcoef" (ptr t @-> t @-> returning void)
-  let stubs_cos = foreign "atg_cos" (ptr t @-> t @-> returning void)
-  let stubs_cos_ = foreign "atg_cos_" (ptr t @-> t @-> returning void)
-  let stubs_cos_out = foreign "atg_cos_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_cosh = foreign "atg_cosh" (ptr t @-> t @-> returning void)
-  let stubs_cosh_ = foreign "atg_cosh_" (ptr t @-> t @-> returning void)
-  let stubs_cosh_out = foreign "atg_cosh_out" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_cosine_embedding_loss =
-    foreign
-      "atg_cosine_embedding_loss"
-      (ptr t @-> t @-> t @-> t @-> double @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_cosine_similarity =
-    foreign
-      "atg_cosine_similarity"
-      (ptr t @-> t @-> t @-> int64_t @-> double @-> returning void)
-  ;;
-
-  let stubs_count_nonzero =
-    foreign
-      "atg_count_nonzero"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_count_nonzero_out =
-    foreign
-      "atg_count_nonzero_out"
-      (ptr t @-> t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_cov =
-    foreign "atg_cov" (ptr t @-> t @-> int64_t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_cross =
-    foreign "atg_cross" (ptr t @-> t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_cross_entropy_loss =
-    foreign
-      "atg_cross_entropy_loss"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> int64_t @-> double @-> returning void)
-  ;;
-
-  let stubs_cross_out =
-    foreign
-      "atg_cross_out"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_crow_indices = foreign "atg_crow_indices" (ptr t @-> t @-> returning void)
-
-  let stubs_crow_indices_copy =
-    foreign "atg_crow_indices_copy" (ptr t @-> t @-> returning void)
-  ;;
-
-  let stubs_crow_indices_copy_out =
-    foreign "atg_crow_indices_copy_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_ctc_loss =
-    foreign
-      "atg_ctc_loss"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_ctc_loss_tensor =
-    foreign
-      "atg_ctc_loss_tensor"
-      (ptr t @-> t @-> t @-> t @-> t @-> int64_t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_cudnn_affine_grid_generator =
-    foreign
-      "atg_cudnn_affine_grid_generator"
-      (ptr t @-> t @-> int64_t @-> int64_t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_cudnn_affine_grid_generator_backward =
-    foreign
-      "atg_cudnn_affine_grid_generator_backward"
-      (ptr t @-> t @-> int64_t @-> int64_t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_cudnn_affine_grid_generator_backward_out =
-    foreign
-      "atg_cudnn_affine_grid_generator_backward_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> int64_t
-       @-> int64_t
-       @-> int64_t
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs_cudnn_affine_grid_generator_out =
-    foreign
-      "atg_cudnn_affine_grid_generator_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> int64_t
-       @-> int64_t
-       @-> int64_t
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs_cudnn_batch_norm =
-    foreign
-      "atg_cudnn_batch_norm"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int
-       @-> double
-       @-> double
-       @-> returning void)
-  ;;
-
-  let stubs_cudnn_batch_norm_backward =
-    foreign
-      "atg_cudnn_batch_norm_backward"
-      (ptr t @-> t @-> t @-> t @-> t @-> t @-> t @-> t @-> double @-> t @-> returning void)
-  ;;
-
-  let stubs_cudnn_batch_norm_backward_out =
-    foreign
-      "atg_cudnn_batch_norm_backward_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> double
-       @-> t
-       @-> returning void)
-  ;;
-
-  let stubs_cudnn_batch_norm_out =
-    foreign
-      "atg_cudnn_batch_norm_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int
-       @-> double
-       @-> double
-       @-> returning void)
-  ;;
-
-  let stubs_cudnn_convolution =
-    foreign
-      "atg_cudnn_convolution"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_cudnn_convolution_add_relu =
-    foreign
-      "atg_cudnn_convolution_add_relu"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> scalar
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs_cudnn_convolution_add_relu_out =
-    foreign
-      "atg_cudnn_convolution_add_relu_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> scalar
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs_cudnn_convolution_out =
-    foreign
-      "atg_cudnn_convolution_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_cudnn_convolution_relu =
-    foreign
-      "atg_cudnn_convolution_relu"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs_cudnn_convolution_relu_out =
-    foreign
-      "atg_cudnn_convolution_relu_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs_cudnn_convolution_transpose =
-    foreign
-      "atg_cudnn_convolution_transpose"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_cudnn_convolution_transpose_out =
-    foreign
-      "atg_cudnn_convolution_transpose_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_cudnn_grid_sampler =
-    foreign "atg_cudnn_grid_sampler" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_cudnn_grid_sampler_backward =
-    foreign "atg_cudnn_grid_sampler_backward" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_cudnn_grid_sampler_backward_out =
-    foreign
-      "atg_cudnn_grid_sampler_backward_out"
-      (ptr t @-> t @-> t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_cudnn_grid_sampler_out =
-    foreign "atg_cudnn_grid_sampler_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_cudnn_is_acceptable = foreign "atg_cudnn_is_acceptable" (t @-> returning bool)
-  let stubs_cummax = foreign "atg_cummax" (ptr t @-> t @-> int64_t @-> returning void)
-
-  let stubs_cummax_out =
-    foreign "atg_cummax_out" (ptr t @-> t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_cummaxmin_backward =
-    foreign
-      "atg_cummaxmin_backward"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_cummin = foreign "atg_cummin" (ptr t @-> t @-> int64_t @-> returning void)
-
-  let stubs_cummin_out =
-    foreign "atg_cummin_out" (ptr t @-> t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_cumprod =
-    foreign "atg_cumprod" (ptr t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_cumprod_ =
-    foreign "atg_cumprod_" (ptr t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_cumprod_backward =
-    foreign "atg_cumprod_backward" (ptr t @-> t @-> t @-> int64_t @-> t @-> returning void)
-  ;;
-
-  let stubs_cumprod_out =
-    foreign "atg_cumprod_out" (ptr t @-> t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_cumsum =
-    foreign "atg_cumsum" (ptr t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_cumsum_ =
-    foreign "atg_cumsum_" (ptr t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_cumsum_out =
-    foreign "atg_cumsum_out" (ptr t @-> t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_cumulative_trapezoid =
-    foreign "atg_cumulative_trapezoid" (ptr t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_cumulative_trapezoid_x =
-    foreign "atg_cumulative_trapezoid_x" (ptr t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_data = foreign "atg_data" (ptr t @-> t @-> returning void)
-  let stubs_deg2rad = foreign "atg_deg2rad" (ptr t @-> t @-> returning void)
-  let stubs_deg2rad_ = foreign "atg_deg2rad_" (ptr t @-> t @-> returning void)
-  let stubs_deg2rad_out = foreign "atg_deg2rad_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_dense_dim = foreign "atg_dense_dim" (t @-> returning int64_t)
-  let stubs_dequantize = foreign "atg_dequantize" (ptr t @-> t @-> returning void)
-
-  let stubs_dequantize_self_out =
-    foreign "atg_dequantize_self_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_dequantize_tensors =
-    foreign "atg_dequantize_tensors" (ptr t @-> int @-> returning (ptr t))
-  ;;
-
-  let stubs_dequantize_tensors_out =
-    foreign
-      "atg_dequantize_tensors_out"
-      (ptr t @-> int @-> ptr t @-> int @-> returning void)
-  ;;
-
-  let stubs_det = foreign "atg_det" (ptr t @-> t @-> returning void)
-  let stubs_detach = foreign "atg_detach" (ptr t @-> t @-> returning void)
-  let stubs_detach_ = foreign "atg_detach_" (ptr t @-> t @-> returning void)
-  let stubs_detach_copy = foreign "atg_detach_copy" (ptr t @-> t @-> returning void)
-
-  let stubs_detach_copy_out =
-    foreign "atg_detach_copy_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_diag = foreign "atg_diag" (ptr t @-> t @-> int64_t @-> returning void)
-
-  let stubs_diag_embed =
-    foreign
-      "atg_diag_embed"
-      (ptr t @-> t @-> int64_t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_diag_embed_out =
-    foreign
-      "atg_diag_embed_out"
-      (ptr t @-> t @-> t @-> int64_t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_diag_out =
-    foreign "atg_diag_out" (ptr t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_diagflat = foreign "atg_diagflat" (ptr t @-> t @-> int64_t @-> returning void)
-
-  let stubs_diagonal =
-    foreign
-      "atg_diagonal"
-      (ptr t @-> t @-> int64_t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_diagonal_backward =
-    foreign
-      "atg_diagonal_backward"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> int64_t
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs_diagonal_backward_out =
-    foreign
-      "atg_diagonal_backward_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> int64_t
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs_diagonal_copy =
-    foreign
-      "atg_diagonal_copy"
-      (ptr t @-> t @-> int64_t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_diagonal_copy_out =
-    foreign
-      "atg_diagonal_copy_out"
-      (ptr t @-> t @-> t @-> int64_t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_diagonal_scatter =
-    foreign
-      "atg_diagonal_scatter"
-      (ptr t @-> t @-> t @-> int64_t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_diagonal_scatter_out =
-    foreign
-      "atg_diagonal_scatter_out"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_diff =
-    foreign "atg_diff" (ptr t @-> t @-> int64_t @-> int64_t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_diff_out =
-    foreign
-      "atg_diff_out"
-      (ptr t @-> t @-> t @-> int64_t @-> int64_t @-> t @-> t @-> returning void)
-  ;;
-end
-
-module C9 (F : Cstubs.FOREIGN) = struct
-  open F
-
-  type t = unit ptr
-
-  let t : t typ = ptr void
-
-  type scalar = unit ptr
-
-  let scalar : scalar typ = ptr void
-  let stubs_digamma = foreign "atg_digamma" (ptr t @-> t @-> returning void)
-  let stubs_digamma_ = foreign "atg_digamma_" (ptr t @-> t @-> returning void)
-  let stubs_digamma_out = foreign "atg_digamma_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_dist = foreign "atg_dist" (ptr t @-> t @-> t @-> returning void)
-  let stubs_dist_out = foreign "atg_dist_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  let stubs_div = foreign "atg_div" (ptr t @-> t @-> t @-> returning void)
-  let stubs_div_ = foreign "atg_div_" (ptr t @-> t @-> t @-> returning void)
-  let stubs_div_out = foreign "atg_div_out" (ptr t @-> t @-> t @-> t @-> returning void)
-
-  let stubs_div_out_mode =
-    foreign "atg_div_out_mode" (ptr t @-> t @-> t @-> t @-> string @-> returning void)
-  ;;
-
-  let stubs_div_scalar =
-    foreign "atg_div_scalar" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_div_scalar_ =
-    foreign "atg_div_scalar_" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_div_scalar_mode =
-    foreign "atg_div_scalar_mode" (ptr t @-> t @-> scalar @-> string @-> returning void)
-  ;;
-
-  let stubs_div_scalar_mode_ =
-    foreign "atg_div_scalar_mode_" (ptr t @-> t @-> scalar @-> string @-> returning void)
-  ;;
-
-  let stubs_div_scalar_mode_out =
-    foreign
-      "atg_div_scalar_mode_out"
-      (ptr t @-> t @-> t @-> scalar @-> string @-> returning void)
-  ;;
-
-  let stubs_div_scalar_out =
-    foreign "atg_div_scalar_out" (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_div_tensor_mode =
-    foreign "atg_div_tensor_mode" (ptr t @-> t @-> t @-> string @-> returning void)
-  ;;
-
-  let stubs_div_tensor_mode_ =
-    foreign "atg_div_tensor_mode_" (ptr t @-> t @-> t @-> string @-> returning void)
-  ;;
-
-  let stubs_divide = foreign "atg_divide" (ptr t @-> t @-> t @-> returning void)
-  let stubs_divide_ = foreign "atg_divide_" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_divide_out =
-    foreign "atg_divide_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_divide_out_mode =
-    foreign "atg_divide_out_mode" (ptr t @-> t @-> t @-> t @-> string @-> returning void)
-  ;;
-
-  let stubs_divide_scalar =
-    foreign "atg_divide_scalar" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_divide_scalar_ =
-    foreign "atg_divide_scalar_" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_divide_scalar_mode =
-    foreign "atg_divide_scalar_mode" (ptr t @-> t @-> scalar @-> string @-> returning void)
-  ;;
-
-  let stubs_divide_scalar_mode_ =
-    foreign
-      "atg_divide_scalar_mode_"
-      (ptr t @-> t @-> scalar @-> string @-> returning void)
-  ;;
-
-  let stubs_divide_tensor_mode =
-    foreign "atg_divide_tensor_mode" (ptr t @-> t @-> t @-> string @-> returning void)
-  ;;
-
-  let stubs_divide_tensor_mode_ =
-    foreign "atg_divide_tensor_mode_" (ptr t @-> t @-> t @-> string @-> returning void)
-  ;;
-
-  let stubs_dot = foreign "atg_dot" (ptr t @-> t @-> t @-> returning void)
-  let stubs_dot_out = foreign "atg_dot_out" (ptr t @-> t @-> t @-> t @-> returning void)
-
-  let stubs_dropout =
-    foreign "atg_dropout" (ptr t @-> t @-> double @-> int @-> returning void)
-  ;;
-
-  let stubs_dropout_ =
-    foreign "atg_dropout_" (ptr t @-> t @-> double @-> int @-> returning void)
-  ;;
-
-  let stubs_dsplit = foreign "atg_dsplit" (t @-> int64_t @-> returning (ptr t))
-
-  let stubs_dsplit_array =
-    foreign "atg_dsplit_array" (t @-> ptr int64_t @-> int @-> returning (ptr t))
-  ;;
-
-  let stubs_dstack = foreign "atg_dstack" (ptr t @-> ptr t @-> int @-> returning void)
-
-  let stubs_dstack_out =
-    foreign "atg_dstack_out" (ptr t @-> t @-> ptr t @-> int @-> returning void)
-  ;;
-
-  let stubs_einsum =
-    foreign
-      "atg_einsum"
-      (ptr t @-> string @-> ptr t @-> int @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_elu = foreign "atg_elu" (ptr t @-> t @-> returning void)
-  let stubs_elu_ = foreign "atg_elu_" (ptr t @-> t @-> returning void)
-
-  let stubs_elu_backward =
-    foreign
-      "atg_elu_backward"
-      (ptr t @-> t @-> scalar @-> scalar @-> scalar @-> int @-> t @-> returning void)
-  ;;
-
-  let stubs_elu_backward_grad_input =
-    foreign
-      "atg_elu_backward_grad_input"
-      (ptr t @-> t @-> t @-> scalar @-> scalar @-> scalar @-> int @-> t @-> returning void)
-  ;;
-
-  let stubs_elu_out = foreign "atg_elu_out" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_embedding =
-    foreign
-      "atg_embedding"
-      (ptr t @-> t @-> t @-> int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_embedding_backward =
-    foreign
-      "atg_embedding_backward"
-      (ptr t @-> t @-> t @-> int64_t @-> int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_embedding_bag =
-    foreign
-      "atg_embedding_bag"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_embedding_bag_padding_idx =
-    foreign
-      "atg_embedding_bag_padding_idx"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> t
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_embedding_dense_backward =
-    foreign
-      "atg_embedding_dense_backward"
-      (ptr t @-> t @-> t @-> int64_t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_embedding_dense_backward_out =
-    foreign
-      "atg_embedding_dense_backward_out"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_embedding_out =
-    foreign
-      "atg_embedding_out"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_embedding_renorm =
-    foreign
-      "atg_embedding_renorm"
-      (ptr t @-> t @-> t @-> double @-> double @-> returning void)
-  ;;
-
-  let stubs_embedding_renorm_ =
-    foreign
-      "atg_embedding_renorm_"
-      (ptr t @-> t @-> t @-> double @-> double @-> returning void)
-  ;;
-
-  let stubs_embedding_renorm_out =
-    foreign
-      "atg_embedding_renorm_out"
-      (ptr t @-> t @-> t @-> t @-> double @-> double @-> returning void)
-  ;;
-
-  let stubs_embedding_sparse_backward =
-    foreign
-      "atg_embedding_sparse_backward"
-      (ptr t @-> t @-> t @-> int64_t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_empty =
-    foreign "atg_empty" (ptr t @-> ptr int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_empty_like = foreign "atg_empty_like" (ptr t @-> t @-> returning void)
-
-  let stubs_empty_like_out =
-    foreign "atg_empty_like_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_empty_out =
-    foreign "atg_empty_out" (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_empty_permuted =
-    foreign
-      "atg_empty_permuted"
-      (ptr t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_empty_permuted_out =
-    foreign
-      "atg_empty_permuted_out"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_empty_quantized =
-    foreign
-      "atg_empty_quantized"
-      (ptr t @-> ptr int64_t @-> int @-> t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_empty_quantized_out =
-    foreign
-      "atg_empty_quantized_out"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> t @-> returning void)
-  ;;
-
-  let stubs_empty_strided =
-    foreign
-      "atg_empty_strided"
-      (ptr t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_empty_strided_out =
-    foreign
-      "atg_empty_strided_out"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_eq = foreign "atg_eq" (ptr t @-> t @-> scalar @-> returning void)
-  let stubs_eq_ = foreign "atg_eq_" (ptr t @-> t @-> scalar @-> returning void)
-
-  let stubs_eq_scalar_out =
-    foreign "atg_eq_scalar_out" (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_eq_tensor = foreign "atg_eq_tensor" (ptr t @-> t @-> t @-> returning void)
-  let stubs_eq_tensor_ = foreign "atg_eq_tensor_" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_eq_tensor_out =
-    foreign "atg_eq_tensor_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_equal = foreign "atg_equal" (t @-> t @-> returning bool)
-  let stubs_erf = foreign "atg_erf" (ptr t @-> t @-> returning void)
-  let stubs_erf_ = foreign "atg_erf_" (ptr t @-> t @-> returning void)
-  let stubs_erf_out = foreign "atg_erf_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_erfc = foreign "atg_erfc" (ptr t @-> t @-> returning void)
-  let stubs_erfc_ = foreign "atg_erfc_" (ptr t @-> t @-> returning void)
-  let stubs_erfc_out = foreign "atg_erfc_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_erfinv = foreign "atg_erfinv" (ptr t @-> t @-> returning void)
-  let stubs_erfinv_ = foreign "atg_erfinv_" (ptr t @-> t @-> returning void)
-  let stubs_erfinv_out = foreign "atg_erfinv_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_exp = foreign "atg_exp" (ptr t @-> t @-> returning void)
-  let stubs_exp2 = foreign "atg_exp2" (ptr t @-> t @-> returning void)
-  let stubs_exp2_ = foreign "atg_exp2_" (ptr t @-> t @-> returning void)
-  let stubs_exp2_out = foreign "atg_exp2_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_exp_ = foreign "atg_exp_" (ptr t @-> t @-> returning void)
-  let stubs_exp_out = foreign "atg_exp_out" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_expand =
-    foreign "atg_expand" (ptr t @-> t @-> ptr int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_expand_as = foreign "atg_expand_as" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_expand_copy =
-    foreign
-      "atg_expand_copy"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_expand_copy_out =
-    foreign
-      "atg_expand_copy_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_expm1 = foreign "atg_expm1" (ptr t @-> t @-> returning void)
-  let stubs_expm1_ = foreign "atg_expm1_" (ptr t @-> t @-> returning void)
-  let stubs_expm1_out = foreign "atg_expm1_out" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_exponential =
-    foreign "atg_exponential" (ptr t @-> t @-> double @-> returning void)
-  ;;
-
-  let stubs_exponential_ =
-    foreign "atg_exponential_" (ptr t @-> t @-> double @-> returning void)
-  ;;
-
-  let stubs_exponential_out =
-    foreign "atg_exponential_out" (ptr t @-> t @-> t @-> double @-> returning void)
-  ;;
-
-  let stubs_eye = foreign "atg_eye" (ptr t @-> int64_t @-> int @-> int @-> returning void)
-
-  let stubs_eye_m =
-    foreign "atg_eye_m" (ptr t @-> int64_t @-> int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_eye_m_out =
-    foreign "atg_eye_m_out" (ptr t @-> t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_eye_out = foreign "atg_eye_out" (ptr t @-> t @-> int64_t @-> returning void)
-
-  let stubs_fake_quantize_per_channel_affine =
-    foreign
-      "atg_fake_quantize_per_channel_affine"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_fake_quantize_per_channel_affine_cachemask =
-    foreign
-      "atg_fake_quantize_per_channel_affine_cachemask"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-end
-
-module C10 (F : Cstubs.FOREIGN) = struct
-  open F
-
-  type t = unit ptr
-
-  let t : t typ = ptr void
-
-  type scalar = unit ptr
-
-  let scalar : scalar typ = ptr void
-
-  let stubs_fake_quantize_per_channel_affine_cachemask_backward =
-    foreign
-      "atg_fake_quantize_per_channel_affine_cachemask_backward"
-      (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_fake_quantize_per_channel_affine_cachemask_out =
-    foreign
-      "atg_fake_quantize_per_channel_affine_cachemask_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int64_t
-       @-> int64_t
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs_fake_quantize_per_tensor_affine =
-    foreign
-      "atg_fake_quantize_per_tensor_affine"
-      (ptr t @-> t @-> double @-> int64_t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_fake_quantize_per_tensor_affine_cachemask =
-    foreign
-      "atg_fake_quantize_per_tensor_affine_cachemask"
-      (ptr t @-> t @-> double @-> int64_t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_fake_quantize_per_tensor_affine_cachemask_backward =
-    foreign
-      "atg_fake_quantize_per_tensor_affine_cachemask_backward"
-      (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_fake_quantize_per_tensor_affine_cachemask_out =
-    foreign
-      "atg_fake_quantize_per_tensor_affine_cachemask_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> double
-       @-> int64_t
-       @-> int64_t
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs_fake_quantize_per_tensor_affine_tensor_qparams =
-    foreign
-      "atg_fake_quantize_per_tensor_affine_tensor_qparams"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_fbgemm_linear_fp16_weight =
-    foreign "atg_fbgemm_linear_fp16_weight" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_fbgemm_linear_fp16_weight_fp32_activation =
-    foreign
-      "atg_fbgemm_linear_fp16_weight_fp32_activation"
-      (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_fbgemm_linear_int8_weight =
-    foreign
-      "atg_fbgemm_linear_int8_weight"
-      (ptr t @-> t @-> t @-> t @-> t @-> scalar @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_fbgemm_linear_int8_weight_fp32_activation =
-    foreign
-      "atg_fbgemm_linear_int8_weight_fp32_activation"
-      (ptr t @-> t @-> t @-> t @-> t @-> scalar @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_fbgemm_pack_gemm_matrix_fp16 =
-    foreign "atg_fbgemm_pack_gemm_matrix_fp16" (ptr t @-> t @-> returning void)
-  ;;
-
-  let stubs_fbgemm_pack_quantized_matrix =
-    foreign "atg_fbgemm_pack_quantized_matrix" (ptr t @-> t @-> returning void)
-  ;;
-
-  let stubs_fbgemm_pack_quantized_matrix_kn =
-    foreign
-      "atg_fbgemm_pack_quantized_matrix_kn"
-      (ptr t @-> t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_feature_alpha_dropout =
-    foreign "atg_feature_alpha_dropout" (ptr t @-> t @-> double @-> int @-> returning void)
-  ;;
-
-  let stubs_feature_alpha_dropout_ =
-    foreign
-      "atg_feature_alpha_dropout_"
-      (ptr t @-> t @-> double @-> int @-> returning void)
-  ;;
-
-  let stubs_feature_dropout =
-    foreign "atg_feature_dropout" (ptr t @-> t @-> double @-> int @-> returning void)
-  ;;
-
-  let stubs_feature_dropout_ =
-    foreign "atg_feature_dropout_" (ptr t @-> t @-> double @-> int @-> returning void)
-  ;;
-
-  let stubs_fft_fft =
-    foreign
-      "atg_fft_fft"
-      (ptr t @-> t @-> int64_t @-> int @-> int64_t @-> string @-> returning void)
-  ;;
-
-  let stubs_fft_fft2 =
-    foreign
-      "atg_fft_fft2"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> string
-       @-> returning void)
-  ;;
-
-  let stubs_fft_fft2_out =
-    foreign
-      "atg_fft_fft2_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> string
-       @-> returning void)
-  ;;
-
-  let stubs_fft_fft_out =
-    foreign
-      "atg_fft_fft_out"
-      (ptr t @-> t @-> t @-> int64_t @-> int @-> int64_t @-> string @-> returning void)
-  ;;
-
-  let stubs_fft_fftfreq =
-    foreign
-      "atg_fft_fftfreq"
-      (ptr t @-> int64_t @-> double @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_fft_fftfreq_out =
-    foreign "atg_fft_fftfreq_out" (ptr t @-> t @-> int64_t @-> double @-> returning void)
-  ;;
-
-  let stubs_fft_fftn =
-    foreign
-      "atg_fft_fftn"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> string
-       @-> returning void)
-  ;;
-
-  let stubs_fft_fftn_out =
-    foreign
-      "atg_fft_fftn_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> string
-       @-> returning void)
-  ;;
-
-  let stubs_fft_fftshift =
-    foreign "atg_fft_fftshift" (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_fft_hfft =
-    foreign
-      "atg_fft_hfft"
-      (ptr t @-> t @-> int64_t @-> int @-> int64_t @-> string @-> returning void)
-  ;;
-
-  let stubs_fft_hfft2 =
-    foreign
-      "atg_fft_hfft2"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> string
-       @-> returning void)
-  ;;
-
-  let stubs_fft_hfft2_out =
-    foreign
-      "atg_fft_hfft2_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> string
-       @-> returning void)
-  ;;
-
-  let stubs_fft_hfft_out =
-    foreign
-      "atg_fft_hfft_out"
-      (ptr t @-> t @-> t @-> int64_t @-> int @-> int64_t @-> string @-> returning void)
-  ;;
-
-  let stubs_fft_hfftn =
-    foreign
-      "atg_fft_hfftn"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> string
-       @-> returning void)
-  ;;
-
-  let stubs_fft_hfftn_out =
-    foreign
-      "atg_fft_hfftn_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> string
-       @-> returning void)
-  ;;
-
-  let stubs_fft_ifft =
-    foreign
-      "atg_fft_ifft"
-      (ptr t @-> t @-> int64_t @-> int @-> int64_t @-> string @-> returning void)
-  ;;
-
-  let stubs_fft_ifft2 =
-    foreign
-      "atg_fft_ifft2"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> string
-       @-> returning void)
-  ;;
-
-  let stubs_fft_ifft2_out =
-    foreign
-      "atg_fft_ifft2_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> string
-       @-> returning void)
-  ;;
-
-  let stubs_fft_ifft_out =
-    foreign
-      "atg_fft_ifft_out"
-      (ptr t @-> t @-> t @-> int64_t @-> int @-> int64_t @-> string @-> returning void)
-  ;;
-
-  let stubs_fft_ifftn =
-    foreign
-      "atg_fft_ifftn"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> string
-       @-> returning void)
-  ;;
-
-  let stubs_fft_ifftn_out =
-    foreign
-      "atg_fft_ifftn_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> string
-       @-> returning void)
-  ;;
-
-  let stubs_fft_ifftshift =
-    foreign "atg_fft_ifftshift" (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_fft_ihfft =
-    foreign
-      "atg_fft_ihfft"
-      (ptr t @-> t @-> int64_t @-> int @-> int64_t @-> string @-> returning void)
-  ;;
-
-  let stubs_fft_ihfft2 =
-    foreign
-      "atg_fft_ihfft2"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> string
-       @-> returning void)
-  ;;
-
-  let stubs_fft_ihfft2_out =
-    foreign
-      "atg_fft_ihfft2_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> string
-       @-> returning void)
-  ;;
-
-  let stubs_fft_ihfft_out =
-    foreign
-      "atg_fft_ihfft_out"
-      (ptr t @-> t @-> t @-> int64_t @-> int @-> int64_t @-> string @-> returning void)
-  ;;
-
-  let stubs_fft_ihfftn =
-    foreign
-      "atg_fft_ihfftn"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> string
-       @-> returning void)
-  ;;
-
-  let stubs_fft_ihfftn_out =
-    foreign
-      "atg_fft_ihfftn_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> string
-       @-> returning void)
-  ;;
-
-  let stubs_fft_irfft =
-    foreign
-      "atg_fft_irfft"
-      (ptr t @-> t @-> int64_t @-> int @-> int64_t @-> string @-> returning void)
-  ;;
-
-  let stubs_fft_irfft2 =
-    foreign
-      "atg_fft_irfft2"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> string
-       @-> returning void)
-  ;;
-
-  let stubs_fft_irfft2_out =
-    foreign
-      "atg_fft_irfft2_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> string
-       @-> returning void)
-  ;;
-
-  let stubs_fft_irfft_out =
-    foreign
-      "atg_fft_irfft_out"
-      (ptr t @-> t @-> t @-> int64_t @-> int @-> int64_t @-> string @-> returning void)
-  ;;
-
-  let stubs_fft_irfftn =
-    foreign
-      "atg_fft_irfftn"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> string
-       @-> returning void)
-  ;;
-
-  let stubs_fft_irfftn_out =
-    foreign
-      "atg_fft_irfftn_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> string
-       @-> returning void)
-  ;;
-
-  let stubs_fft_rfft =
-    foreign
-      "atg_fft_rfft"
-      (ptr t @-> t @-> int64_t @-> int @-> int64_t @-> string @-> returning void)
-  ;;
-
-  let stubs_fft_rfft2 =
-    foreign
-      "atg_fft_rfft2"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> string
-       @-> returning void)
-  ;;
-
-  let stubs_fft_rfft2_out =
-    foreign
-      "atg_fft_rfft2_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> string
-       @-> returning void)
-  ;;
-
-  let stubs_fft_rfft_out =
-    foreign
-      "atg_fft_rfft_out"
-      (ptr t @-> t @-> t @-> int64_t @-> int @-> int64_t @-> string @-> returning void)
-  ;;
-
-  let stubs_fft_rfftfreq =
-    foreign
-      "atg_fft_rfftfreq"
-      (ptr t @-> int64_t @-> double @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_fft_rfftfreq_out =
-    foreign "atg_fft_rfftfreq_out" (ptr t @-> t @-> int64_t @-> double @-> returning void)
-  ;;
-
-  let stubs_fft_rfftn =
-    foreign
-      "atg_fft_rfftn"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> string
-       @-> returning void)
-  ;;
-
-  let stubs_fft_rfftn_out =
-    foreign
-      "atg_fft_rfftn_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> string
-       @-> returning void)
-  ;;
-
-  let stubs_fill = foreign "atg_fill" (ptr t @-> t @-> scalar @-> returning void)
-  let stubs_fill_ = foreign "atg_fill_" (ptr t @-> t @-> scalar @-> returning void)
-
-  let stubs_fill_diagonal_ =
-    foreign "atg_fill_diagonal_" (ptr t @-> t @-> scalar @-> int @-> returning void)
-  ;;
-
-  let stubs_fill_scalar_out =
-    foreign "atg_fill_scalar_out" (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_fill_tensor = foreign "atg_fill_tensor" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_fill_tensor_ =
-    foreign "atg_fill_tensor_" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_fill_tensor_out =
-    foreign "atg_fill_tensor_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_fix = foreign "atg_fix" (ptr t @-> t @-> returning void)
-  let stubs_fix_ = foreign "atg_fix_" (ptr t @-> t @-> returning void)
-  let stubs_fix_out = foreign "atg_fix_out" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_flatten =
-    foreign "atg_flatten" (ptr t @-> t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_flatten_dense_tensors =
-    foreign "atg_flatten_dense_tensors" (ptr t @-> ptr t @-> int @-> returning void)
-  ;;
-
-  let stubs_flip =
-    foreign "atg_flip" (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_flip_out =
-    foreign "atg_flip_out" (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_fliplr = foreign "atg_fliplr" (ptr t @-> t @-> returning void)
-  let stubs_flipud = foreign "atg_flipud" (ptr t @-> t @-> returning void)
-  let stubs_float_power = foreign "atg_float_power" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_float_power_ =
-    foreign "atg_float_power_" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_float_power_scalar =
-    foreign "atg_float_power_scalar" (ptr t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_float_power_scalar_out =
-    foreign "atg_float_power_scalar_out" (ptr t @-> t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_float_power_tensor_ =
-    foreign "atg_float_power_tensor_" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_float_power_tensor_scalar =
-    foreign "atg_float_power_tensor_scalar" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_float_power_tensor_scalar_out =
-    foreign
-      "atg_float_power_tensor_scalar_out"
-      (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_float_power_tensor_tensor_out =
-    foreign
-      "atg_float_power_tensor_tensor_out"
-      (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_floor = foreign "atg_floor" (ptr t @-> t @-> returning void)
-  let stubs_floor_ = foreign "atg_floor_" (ptr t @-> t @-> returning void)
-
-  let stubs_floor_divide =
-    foreign "atg_floor_divide" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_floor_divide_ =
-    foreign "atg_floor_divide_" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_floor_divide_out =
-    foreign "atg_floor_divide_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_floor_divide_scalar =
-    foreign "atg_floor_divide_scalar" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_floor_divide_scalar_ =
-    foreign "atg_floor_divide_scalar_" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_floor_out = foreign "atg_floor_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_fmax = foreign "atg_fmax" (ptr t @-> t @-> t @-> returning void)
-  let stubs_fmax_out = foreign "atg_fmax_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  let stubs_fmin = foreign "atg_fmin" (ptr t @-> t @-> t @-> returning void)
-  let stubs_fmin_out = foreign "atg_fmin_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  let stubs_fmod = foreign "atg_fmod" (ptr t @-> t @-> scalar @-> returning void)
-  let stubs_fmod_ = foreign "atg_fmod_" (ptr t @-> t @-> scalar @-> returning void)
-
-  let stubs_fmod_scalar_out =
-    foreign "atg_fmod_scalar_out" (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_fmod_tensor = foreign "atg_fmod_tensor" (ptr t @-> t @-> t @-> returning void)
-end
-
-module C11 (F : Cstubs.FOREIGN) = struct
-  open F
-
-  type t = unit ptr
-
-  let t : t typ = ptr void
-
-  type scalar = unit ptr
-
-  let scalar : scalar typ = ptr void
-
-  let stubs_fmod_tensor_ =
-    foreign "atg_fmod_tensor_" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_fmod_tensor_out =
-    foreign "atg_fmod_tensor_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_frac = foreign "atg_frac" (ptr t @-> t @-> returning void)
-  let stubs_frac_ = foreign "atg_frac_" (ptr t @-> t @-> returning void)
-  let stubs_frac_out = foreign "atg_frac_out" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_fractional_max_pool2d =
-    foreign
-      "atg_fractional_max_pool2d"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> t
-       @-> returning void)
-  ;;
-
-  let stubs_fractional_max_pool2d_backward =
-    foreign
-      "atg_fractional_max_pool2d_backward"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> t
-       @-> returning void)
-  ;;
-
-  let stubs_fractional_max_pool2d_backward_grad_input =
-    foreign
-      "atg_fractional_max_pool2d_backward_grad_input"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> t
-       @-> returning void)
-  ;;
-
-  let stubs_fractional_max_pool2d_output =
-    foreign
-      "atg_fractional_max_pool2d_output"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> t
-       @-> returning void)
-  ;;
-
-  let stubs_fractional_max_pool3d =
-    foreign
-      "atg_fractional_max_pool3d"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> t
-       @-> returning void)
-  ;;
-
-  let stubs_fractional_max_pool3d_backward =
-    foreign
-      "atg_fractional_max_pool3d_backward"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> t
-       @-> returning void)
-  ;;
-
-  let stubs_fractional_max_pool3d_backward_grad_input =
-    foreign
-      "atg_fractional_max_pool3d_backward_grad_input"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> t
-       @-> returning void)
-  ;;
-
-  let stubs_fractional_max_pool3d_output =
-    foreign
-      "atg_fractional_max_pool3d_output"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> t
-       @-> returning void)
-  ;;
-
-  let stubs_frexp = foreign "atg_frexp" (ptr t @-> t @-> returning void)
-
-  let stubs_frexp_tensor_out =
-    foreign "atg_frexp_tensor_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_frobenius_norm =
-    foreign
-      "atg_frobenius_norm"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_frobenius_norm_out =
-    foreign
-      "atg_frobenius_norm_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_from_file =
-    foreign
-      "atg_from_file"
-      (ptr t @-> string @-> int @-> int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_from_file_out =
-    foreign
-      "atg_from_file_out"
-      (ptr t @-> t @-> string @-> int @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_full =
-    foreign
-      "atg_full"
-      (ptr t @-> ptr int64_t @-> int @-> scalar @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_full_like = foreign "atg_full_like" (ptr t @-> t @-> scalar @-> returning void)
-
-  let stubs_full_like_out =
-    foreign "atg_full_like_out" (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_full_out =
-    foreign
-      "atg_full_out"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> scalar @-> returning void)
-  ;;
-
-  let stubs_fused_moving_avg_obs_fake_quant =
-    foreign
-      "atg_fused_moving_avg_obs_fake_quant"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> double
-       @-> int64_t
-       @-> int64_t
-       @-> int64_t
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_gather =
-    foreign "atg_gather" (ptr t @-> t @-> int64_t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_gather_backward =
-    foreign
-      "atg_gather_backward"
-      (ptr t @-> t @-> t @-> int64_t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_gather_out =
-    foreign
-      "atg_gather_out"
-      (ptr t @-> t @-> t @-> int64_t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_gcd = foreign "atg_gcd" (ptr t @-> t @-> t @-> returning void)
-  let stubs_gcd_ = foreign "atg_gcd_" (ptr t @-> t @-> t @-> returning void)
-  let stubs_gcd_out = foreign "atg_gcd_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  let stubs_ge = foreign "atg_ge" (ptr t @-> t @-> scalar @-> returning void)
-  let stubs_ge_ = foreign "atg_ge_" (ptr t @-> t @-> scalar @-> returning void)
-
-  let stubs_ge_scalar_out =
-    foreign "atg_ge_scalar_out" (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_ge_tensor = foreign "atg_ge_tensor" (ptr t @-> t @-> t @-> returning void)
-  let stubs_ge_tensor_ = foreign "atg_ge_tensor_" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_ge_tensor_out =
-    foreign "atg_ge_tensor_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_gelu = foreign "atg_gelu" (ptr t @-> t @-> string @-> returning void)
-  let stubs_gelu_ = foreign "atg_gelu_" (ptr t @-> t @-> string @-> returning void)
-
-  let stubs_gelu_backward =
-    foreign "atg_gelu_backward" (ptr t @-> t @-> t @-> string @-> returning void)
-  ;;
-
-  let stubs_gelu_backward_grad_input =
-    foreign
-      "atg_gelu_backward_grad_input"
-      (ptr t @-> t @-> t @-> t @-> string @-> returning void)
-  ;;
-
-  let stubs_gelu_out =
-    foreign "atg_gelu_out" (ptr t @-> t @-> t @-> string @-> returning void)
-  ;;
-
-  let stubs_geometric = foreign "atg_geometric" (ptr t @-> t @-> double @-> returning void)
-
-  let stubs_geometric_ =
-    foreign "atg_geometric_" (ptr t @-> t @-> double @-> returning void)
-  ;;
-
-  let stubs_geometric_out =
-    foreign "atg_geometric_out" (ptr t @-> t @-> t @-> double @-> returning void)
-  ;;
-
-  let stubs_geqrf = foreign "atg_geqrf" (ptr t @-> t @-> returning void)
-  let stubs_geqrf_a = foreign "atg_geqrf_a" (ptr t @-> t @-> t @-> t @-> returning void)
-  let stubs_ger = foreign "atg_ger" (ptr t @-> t @-> t @-> returning void)
-  let stubs_ger_out = foreign "atg_ger_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  let stubs_glu = foreign "atg_glu" (ptr t @-> t @-> int64_t @-> returning void)
-
-  let stubs_glu_backward =
-    foreign "atg_glu_backward" (ptr t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_glu_backward_grad_input =
-    foreign
-      "atg_glu_backward_grad_input"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_glu_backward_jvp =
-    foreign
-      "atg_glu_backward_jvp"
-      (ptr t @-> t @-> t @-> t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_glu_backward_jvp_out =
-    foreign
-      "atg_glu_backward_jvp_out"
-      (ptr t @-> t @-> t @-> t @-> t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_glu_jvp =
-    foreign "atg_glu_jvp" (ptr t @-> t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_glu_jvp_out =
-    foreign
-      "atg_glu_jvp_out"
-      (ptr t @-> t @-> t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_glu_out =
-    foreign "atg_glu_out" (ptr t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_grad = foreign "atg_grad" (ptr t @-> t @-> returning void)
-  let stubs_greater = foreign "atg_greater" (ptr t @-> t @-> scalar @-> returning void)
-  let stubs_greater_ = foreign "atg_greater_" (ptr t @-> t @-> scalar @-> returning void)
-
-  let stubs_greater_equal =
-    foreign "atg_greater_equal" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_greater_equal_ =
-    foreign "atg_greater_equal_" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_greater_equal_scalar_out =
-    foreign
-      "atg_greater_equal_scalar_out"
-      (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_greater_equal_tensor =
-    foreign "atg_greater_equal_tensor" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_greater_equal_tensor_ =
-    foreign "atg_greater_equal_tensor_" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_greater_equal_tensor_out =
-    foreign "atg_greater_equal_tensor_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_greater_scalar_out =
-    foreign "atg_greater_scalar_out" (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_greater_tensor =
-    foreign "atg_greater_tensor" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_greater_tensor_ =
-    foreign "atg_greater_tensor_" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_greater_tensor_out =
-    foreign "atg_greater_tensor_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_grid_sampler =
-    foreign
-      "atg_grid_sampler"
-      (ptr t @-> t @-> t @-> int64_t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_grid_sampler_2d =
-    foreign
-      "atg_grid_sampler_2d"
-      (ptr t @-> t @-> t @-> int64_t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_grid_sampler_2d_out =
-    foreign
-      "atg_grid_sampler_2d_out"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_grid_sampler_3d =
-    foreign
-      "atg_grid_sampler_3d"
-      (ptr t @-> t @-> t @-> int64_t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_grid_sampler_3d_out =
-    foreign
-      "atg_grid_sampler_3d_out"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_group_norm =
-    foreign
-      "atg_group_norm"
-      (ptr t @-> t @-> int64_t @-> t @-> t @-> double @-> int @-> returning void)
-  ;;
-
-  let stubs_gru =
-    foreign
-      "atg_gru"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr t
-       @-> int
-       @-> int
-       @-> int64_t
-       @-> double
-       @-> int
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_gru_cell =
-    foreign "atg_gru_cell" (ptr t @-> t @-> t @-> t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_gru_data =
-    foreign
-      "atg_gru_data"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr t
-       @-> int
-       @-> int
-       @-> int64_t
-       @-> double
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_gt = foreign "atg_gt" (ptr t @-> t @-> scalar @-> returning void)
-  let stubs_gt_ = foreign "atg_gt_" (ptr t @-> t @-> scalar @-> returning void)
-
-  let stubs_gt_scalar_out =
-    foreign "atg_gt_scalar_out" (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_gt_tensor = foreign "atg_gt_tensor" (ptr t @-> t @-> t @-> returning void)
-  let stubs_gt_tensor_ = foreign "atg_gt_tensor_" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_gt_tensor_out =
-    foreign "atg_gt_tensor_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_hamming_window =
-    foreign "atg_hamming_window" (ptr t @-> int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_hamming_window_out =
-    foreign "atg_hamming_window_out" (ptr t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_hamming_window_periodic =
-    foreign
-      "atg_hamming_window_periodic"
-      (ptr t @-> int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_hamming_window_periodic_alpha =
-    foreign
-      "atg_hamming_window_periodic_alpha"
-      (ptr t @-> int64_t @-> int @-> double @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_hamming_window_periodic_alpha_beta =
-    foreign
-      "atg_hamming_window_periodic_alpha_beta"
-      (ptr t @-> int64_t @-> int @-> double @-> double @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_hamming_window_periodic_alpha_beta_out =
-    foreign
-      "atg_hamming_window_periodic_alpha_beta_out"
-      (ptr t @-> t @-> int64_t @-> int @-> double @-> double @-> returning void)
-  ;;
-
-  let stubs_hamming_window_periodic_alpha_out =
-    foreign
-      "atg_hamming_window_periodic_alpha_out"
-      (ptr t @-> t @-> int64_t @-> int @-> double @-> returning void)
-  ;;
-
-  let stubs_hamming_window_periodic_out =
-    foreign
-      "atg_hamming_window_periodic_out"
-      (ptr t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_hann_window =
-    foreign "atg_hann_window" (ptr t @-> int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_hann_window_out =
-    foreign "atg_hann_window_out" (ptr t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_hann_window_periodic =
-    foreign
-      "atg_hann_window_periodic"
-      (ptr t @-> int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_hann_window_periodic_out =
-    foreign
-      "atg_hann_window_periodic_out"
-      (ptr t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_hardshrink = foreign "atg_hardshrink" (ptr t @-> t @-> returning void)
-
-  let stubs_hardshrink_backward =
-    foreign "atg_hardshrink_backward" (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_hardshrink_backward_grad_input =
-    foreign
-      "atg_hardshrink_backward_grad_input"
-      (ptr t @-> t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_hardshrink_out =
-    foreign "atg_hardshrink_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-end
-
-module C12 (F : Cstubs.FOREIGN) = struct
-  open F
-
-  type t = unit ptr
-
-  let t : t typ = ptr void
-
-  type scalar = unit ptr
-
-  let scalar : scalar typ = ptr void
-  let stubs_hardsigmoid = foreign "atg_hardsigmoid" (ptr t @-> t @-> returning void)
-  let stubs_hardsigmoid_ = foreign "atg_hardsigmoid_" (ptr t @-> t @-> returning void)
-
-  let stubs_hardsigmoid_backward =
-    foreign "atg_hardsigmoid_backward" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_hardsigmoid_backward_grad_input =
-    foreign
-      "atg_hardsigmoid_backward_grad_input"
-      (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_hardsigmoid_out =
-    foreign "atg_hardsigmoid_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_hardswish = foreign "atg_hardswish" (ptr t @-> t @-> returning void)
-  let stubs_hardswish_ = foreign "atg_hardswish_" (ptr t @-> t @-> returning void)
-
-  let stubs_hardswish_backward =
-    foreign "atg_hardswish_backward" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_hardswish_backward_out =
-    foreign "atg_hardswish_backward_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_hardswish_out =
-    foreign "atg_hardswish_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_hardtanh = foreign "atg_hardtanh" (ptr t @-> t @-> returning void)
-  let stubs_hardtanh_ = foreign "atg_hardtanh_" (ptr t @-> t @-> returning void)
-
-  let stubs_hardtanh_backward =
-    foreign
-      "atg_hardtanh_backward"
-      (ptr t @-> t @-> t @-> scalar @-> scalar @-> returning void)
-  ;;
-
-  let stubs_hardtanh_backward_grad_input =
-    foreign
-      "atg_hardtanh_backward_grad_input"
-      (ptr t @-> t @-> t @-> t @-> scalar @-> scalar @-> returning void)
-  ;;
-
-  let stubs_hardtanh_out =
-    foreign "atg_hardtanh_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_heaviside = foreign "atg_heaviside" (ptr t @-> t @-> t @-> returning void)
-  let stubs_heaviside_ = foreign "atg_heaviside_" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_heaviside_out =
-    foreign "atg_heaviside_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_hinge_embedding_loss =
-    foreign
-      "atg_hinge_embedding_loss"
-      (ptr t @-> t @-> t @-> double @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_histc = foreign "atg_histc" (ptr t @-> t @-> int64_t @-> returning void)
-
-  let stubs_histc_out =
-    foreign "atg_histc_out" (ptr t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_hsplit = foreign "atg_hsplit" (t @-> int64_t @-> returning (ptr t))
-
-  let stubs_hsplit_array =
-    foreign "atg_hsplit_array" (t @-> ptr int64_t @-> int @-> returning (ptr t))
-  ;;
-
-  let stubs_hspmm = foreign "atg_hspmm" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_hspmm_out =
-    foreign "atg_hspmm_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_hstack = foreign "atg_hstack" (ptr t @-> ptr t @-> int @-> returning void)
-
-  let stubs_hstack_out =
-    foreign "atg_hstack_out" (ptr t @-> t @-> ptr t @-> int @-> returning void)
-  ;;
-
-  let stubs_huber_loss =
-    foreign "atg_huber_loss" (ptr t @-> t @-> t @-> int64_t @-> double @-> returning void)
-  ;;
-
-  let stubs_huber_loss_backward =
-    foreign
-      "atg_huber_loss_backward"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> double @-> returning void)
-  ;;
-
-  let stubs_huber_loss_backward_out =
-    foreign
-      "atg_huber_loss_backward_out"
-      (ptr t @-> t @-> t @-> t @-> t @-> int64_t @-> double @-> returning void)
-  ;;
-
-  let stubs_huber_loss_out =
-    foreign
-      "atg_huber_loss_out"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> double @-> returning void)
-  ;;
-
-  let stubs_hypot = foreign "atg_hypot" (ptr t @-> t @-> t @-> returning void)
-  let stubs_hypot_ = foreign "atg_hypot_" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_hypot_out =
-    foreign "atg_hypot_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_i0 = foreign "atg_i0" (ptr t @-> t @-> returning void)
-  let stubs_i0_ = foreign "atg_i0_" (ptr t @-> t @-> returning void)
-  let stubs_i0_out = foreign "atg_i0_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_igamma = foreign "atg_igamma" (ptr t @-> t @-> t @-> returning void)
-  let stubs_igamma_ = foreign "atg_igamma_" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_igamma_out =
-    foreign "atg_igamma_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_igammac = foreign "atg_igammac" (ptr t @-> t @-> t @-> returning void)
-  let stubs_igammac_ = foreign "atg_igammac_" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_igammac_out =
-    foreign "atg_igammac_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_im2col =
-    foreign
-      "atg_im2col"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_im2col_out =
-    foreign
-      "atg_im2col_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_imag = foreign "atg_imag" (ptr t @-> t @-> returning void)
-  let stubs_index = foreign "atg_index" (ptr t @-> t @-> ptr t @-> int @-> returning void)
-
-  let stubs_index_add =
-    foreign "atg_index_add" (ptr t @-> t @-> int64_t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_index_add_ =
-    foreign "atg_index_add_" (ptr t @-> t @-> int64_t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_index_add_out =
-    foreign
-      "atg_index_add_out"
-      (ptr t @-> t @-> t @-> int64_t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_index_copy =
-    foreign "atg_index_copy" (ptr t @-> t @-> int64_t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_index_copy_ =
-    foreign "atg_index_copy_" (ptr t @-> t @-> int64_t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_index_copy_out =
-    foreign
-      "atg_index_copy_out"
-      (ptr t @-> t @-> t @-> int64_t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_index_fill =
-    foreign "atg_index_fill" (ptr t @-> t @-> int64_t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_index_fill_ =
-    foreign "atg_index_fill_" (ptr t @-> t @-> int64_t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_index_fill_int_scalar_out =
-    foreign
-      "atg_index_fill_int_scalar_out"
-      (ptr t @-> t @-> t @-> int64_t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_index_fill_int_tensor =
-    foreign
-      "atg_index_fill_int_tensor"
-      (ptr t @-> t @-> int64_t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_index_fill_int_tensor_ =
-    foreign
-      "atg_index_fill_int_tensor_"
-      (ptr t @-> t @-> int64_t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_index_fill_int_tensor_out =
-    foreign
-      "atg_index_fill_int_tensor_out"
-      (ptr t @-> t @-> t @-> int64_t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_index_put =
-    foreign
-      "atg_index_put"
-      (ptr t @-> t @-> ptr t @-> int @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_index_put_ =
-    foreign
-      "atg_index_put_"
-      (ptr t @-> t @-> ptr t @-> int @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_index_put_out =
-    foreign
-      "atg_index_put_out"
-      (ptr t @-> t @-> t @-> ptr t @-> int @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_index_reduce =
-    foreign
-      "atg_index_reduce"
-      (ptr t @-> t @-> int64_t @-> t @-> t @-> string @-> int @-> returning void)
-  ;;
-
-  let stubs_index_reduce_ =
-    foreign
-      "atg_index_reduce_"
-      (ptr t @-> t @-> int64_t @-> t @-> t @-> string @-> int @-> returning void)
-  ;;
-
-  let stubs_index_reduce_out =
-    foreign
-      "atg_index_reduce_out"
-      (ptr t @-> t @-> t @-> int64_t @-> t @-> t @-> string @-> int @-> returning void)
-  ;;
-
-  let stubs_index_select =
-    foreign "atg_index_select" (ptr t @-> t @-> int64_t @-> t @-> returning void)
-  ;;
-
-  let stubs_index_select_backward =
-    foreign
-      "atg_index_select_backward"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> int64_t @-> t @-> returning void)
-  ;;
-
-  let stubs_index_select_out =
-    foreign "atg_index_select_out" (ptr t @-> t @-> t @-> int64_t @-> t @-> returning void)
-  ;;
-
-  let stubs_index_tensor_out =
-    foreign "atg_index_tensor_out" (ptr t @-> t @-> t @-> ptr t @-> int @-> returning void)
-  ;;
-
-  let stubs_indices = foreign "atg_indices" (ptr t @-> t @-> returning void)
-  let stubs_indices_copy = foreign "atg_indices_copy" (ptr t @-> t @-> returning void)
-
-  let stubs_indices_copy_out =
-    foreign "atg_indices_copy_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_infinitely_differentiable_gelu_backward =
-    foreign
-      "atg_infinitely_differentiable_gelu_backward"
-      (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_inner = foreign "atg_inner" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_inner_out =
-    foreign "atg_inner_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_instance_norm =
-    foreign
-      "atg_instance_norm"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int
-       @-> double
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_int_repr = foreign "atg_int_repr" (ptr t @-> t @-> returning void)
-
-  let stubs_int_repr_out =
-    foreign "atg_int_repr_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_inverse = foreign "atg_inverse" (ptr t @-> t @-> returning void)
-  let stubs_inverse_out = foreign "atg_inverse_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_is_coalesced = foreign "atg_is_coalesced" (t @-> returning bool)
-  let stubs_is_complex = foreign "atg_is_complex" (t @-> returning bool)
-  let stubs_is_conj = foreign "atg_is_conj" (t @-> returning bool)
-  let stubs_is_distributed = foreign "atg_is_distributed" (t @-> returning bool)
-  let stubs_is_floating_point = foreign "atg_is_floating_point" (t @-> returning bool)
-  let stubs_is_inference = foreign "atg_is_inference" (t @-> returning bool)
-  let stubs_is_leaf = foreign "atg_is_leaf" (t @-> returning bool)
-  let stubs_is_neg = foreign "atg_is_neg" (t @-> returning bool)
-  let stubs_is_nonzero = foreign "atg_is_nonzero" (t @-> returning bool)
-  let stubs_is_pinned = foreign "atg_is_pinned" (t @-> int @-> returning bool)
-  let stubs_is_same_size = foreign "atg_is_same_size" (t @-> t @-> returning bool)
-  let stubs_is_set_to = foreign "atg_is_set_to" (t @-> t @-> returning bool)
-  let stubs_is_signed = foreign "atg_is_signed" (t @-> returning bool)
-
-  let stubs_is_vulkan_available =
-    foreign "atg_is_vulkan_available" (void @-> returning bool)
-  ;;
-
-  let stubs_isclose =
-    foreign
-      "atg_isclose"
-      (ptr t @-> t @-> t @-> double @-> double @-> int @-> returning void)
-  ;;
-
-  let stubs_isfinite = foreign "atg_isfinite" (ptr t @-> t @-> returning void)
-
-  let stubs_isin =
-    foreign "atg_isin" (ptr t @-> t @-> t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_isin_scalar_tensor =
-    foreign
-      "atg_isin_scalar_tensor"
-      (ptr t @-> scalar @-> t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_isin_scalar_tensor_out =
-    foreign
-      "atg_isin_scalar_tensor_out"
-      (ptr t @-> t @-> scalar @-> t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_isin_tensor_scalar =
-    foreign
-      "atg_isin_tensor_scalar"
-      (ptr t @-> t @-> scalar @-> int @-> int @-> returning void)
-  ;;
-end
-
-module C13 (F : Cstubs.FOREIGN) = struct
-  open F
-
-  type t = unit ptr
-
-  let t : t typ = ptr void
-
-  type scalar = unit ptr
-
-  let scalar : scalar typ = ptr void
-
-  let stubs_isin_tensor_scalar_out =
-    foreign
-      "atg_isin_tensor_scalar_out"
-      (ptr t @-> t @-> t @-> scalar @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_isin_tensor_tensor_out =
-    foreign
-      "atg_isin_tensor_tensor_out"
-      (ptr t @-> t @-> t @-> t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_isinf = foreign "atg_isinf" (ptr t @-> t @-> returning void)
-  let stubs_isinf_out = foreign "atg_isinf_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_isnan = foreign "atg_isnan" (ptr t @-> t @-> returning void)
-  let stubs_isnan_out = foreign "atg_isnan_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_isneginf = foreign "atg_isneginf" (ptr t @-> t @-> returning void)
-
-  let stubs_isneginf_out =
-    foreign "atg_isneginf_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_isposinf = foreign "atg_isposinf" (ptr t @-> t @-> returning void)
-
-  let stubs_isposinf_out =
-    foreign "atg_isposinf_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_isreal = foreign "atg_isreal" (ptr t @-> t @-> returning void)
-
-  let stubs_istft =
-    foreign
-      "atg_istft"
-      (ptr t
-       @-> t
-       @-> int64_t
-       @-> int64_t
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> t
-       @-> int
-       @-> int
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_kaiser_window =
-    foreign "atg_kaiser_window" (ptr t @-> int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_kaiser_window_beta =
-    foreign
-      "atg_kaiser_window_beta"
-      (ptr t @-> int64_t @-> int @-> double @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_kaiser_window_beta_out =
-    foreign
-      "atg_kaiser_window_beta_out"
-      (ptr t @-> t @-> int64_t @-> int @-> double @-> returning void)
-  ;;
-
-  let stubs_kaiser_window_out =
-    foreign "atg_kaiser_window_out" (ptr t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_kaiser_window_periodic =
-    foreign
-      "atg_kaiser_window_periodic"
-      (ptr t @-> int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_kaiser_window_periodic_out =
-    foreign
-      "atg_kaiser_window_periodic_out"
-      (ptr t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_kl_div =
-    foreign "atg_kl_div" (ptr t @-> t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_kron = foreign "atg_kron" (ptr t @-> t @-> t @-> returning void)
-  let stubs_kron_out = foreign "atg_kron_out" (ptr t @-> t @-> t @-> t @-> returning void)
-
-  let stubs_kthvalue =
-    foreign "atg_kthvalue" (ptr t @-> t @-> int64_t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_kthvalue_values =
-    foreign
-      "atg_kthvalue_values"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_l1_loss =
-    foreign "atg_l1_loss" (ptr t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_layer_norm =
-    foreign
-      "atg_layer_norm"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> t
-       @-> t
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_lcm = foreign "atg_lcm" (ptr t @-> t @-> t @-> returning void)
-  let stubs_lcm_ = foreign "atg_lcm_" (ptr t @-> t @-> t @-> returning void)
-  let stubs_lcm_out = foreign "atg_lcm_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  let stubs_ldexp = foreign "atg_ldexp" (ptr t @-> t @-> t @-> returning void)
-  let stubs_ldexp_ = foreign "atg_ldexp_" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_ldexp_out =
-    foreign "atg_ldexp_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_le = foreign "atg_le" (ptr t @-> t @-> scalar @-> returning void)
-  let stubs_le_ = foreign "atg_le_" (ptr t @-> t @-> scalar @-> returning void)
-
-  let stubs_le_scalar_out =
-    foreign "atg_le_scalar_out" (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_le_tensor = foreign "atg_le_tensor" (ptr t @-> t @-> t @-> returning void)
-  let stubs_le_tensor_ = foreign "atg_le_tensor_" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_le_tensor_out =
-    foreign "atg_le_tensor_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_leaky_relu = foreign "atg_leaky_relu" (ptr t @-> t @-> returning void)
-  let stubs_leaky_relu_ = foreign "atg_leaky_relu_" (ptr t @-> t @-> returning void)
-
-  let stubs_leaky_relu_backward =
-    foreign
-      "atg_leaky_relu_backward"
-      (ptr t @-> t @-> t @-> scalar @-> int @-> returning void)
-  ;;
-
-  let stubs_leaky_relu_backward_grad_input =
-    foreign
-      "atg_leaky_relu_backward_grad_input"
-      (ptr t @-> t @-> t @-> t @-> scalar @-> int @-> returning void)
-  ;;
-
-  let stubs_leaky_relu_out =
-    foreign "atg_leaky_relu_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_lerp = foreign "atg_lerp" (ptr t @-> t @-> t @-> scalar @-> returning void)
-  let stubs_lerp_ = foreign "atg_lerp_" (ptr t @-> t @-> t @-> scalar @-> returning void)
-
-  let stubs_lerp_scalar_out =
-    foreign "atg_lerp_scalar_out" (ptr t @-> t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_lerp_tensor =
-    foreign "atg_lerp_tensor" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_lerp_tensor_ =
-    foreign "atg_lerp_tensor_" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_lerp_tensor_out =
-    foreign "atg_lerp_tensor_out" (ptr t @-> t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_less = foreign "atg_less" (ptr t @-> t @-> scalar @-> returning void)
-  let stubs_less_ = foreign "atg_less_" (ptr t @-> t @-> scalar @-> returning void)
-
-  let stubs_less_equal =
-    foreign "atg_less_equal" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_less_equal_ =
-    foreign "atg_less_equal_" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_less_equal_scalar_out =
-    foreign "atg_less_equal_scalar_out" (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_less_equal_tensor =
-    foreign "atg_less_equal_tensor" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_less_equal_tensor_ =
-    foreign "atg_less_equal_tensor_" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_less_equal_tensor_out =
-    foreign "atg_less_equal_tensor_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_less_scalar_out =
-    foreign "atg_less_scalar_out" (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_less_tensor = foreign "atg_less_tensor" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_less_tensor_ =
-    foreign "atg_less_tensor_" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_less_tensor_out =
-    foreign "atg_less_tensor_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_lgamma = foreign "atg_lgamma" (ptr t @-> t @-> returning void)
-  let stubs_lgamma_ = foreign "atg_lgamma_" (ptr t @-> t @-> returning void)
-  let stubs_lgamma_out = foreign "atg_lgamma_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_lift = foreign "atg_lift" (ptr t @-> t @-> returning void)
-  let stubs_lift_fresh = foreign "atg_lift_fresh" (ptr t @-> t @-> returning void)
-
-  let stubs_lift_fresh_copy =
-    foreign "atg_lift_fresh_copy" (ptr t @-> t @-> returning void)
-  ;;
-
-  let stubs_lift_fresh_copy_out =
-    foreign "atg_lift_fresh_copy_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_lift_out = foreign "atg_lift_out" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_linalg_cholesky =
-    foreign "atg_linalg_cholesky" (ptr t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_linalg_cholesky_ex =
-    foreign "atg_linalg_cholesky_ex" (ptr t @-> t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_linalg_cholesky_ex_l =
-    foreign
-      "atg_linalg_cholesky_ex_l"
-      (ptr t @-> t @-> t @-> t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_linalg_cholesky_out =
-    foreign "atg_linalg_cholesky_out" (ptr t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_linalg_cond =
-    foreign "atg_linalg_cond" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_linalg_cond_out =
-    foreign "atg_linalg_cond_out" (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_linalg_cond_p_str =
-    foreign "atg_linalg_cond_p_str" (ptr t @-> t @-> string @-> returning void)
-  ;;
-
-  let stubs_linalg_cond_p_str_out =
-    foreign "atg_linalg_cond_p_str_out" (ptr t @-> t @-> t @-> string @-> returning void)
-  ;;
-
-  let stubs_linalg_cross =
-    foreign "atg_linalg_cross" (ptr t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_linalg_cross_out =
-    foreign "atg_linalg_cross_out" (ptr t @-> t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_linalg_det = foreign "atg_linalg_det" (ptr t @-> t @-> returning void)
-
-  let stubs_linalg_det_out =
-    foreign "atg_linalg_det_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_linalg_diagonal =
-    foreign
-      "atg_linalg_diagonal"
-      (ptr t @-> t @-> int64_t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_linalg_eig = foreign "atg_linalg_eig" (ptr t @-> t @-> returning void)
-
-  let stubs_linalg_eig_out =
-    foreign "atg_linalg_eig_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_linalg_eigh =
-    foreign "atg_linalg_eigh" (ptr t @-> t @-> string @-> returning void)
-  ;;
-
-  let stubs_linalg_eigh_eigvals =
-    foreign
-      "atg_linalg_eigh_eigvals"
-      (ptr t @-> t @-> t @-> t @-> string @-> returning void)
-  ;;
-
-  let stubs_linalg_eigvals = foreign "atg_linalg_eigvals" (ptr t @-> t @-> returning void)
-
-  let stubs_linalg_eigvals_out =
-    foreign "atg_linalg_eigvals_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_linalg_eigvalsh =
-    foreign "atg_linalg_eigvalsh" (ptr t @-> t @-> string @-> returning void)
-  ;;
-
-  let stubs_linalg_eigvalsh_out =
-    foreign "atg_linalg_eigvalsh_out" (ptr t @-> t @-> t @-> string @-> returning void)
-  ;;
-
-  let stubs_linalg_householder_product =
-    foreign "atg_linalg_householder_product" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_linalg_householder_product_out =
-    foreign
-      "atg_linalg_householder_product_out"
-      (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_linalg_inv = foreign "atg_linalg_inv" (ptr t @-> t @-> returning void)
-
-  let stubs_linalg_inv_ex =
-    foreign "atg_linalg_inv_ex" (ptr t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_linalg_inv_ex_inverse =
-    foreign
-      "atg_linalg_inv_ex_inverse"
-      (ptr t @-> t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_linalg_inv_out =
-    foreign "atg_linalg_inv_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_linalg_ldl_factor =
-    foreign "atg_linalg_ldl_factor" (ptr t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_linalg_ldl_factor_ex =
-    foreign "atg_linalg_ldl_factor_ex" (ptr t @-> t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_linalg_ldl_factor_ex_out =
-    foreign
-      "atg_linalg_ldl_factor_ex_out"
-      (ptr t @-> t @-> t @-> t @-> t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_linalg_ldl_factor_out =
-    foreign
-      "atg_linalg_ldl_factor_out"
-      (ptr t @-> t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_linalg_ldl_solve =
-    foreign "atg_linalg_ldl_solve" (ptr t @-> t @-> t @-> t @-> int @-> returning void)
-  ;;
-end
-
-module C14 (F : Cstubs.FOREIGN) = struct
-  open F
-
-  type t = unit ptr
-
-  let t : t typ = ptr void
-
-  type scalar = unit ptr
-
-  let scalar : scalar typ = ptr void
-
-  let stubs_linalg_ldl_solve_out =
-    foreign
-      "atg_linalg_ldl_solve_out"
-      (ptr t @-> t @-> t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_linalg_lstsq =
-    foreign
-      "atg_linalg_lstsq"
-      (ptr t @-> t @-> t @-> double @-> int @-> string @-> returning void)
-  ;;
-
-  let stubs_linalg_lstsq_out =
-    foreign
-      "atg_linalg_lstsq_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> double
-       @-> int
-       @-> string
-       @-> returning void)
-  ;;
-
-  let stubs_linalg_lu = foreign "atg_linalg_lu" (ptr t @-> t @-> int @-> returning void)
-
-  let stubs_linalg_lu_factor =
-    foreign "atg_linalg_lu_factor" (ptr t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_linalg_lu_factor_ex =
-    foreign "atg_linalg_lu_factor_ex" (ptr t @-> t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_linalg_lu_factor_ex_out =
-    foreign
-      "atg_linalg_lu_factor_ex_out"
-      (ptr t @-> t @-> t @-> t @-> t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_linalg_lu_factor_out =
-    foreign "atg_linalg_lu_factor_out" (ptr t @-> t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_linalg_lu_out =
-    foreign "atg_linalg_lu_out" (ptr t @-> t @-> t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_linalg_lu_solve =
-    foreign
-      "atg_linalg_lu_solve"
-      (ptr t @-> t @-> t @-> t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_linalg_lu_solve_out =
-    foreign
-      "atg_linalg_lu_solve_out"
-      (ptr t @-> t @-> t @-> t @-> t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_linalg_matmul =
-    foreign "atg_linalg_matmul" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_linalg_matmul_out =
-    foreign "atg_linalg_matmul_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_linalg_matrix_exp =
-    foreign "atg_linalg_matrix_exp" (ptr t @-> t @-> returning void)
-  ;;
-
-  let stubs_linalg_matrix_exp_out =
-    foreign "atg_linalg_matrix_exp_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_linalg_matrix_power =
-    foreign "atg_linalg_matrix_power" (ptr t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_linalg_matrix_power_out =
-    foreign
-      "atg_linalg_matrix_power_out"
-      (ptr t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_linalg_matrix_rank =
-    foreign "atg_linalg_matrix_rank" (ptr t @-> t @-> double @-> int @-> returning void)
-  ;;
-
-  let stubs_linalg_matrix_rank_atol_rtol_float =
-    foreign
-      "atg_linalg_matrix_rank_atol_rtol_float"
-      (ptr t @-> t @-> double @-> int @-> double @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_linalg_matrix_rank_atol_rtol_float_out =
-    foreign
-      "atg_linalg_matrix_rank_atol_rtol_float_out"
-      (ptr t @-> t @-> t @-> double @-> int @-> double @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_linalg_matrix_rank_atol_rtol_tensor =
-    foreign
-      "atg_linalg_matrix_rank_atol_rtol_tensor"
-      (ptr t @-> t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_linalg_matrix_rank_atol_rtol_tensor_out =
-    foreign
-      "atg_linalg_matrix_rank_atol_rtol_tensor_out"
-      (ptr t @-> t @-> t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_linalg_matrix_rank_out =
-    foreign
-      "atg_linalg_matrix_rank_out"
-      (ptr t @-> t @-> t @-> double @-> int @-> returning void)
-  ;;
-
-  let stubs_linalg_matrix_rank_out_tol_tensor =
-    foreign
-      "atg_linalg_matrix_rank_out_tol_tensor"
-      (ptr t @-> t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_linalg_matrix_rank_tol_tensor =
-    foreign
-      "atg_linalg_matrix_rank_tol_tensor"
-      (ptr t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_linalg_multi_dot =
-    foreign "atg_linalg_multi_dot" (ptr t @-> ptr t @-> int @-> returning void)
-  ;;
-
-  let stubs_linalg_multi_dot_out =
-    foreign "atg_linalg_multi_dot_out" (ptr t @-> t @-> ptr t @-> int @-> returning void)
-  ;;
-
-  let stubs_linalg_pinv =
-    foreign "atg_linalg_pinv" (ptr t @-> t @-> double @-> int @-> returning void)
-  ;;
-
-  let stubs_linalg_pinv_atol_rtol_float =
-    foreign
-      "atg_linalg_pinv_atol_rtol_float"
-      (ptr t @-> t @-> double @-> int @-> double @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_linalg_pinv_atol_rtol_float_out =
-    foreign
-      "atg_linalg_pinv_atol_rtol_float_out"
-      (ptr t @-> t @-> t @-> double @-> int @-> double @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_linalg_pinv_atol_rtol_tensor =
-    foreign
-      "atg_linalg_pinv_atol_rtol_tensor"
-      (ptr t @-> t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_linalg_pinv_atol_rtol_tensor_out =
-    foreign
-      "atg_linalg_pinv_atol_rtol_tensor_out"
-      (ptr t @-> t @-> t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_linalg_pinv_out =
-    foreign "atg_linalg_pinv_out" (ptr t @-> t @-> t @-> double @-> int @-> returning void)
-  ;;
-
-  let stubs_linalg_pinv_out_rcond_tensor =
-    foreign
-      "atg_linalg_pinv_out_rcond_tensor"
-      (ptr t @-> t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_linalg_pinv_rcond_tensor =
-    foreign "atg_linalg_pinv_rcond_tensor" (ptr t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_linalg_qr = foreign "atg_linalg_qr" (ptr t @-> t @-> string @-> returning void)
-
-  let stubs_linalg_qr_out =
-    foreign "atg_linalg_qr_out" (ptr t @-> t @-> t @-> t @-> string @-> returning void)
-  ;;
-
-  let stubs_linalg_slogdet = foreign "atg_linalg_slogdet" (ptr t @-> t @-> returning void)
-
-  let stubs_linalg_slogdet_out =
-    foreign "atg_linalg_slogdet_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_linalg_solve =
-    foreign "atg_linalg_solve" (ptr t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_linalg_solve_ex =
-    foreign "atg_linalg_solve_ex" (ptr t @-> t @-> t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_linalg_solve_ex_out =
-    foreign
-      "atg_linalg_solve_ex_out"
-      (ptr t @-> t @-> t @-> t @-> t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_linalg_solve_out =
-    foreign "atg_linalg_solve_out" (ptr t @-> t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_linalg_solve_triangular =
-    foreign
-      "atg_linalg_solve_triangular"
-      (ptr t @-> t @-> t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_linalg_solve_triangular_out =
-    foreign
-      "atg_linalg_solve_triangular_out"
-      (ptr t @-> t @-> t @-> t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_linalg_svd =
-    foreign "atg_linalg_svd" (ptr t @-> t @-> int @-> string @-> returning void)
-  ;;
-
-  let stubs_linalg_svd_u =
-    foreign
-      "atg_linalg_svd_u"
-      (ptr t @-> t @-> t @-> t @-> t @-> int @-> string @-> returning void)
-  ;;
-
-  let stubs_linalg_svdvals =
-    foreign "atg_linalg_svdvals" (ptr t @-> t @-> string @-> returning void)
-  ;;
-
-  let stubs_linalg_svdvals_out =
-    foreign "atg_linalg_svdvals_out" (ptr t @-> t @-> t @-> string @-> returning void)
-  ;;
-
-  let stubs_linalg_tensorinv =
-    foreign "atg_linalg_tensorinv" (ptr t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_linalg_tensorinv_out =
-    foreign "atg_linalg_tensorinv_out" (ptr t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_linalg_tensorsolve =
-    foreign
-      "atg_linalg_tensorsolve"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_linalg_tensorsolve_out =
-    foreign
-      "atg_linalg_tensorsolve_out"
-      (ptr t @-> t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_linalg_vander =
-    foreign "atg_linalg_vander" (ptr t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_linalg_vecdot =
-    foreign "atg_linalg_vecdot" (ptr t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_linalg_vecdot_out =
-    foreign
-      "atg_linalg_vecdot_out"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_linear = foreign "atg_linear" (ptr t @-> t @-> t @-> t @-> returning void)
-
-  let stubs_linear_out =
-    foreign "atg_linear_out" (ptr t @-> t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_linspace =
-    foreign
-      "atg_linspace"
-      (ptr t @-> scalar @-> scalar @-> int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_linspace_out =
-    foreign
-      "atg_linspace_out"
-      (ptr t @-> t @-> scalar @-> scalar @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_log = foreign "atg_log" (ptr t @-> t @-> returning void)
-  let stubs_log10 = foreign "atg_log10" (ptr t @-> t @-> returning void)
-  let stubs_log10_ = foreign "atg_log10_" (ptr t @-> t @-> returning void)
-  let stubs_log10_out = foreign "atg_log10_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_log1p = foreign "atg_log1p" (ptr t @-> t @-> returning void)
-  let stubs_log1p_ = foreign "atg_log1p_" (ptr t @-> t @-> returning void)
-  let stubs_log1p_out = foreign "atg_log1p_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_log2 = foreign "atg_log2" (ptr t @-> t @-> returning void)
-  let stubs_log2_ = foreign "atg_log2_" (ptr t @-> t @-> returning void)
-  let stubs_log2_out = foreign "atg_log2_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_log_ = foreign "atg_log_" (ptr t @-> t @-> returning void)
-
-  let stubs_log_normal =
-    foreign "atg_log_normal" (ptr t @-> t @-> double @-> double @-> returning void)
-  ;;
-
-  let stubs_log_normal_ =
-    foreign "atg_log_normal_" (ptr t @-> t @-> double @-> double @-> returning void)
-  ;;
-
-  let stubs_log_normal_out =
-    foreign
-      "atg_log_normal_out"
-      (ptr t @-> t @-> t @-> double @-> double @-> returning void)
-  ;;
-
-  let stubs_log_out = foreign "atg_log_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_log_sigmoid = foreign "atg_log_sigmoid" (ptr t @-> t @-> returning void)
-
-  let stubs_log_sigmoid_backward =
-    foreign "atg_log_sigmoid_backward" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_log_sigmoid_backward_grad_input =
-    foreign
-      "atg_log_sigmoid_backward_grad_input"
-      (ptr t @-> t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_log_sigmoid_out =
-    foreign "atg_log_sigmoid_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_log_softmax =
-    foreign "atg_log_softmax" (ptr t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_log_softmax_int_out =
-    foreign
-      "atg_log_softmax_int_out"
-      (ptr t @-> t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_logaddexp = foreign "atg_logaddexp" (ptr t @-> t @-> t @-> returning void)
-  let stubs_logaddexp2 = foreign "atg_logaddexp2" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_logaddexp2_out =
-    foreign "atg_logaddexp2_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_logaddexp_out =
-    foreign "atg_logaddexp_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_logcumsumexp =
-    foreign "atg_logcumsumexp" (ptr t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_logcumsumexp_out =
-    foreign "atg_logcumsumexp_out" (ptr t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_logdet = foreign "atg_logdet" (ptr t @-> t @-> returning void)
-  let stubs_logical_and = foreign "atg_logical_and" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_logical_and_ =
-    foreign "atg_logical_and_" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_logical_and_out =
-    foreign "atg_logical_and_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_logical_not = foreign "atg_logical_not" (ptr t @-> t @-> returning void)
-  let stubs_logical_not_ = foreign "atg_logical_not_" (ptr t @-> t @-> returning void)
-
-  let stubs_logical_not_out =
-    foreign "atg_logical_not_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_logical_or = foreign "atg_logical_or" (ptr t @-> t @-> t @-> returning void)
-  let stubs_logical_or_ = foreign "atg_logical_or_" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_logical_or_out =
-    foreign "atg_logical_or_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_logical_xor = foreign "atg_logical_xor" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_logical_xor_ =
-    foreign "atg_logical_xor_" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_logical_xor_out =
-    foreign "atg_logical_xor_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-end
-
-module C15 (F : Cstubs.FOREIGN) = struct
-  open F
-
-  type t = unit ptr
-
-  let t : t typ = ptr void
-
-  type scalar = unit ptr
-
-  let scalar : scalar typ = ptr void
-  let stubs_logit = foreign "atg_logit" (ptr t @-> t @-> double @-> int @-> returning void)
-
-  let stubs_logit_ =
-    foreign "atg_logit_" (ptr t @-> t @-> double @-> int @-> returning void)
-  ;;
-
-  let stubs_logit_backward =
-    foreign "atg_logit_backward" (ptr t @-> t @-> t @-> double @-> int @-> returning void)
-  ;;
-
-  let stubs_logit_backward_grad_input =
-    foreign
-      "atg_logit_backward_grad_input"
-      (ptr t @-> t @-> t @-> t @-> double @-> int @-> returning void)
-  ;;
-
-  let stubs_logit_out =
-    foreign "atg_logit_out" (ptr t @-> t @-> t @-> double @-> int @-> returning void)
-  ;;
-
-  let stubs_logspace =
-    foreign
-      "atg_logspace"
-      (ptr t
-       @-> scalar
-       @-> scalar
-       @-> int64_t
-       @-> double
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_logspace_out =
-    foreign
-      "atg_logspace_out"
-      (ptr t @-> t @-> scalar @-> scalar @-> int64_t @-> double @-> returning void)
-  ;;
-
-  let stubs_logsumexp =
-    foreign
-      "atg_logsumexp"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_logsumexp_out =
-    foreign
-      "atg_logsumexp_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_lstm =
-    foreign
-      "atg_lstm"
-      (ptr t
-       @-> t
-       @-> ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> int
-       @-> int64_t
-       @-> double
-       @-> int
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_lstm_cell =
-    foreign
-      "atg_lstm_cell"
-      (ptr t @-> t @-> ptr t @-> int @-> t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_lstm_data =
-    foreign
-      "atg_lstm_data"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> int
-       @-> int64_t
-       @-> double
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_lstm_mps_backward =
-    foreign
-      "atg_lstm_mps_backward"
-      (t
-       @-> ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr t
-       @-> int
-       @-> ptr t
-       @-> int
-       @-> int
-       @-> int64_t
-       @-> double
-       @-> int
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_lt = foreign "atg_lt" (ptr t @-> t @-> scalar @-> returning void)
-  let stubs_lt_ = foreign "atg_lt_" (ptr t @-> t @-> scalar @-> returning void)
-
-  let stubs_lt_scalar_out =
-    foreign "atg_lt_scalar_out" (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_lt_tensor = foreign "atg_lt_tensor" (ptr t @-> t @-> t @-> returning void)
-  let stubs_lt_tensor_ = foreign "atg_lt_tensor_" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_lt_tensor_out =
-    foreign "atg_lt_tensor_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_lu_solve = foreign "atg_lu_solve" (ptr t @-> t @-> t @-> t @-> returning void)
-
-  let stubs_lu_solve_out =
-    foreign "atg_lu_solve_out" (ptr t @-> t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_lu_unpack =
-    foreign "atg_lu_unpack" (ptr t @-> t @-> t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_lu_unpack_out =
-    foreign
-      "atg_lu_unpack_out"
-      (ptr t @-> t @-> t @-> t @-> t @-> t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_margin_ranking_loss =
-    foreign
-      "atg_margin_ranking_loss"
-      (ptr t @-> t @-> t @-> t @-> double @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_masked_fill =
-    foreign "atg_masked_fill" (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_masked_fill_ =
-    foreign "atg_masked_fill_" (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_masked_fill_scalar_out =
-    foreign
-      "atg_masked_fill_scalar_out"
-      (ptr t @-> t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_masked_fill_tensor =
-    foreign "atg_masked_fill_tensor" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_masked_fill_tensor_ =
-    foreign "atg_masked_fill_tensor_" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_masked_fill_tensor_out =
-    foreign "atg_masked_fill_tensor_out" (ptr t @-> t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_masked_scatter =
-    foreign "atg_masked_scatter" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_masked_scatter_ =
-    foreign "atg_masked_scatter_" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_masked_scatter_out =
-    foreign "atg_masked_scatter_out" (ptr t @-> t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_masked_select =
-    foreign "atg_masked_select" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_masked_select_backward =
-    foreign "atg_masked_select_backward" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_masked_select_out =
-    foreign "atg_masked_select_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_matmul = foreign "atg_matmul" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_matmul_out =
-    foreign "atg_matmul_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_matrix_exp = foreign "atg_matrix_exp" (ptr t @-> t @-> returning void)
-
-  let stubs_matrix_exp_backward =
-    foreign "atg_matrix_exp_backward" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_matrix_h = foreign "atg_matrix_h" (ptr t @-> t @-> returning void)
-
-  let stubs_matrix_power =
-    foreign "atg_matrix_power" (ptr t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_matrix_power_out =
-    foreign "atg_matrix_power_out" (ptr t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_max = foreign "atg_max" (ptr t @-> t @-> returning void)
-
-  let stubs_max_dim =
-    foreign "atg_max_dim" (ptr t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_max_dim_max =
-    foreign
-      "atg_max_dim_max"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_max_other = foreign "atg_max_other" (ptr t @-> t @-> t @-> returning void)
-  let stubs_max_out = foreign "atg_max_out" (ptr t @-> t @-> t @-> t @-> returning void)
-
-  let stubs_max_pool1d =
-    foreign
-      "atg_max_pool1d"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_max_pool1d_with_indices =
-    foreign
-      "atg_max_pool1d_with_indices"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_max_pool2d =
-    foreign
-      "atg_max_pool2d"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_max_pool2d_backward =
-    foreign
-      "atg_max_pool2d_backward"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_max_pool2d_backward_out =
-    foreign
-      "atg_max_pool2d_backward_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_max_pool2d_with_indices =
-    foreign
-      "atg_max_pool2d_with_indices"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_max_pool2d_with_indices_backward =
-    foreign
-      "atg_max_pool2d_with_indices_backward"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> t
-       @-> returning void)
-  ;;
-
-  let stubs_max_pool2d_with_indices_backward_grad_input =
-    foreign
-      "atg_max_pool2d_with_indices_backward_grad_input"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> t
-       @-> returning void)
-  ;;
-
-  let stubs_max_pool2d_with_indices_out =
-    foreign
-      "atg_max_pool2d_with_indices_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_max_pool3d =
-    foreign
-      "atg_max_pool3d"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_max_pool3d_with_indices =
-    foreign
-      "atg_max_pool3d_with_indices"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_max_pool3d_with_indices_backward =
-    foreign
-      "atg_max_pool3d_with_indices_backward"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> t
-       @-> returning void)
-  ;;
-
-  let stubs_max_pool3d_with_indices_backward_grad_input =
-    foreign
-      "atg_max_pool3d_with_indices_backward_grad_input"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> t
-       @-> returning void)
-  ;;
-
-  let stubs_max_pool3d_with_indices_out =
-    foreign
-      "atg_max_pool3d_with_indices_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_max_unary_out =
-    foreign "atg_max_unary_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_max_unpool2d =
-    foreign
-      "atg_max_unpool2d"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_max_unpool2d_out =
-    foreign
-      "atg_max_unpool2d_out"
-      (ptr t @-> t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_max_unpool3d =
-    foreign
-      "atg_max_unpool3d"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_max_unpool3d_out =
-    foreign
-      "atg_max_unpool3d_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_maximum = foreign "atg_maximum" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_maximum_out =
-    foreign "atg_maximum_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_mean = foreign "atg_mean" (ptr t @-> t @-> int @-> returning void)
-
-  let stubs_mean_dim =
-    foreign
-      "atg_mean_dim"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_mean_out =
-    foreign
-      "atg_mean_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_median = foreign "atg_median" (ptr t @-> t @-> returning void)
-
-  let stubs_median_dim =
-    foreign "atg_median_dim" (ptr t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_median_dim_values =
-    foreign
-      "atg_median_dim_values"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_median_out = foreign "atg_median_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_meshgrid = foreign "atg_meshgrid" (ptr t @-> int @-> returning (ptr t))
-
-  let stubs_meshgrid_indexing =
-    foreign "atg_meshgrid_indexing" (ptr t @-> int @-> string @-> returning (ptr t))
-  ;;
-
-  let stubs_mh = foreign "atg_mh" (ptr t @-> t @-> returning void)
-  let stubs_min = foreign "atg_min" (ptr t @-> t @-> returning void)
-
-  let stubs_min_dim =
-    foreign "atg_min_dim" (ptr t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_min_dim_min =
-    foreign
-      "atg_min_dim_min"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_min_other = foreign "atg_min_other" (ptr t @-> t @-> t @-> returning void)
-  let stubs_min_out = foreign "atg_min_out" (ptr t @-> t @-> t @-> t @-> returning void)
-
-  let stubs_min_unary_out =
-    foreign "atg_min_unary_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_minimum = foreign "atg_minimum" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_minimum_out =
-    foreign "atg_minimum_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_miopen_batch_norm =
-    foreign
-      "atg_miopen_batch_norm"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int
-       @-> double
-       @-> double
-       @-> returning void)
-  ;;
-
-  let stubs_miopen_batch_norm_backward =
-    foreign
-      "atg_miopen_batch_norm_backward"
-      (ptr t @-> t @-> t @-> t @-> t @-> t @-> t @-> t @-> double @-> returning void)
-  ;;
-
-  let stubs_miopen_batch_norm_backward_out =
-    foreign
-      "atg_miopen_batch_norm_backward_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> double
-       @-> returning void)
-  ;;
-
-  let stubs_miopen_batch_norm_out =
-    foreign
-      "atg_miopen_batch_norm_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int
-       @-> double
-       @-> double
-       @-> returning void)
-  ;;
-
-  let stubs_miopen_convolution =
-    foreign
-      "atg_miopen_convolution"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_miopen_convolution_add_relu =
-    foreign
-      "atg_miopen_convolution_add_relu"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> scalar
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs_miopen_convolution_out =
-    foreign
-      "atg_miopen_convolution_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_miopen_convolution_relu =
-    foreign
-      "atg_miopen_convolution_relu"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs_miopen_convolution_transpose =
-    foreign
-      "atg_miopen_convolution_transpose"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_miopen_convolution_transpose_out =
-    foreign
-      "atg_miopen_convolution_transpose_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_miopen_depthwise_convolution =
-    foreign
-      "atg_miopen_depthwise_convolution"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_miopen_depthwise_convolution_out =
-    foreign
-      "atg_miopen_depthwise_convolution_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_miopen_rnn =
-    foreign
-      "atg_miopen_rnn"
-      (ptr t
-       @-> t
-       @-> ptr t
-       @-> int
-       @-> int64_t
-       @-> t
-       @-> t
-       @-> int64_t
-       @-> int64_t
-       @-> int64_t
-       @-> int
-       @-> double
-       @-> int
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> t
-       @-> returning void)
-  ;;
-end
-
-module C16 (F : Cstubs.FOREIGN) = struct
-  open F
-
-  type t = unit ptr
-
-  let t : t typ = ptr void
-
-  type scalar = unit ptr
-
-  let scalar : scalar typ = ptr void
-
-  let stubs_miopen_rnn_out =
-    foreign
-      "atg_miopen_rnn_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr t
-       @-> int
-       @-> int64_t
-       @-> t
-       @-> t
-       @-> int64_t
-       @-> int64_t
-       @-> int64_t
-       @-> int
-       @-> double
-       @-> int
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> t
-       @-> returning void)
-  ;;
-
-  let stubs_mish = foreign "atg_mish" (ptr t @-> t @-> returning void)
-  let stubs_mish_ = foreign "atg_mish_" (ptr t @-> t @-> returning void)
-
-  let stubs_mish_backward =
-    foreign "atg_mish_backward" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_mish_out = foreign "atg_mish_out" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_mkldnn_adaptive_avg_pool2d =
-    foreign
-      "atg_mkldnn_adaptive_avg_pool2d"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_mkldnn_adaptive_avg_pool2d_backward =
-    foreign
-      "atg_mkldnn_adaptive_avg_pool2d_backward"
-      (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_mkldnn_adaptive_avg_pool2d_backward_out =
-    foreign
-      "atg_mkldnn_adaptive_avg_pool2d_backward_out"
-      (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_mkldnn_adaptive_avg_pool2d_out =
-    foreign
-      "atg_mkldnn_adaptive_avg_pool2d_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_mkldnn_convolution =
-    foreign
-      "atg_mkldnn_convolution"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs_mkldnn_convolution_out =
-    foreign
-      "atg_mkldnn_convolution_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs_mkldnn_linear =
-    foreign "atg_mkldnn_linear" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_mkldnn_linear_backward_input =
-    foreign
-      "atg_mkldnn_linear_backward_input"
-      (ptr t @-> ptr int64_t @-> int @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_mkldnn_linear_backward_input_out =
-    foreign
-      "atg_mkldnn_linear_backward_input_out"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_mkldnn_linear_backward_weights =
-    foreign
-      "atg_mkldnn_linear_backward_weights"
-      (ptr t @-> t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_mkldnn_linear_backward_weights_out =
-    foreign
-      "atg_mkldnn_linear_backward_weights_out"
-      (ptr t @-> t @-> t @-> t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_mkldnn_linear_out =
-    foreign "atg_mkldnn_linear_out" (ptr t @-> t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_mkldnn_max_pool2d =
-    foreign
-      "atg_mkldnn_max_pool2d"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_mkldnn_max_pool2d_backward =
-    foreign
-      "atg_mkldnn_max_pool2d_backward"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_mkldnn_max_pool2d_backward_out =
-    foreign
-      "atg_mkldnn_max_pool2d_backward_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_mkldnn_max_pool2d_out =
-    foreign
-      "atg_mkldnn_max_pool2d_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_mkldnn_max_pool3d =
-    foreign
-      "atg_mkldnn_max_pool3d"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_mkldnn_max_pool3d_backward =
-    foreign
-      "atg_mkldnn_max_pool3d_backward"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_mkldnn_max_pool3d_backward_out =
-    foreign
-      "atg_mkldnn_max_pool3d_backward_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_mkldnn_max_pool3d_out =
-    foreign
-      "atg_mkldnn_max_pool3d_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_mkldnn_reorder_conv2d_weight =
-    foreign
-      "atg_mkldnn_reorder_conv2d_weight"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> ptr int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_mkldnn_reorder_conv2d_weight_out =
-    foreign
-      "atg_mkldnn_reorder_conv2d_weight_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> ptr int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_mkldnn_reorder_conv3d_weight =
-    foreign
-      "atg_mkldnn_reorder_conv3d_weight"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs_mkldnn_reorder_conv3d_weight_out =
-    foreign
-      "atg_mkldnn_reorder_conv3d_weight_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs_mkldnn_rnn_layer =
-    foreign
-      "atg_mkldnn_rnn_layer"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> int64_t
-       @-> int64_t
-       @-> int
-       @-> int
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_mkldnn_rnn_layer_backward =
-    foreign
-      "atg_mkldnn_rnn_layer_backward"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int
-       @-> int64_t
-       @-> int64_t
-       @-> int64_t
-       @-> int
-       @-> int
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> t
-       @-> returning void)
-  ;;
-
-  let stubs_mkldnn_rnn_layer_backward_out =
-    foreign
-      "atg_mkldnn_rnn_layer_backward_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int
-       @-> int64_t
-       @-> int64_t
-       @-> int64_t
-       @-> int
-       @-> int
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> t
-       @-> returning void)
-  ;;
-
-  let stubs_mkldnn_rnn_layer_out =
-    foreign
-      "atg_mkldnn_rnn_layer_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> int64_t
-       @-> int64_t
-       @-> int
-       @-> int
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_mm = foreign "atg_mm" (ptr t @-> t @-> t @-> returning void)
-  let stubs_mm_out = foreign "atg_mm_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  let stubs_mode = foreign "atg_mode" (ptr t @-> t @-> int64_t @-> int @-> returning void)
-
-  let stubs_mode_values =
-    foreign
-      "atg_mode_values"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_moveaxis =
-    foreign
-      "atg_moveaxis"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_moveaxis_int =
-    foreign "atg_moveaxis_int" (ptr t @-> t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_movedim =
-    foreign
-      "atg_movedim"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_movedim_int =
-    foreign "atg_movedim_int" (ptr t @-> t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_mse_loss =
-    foreign "atg_mse_loss" (ptr t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_mse_loss_backward =
-    foreign
-      "atg_mse_loss_backward"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_mse_loss_backward_grad_input =
-    foreign
-      "atg_mse_loss_backward_grad_input"
-      (ptr t @-> t @-> t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_mse_loss_out =
-    foreign "atg_mse_loss_out" (ptr t @-> t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_msort = foreign "atg_msort" (ptr t @-> t @-> returning void)
-  let stubs_msort_out = foreign "atg_msort_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_mt = foreign "atg_mt" (ptr t @-> t @-> returning void)
-  let stubs_mul = foreign "atg_mul" (ptr t @-> t @-> t @-> returning void)
-  let stubs_mul_ = foreign "atg_mul_" (ptr t @-> t @-> t @-> returning void)
-  let stubs_mul_out = foreign "atg_mul_out" (ptr t @-> t @-> t @-> t @-> returning void)
-
-  let stubs_mul_scalar =
-    foreign "atg_mul_scalar" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_mul_scalar_ =
-    foreign "atg_mul_scalar_" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_mul_scalar_out =
-    foreign "atg_mul_scalar_out" (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_multi_margin_loss_backward =
-    foreign
-      "atg_multi_margin_loss_backward"
-      (ptr t @-> t @-> t @-> t @-> scalar @-> scalar @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_multi_margin_loss_backward_grad_input =
-    foreign
-      "atg_multi_margin_loss_backward_grad_input"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> scalar
-       @-> scalar
-       @-> t
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs_multilabel_margin_loss =
-    foreign "atg_multilabel_margin_loss" (ptr t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_multilabel_margin_loss_backward =
-    foreign
-      "atg_multilabel_margin_loss_backward"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> t @-> returning void)
-  ;;
-
-  let stubs_multilabel_margin_loss_backward_grad_input =
-    foreign
-      "atg_multilabel_margin_loss_backward_grad_input"
-      (ptr t @-> t @-> t @-> t @-> t @-> int64_t @-> t @-> returning void)
-  ;;
-
-  let stubs_multilabel_margin_loss_out =
-    foreign
-      "atg_multilabel_margin_loss_out"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_multinomial =
-    foreign "atg_multinomial" (ptr t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_multinomial_out =
-    foreign
-      "atg_multinomial_out"
-      (ptr t @-> t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_multiply = foreign "atg_multiply" (ptr t @-> t @-> t @-> returning void)
-  let stubs_multiply_ = foreign "atg_multiply_" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_multiply_out =
-    foreign "atg_multiply_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_multiply_scalar =
-    foreign "atg_multiply_scalar" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_multiply_scalar_ =
-    foreign "atg_multiply_scalar_" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_mv = foreign "atg_mv" (ptr t @-> t @-> t @-> returning void)
-  let stubs_mv_out = foreign "atg_mv_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  let stubs_mvlgamma = foreign "atg_mvlgamma" (ptr t @-> t @-> int64_t @-> returning void)
-
-  let stubs_mvlgamma_ =
-    foreign "atg_mvlgamma_" (ptr t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_mvlgamma_out =
-    foreign "atg_mvlgamma_out" (ptr t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_nan_to_num =
-    foreign
-      "atg_nan_to_num"
-      (ptr t
-       @-> t
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_nan_to_num_ =
-    foreign
-      "atg_nan_to_num_"
-      (ptr t
-       @-> t
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_nan_to_num_out =
-    foreign
-      "atg_nan_to_num_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_nanmean =
-    foreign
-      "atg_nanmean"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_nanmean_out =
-    foreign
-      "atg_nanmean_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_nanmedian = foreign "atg_nanmedian" (ptr t @-> t @-> returning void)
-
-  let stubs_nanmedian_dim =
-    foreign "atg_nanmedian_dim" (ptr t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_nanmedian_dim_values =
-    foreign
-      "atg_nanmedian_dim_values"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_nanmedian_out =
-    foreign "atg_nanmedian_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_nanquantile =
-    foreign
-      "atg_nanquantile"
-      (ptr t @-> t @-> t @-> int64_t @-> int @-> int @-> string @-> returning void)
-  ;;
-
-  let stubs_nanquantile_out =
-    foreign
-      "atg_nanquantile_out"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> int @-> int @-> string @-> returning void)
-  ;;
-
-  let stubs_nanquantile_scalar =
-    foreign
-      "atg_nanquantile_scalar"
-      (ptr t @-> t @-> double @-> int64_t @-> int @-> int @-> string @-> returning void)
-  ;;
-
-  let stubs_nanquantile_scalar_out =
-    foreign
-      "atg_nanquantile_scalar_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> double
-       @-> int64_t
-       @-> int
-       @-> int
-       @-> string
-       @-> returning void)
-  ;;
-
-  let stubs_nansum =
-    foreign
-      "atg_nansum"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_nansum_out =
-    foreign
-      "atg_nansum_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_narrow =
-    foreign
-      "atg_narrow"
-      (ptr t @-> t @-> int64_t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_narrow_copy =
-    foreign
-      "atg_narrow_copy"
-      (ptr t @-> t @-> int64_t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_narrow_copy_out =
-    foreign
-      "atg_narrow_copy_out"
-      (ptr t @-> t @-> t @-> int64_t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_narrow_tensor =
-    foreign
-      "atg_narrow_tensor"
-      (ptr t @-> t @-> int64_t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_native_batch_norm =
-    foreign
-      "atg_native_batch_norm"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int
-       @-> double
-       @-> double
-       @-> returning void)
-  ;;
-
-  let stubs_native_batch_norm_out =
-    foreign
-      "atg_native_batch_norm_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int
-       @-> double
-       @-> double
-       @-> returning void)
-  ;;
-
-  let stubs_native_channel_shuffle =
-    foreign "atg_native_channel_shuffle" (ptr t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_native_dropout =
-    foreign "atg_native_dropout" (ptr t @-> t @-> double @-> int @-> returning void)
-  ;;
-
-  let stubs_native_dropout_backward =
-    foreign "atg_native_dropout_backward" (ptr t @-> t @-> t @-> double @-> returning void)
-  ;;
-
-  let stubs_native_dropout_backward_out =
-    foreign
-      "atg_native_dropout_backward_out"
-      (ptr t @-> t @-> t @-> t @-> double @-> returning void)
-  ;;
-
-  let stubs_native_dropout_out =
-    foreign
-      "atg_native_dropout_out"
-      (ptr t @-> t @-> t @-> t @-> double @-> int @-> returning void)
-  ;;
-
-  let stubs_native_group_norm =
-    foreign
-      "atg_native_group_norm"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> int64_t
-       @-> int64_t
-       @-> int64_t
-       @-> int64_t
-       @-> double
-       @-> returning void)
-  ;;
-
-  let stubs_native_group_norm_out =
-    foreign
-      "atg_native_group_norm_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int64_t
-       @-> int64_t
-       @-> int64_t
-       @-> int64_t
-       @-> double
-       @-> returning void)
-  ;;
-end
-
-module C17 (F : Cstubs.FOREIGN) = struct
-  open F
-
-  type t = unit ptr
-
-  let t : t typ = ptr void
-
-  type scalar = unit ptr
-
-  let scalar : scalar typ = ptr void
-
-  let stubs_native_layer_norm =
-    foreign
-      "atg_native_layer_norm"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> t @-> t @-> double @-> returning void)
-  ;;
-
-  let stubs_native_layer_norm_out =
-    foreign
-      "atg_native_layer_norm_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> t
-       @-> t
-       @-> double
-       @-> returning void)
-  ;;
-
-  let stubs_native_norm = foreign "atg_native_norm" (ptr t @-> t @-> returning void)
-
-  let stubs_native_norm_out =
-    foreign "atg_native_norm_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_native_norm_scalaropt_dim_dtype =
-    foreign
-      "atg_native_norm_scalaropt_dim_dtype"
-      (ptr t @-> t @-> scalar @-> ptr int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_native_norm_scalaropt_dim_dtype_out =
-    foreign
-      "atg_native_norm_scalaropt_dim_dtype_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> scalar
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_ne = foreign "atg_ne" (ptr t @-> t @-> scalar @-> returning void)
-  let stubs_ne_ = foreign "atg_ne_" (ptr t @-> t @-> scalar @-> returning void)
-
-  let stubs_ne_scalar_out =
-    foreign "atg_ne_scalar_out" (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_ne_tensor = foreign "atg_ne_tensor" (ptr t @-> t @-> t @-> returning void)
-  let stubs_ne_tensor_ = foreign "atg_ne_tensor_" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_ne_tensor_out =
-    foreign "atg_ne_tensor_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_neg = foreign "atg_neg" (ptr t @-> t @-> returning void)
-  let stubs_neg_ = foreign "atg_neg_" (ptr t @-> t @-> returning void)
-  let stubs_neg_out = foreign "atg_neg_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_negative = foreign "atg_negative" (ptr t @-> t @-> returning void)
-  let stubs_negative_ = foreign "atg_negative_" (ptr t @-> t @-> returning void)
-
-  let stubs_negative_out =
-    foreign "atg_negative_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_nested_to_padded_tensor =
-    foreign
-      "atg_nested_to_padded_tensor"
-      (ptr t @-> t @-> double @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_new_empty =
-    foreign
-      "atg_new_empty"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_new_empty_out =
-    foreign
-      "atg_new_empty_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_new_empty_strided =
-    foreign
-      "atg_new_empty_strided"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_new_empty_strided_out =
-    foreign
-      "atg_new_empty_strided_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_new_full =
-    foreign
-      "atg_new_full"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> scalar @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_new_full_out =
-    foreign
-      "atg_new_full_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> scalar @-> returning void)
-  ;;
-
-  let stubs_new_ones =
-    foreign
-      "atg_new_ones"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_new_ones_out =
-    foreign
-      "atg_new_ones_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_new_zeros =
-    foreign
-      "atg_new_zeros"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_new_zeros_out =
-    foreign
-      "atg_new_zeros_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_nextafter = foreign "atg_nextafter" (ptr t @-> t @-> t @-> returning void)
-  let stubs_nextafter_ = foreign "atg_nextafter_" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_nextafter_out =
-    foreign "atg_nextafter_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_nll_loss =
-    foreign
-      "atg_nll_loss"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_nll_loss2d =
-    foreign
-      "atg_nll_loss2d"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_nll_loss2d_backward =
-    foreign
-      "atg_nll_loss2d_backward"
-      (ptr t @-> t @-> t @-> t @-> t @-> int64_t @-> int64_t @-> t @-> returning void)
-  ;;
-
-  let stubs_nll_loss2d_backward_grad_input =
-    foreign
-      "atg_nll_loss2d_backward_grad_input"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int64_t
-       @-> int64_t
-       @-> t
-       @-> returning void)
-  ;;
-
-  let stubs_nll_loss2d_out =
-    foreign
-      "atg_nll_loss2d_out"
-      (ptr t @-> t @-> t @-> t @-> t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_nll_loss_backward =
-    foreign
-      "atg_nll_loss_backward"
-      (ptr t @-> t @-> t @-> t @-> t @-> int64_t @-> int64_t @-> t @-> returning void)
-  ;;
-
-  let stubs_nll_loss_backward_grad_input =
-    foreign
-      "atg_nll_loss_backward_grad_input"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int64_t
-       @-> int64_t
-       @-> t
-       @-> returning void)
-  ;;
-
-  let stubs_nll_loss_nd =
-    foreign
-      "atg_nll_loss_nd"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_nll_loss_out =
-    foreign
-      "atg_nll_loss_out"
-      (ptr t @-> t @-> t @-> t @-> t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_nonzero = foreign "atg_nonzero" (ptr t @-> t @-> returning void)
-  let stubs_nonzero_numpy = foreign "atg_nonzero_numpy" (t @-> returning (ptr t))
-  let stubs_nonzero_out = foreign "atg_nonzero_out" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_nonzero_static =
-    foreign "atg_nonzero_static" (ptr t @-> t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_nonzero_static_out =
-    foreign
-      "atg_nonzero_static_out"
-      (ptr t @-> t @-> t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_norm = foreign "atg_norm" (ptr t @-> t @-> returning void)
-
-  let stubs_norm_dtype_out =
-    foreign
-      "atg_norm_dtype_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> scalar
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_norm_except_dim =
-    foreign "atg_norm_except_dim" (ptr t @-> t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_norm_out =
-    foreign
-      "atg_norm_out"
-      (ptr t @-> t @-> t @-> scalar @-> ptr int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_norm_scalar_out =
-    foreign "atg_norm_scalar_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_norm_scalaropt_dim =
-    foreign
-      "atg_norm_scalaropt_dim"
-      (ptr t @-> t @-> scalar @-> ptr int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_norm_scalaropt_dim_dtype =
-    foreign
-      "atg_norm_scalaropt_dim_dtype"
-      (ptr t @-> t @-> scalar @-> ptr int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_norm_scalaropt_dtype =
-    foreign "atg_norm_scalaropt_dtype" (ptr t @-> t @-> scalar @-> int @-> returning void)
-  ;;
-
-  let stubs_norm_scalaropt_dtype_out =
-    foreign
-      "atg_norm_scalaropt_dtype_out"
-      (ptr t @-> t @-> t @-> scalar @-> int @-> returning void)
-  ;;
-
-  let stubs_normal_ =
-    foreign "atg_normal_" (ptr t @-> t @-> double @-> double @-> returning void)
-  ;;
-
-  let stubs_normal_functional =
-    foreign "atg_normal_functional" (ptr t @-> t @-> double @-> double @-> returning void)
-  ;;
-
-  let stubs_not_equal = foreign "atg_not_equal" (ptr t @-> t @-> scalar @-> returning void)
-
-  let stubs_not_equal_ =
-    foreign "atg_not_equal_" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_not_equal_scalar_out =
-    foreign "atg_not_equal_scalar_out" (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_not_equal_tensor =
-    foreign "atg_not_equal_tensor" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_not_equal_tensor_ =
-    foreign "atg_not_equal_tensor_" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_not_equal_tensor_out =
-    foreign "atg_not_equal_tensor_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_nuclear_norm =
-    foreign "atg_nuclear_norm" (ptr t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_nuclear_norm_dim =
-    foreign
-      "atg_nuclear_norm_dim"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_nuclear_norm_dim_out =
-    foreign
-      "atg_nuclear_norm_dim_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_nuclear_norm_out =
-    foreign "atg_nuclear_norm_out" (ptr t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_numpy_t = foreign "atg_numpy_t" (ptr t @-> t @-> returning void)
-  let stubs_one_hot = foreign "atg_one_hot" (ptr t @-> t @-> int64_t @-> returning void)
-
-  let stubs_ones =
-    foreign "atg_ones" (ptr t @-> ptr int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_ones_like = foreign "atg_ones_like" (ptr t @-> t @-> returning void)
-
-  let stubs_ones_like_out =
-    foreign "atg_ones_like_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_ones_out =
-    foreign "atg_ones_out" (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_orgqr = foreign "atg_orgqr" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_orgqr_out =
-    foreign "atg_orgqr_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_ormqr =
-    foreign "atg_ormqr" (ptr t @-> t @-> t @-> t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_ormqr_out =
-    foreign
-      "atg_ormqr_out"
-      (ptr t @-> t @-> t @-> t @-> t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_outer = foreign "atg_outer" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_outer_out =
-    foreign "atg_outer_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_output_nr = foreign "atg_output_nr" (t @-> returning int64_t)
-
-  let stubs_pad =
-    foreign
-      "atg_pad"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> string
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_pad_sequence =
-    foreign
-      "atg_pad_sequence"
-      (ptr t @-> ptr t @-> int @-> int @-> double @-> returning void)
-  ;;
-
-  let stubs_pairwise_distance =
-    foreign
-      "atg_pairwise_distance"
-      (ptr t @-> t @-> t @-> double @-> double @-> int @-> returning void)
-  ;;
-
-  let stubs_pdist = foreign "atg_pdist" (ptr t @-> t @-> double @-> returning void)
-
-  let stubs_permute =
-    foreign "atg_permute" (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_permute_copy =
-    foreign "atg_permute_copy" (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_permute_copy_out =
-    foreign
-      "atg_permute_copy_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_pin_memory = foreign "atg_pin_memory" (ptr t @-> t @-> int @-> returning void)
-  let stubs_pinverse = foreign "atg_pinverse" (ptr t @-> t @-> double @-> returning void)
-
-  let stubs_pixel_shuffle =
-    foreign "atg_pixel_shuffle" (ptr t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_pixel_shuffle_out =
-    foreign "atg_pixel_shuffle_out" (ptr t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_pixel_unshuffle =
-    foreign "atg_pixel_unshuffle" (ptr t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_pixel_unshuffle_out =
-    foreign "atg_pixel_unshuffle_out" (ptr t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_poisson = foreign "atg_poisson" (ptr t @-> t @-> returning void)
-
-  let stubs_poisson_nll_loss =
-    foreign
-      "atg_poisson_nll_loss"
-      (ptr t @-> t @-> t @-> int @-> int @-> double @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_poisson_out = foreign "atg_poisson_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_polar = foreign "atg_polar" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_polar_out =
-    foreign "atg_polar_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_polygamma =
-    foreign "atg_polygamma" (ptr t @-> int64_t @-> t @-> returning void)
-  ;;
-
-  let stubs_polygamma_ =
-    foreign "atg_polygamma_" (ptr t @-> t @-> int64_t @-> returning void)
-  ;;
-end
-
-module C18 (F : Cstubs.FOREIGN) = struct
-  open F
-
-  type t = unit ptr
-
-  let t : t typ = ptr void
-
-  type scalar = unit ptr
-
-  let scalar : scalar typ = ptr void
-
-  let stubs_polygamma_out =
-    foreign "atg_polygamma_out" (ptr t @-> t @-> int64_t @-> t @-> returning void)
-  ;;
-
-  let stubs_positive = foreign "atg_positive" (ptr t @-> t @-> returning void)
-  let stubs_pow = foreign "atg_pow" (ptr t @-> t @-> t @-> returning void)
-  let stubs_pow_ = foreign "atg_pow_" (ptr t @-> t @-> scalar @-> returning void)
-
-  let stubs_pow_scalar =
-    foreign "atg_pow_scalar" (ptr t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_pow_scalar_out =
-    foreign "atg_pow_scalar_out" (ptr t @-> t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_pow_tensor_ = foreign "atg_pow_tensor_" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_pow_tensor_scalar =
-    foreign "atg_pow_tensor_scalar" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_pow_tensor_scalar_out =
-    foreign "atg_pow_tensor_scalar_out" (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_pow_tensor_tensor_out =
-    foreign "atg_pow_tensor_tensor_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_prelu = foreign "atg_prelu" (ptr t @-> t @-> t @-> returning void)
-  let stubs_prod = foreign "atg_prod" (ptr t @-> t @-> int @-> returning void)
-
-  let stubs_prod_dim_int =
-    foreign "atg_prod_dim_int" (ptr t @-> t @-> int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_prod_int_out =
-    foreign
-      "atg_prod_int_out"
-      (ptr t @-> t @-> t @-> int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_prod_out =
-    foreign "atg_prod_out" (ptr t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_put = foreign "atg_put" (ptr t @-> t @-> t @-> t @-> int @-> returning void)
-  let stubs_put_ = foreign "atg_put_" (ptr t @-> t @-> t @-> t @-> int @-> returning void)
-
-  let stubs_put_out =
-    foreign "atg_put_out" (ptr t @-> t @-> t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_q_per_channel_axis = foreign "atg_q_per_channel_axis" (t @-> returning int64_t)
-
-  let stubs_q_per_channel_scales =
-    foreign "atg_q_per_channel_scales" (ptr t @-> t @-> returning void)
-  ;;
-
-  let stubs_q_per_channel_scales_out =
-    foreign "atg_q_per_channel_scales_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_q_per_channel_zero_points =
-    foreign "atg_q_per_channel_zero_points" (ptr t @-> t @-> returning void)
-  ;;
-
-  let stubs_q_per_channel_zero_points_out =
-    foreign "atg_q_per_channel_zero_points_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_q_scale = foreign "atg_q_scale" (t @-> returning double)
-  let stubs_q_zero_point = foreign "atg_q_zero_point" (t @-> returning int64_t)
-  let stubs_qr = foreign "atg_qr" (ptr t @-> t @-> int @-> returning void)
-  let stubs_qr_q = foreign "atg_qr_q" (ptr t @-> t @-> t @-> t @-> int @-> returning void)
-
-  let stubs_quantile =
-    foreign
-      "atg_quantile"
-      (ptr t @-> t @-> t @-> int64_t @-> int @-> int @-> string @-> returning void)
-  ;;
-
-  let stubs_quantile_out =
-    foreign
-      "atg_quantile_out"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> int @-> int @-> string @-> returning void)
-  ;;
-
-  let stubs_quantile_scalar =
-    foreign
-      "atg_quantile_scalar"
-      (ptr t @-> t @-> double @-> int64_t @-> int @-> int @-> string @-> returning void)
-  ;;
-
-  let stubs_quantile_scalar_out =
-    foreign
-      "atg_quantile_scalar_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> double
-       @-> int64_t
-       @-> int
-       @-> int
-       @-> string
-       @-> returning void)
-  ;;
-
-  let stubs_quantize_per_channel =
-    foreign
-      "atg_quantize_per_channel"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_quantize_per_channel_out =
-    foreign
-      "atg_quantize_per_channel_out"
-      (ptr t @-> t @-> t @-> t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_quantize_per_tensor =
-    foreign
-      "atg_quantize_per_tensor"
-      (ptr t @-> t @-> double @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_quantize_per_tensor_dynamic =
-    foreign
-      "atg_quantize_per_tensor_dynamic"
-      (ptr t @-> t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_quantize_per_tensor_dynamic_out =
-    foreign
-      "atg_quantize_per_tensor_dynamic_out"
-      (ptr t @-> t @-> t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_quantize_per_tensor_out =
-    foreign
-      "atg_quantize_per_tensor_out"
-      (ptr t @-> t @-> t @-> double @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_quantize_per_tensor_tensor_qparams =
-    foreign
-      "atg_quantize_per_tensor_tensor_qparams"
-      (ptr t @-> t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_quantize_per_tensor_tensor_qparams_out =
-    foreign
-      "atg_quantize_per_tensor_tensor_qparams_out"
-      (ptr t @-> t @-> t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_quantize_per_tensor_tensors =
-    foreign
-      "atg_quantize_per_tensor_tensors"
-      (ptr t @-> int @-> t @-> t @-> int @-> returning (ptr t))
-  ;;
-
-  let stubs_quantize_per_tensor_tensors_out =
-    foreign
-      "atg_quantize_per_tensor_tensors_out"
-      (ptr t @-> int @-> ptr t @-> int @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_quantized_batch_norm =
-    foreign
-      "atg_quantized_batch_norm"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> double
-       @-> double
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs_quantized_batch_norm_out =
-    foreign
-      "atg_quantized_batch_norm_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> double
-       @-> double
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs_quantized_gru_cell =
-    foreign
-      "atg_quantized_gru_cell"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> scalar
-       @-> scalar
-       @-> scalar
-       @-> scalar
-       @-> returning void)
-  ;;
-
-  let stubs_quantized_lstm_cell =
-    foreign
-      "atg_quantized_lstm_cell"
-      (ptr t
-       @-> t
-       @-> ptr t
-       @-> int
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> scalar
-       @-> scalar
-       @-> scalar
-       @-> scalar
-       @-> returning void)
-  ;;
-
-  let stubs_quantized_max_pool1d =
-    foreign
-      "atg_quantized_max_pool1d"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_quantized_max_pool1d_out =
-    foreign
-      "atg_quantized_max_pool1d_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_quantized_max_pool2d =
-    foreign
-      "atg_quantized_max_pool2d"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_quantized_max_pool2d_out =
-    foreign
-      "atg_quantized_max_pool2d_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_quantized_max_pool3d =
-    foreign
-      "atg_quantized_max_pool3d"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_quantized_max_pool3d_out =
-    foreign
-      "atg_quantized_max_pool3d_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_quantized_rnn_relu_cell =
-    foreign
-      "atg_quantized_rnn_relu_cell"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> scalar
-       @-> scalar
-       @-> scalar
-       @-> scalar
-       @-> returning void)
-  ;;
-
-  let stubs_quantized_rnn_tanh_cell =
-    foreign
-      "atg_quantized_rnn_tanh_cell"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> scalar
-       @-> scalar
-       @-> scalar
-       @-> scalar
-       @-> returning void)
-  ;;
-
-  let stubs_rad2deg = foreign "atg_rad2deg" (ptr t @-> t @-> returning void)
-  let stubs_rad2deg_ = foreign "atg_rad2deg_" (ptr t @-> t @-> returning void)
-  let stubs_rad2deg_out = foreign "atg_rad2deg_out" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_rand =
-    foreign "atg_rand" (ptr t @-> ptr int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_rand_like = foreign "atg_rand_like" (ptr t @-> t @-> returning void)
-
-  let stubs_rand_like_out =
-    foreign "atg_rand_like_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_rand_out =
-    foreign "atg_rand_out" (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_randint =
-    foreign
-      "atg_randint"
-      (ptr t @-> int64_t @-> ptr int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_randint_like =
-    foreign "atg_randint_like" (ptr t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_randint_like_low_dtype =
-    foreign
-      "atg_randint_like_low_dtype"
-      (ptr t @-> t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_randint_like_low_dtype_out =
-    foreign
-      "atg_randint_like_low_dtype_out"
-      (ptr t @-> t @-> t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_randint_like_out =
-    foreign "atg_randint_like_out" (ptr t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_randint_low =
-    foreign
-      "atg_randint_low"
-      (ptr t
-       @-> int64_t
-       @-> int64_t
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_randint_low_out =
-    foreign
-      "atg_randint_low_out"
-      (ptr t @-> t @-> int64_t @-> int64_t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_randint_out =
-    foreign
-      "atg_randint_out"
-      (ptr t @-> t @-> int64_t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_randn =
-    foreign "atg_randn" (ptr t @-> ptr int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_randn_like = foreign "atg_randn_like" (ptr t @-> t @-> returning void)
-
-  let stubs_randn_like_out =
-    foreign "atg_randn_like_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_randn_out =
-    foreign "atg_randn_out" (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_random = foreign "atg_random" (ptr t @-> t @-> returning void)
-  let stubs_random_ = foreign "atg_random_" (ptr t @-> t @-> returning void)
-
-  let stubs_random_from =
-    foreign
-      "atg_random_from"
-      (ptr t @-> t @-> int64_t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_random_from_ =
-    foreign
-      "atg_random_from_"
-      (ptr t @-> t @-> int64_t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_random_from_out =
-    foreign
-      "atg_random_from_out"
-      (ptr t @-> t @-> t @-> int64_t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_random_out = foreign "atg_random_out" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_random_to =
-    foreign "atg_random_to" (ptr t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_random_to_ =
-    foreign "atg_random_to_" (ptr t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_random_to_out =
-    foreign "atg_random_to_out" (ptr t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_randperm =
-    foreign "atg_randperm" (ptr t @-> int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_randperm_out =
-    foreign "atg_randperm_out" (ptr t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_range =
-    foreign "atg_range" (ptr t @-> scalar @-> scalar @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_range_out =
-    foreign "atg_range_out" (ptr t @-> t @-> scalar @-> scalar @-> returning void)
-  ;;
-
-  let stubs_range_out_ =
-    foreign "atg_range_out_" (ptr t @-> t @-> scalar @-> scalar @-> returning void)
-  ;;
-
-  let stubs_range_step =
-    foreign
-      "atg_range_step"
-      (ptr t @-> scalar @-> scalar @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_ravel = foreign "atg_ravel" (ptr t @-> t @-> returning void)
-  let stubs_real = foreign "atg_real" (ptr t @-> t @-> returning void)
-  let stubs_reciprocal = foreign "atg_reciprocal" (ptr t @-> t @-> returning void)
-  let stubs_reciprocal_ = foreign "atg_reciprocal_" (ptr t @-> t @-> returning void)
-
-  let stubs_reciprocal_out =
-    foreign "atg_reciprocal_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_reflection_pad1d =
-    foreign "atg_reflection_pad1d" (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_reflection_pad1d_backward =
-    foreign
-      "atg_reflection_pad1d_backward"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_reflection_pad1d_backward_grad_input =
-    foreign
-      "atg_reflection_pad1d_backward_grad_input"
-      (ptr t @-> t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_reflection_pad1d_out =
-    foreign
-      "atg_reflection_pad1d_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_reflection_pad2d =
-    foreign "atg_reflection_pad2d" (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_reflection_pad2d_backward =
-    foreign
-      "atg_reflection_pad2d_backward"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_reflection_pad2d_backward_grad_input =
-    foreign
-      "atg_reflection_pad2d_backward_grad_input"
-      (ptr t @-> t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_reflection_pad2d_out =
-    foreign
-      "atg_reflection_pad2d_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-end
-
-module C19 (F : Cstubs.FOREIGN) = struct
-  open F
-
-  type t = unit ptr
-
-  let t : t typ = ptr void
-
-  type scalar = unit ptr
-
-  let scalar : scalar typ = ptr void
-
-  let stubs_reflection_pad3d =
-    foreign "atg_reflection_pad3d" (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_reflection_pad3d_backward =
-    foreign
-      "atg_reflection_pad3d_backward"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_reflection_pad3d_backward_grad_input =
-    foreign
-      "atg_reflection_pad3d_backward_grad_input"
-      (ptr t @-> t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_reflection_pad3d_out =
-    foreign
-      "atg_reflection_pad3d_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_relu = foreign "atg_relu" (ptr t @-> t @-> returning void)
-  let stubs_relu6 = foreign "atg_relu6" (ptr t @-> t @-> returning void)
-  let stubs_relu6_ = foreign "atg_relu6_" (ptr t @-> t @-> returning void)
-  let stubs_relu_ = foreign "atg_relu_" (ptr t @-> t @-> returning void)
-  let stubs_relu_out = foreign "atg_relu_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_remainder = foreign "atg_remainder" (ptr t @-> t @-> scalar @-> returning void)
-
-  let stubs_remainder_ =
-    foreign "atg_remainder_" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_remainder_scalar_out =
-    foreign "atg_remainder_scalar_out" (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_remainder_scalar_tensor =
-    foreign "atg_remainder_scalar_tensor" (ptr t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_remainder_scalar_tensor_out =
-    foreign
-      "atg_remainder_scalar_tensor_out"
-      (ptr t @-> t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_remainder_tensor =
-    foreign "atg_remainder_tensor" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_remainder_tensor_ =
-    foreign "atg_remainder_tensor_" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_remainder_tensor_out =
-    foreign "atg_remainder_tensor_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_renorm =
-    foreign "atg_renorm" (ptr t @-> t @-> scalar @-> int64_t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_renorm_ =
-    foreign
-      "atg_renorm_"
-      (ptr t @-> t @-> scalar @-> int64_t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_renorm_out =
-    foreign
-      "atg_renorm_out"
-      (ptr t @-> t @-> t @-> scalar @-> int64_t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_repeat =
-    foreign "atg_repeat" (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_repeat_interleave =
-    foreign "atg_repeat_interleave" (ptr t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_repeat_interleave_self_int =
-    foreign
-      "atg_repeat_interleave_self_int"
-      (ptr t @-> t @-> int64_t @-> int64_t @-> int @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_repeat_interleave_self_tensor =
-    foreign
-      "atg_repeat_interleave_self_tensor"
-      (ptr t @-> t @-> t @-> int64_t @-> int @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_repeat_interleave_tensor_out =
-    foreign
-      "atg_repeat_interleave_tensor_out"
-      (ptr t @-> t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_repeat_out =
-    foreign "atg_repeat_out" (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_replication_pad1d =
-    foreign
-      "atg_replication_pad1d"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_replication_pad1d_backward =
-    foreign
-      "atg_replication_pad1d_backward"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_replication_pad1d_backward_grad_input =
-    foreign
-      "atg_replication_pad1d_backward_grad_input"
-      (ptr t @-> t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_replication_pad1d_out =
-    foreign
-      "atg_replication_pad1d_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_replication_pad2d =
-    foreign
-      "atg_replication_pad2d"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_replication_pad2d_backward =
-    foreign
-      "atg_replication_pad2d_backward"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_replication_pad2d_backward_grad_input =
-    foreign
-      "atg_replication_pad2d_backward_grad_input"
-      (ptr t @-> t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_replication_pad2d_out =
-    foreign
-      "atg_replication_pad2d_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_replication_pad3d =
-    foreign
-      "atg_replication_pad3d"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_replication_pad3d_backward =
-    foreign
-      "atg_replication_pad3d_backward"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_replication_pad3d_backward_grad_input =
-    foreign
-      "atg_replication_pad3d_backward_grad_input"
-      (ptr t @-> t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_replication_pad3d_out =
-    foreign
-      "atg_replication_pad3d_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_requires_grad_ =
-    foreign "atg_requires_grad_" (ptr t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_reshape =
-    foreign "atg_reshape" (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_reshape_as = foreign "atg_reshape_as" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_resize =
-    foreign "atg_resize" (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_resize_ =
-    foreign "atg_resize_" (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_resize_as = foreign "atg_resize_as" (ptr t @-> t @-> t @-> returning void)
-  let stubs_resize_as_ = foreign "atg_resize_as_" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_resize_as_out =
-    foreign "atg_resize_as_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_resize_as_sparse =
-    foreign "atg_resize_as_sparse" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_resize_as_sparse_ =
-    foreign "atg_resize_as_sparse_" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_resize_as_sparse_out =
-    foreign "atg_resize_as_sparse_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_resize_out =
-    foreign "atg_resize_out" (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_resolve_conj = foreign "atg_resolve_conj" (ptr t @-> t @-> returning void)
-  let stubs_resolve_neg = foreign "atg_resolve_neg" (ptr t @-> t @-> returning void)
-  let stubs_retains_grad = foreign "atg_retains_grad" (t @-> returning bool)
-
-  let stubs_rnn_relu =
-    foreign
-      "atg_rnn_relu"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr t
-       @-> int
-       @-> int
-       @-> int64_t
-       @-> double
-       @-> int
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_rnn_relu_cell =
-    foreign
-      "atg_rnn_relu_cell"
-      (ptr t @-> t @-> t @-> t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_rnn_relu_data =
-    foreign
-      "atg_rnn_relu_data"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr t
-       @-> int
-       @-> int
-       @-> int64_t
-       @-> double
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_rnn_tanh =
-    foreign
-      "atg_rnn_tanh"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr t
-       @-> int
-       @-> int
-       @-> int64_t
-       @-> double
-       @-> int
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_rnn_tanh_cell =
-    foreign
-      "atg_rnn_tanh_cell"
-      (ptr t @-> t @-> t @-> t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_rnn_tanh_data =
-    foreign
-      "atg_rnn_tanh_data"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr t
-       @-> int
-       @-> int
-       @-> int64_t
-       @-> double
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_roll =
-    foreign
-      "atg_roll"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_roll_out =
-    foreign
-      "atg_roll_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_rot90 =
-    foreign
-      "atg_rot90"
-      (ptr t @-> t @-> int64_t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_rot90_out =
-    foreign
-      "atg_rot90_out"
-      (ptr t @-> t @-> t @-> int64_t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_round = foreign "atg_round" (ptr t @-> t @-> returning void)
-  let stubs_round_ = foreign "atg_round_" (ptr t @-> t @-> returning void)
-
-  let stubs_round_decimals =
-    foreign "atg_round_decimals" (ptr t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_round_decimals_ =
-    foreign "atg_round_decimals_" (ptr t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_round_decimals_out =
-    foreign "atg_round_decimals_out" (ptr t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_round_out = foreign "atg_round_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_row_indices = foreign "atg_row_indices" (ptr t @-> t @-> returning void)
-
-  let stubs_row_indices_copy =
-    foreign "atg_row_indices_copy" (ptr t @-> t @-> returning void)
-  ;;
-
-  let stubs_row_indices_copy_out =
-    foreign "atg_row_indices_copy_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_row_stack =
-    foreign "atg_row_stack" (ptr t @-> ptr t @-> int @-> returning void)
-  ;;
-
-  let stubs_row_stack_out =
-    foreign "atg_row_stack_out" (ptr t @-> t @-> ptr t @-> int @-> returning void)
-  ;;
-
-  let stubs_rrelu = foreign "atg_rrelu" (ptr t @-> t @-> int @-> returning void)
-  let stubs_rrelu_ = foreign "atg_rrelu_" (ptr t @-> t @-> int @-> returning void)
-
-  let stubs_rrelu_with_noise =
-    foreign "atg_rrelu_with_noise" (ptr t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_rrelu_with_noise_ =
-    foreign "atg_rrelu_with_noise_" (ptr t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_rrelu_with_noise_backward =
-    foreign
-      "atg_rrelu_with_noise_backward"
-      (ptr t @-> t @-> t @-> t @-> scalar @-> scalar @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_rrelu_with_noise_backward_out =
-    foreign
-      "atg_rrelu_with_noise_backward_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> scalar
-       @-> scalar
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_rrelu_with_noise_out =
-    foreign "atg_rrelu_with_noise_out" (ptr t @-> t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_rsqrt = foreign "atg_rsqrt" (ptr t @-> t @-> returning void)
-  let stubs_rsqrt_ = foreign "atg_rsqrt_" (ptr t @-> t @-> returning void)
-  let stubs_rsqrt_out = foreign "atg_rsqrt_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_rsub = foreign "atg_rsub" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_rsub_scalar =
-    foreign "atg_rsub_scalar" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_rsub_scalar_out =
-    foreign "atg_rsub_scalar_out" (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_rsub_tensor_out =
-    foreign "atg_rsub_tensor_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_scalar_tensor =
-    foreign "atg_scalar_tensor" (ptr t @-> scalar @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_scalar_tensor_out =
-    foreign "atg_scalar_tensor_out" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_scaled_dot_product_attention =
-    foreign
-      "atg_scaled_dot_product_attention"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_scatter =
-    foreign "atg_scatter" (ptr t @-> t @-> int64_t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_scatter_ =
-    foreign "atg_scatter_" (ptr t @-> t @-> int64_t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_scatter_add =
-    foreign "atg_scatter_add" (ptr t @-> t @-> int64_t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_scatter_add_ =
-    foreign "atg_scatter_add_" (ptr t @-> t @-> int64_t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_scatter_add_out =
-    foreign
-      "atg_scatter_add_out"
-      (ptr t @-> t @-> t @-> int64_t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_scatter_reduce =
-    foreign
-      "atg_scatter_reduce"
-      (ptr t @-> t @-> int64_t @-> t @-> t @-> string @-> returning void)
-  ;;
-
-  let stubs_scatter_reduce_ =
-    foreign
-      "atg_scatter_reduce_"
-      (ptr t @-> t @-> int64_t @-> t @-> t @-> string @-> returning void)
-  ;;
-
-  let stubs_scatter_reduce_out =
-    foreign
-      "atg_scatter_reduce_out"
-      (ptr t @-> t @-> t @-> int64_t @-> t @-> t @-> string @-> returning void)
-  ;;
-
-  let stubs_scatter_src_out =
-    foreign
-      "atg_scatter_src_out"
-      (ptr t @-> t @-> t @-> int64_t @-> t @-> t @-> returning void)
-  ;;
-end
-
-module C20 (F : Cstubs.FOREIGN) = struct
-  open F
-
-  type t = unit ptr
-
-  let t : t typ = ptr void
-
-  type scalar = unit ptr
-
-  let scalar : scalar typ = ptr void
-
-  let stubs_scatter_value =
-    foreign
-      "atg_scatter_value"
-      (ptr t @-> t @-> int64_t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_scatter_value_ =
-    foreign
-      "atg_scatter_value_"
-      (ptr t @-> t @-> int64_t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_scatter_value_out =
-    foreign
-      "atg_scatter_value_out"
-      (ptr t @-> t @-> t @-> int64_t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_scatter_value_reduce =
-    foreign
-      "atg_scatter_value_reduce"
-      (ptr t @-> t @-> int64_t @-> t @-> scalar @-> string @-> returning void)
-  ;;
-
-  let stubs_scatter_value_reduce_ =
-    foreign
-      "atg_scatter_value_reduce_"
-      (ptr t @-> t @-> int64_t @-> t @-> scalar @-> string @-> returning void)
-  ;;
-
-  let stubs_scatter_value_reduce_out =
-    foreign
-      "atg_scatter_value_reduce_out"
-      (ptr t @-> t @-> t @-> int64_t @-> t @-> scalar @-> string @-> returning void)
-  ;;
-
-  let stubs_searchsorted =
-    foreign
-      "atg_searchsorted"
-      (ptr t @-> t @-> t @-> int @-> int @-> string @-> t @-> returning void)
-  ;;
-
-  let stubs_searchsorted_scalar =
-    foreign
-      "atg_searchsorted_scalar"
-      (ptr t @-> t @-> scalar @-> int @-> int @-> string @-> t @-> returning void)
-  ;;
-
-  let stubs_searchsorted_scalar_out =
-    foreign
-      "atg_searchsorted_scalar_out"
-      (ptr t @-> t @-> t @-> scalar @-> int @-> int @-> string @-> t @-> returning void)
-  ;;
-
-  let stubs_searchsorted_tensor_out =
-    foreign
-      "atg_searchsorted_tensor_out"
-      (ptr t @-> t @-> t @-> t @-> int @-> int @-> string @-> t @-> returning void)
-  ;;
-
-  let stubs_segment_reduce =
-    foreign
-      "atg_segment_reduce"
-      (ptr t
-       @-> t
-       @-> string
-       @-> t
-       @-> t
-       @-> t
-       @-> int64_t
-       @-> int
-       @-> scalar
-       @-> returning void)
-  ;;
-
-  let stubs_segment_reduce_out =
-    foreign
-      "atg_segment_reduce_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> string
-       @-> t
-       @-> t
-       @-> t
-       @-> int64_t
-       @-> int
-       @-> scalar
-       @-> returning void)
-  ;;
-
-  let stubs_select =
-    foreign "atg_select" (ptr t @-> t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_select_backward =
-    foreign
-      "atg_select_backward"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_select_backward_out =
-    foreign
-      "atg_select_backward_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs_select_copy =
-    foreign "atg_select_copy" (ptr t @-> t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_select_copy_int_out =
-    foreign
-      "atg_select_copy_int_out"
-      (ptr t @-> t @-> t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_select_scatter =
-    foreign
-      "atg_select_scatter"
-      (ptr t @-> t @-> t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_select_scatter_out =
-    foreign
-      "atg_select_scatter_out"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_selu = foreign "atg_selu" (ptr t @-> t @-> returning void)
-  let stubs_selu_ = foreign "atg_selu_" (ptr t @-> t @-> returning void)
-  let stubs_set = foreign "atg_set" (ptr t @-> t @-> returning void)
-  let stubs_set_ = foreign "atg_set_" (ptr t @-> t @-> returning void)
-  let stubs_set_out = foreign "atg_set_out" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_set_requires_grad =
-    foreign "atg_set_requires_grad" (ptr t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_set_source_tensor =
-    foreign "atg_set_source_tensor" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_set_source_tensor_ =
-    foreign "atg_set_source_tensor_" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_set_source_tensor_out =
-    foreign "atg_set_source_tensor_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_set_source_tensor_storage_offset_ =
-    foreign
-      "atg_set_source_tensor_storage_offset_"
-      (ptr t
-       @-> t
-       @-> t
-       @-> int64_t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_sgn = foreign "atg_sgn" (ptr t @-> t @-> returning void)
-  let stubs_sgn_ = foreign "atg_sgn_" (ptr t @-> t @-> returning void)
-  let stubs_sgn_out = foreign "atg_sgn_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_sigmoid = foreign "atg_sigmoid" (ptr t @-> t @-> returning void)
-  let stubs_sigmoid_ = foreign "atg_sigmoid_" (ptr t @-> t @-> returning void)
-
-  let stubs_sigmoid_backward =
-    foreign "atg_sigmoid_backward" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_sigmoid_backward_grad_input =
-    foreign "atg_sigmoid_backward_grad_input" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_sigmoid_out = foreign "atg_sigmoid_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_sign = foreign "atg_sign" (ptr t @-> t @-> returning void)
-  let stubs_sign_ = foreign "atg_sign_" (ptr t @-> t @-> returning void)
-  let stubs_sign_out = foreign "atg_sign_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_signbit = foreign "atg_signbit" (ptr t @-> t @-> returning void)
-  let stubs_signbit_out = foreign "atg_signbit_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_silu = foreign "atg_silu" (ptr t @-> t @-> returning void)
-  let stubs_silu_ = foreign "atg_silu_" (ptr t @-> t @-> returning void)
-
-  let stubs_silu_backward =
-    foreign "atg_silu_backward" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_silu_backward_grad_input =
-    foreign "atg_silu_backward_grad_input" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_silu_out = foreign "atg_silu_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_sin = foreign "atg_sin" (ptr t @-> t @-> returning void)
-  let stubs_sin_ = foreign "atg_sin_" (ptr t @-> t @-> returning void)
-  let stubs_sin_out = foreign "atg_sin_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_sinc = foreign "atg_sinc" (ptr t @-> t @-> returning void)
-  let stubs_sinc_ = foreign "atg_sinc_" (ptr t @-> t @-> returning void)
-  let stubs_sinc_out = foreign "atg_sinc_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_sinh = foreign "atg_sinh" (ptr t @-> t @-> returning void)
-  let stubs_sinh_ = foreign "atg_sinh_" (ptr t @-> t @-> returning void)
-  let stubs_sinh_out = foreign "atg_sinh_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_size = foreign "atg_size" (t @-> int64_t @-> returning int64_t)
-
-  let stubs_slice =
-    foreign
-      "atg_slice"
-      (ptr t
-       @-> t
-       @-> int64_t
-       @-> int64_t
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs_slice_backward =
-    foreign
-      "atg_slice_backward"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> int64_t
-       @-> int64_t
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs_slice_backward_out =
-    foreign
-      "atg_slice_backward_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> int64_t
-       @-> int64_t
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs_slice_copy =
-    foreign
-      "atg_slice_copy"
-      (ptr t
-       @-> t
-       @-> int64_t
-       @-> int64_t
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs_slice_copy_tensor_out =
-    foreign
-      "atg_slice_copy_tensor_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> int64_t
-       @-> int64_t
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs_slice_scatter =
-    foreign
-      "atg_slice_scatter"
-      (ptr t
-       @-> t
-       @-> t
-       @-> int64_t
-       @-> int64_t
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs_slice_scatter_out =
-    foreign
-      "atg_slice_scatter_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> int64_t
-       @-> int64_t
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs_slogdet = foreign "atg_slogdet" (ptr t @-> t @-> returning void)
-
-  let stubs_slogdet_out =
-    foreign "atg_slogdet_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_slow_conv3d =
-    foreign
-      "atg_slow_conv3d"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_slow_conv3d_out =
-    foreign
-      "atg_slow_conv3d_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_slow_conv_dilated2d =
-    foreign
-      "atg_slow_conv_dilated2d"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_slow_conv_dilated2d_out =
-    foreign
-      "atg_slow_conv_dilated2d_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_slow_conv_dilated3d =
-    foreign
-      "atg_slow_conv_dilated3d"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_slow_conv_dilated3d_out =
-    foreign
-      "atg_slow_conv_dilated3d_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_slow_conv_transpose2d =
-    foreign
-      "atg_slow_conv_transpose2d"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_slow_conv_transpose2d_out =
-    foreign
-      "atg_slow_conv_transpose2d_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_slow_conv_transpose3d =
-    foreign
-      "atg_slow_conv_transpose3d"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_slow_conv_transpose3d_out =
-    foreign
-      "atg_slow_conv_transpose3d_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_smm = foreign "atg_smm" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_smooth_l1_loss =
-    foreign
-      "atg_smooth_l1_loss"
-      (ptr t @-> t @-> t @-> int64_t @-> double @-> returning void)
-  ;;
-
-  let stubs_smooth_l1_loss_backward =
-    foreign
-      "atg_smooth_l1_loss_backward"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> double @-> returning void)
-  ;;
-
-  let stubs_smooth_l1_loss_backward_grad_input =
-    foreign
-      "atg_smooth_l1_loss_backward_grad_input"
-      (ptr t @-> t @-> t @-> t @-> t @-> int64_t @-> double @-> returning void)
-  ;;
-
-  let stubs_smooth_l1_loss_out =
-    foreign
-      "atg_smooth_l1_loss_out"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> double @-> returning void)
-  ;;
-
-  let stubs_soft_margin_loss =
-    foreign "atg_soft_margin_loss" (ptr t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_soft_margin_loss_backward =
-    foreign
-      "atg_soft_margin_loss_backward"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_soft_margin_loss_backward_grad_input =
-    foreign
-      "atg_soft_margin_loss_backward_grad_input"
-      (ptr t @-> t @-> t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_soft_margin_loss_out =
-    foreign
-      "atg_soft_margin_loss_out"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_softmax =
-    foreign "atg_softmax" (ptr t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_softmax_int_out =
-    foreign
-      "atg_softmax_int_out"
-      (ptr t @-> t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_softplus = foreign "atg_softplus" (ptr t @-> t @-> returning void)
-
-  let stubs_softplus_backward =
-    foreign
-      "atg_softplus_backward"
-      (ptr t @-> t @-> t @-> scalar @-> scalar @-> returning void)
-  ;;
-
-  let stubs_softplus_backward_grad_input =
-    foreign
-      "atg_softplus_backward_grad_input"
-      (ptr t @-> t @-> t @-> t @-> scalar @-> scalar @-> returning void)
-  ;;
-
-  let stubs_softplus_out =
-    foreign "atg_softplus_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_softshrink = foreign "atg_softshrink" (ptr t @-> t @-> returning void)
-
-  let stubs_softshrink_backward =
-    foreign "atg_softshrink_backward" (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_softshrink_backward_grad_input =
-    foreign
-      "atg_softshrink_backward_grad_input"
-      (ptr t @-> t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_softshrink_out =
-    foreign "atg_softshrink_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_sort = foreign "atg_sort" (ptr t @-> t @-> int64_t @-> int @-> returning void)
-
-  let stubs_sort_stable =
-    foreign "atg_sort_stable" (ptr t @-> t @-> int @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_sort_values =
-    foreign
-      "atg_sort_values"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_sort_values_stable =
-    foreign
-      "atg_sort_values_stable"
-      (ptr t @-> t @-> t @-> t @-> int @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_sparse_bsc_tensor =
-    foreign
-      "atg_sparse_bsc_tensor"
-      (ptr t @-> t @-> t @-> t @-> int @-> int @-> returning void)
-  ;;
-end
-
-module C21 (F : Cstubs.FOREIGN) = struct
-  open F
-
-  type t = unit ptr
-
-  let t : t typ = ptr void
-
-  type scalar = unit ptr
-
-  let scalar : scalar typ = ptr void
-
-  let stubs_sparse_bsc_tensor_ccol_row_value_size =
-    foreign
-      "atg_sparse_bsc_tensor_ccol_row_value_size"
-      (ptr t @-> t @-> t @-> t @-> ptr int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_sparse_bsr_tensor =
-    foreign
-      "atg_sparse_bsr_tensor"
-      (ptr t @-> t @-> t @-> t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_sparse_bsr_tensor_crow_col_value_size =
-    foreign
-      "atg_sparse_bsr_tensor_crow_col_value_size"
-      (ptr t @-> t @-> t @-> t @-> ptr int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_sparse_compressed_tensor =
-    foreign
-      "atg_sparse_compressed_tensor"
-      (ptr t @-> t @-> t @-> t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_sparse_compressed_tensor_comp_plain_value_size =
-    foreign
-      "atg_sparse_compressed_tensor_comp_plain_value_size"
-      (ptr t @-> t @-> t @-> t @-> ptr int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_sparse_coo_tensor =
-    foreign
-      "atg_sparse_coo_tensor"
-      (ptr t @-> ptr int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_sparse_coo_tensor_indices =
-    foreign
-      "atg_sparse_coo_tensor_indices"
-      (ptr t @-> t @-> t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_sparse_coo_tensor_indices_size =
-    foreign
-      "atg_sparse_coo_tensor_indices_size"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_sparse_coo_tensor_size_out =
-    foreign
-      "atg_sparse_coo_tensor_size_out"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_sparse_csc_tensor =
-    foreign
-      "atg_sparse_csc_tensor"
-      (ptr t @-> t @-> t @-> t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_sparse_csc_tensor_ccol_row_value_size =
-    foreign
-      "atg_sparse_csc_tensor_ccol_row_value_size"
-      (ptr t @-> t @-> t @-> t @-> ptr int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_sparse_csr_tensor =
-    foreign
-      "atg_sparse_csr_tensor"
-      (ptr t @-> t @-> t @-> t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_sparse_csr_tensor_crow_col_value_size =
-    foreign
-      "atg_sparse_csr_tensor_crow_col_value_size"
-      (ptr t @-> t @-> t @-> t @-> ptr int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_sparse_dim = foreign "atg_sparse_dim" (t @-> returning int64_t)
-  let stubs_sparse_mask = foreign "atg_sparse_mask" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_sparse_mask_out =
-    foreign "atg_sparse_mask_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_sparse_resize =
-    foreign
-      "atg_sparse_resize"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_sparse_resize_ =
-    foreign
-      "atg_sparse_resize_"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_sparse_resize_and_clear =
-    foreign
-      "atg_sparse_resize_and_clear"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_sparse_resize_and_clear_ =
-    foreign
-      "atg_sparse_resize_and_clear_"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_sparse_resize_and_clear_out =
-    foreign
-      "atg_sparse_resize_and_clear_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs_sparse_resize_out =
-    foreign
-      "atg_sparse_resize_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs_sparse_sampled_addmm =
-    foreign "atg_sparse_sampled_addmm" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_sparse_sampled_addmm_out =
-    foreign
-      "atg_sparse_sampled_addmm_out"
-      (ptr t @-> t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_airy_ai =
-    foreign "atg_special_airy_ai" (ptr t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_airy_ai_out =
-    foreign "atg_special_airy_ai_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_bessel_j0 =
-    foreign "atg_special_bessel_j0" (ptr t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_bessel_j0_out =
-    foreign "atg_special_bessel_j0_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_bessel_j1 =
-    foreign "atg_special_bessel_j1" (ptr t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_bessel_j1_out =
-    foreign "atg_special_bessel_j1_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_bessel_y0 =
-    foreign "atg_special_bessel_y0" (ptr t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_bessel_y0_out =
-    foreign "atg_special_bessel_y0_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_bessel_y1 =
-    foreign "atg_special_bessel_y1" (ptr t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_bessel_y1_out =
-    foreign "atg_special_bessel_y1_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_chebyshev_polynomial_t =
-    foreign "atg_special_chebyshev_polynomial_t" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_chebyshev_polynomial_t_n_scalar =
-    foreign
-      "atg_special_chebyshev_polynomial_t_n_scalar"
-      (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_special_chebyshev_polynomial_t_n_scalar_out =
-    foreign
-      "atg_special_chebyshev_polynomial_t_n_scalar_out"
-      (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_special_chebyshev_polynomial_t_out =
-    foreign
-      "atg_special_chebyshev_polynomial_t_out"
-      (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_chebyshev_polynomial_t_x_scalar =
-    foreign
-      "atg_special_chebyshev_polynomial_t_x_scalar"
-      (ptr t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_special_chebyshev_polynomial_t_x_scalar_out =
-    foreign
-      "atg_special_chebyshev_polynomial_t_x_scalar_out"
-      (ptr t @-> t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_special_chebyshev_polynomial_u =
-    foreign "atg_special_chebyshev_polynomial_u" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_chebyshev_polynomial_u_n_scalar =
-    foreign
-      "atg_special_chebyshev_polynomial_u_n_scalar"
-      (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_special_chebyshev_polynomial_u_n_scalar_out =
-    foreign
-      "atg_special_chebyshev_polynomial_u_n_scalar_out"
-      (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_special_chebyshev_polynomial_u_out =
-    foreign
-      "atg_special_chebyshev_polynomial_u_out"
-      (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_chebyshev_polynomial_u_x_scalar =
-    foreign
-      "atg_special_chebyshev_polynomial_u_x_scalar"
-      (ptr t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_special_chebyshev_polynomial_u_x_scalar_out =
-    foreign
-      "atg_special_chebyshev_polynomial_u_x_scalar_out"
-      (ptr t @-> t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_special_chebyshev_polynomial_v =
-    foreign "atg_special_chebyshev_polynomial_v" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_chebyshev_polynomial_v_n_scalar =
-    foreign
-      "atg_special_chebyshev_polynomial_v_n_scalar"
-      (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_special_chebyshev_polynomial_v_n_scalar_out =
-    foreign
-      "atg_special_chebyshev_polynomial_v_n_scalar_out"
-      (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_special_chebyshev_polynomial_v_out =
-    foreign
-      "atg_special_chebyshev_polynomial_v_out"
-      (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_chebyshev_polynomial_v_x_scalar =
-    foreign
-      "atg_special_chebyshev_polynomial_v_x_scalar"
-      (ptr t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_special_chebyshev_polynomial_v_x_scalar_out =
-    foreign
-      "atg_special_chebyshev_polynomial_v_x_scalar_out"
-      (ptr t @-> t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_special_chebyshev_polynomial_w =
-    foreign "atg_special_chebyshev_polynomial_w" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_chebyshev_polynomial_w_n_scalar =
-    foreign
-      "atg_special_chebyshev_polynomial_w_n_scalar"
-      (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_special_chebyshev_polynomial_w_n_scalar_out =
-    foreign
-      "atg_special_chebyshev_polynomial_w_n_scalar_out"
-      (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_special_chebyshev_polynomial_w_out =
-    foreign
-      "atg_special_chebyshev_polynomial_w_out"
-      (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_chebyshev_polynomial_w_x_scalar =
-    foreign
-      "atg_special_chebyshev_polynomial_w_x_scalar"
-      (ptr t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_special_chebyshev_polynomial_w_x_scalar_out =
-    foreign
-      "atg_special_chebyshev_polynomial_w_x_scalar_out"
-      (ptr t @-> t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_special_digamma =
-    foreign "atg_special_digamma" (ptr t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_digamma_out =
-    foreign "atg_special_digamma_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_entr = foreign "atg_special_entr" (ptr t @-> t @-> returning void)
-
-  let stubs_special_entr_out =
-    foreign "atg_special_entr_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_erf = foreign "atg_special_erf" (ptr t @-> t @-> returning void)
-
-  let stubs_special_erf_out =
-    foreign "atg_special_erf_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_erfc = foreign "atg_special_erfc" (ptr t @-> t @-> returning void)
-
-  let stubs_special_erfc_out =
-    foreign "atg_special_erfc_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_erfcx = foreign "atg_special_erfcx" (ptr t @-> t @-> returning void)
-
-  let stubs_special_erfcx_out =
-    foreign "atg_special_erfcx_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_erfinv = foreign "atg_special_erfinv" (ptr t @-> t @-> returning void)
-
-  let stubs_special_erfinv_out =
-    foreign "atg_special_erfinv_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_exp2 = foreign "atg_special_exp2" (ptr t @-> t @-> returning void)
-
-  let stubs_special_exp2_out =
-    foreign "atg_special_exp2_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_expit = foreign "atg_special_expit" (ptr t @-> t @-> returning void)
-
-  let stubs_special_expit_out =
-    foreign "atg_special_expit_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_expm1 = foreign "atg_special_expm1" (ptr t @-> t @-> returning void)
-
-  let stubs_special_expm1_out =
-    foreign "atg_special_expm1_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_gammainc =
-    foreign "atg_special_gammainc" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_gammainc_out =
-    foreign "atg_special_gammainc_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_gammaincc =
-    foreign "atg_special_gammaincc" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_gammaincc_out =
-    foreign "atg_special_gammaincc_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_gammaln =
-    foreign "atg_special_gammaln" (ptr t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_gammaln_out =
-    foreign "atg_special_gammaln_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_hermite_polynomial_h =
-    foreign "atg_special_hermite_polynomial_h" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_hermite_polynomial_h_n_scalar =
-    foreign
-      "atg_special_hermite_polynomial_h_n_scalar"
-      (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_special_hermite_polynomial_h_n_scalar_out =
-    foreign
-      "atg_special_hermite_polynomial_h_n_scalar_out"
-      (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_special_hermite_polynomial_h_out =
-    foreign
-      "atg_special_hermite_polynomial_h_out"
-      (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_hermite_polynomial_h_x_scalar =
-    foreign
-      "atg_special_hermite_polynomial_h_x_scalar"
-      (ptr t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_special_hermite_polynomial_h_x_scalar_out =
-    foreign
-      "atg_special_hermite_polynomial_h_x_scalar_out"
-      (ptr t @-> t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_special_hermite_polynomial_he =
-    foreign "atg_special_hermite_polynomial_he" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_hermite_polynomial_he_n_scalar =
-    foreign
-      "atg_special_hermite_polynomial_he_n_scalar"
-      (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_special_hermite_polynomial_he_n_scalar_out =
-    foreign
-      "atg_special_hermite_polynomial_he_n_scalar_out"
-      (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_special_hermite_polynomial_he_out =
-    foreign
-      "atg_special_hermite_polynomial_he_out"
-      (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_hermite_polynomial_he_x_scalar =
-    foreign
-      "atg_special_hermite_polynomial_he_x_scalar"
-      (ptr t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_special_hermite_polynomial_he_x_scalar_out =
-    foreign
-      "atg_special_hermite_polynomial_he_x_scalar_out"
-      (ptr t @-> t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_special_i0 = foreign "atg_special_i0" (ptr t @-> t @-> returning void)
-
-  let stubs_special_i0_out =
-    foreign "atg_special_i0_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_i0e = foreign "atg_special_i0e" (ptr t @-> t @-> returning void)
-
-  let stubs_special_i0e_out =
-    foreign "atg_special_i0e_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_i1 = foreign "atg_special_i1" (ptr t @-> t @-> returning void)
-
-  let stubs_special_i1_out =
-    foreign "atg_special_i1_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-end
-
-module C22 (F : Cstubs.FOREIGN) = struct
-  open F
-
-  type t = unit ptr
-
-  let t : t typ = ptr void
-
-  type scalar = unit ptr
-
-  let scalar : scalar typ = ptr void
-  let stubs_special_i1e = foreign "atg_special_i1e" (ptr t @-> t @-> returning void)
-
-  let stubs_special_i1e_out =
-    foreign "atg_special_i1e_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_laguerre_polynomial_l =
-    foreign "atg_special_laguerre_polynomial_l" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_laguerre_polynomial_l_n_scalar =
-    foreign
-      "atg_special_laguerre_polynomial_l_n_scalar"
-      (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_special_laguerre_polynomial_l_n_scalar_out =
-    foreign
-      "atg_special_laguerre_polynomial_l_n_scalar_out"
-      (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_special_laguerre_polynomial_l_out =
-    foreign
-      "atg_special_laguerre_polynomial_l_out"
-      (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_laguerre_polynomial_l_x_scalar =
-    foreign
-      "atg_special_laguerre_polynomial_l_x_scalar"
-      (ptr t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_special_laguerre_polynomial_l_x_scalar_out =
-    foreign
-      "atg_special_laguerre_polynomial_l_x_scalar_out"
-      (ptr t @-> t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_special_legendre_polynomial_p =
-    foreign "atg_special_legendre_polynomial_p" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_legendre_polynomial_p_n_scalar =
-    foreign
-      "atg_special_legendre_polynomial_p_n_scalar"
-      (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_special_legendre_polynomial_p_n_scalar_out =
-    foreign
-      "atg_special_legendre_polynomial_p_n_scalar_out"
-      (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_special_legendre_polynomial_p_out =
-    foreign
-      "atg_special_legendre_polynomial_p_out"
-      (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_legendre_polynomial_p_x_scalar =
-    foreign
-      "atg_special_legendre_polynomial_p_x_scalar"
-      (ptr t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_special_legendre_polynomial_p_x_scalar_out =
-    foreign
-      "atg_special_legendre_polynomial_p_x_scalar_out"
-      (ptr t @-> t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_special_log1p = foreign "atg_special_log1p" (ptr t @-> t @-> returning void)
-
-  let stubs_special_log1p_out =
-    foreign "atg_special_log1p_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_log_ndtr =
-    foreign "atg_special_log_ndtr" (ptr t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_log_ndtr_out =
-    foreign "atg_special_log_ndtr_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_log_softmax =
-    foreign "atg_special_log_softmax" (ptr t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_special_logit =
-    foreign "atg_special_logit" (ptr t @-> t @-> double @-> int @-> returning void)
-  ;;
-
-  let stubs_special_logit_out =
-    foreign
-      "atg_special_logit_out"
-      (ptr t @-> t @-> t @-> double @-> int @-> returning void)
-  ;;
-
-  let stubs_special_logsumexp =
-    foreign
-      "atg_special_logsumexp"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_special_logsumexp_out =
-    foreign
-      "atg_special_logsumexp_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_special_modified_bessel_i0 =
-    foreign "atg_special_modified_bessel_i0" (ptr t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_modified_bessel_i0_out =
-    foreign "atg_special_modified_bessel_i0_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_modified_bessel_i1 =
-    foreign "atg_special_modified_bessel_i1" (ptr t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_modified_bessel_i1_out =
-    foreign "atg_special_modified_bessel_i1_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_modified_bessel_k0 =
-    foreign "atg_special_modified_bessel_k0" (ptr t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_modified_bessel_k0_out =
-    foreign "atg_special_modified_bessel_k0_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_modified_bessel_k1 =
-    foreign "atg_special_modified_bessel_k1" (ptr t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_modified_bessel_k1_out =
-    foreign "atg_special_modified_bessel_k1_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_multigammaln =
-    foreign "atg_special_multigammaln" (ptr t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_special_multigammaln_out =
-    foreign
-      "atg_special_multigammaln_out"
-      (ptr t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_special_ndtr = foreign "atg_special_ndtr" (ptr t @-> t @-> returning void)
-
-  let stubs_special_ndtr_out =
-    foreign "atg_special_ndtr_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_ndtri = foreign "atg_special_ndtri" (ptr t @-> t @-> returning void)
-
-  let stubs_special_ndtri_out =
-    foreign "atg_special_ndtri_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_polygamma =
-    foreign "atg_special_polygamma" (ptr t @-> int64_t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_polygamma_out =
-    foreign "atg_special_polygamma_out" (ptr t @-> t @-> int64_t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_psi = foreign "atg_special_psi" (ptr t @-> t @-> returning void)
-
-  let stubs_special_psi_out =
-    foreign "atg_special_psi_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_round =
-    foreign "atg_special_round" (ptr t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_special_round_out =
-    foreign "atg_special_round_out" (ptr t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_special_scaled_modified_bessel_k0 =
-    foreign "atg_special_scaled_modified_bessel_k0" (ptr t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_scaled_modified_bessel_k0_out =
-    foreign
-      "atg_special_scaled_modified_bessel_k0_out"
-      (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_scaled_modified_bessel_k1 =
-    foreign "atg_special_scaled_modified_bessel_k1" (ptr t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_scaled_modified_bessel_k1_out =
-    foreign
-      "atg_special_scaled_modified_bessel_k1_out"
-      (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_shifted_chebyshev_polynomial_t =
-    foreign
-      "atg_special_shifted_chebyshev_polynomial_t"
-      (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_shifted_chebyshev_polynomial_t_n_scalar =
-    foreign
-      "atg_special_shifted_chebyshev_polynomial_t_n_scalar"
-      (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_special_shifted_chebyshev_polynomial_t_n_scalar_out =
-    foreign
-      "atg_special_shifted_chebyshev_polynomial_t_n_scalar_out"
-      (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_special_shifted_chebyshev_polynomial_t_out =
-    foreign
-      "atg_special_shifted_chebyshev_polynomial_t_out"
-      (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_shifted_chebyshev_polynomial_t_x_scalar =
-    foreign
-      "atg_special_shifted_chebyshev_polynomial_t_x_scalar"
-      (ptr t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_special_shifted_chebyshev_polynomial_t_x_scalar_out =
-    foreign
-      "atg_special_shifted_chebyshev_polynomial_t_x_scalar_out"
-      (ptr t @-> t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_special_shifted_chebyshev_polynomial_u =
-    foreign
-      "atg_special_shifted_chebyshev_polynomial_u"
-      (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_shifted_chebyshev_polynomial_u_n_scalar =
-    foreign
-      "atg_special_shifted_chebyshev_polynomial_u_n_scalar"
-      (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_special_shifted_chebyshev_polynomial_u_n_scalar_out =
-    foreign
-      "atg_special_shifted_chebyshev_polynomial_u_n_scalar_out"
-      (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_special_shifted_chebyshev_polynomial_u_out =
-    foreign
-      "atg_special_shifted_chebyshev_polynomial_u_out"
-      (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_shifted_chebyshev_polynomial_u_x_scalar =
-    foreign
-      "atg_special_shifted_chebyshev_polynomial_u_x_scalar"
-      (ptr t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_special_shifted_chebyshev_polynomial_u_x_scalar_out =
-    foreign
-      "atg_special_shifted_chebyshev_polynomial_u_x_scalar_out"
-      (ptr t @-> t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_special_shifted_chebyshev_polynomial_v =
-    foreign
-      "atg_special_shifted_chebyshev_polynomial_v"
-      (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_shifted_chebyshev_polynomial_v_n_scalar =
-    foreign
-      "atg_special_shifted_chebyshev_polynomial_v_n_scalar"
-      (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_special_shifted_chebyshev_polynomial_v_n_scalar_out =
-    foreign
-      "atg_special_shifted_chebyshev_polynomial_v_n_scalar_out"
-      (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_special_shifted_chebyshev_polynomial_v_out =
-    foreign
-      "atg_special_shifted_chebyshev_polynomial_v_out"
-      (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_shifted_chebyshev_polynomial_v_x_scalar =
-    foreign
-      "atg_special_shifted_chebyshev_polynomial_v_x_scalar"
-      (ptr t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_special_shifted_chebyshev_polynomial_v_x_scalar_out =
-    foreign
-      "atg_special_shifted_chebyshev_polynomial_v_x_scalar_out"
-      (ptr t @-> t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_special_shifted_chebyshev_polynomial_w =
-    foreign
-      "atg_special_shifted_chebyshev_polynomial_w"
-      (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_shifted_chebyshev_polynomial_w_n_scalar =
-    foreign
-      "atg_special_shifted_chebyshev_polynomial_w_n_scalar"
-      (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_special_shifted_chebyshev_polynomial_w_n_scalar_out =
-    foreign
-      "atg_special_shifted_chebyshev_polynomial_w_n_scalar_out"
-      (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_special_shifted_chebyshev_polynomial_w_out =
-    foreign
-      "atg_special_shifted_chebyshev_polynomial_w_out"
-      (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_shifted_chebyshev_polynomial_w_x_scalar =
-    foreign
-      "atg_special_shifted_chebyshev_polynomial_w_x_scalar"
-      (ptr t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_special_shifted_chebyshev_polynomial_w_x_scalar_out =
-    foreign
-      "atg_special_shifted_chebyshev_polynomial_w_x_scalar_out"
-      (ptr t @-> t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_special_sinc = foreign "atg_special_sinc" (ptr t @-> t @-> returning void)
-
-  let stubs_special_sinc_out =
-    foreign "atg_special_sinc_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_softmax =
-    foreign "atg_special_softmax" (ptr t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_special_spherical_bessel_j0 =
-    foreign "atg_special_spherical_bessel_j0" (ptr t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_spherical_bessel_j0_out =
-    foreign "atg_special_spherical_bessel_j0_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_xlog1py =
-    foreign "atg_special_xlog1py" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_xlog1py_other_scalar =
-    foreign "atg_special_xlog1py_other_scalar" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_special_xlog1py_other_scalar_out =
-    foreign
-      "atg_special_xlog1py_other_scalar_out"
-      (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_special_xlog1py_out =
-    foreign "atg_special_xlog1py_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_xlog1py_self_scalar =
-    foreign "atg_special_xlog1py_self_scalar" (ptr t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_special_xlog1py_self_scalar_out =
-    foreign
-      "atg_special_xlog1py_self_scalar_out"
-      (ptr t @-> t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_special_xlogy =
-    foreign "atg_special_xlogy" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_xlogy_other_scalar =
-    foreign "atg_special_xlogy_other_scalar" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_special_xlogy_other_scalar_out =
-    foreign
-      "atg_special_xlogy_other_scalar_out"
-      (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_special_xlogy_out =
-    foreign "atg_special_xlogy_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_xlogy_self_scalar =
-    foreign "atg_special_xlogy_self_scalar" (ptr t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_special_xlogy_self_scalar_out =
-    foreign
-      "atg_special_xlogy_self_scalar_out"
-      (ptr t @-> t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_special_zeta =
-    foreign "atg_special_zeta" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_zeta_other_scalar =
-    foreign "atg_special_zeta_other_scalar" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_special_zeta_other_scalar_out =
-    foreign
-      "atg_special_zeta_other_scalar_out"
-      (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_special_zeta_out =
-    foreign "atg_special_zeta_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_special_zeta_self_scalar =
-    foreign "atg_special_zeta_self_scalar" (ptr t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_special_zeta_self_scalar_out =
-    foreign
-      "atg_special_zeta_self_scalar_out"
-      (ptr t @-> t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_split = foreign "atg_split" (t @-> int64_t @-> int64_t @-> returning (ptr t))
-
-  let stubs_split_copy =
-    foreign "atg_split_copy" (t @-> int64_t @-> int64_t @-> returning (ptr t))
-  ;;
-
-  let stubs_split_copy_tensor_out =
-    foreign
-      "atg_split_copy_tensor_out"
-      (ptr t @-> int @-> t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_split_sizes =
-    foreign "atg_split_sizes" (t @-> ptr int64_t @-> int @-> int64_t @-> returning (ptr t))
-  ;;
-
-  let stubs_split_with_sizes =
-    foreign
-      "atg_split_with_sizes"
-      (t @-> ptr int64_t @-> int @-> int64_t @-> returning (ptr t))
-  ;;
-
-  let stubs_split_with_sizes_copy =
-    foreign
-      "atg_split_with_sizes_copy"
-      (t @-> ptr int64_t @-> int @-> int64_t @-> returning (ptr t))
-  ;;
-end
-
-module C23 (F : Cstubs.FOREIGN) = struct
-  open F
-
-  type t = unit ptr
-
-  let t : t typ = ptr void
-
-  type scalar = unit ptr
-
-  let scalar : scalar typ = ptr void
-
-  let stubs_split_with_sizes_copy_out =
-    foreign
-      "atg_split_with_sizes_copy_out"
-      (ptr t @-> int @-> t @-> ptr int64_t @-> int @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_sqrt = foreign "atg_sqrt" (ptr t @-> t @-> returning void)
-  let stubs_sqrt_ = foreign "atg_sqrt_" (ptr t @-> t @-> returning void)
-  let stubs_sqrt_out = foreign "atg_sqrt_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_square = foreign "atg_square" (ptr t @-> t @-> returning void)
-  let stubs_square_ = foreign "atg_square_" (ptr t @-> t @-> returning void)
-  let stubs_square_out = foreign "atg_square_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_squeeze = foreign "atg_squeeze" (ptr t @-> t @-> returning void)
-  let stubs_squeeze_ = foreign "atg_squeeze_" (ptr t @-> t @-> returning void)
-  let stubs_squeeze_copy = foreign "atg_squeeze_copy" (ptr t @-> t @-> returning void)
-
-  let stubs_squeeze_copy_dim =
-    foreign "atg_squeeze_copy_dim" (ptr t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_squeeze_copy_dim_out =
-    foreign "atg_squeeze_copy_dim_out" (ptr t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_squeeze_copy_dims =
-    foreign
-      "atg_squeeze_copy_dims"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_squeeze_copy_dims_out =
-    foreign
-      "atg_squeeze_copy_dims_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_squeeze_copy_out =
-    foreign "atg_squeeze_copy_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_squeeze_dim =
-    foreign "atg_squeeze_dim" (ptr t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_squeeze_dim_ =
-    foreign "atg_squeeze_dim_" (ptr t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_squeeze_dims =
-    foreign "atg_squeeze_dims" (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_squeeze_dims_ =
-    foreign "atg_squeeze_dims_" (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_sspaddmm = foreign "atg_sspaddmm" (ptr t @-> t @-> t @-> t @-> returning void)
-
-  let stubs_sspaddmm_out =
-    foreign "atg_sspaddmm_out" (ptr t @-> t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_stack =
-    foreign "atg_stack" (ptr t @-> ptr t @-> int @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_stack_out =
-    foreign "atg_stack_out" (ptr t @-> t @-> ptr t @-> int @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_std = foreign "atg_std" (ptr t @-> t @-> int @-> returning void)
-
-  let stubs_std_correction =
-    foreign
-      "atg_std_correction"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> scalar @-> int @-> returning void)
-  ;;
-
-  let stubs_std_correction_out =
-    foreign
-      "atg_std_correction_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> scalar @-> int @-> returning void)
-  ;;
-
-  let stubs_std_dim =
-    foreign
-      "atg_std_dim"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_std_mean = foreign "atg_std_mean" (ptr t @-> t @-> int @-> returning void)
-
-  let stubs_std_mean_correction =
-    foreign
-      "atg_std_mean_correction"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> scalar @-> int @-> returning void)
-  ;;
-
-  let stubs_std_mean_correction_out =
-    foreign
-      "atg_std_mean_correction_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> scalar
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_std_mean_dim =
-    foreign
-      "atg_std_mean_dim"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_std_out =
-    foreign
-      "atg_std_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_stft =
-    foreign
-      "atg_stft"
-      (ptr t
-       @-> t
-       @-> int64_t
-       @-> int64_t
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> t
-       @-> int
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_stft_center =
-    foreign
-      "atg_stft_center"
-      (ptr t
-       @-> t
-       @-> int64_t
-       @-> int64_t
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> t
-       @-> int
-       @-> string
-       @-> int
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_stride = foreign "atg_stride" (t @-> int64_t @-> returning int64_t)
-  let stubs_sub = foreign "atg_sub" (ptr t @-> t @-> t @-> returning void)
-  let stubs_sub_ = foreign "atg_sub_" (ptr t @-> t @-> t @-> returning void)
-  let stubs_sub_out = foreign "atg_sub_out" (ptr t @-> t @-> t @-> t @-> returning void)
-
-  let stubs_sub_scalar =
-    foreign "atg_sub_scalar" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_sub_scalar_ =
-    foreign "atg_sub_scalar_" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_sub_scalar_out =
-    foreign "atg_sub_scalar_out" (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_subtract = foreign "atg_subtract" (ptr t @-> t @-> t @-> returning void)
-  let stubs_subtract_ = foreign "atg_subtract_" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_subtract_out =
-    foreign "atg_subtract_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_subtract_scalar =
-    foreign "atg_subtract_scalar" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_subtract_scalar_ =
-    foreign "atg_subtract_scalar_" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_sum = foreign "atg_sum" (ptr t @-> t @-> int @-> returning void)
-
-  let stubs_sum_dim_intlist =
-    foreign
-      "atg_sum_dim_intlist"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_sum_intlist_out =
-    foreign
-      "atg_sum_intlist_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_sum_out = foreign "atg_sum_out" (ptr t @-> t @-> t @-> int @-> returning void)
-
-  let stubs_sum_to_size =
-    foreign "atg_sum_to_size" (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_svd = foreign "atg_svd" (ptr t @-> t @-> int @-> int @-> returning void)
-
-  let stubs_svd_u =
-    foreign "atg_svd_u" (ptr t @-> t @-> t @-> t @-> t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_swapaxes =
-    foreign "atg_swapaxes" (ptr t @-> t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_swapaxes_ =
-    foreign "atg_swapaxes_" (ptr t @-> t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_swapdims =
-    foreign "atg_swapdims" (ptr t @-> t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_swapdims_ =
-    foreign "atg_swapdims_" (ptr t @-> t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_tr = foreign "atg_t" (ptr t @-> t @-> returning void)
-  let stubs_t_ = foreign "atg_t_" (ptr t @-> t @-> returning void)
-  let stubs_t_copy = foreign "atg_t_copy" (ptr t @-> t @-> returning void)
-  let stubs_t_copy_out = foreign "atg_t_copy_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_take = foreign "atg_take" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_take_along_dim =
-    foreign "atg_take_along_dim" (ptr t @-> t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_take_along_dim_out =
-    foreign
-      "atg_take_along_dim_out"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_take_out = foreign "atg_take_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  let stubs_tan = foreign "atg_tan" (ptr t @-> t @-> returning void)
-  let stubs_tan_ = foreign "atg_tan_" (ptr t @-> t @-> returning void)
-  let stubs_tan_out = foreign "atg_tan_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_tanh = foreign "atg_tanh" (ptr t @-> t @-> returning void)
-  let stubs_tanh_ = foreign "atg_tanh_" (ptr t @-> t @-> returning void)
-
-  let stubs_tanh_backward =
-    foreign "atg_tanh_backward" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_tanh_backward_grad_input =
-    foreign "atg_tanh_backward_grad_input" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_tanh_out = foreign "atg_tanh_out" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_tensor_split =
-    foreign "atg_tensor_split" (t @-> int64_t @-> int64_t @-> returning (ptr t))
-  ;;
-
-  let stubs_tensor_split_indices =
-    foreign
-      "atg_tensor_split_indices"
-      (t @-> ptr int64_t @-> int @-> int64_t @-> returning (ptr t))
-  ;;
-
-  let stubs_tensor_split_tensor_indices_or_sections =
-    foreign
-      "atg_tensor_split_tensor_indices_or_sections"
-      (t @-> t @-> int64_t @-> returning (ptr t))
-  ;;
-
-  let stubs_tensordot =
-    foreign
-      "atg_tensordot"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_tensordot_out =
-    foreign
-      "atg_tensordot_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_threshold =
-    foreign "atg_threshold" (ptr t @-> t @-> scalar @-> scalar @-> returning void)
-  ;;
-
-  let stubs_threshold_ =
-    foreign "atg_threshold_" (ptr t @-> t @-> scalar @-> scalar @-> returning void)
-  ;;
-
-  let stubs_threshold_backward =
-    foreign "atg_threshold_backward" (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_threshold_backward_grad_input =
-    foreign
-      "atg_threshold_backward_grad_input"
-      (ptr t @-> t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_threshold_out =
-    foreign
-      "atg_threshold_out"
-      (ptr t @-> t @-> t @-> scalar @-> scalar @-> returning void)
-  ;;
-
-  let stubs_tile =
-    foreign "atg_tile" (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_to_ = foreign "atg_to" (ptr t @-> t @-> int @-> returning void)
-
-  let stubs_to_dense =
-    foreign "atg_to_dense" (ptr t @-> t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_to_dense_backward =
-    foreign "atg_to_dense_backward" (ptr t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_to_device =
-    foreign
-      "atg_to_device"
-      (ptr t @-> t @-> int @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_to_dtype =
-    foreign "atg_to_dtype" (ptr t @-> t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_to_dtype_layout =
-    foreign
-      "atg_to_dtype_layout"
-      (ptr t @-> t @-> int @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_to_mkldnn = foreign "atg_to_mkldnn" (ptr t @-> t @-> int @-> returning void)
-
-  let stubs_to_mkldnn_backward =
-    foreign "atg_to_mkldnn_backward" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_to_mkldnn_out =
-    foreign "atg_to_mkldnn_out" (ptr t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_to_other =
-    foreign "atg_to_other" (ptr t @-> t @-> t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_to_padded_tensor =
-    foreign
-      "atg_to_padded_tensor"
-      (ptr t @-> t @-> double @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_to_padded_tensor_out =
-    foreign
-      "atg_to_padded_tensor_out"
-      (ptr t @-> t @-> t @-> double @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_topk =
-    foreign
-      "atg_topk"
-      (ptr t @-> t @-> int64_t @-> int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_topk_values =
-    foreign
-      "atg_topk_values"
-      (ptr t @-> t @-> t @-> t @-> int64_t @-> int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_totype = foreign "atg_totype" (ptr t @-> t @-> int @-> returning void)
-  let stubs_trace = foreign "atg_trace" (ptr t @-> t @-> returning void)
-end
-
-module C24 (F : Cstubs.FOREIGN) = struct
-  open F
-
-  type t = unit ptr
-
-  let t : t typ = ptr void
-
-  type scalar = unit ptr
-
-  let scalar : scalar typ = ptr void
-
-  let stubs_trace_backward =
-    foreign "atg_trace_backward" (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_trace_out = foreign "atg_trace_out" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_transpose =
-    foreign "atg_transpose" (ptr t @-> t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_transpose_ =
-    foreign "atg_transpose_" (ptr t @-> t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_transpose_copy =
-    foreign "atg_transpose_copy" (ptr t @-> t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_transpose_copy_int_out =
-    foreign
-      "atg_transpose_copy_int_out"
-      (ptr t @-> t @-> t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_trapezoid =
-    foreign "atg_trapezoid" (ptr t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_trapezoid_x =
-    foreign "atg_trapezoid_x" (ptr t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_trapz = foreign "atg_trapz" (ptr t @-> t @-> t @-> int64_t @-> returning void)
-
-  let stubs_trapz_dx =
-    foreign "atg_trapz_dx" (ptr t @-> t @-> double @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_triangular_solve =
-    foreign
-      "atg_triangular_solve"
-      (ptr t @-> t @-> t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_triangular_solve_x =
-    foreign
-      "atg_triangular_solve_x"
-      (ptr t @-> t @-> t @-> t @-> t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_tril = foreign "atg_tril" (ptr t @-> t @-> int64_t @-> returning void)
-  let stubs_tril_ = foreign "atg_tril_" (ptr t @-> t @-> int64_t @-> returning void)
-
-  let stubs_tril_indices =
-    foreign
-      "atg_tril_indices"
-      (ptr t @-> int64_t @-> int64_t @-> int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_tril_indices_out =
-    foreign
-      "atg_tril_indices_out"
-      (ptr t @-> t @-> int64_t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_tril_out =
-    foreign "atg_tril_out" (ptr t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_triplet_margin_loss =
-    foreign
-      "atg_triplet_margin_loss"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> double
-       @-> double
-       @-> double
-       @-> int
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs_triu = foreign "atg_triu" (ptr t @-> t @-> int64_t @-> returning void)
-  let stubs_triu_ = foreign "atg_triu_" (ptr t @-> t @-> int64_t @-> returning void)
-
-  let stubs_triu_indices =
-    foreign
-      "atg_triu_indices"
-      (ptr t @-> int64_t @-> int64_t @-> int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_triu_indices_out =
-    foreign
-      "atg_triu_indices_out"
-      (ptr t @-> t @-> int64_t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_triu_out =
-    foreign "atg_triu_out" (ptr t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_true_divide = foreign "atg_true_divide" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_true_divide_ =
-    foreign "atg_true_divide_" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_true_divide_out =
-    foreign "atg_true_divide_out" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_true_divide_scalar =
-    foreign "atg_true_divide_scalar" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_true_divide_scalar_ =
-    foreign "atg_true_divide_scalar_" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_trunc = foreign "atg_trunc" (ptr t @-> t @-> returning void)
-  let stubs_trunc_ = foreign "atg_trunc_" (ptr t @-> t @-> returning void)
-  let stubs_trunc_out = foreign "atg_trunc_out" (ptr t @-> t @-> t @-> returning void)
-  let stubs_type_as = foreign "atg_type_as" (ptr t @-> t @-> t @-> returning void)
-  let stubs_unbind = foreign "atg_unbind" (t @-> int64_t @-> returning (ptr t))
-  let stubs_unbind_copy = foreign "atg_unbind_copy" (t @-> int64_t @-> returning (ptr t))
-
-  let stubs_unbind_copy_int_out =
-    foreign "atg_unbind_copy_int_out" (ptr t @-> int @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_unflatten =
-    foreign
-      "atg_unflatten"
-      (ptr t @-> t @-> int64_t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_unflatten_dense_tensors =
-    foreign "atg_unflatten_dense_tensors" (t @-> ptr t @-> int @-> returning (ptr t))
-  ;;
-
-  let stubs_unfold =
-    foreign
-      "atg_unfold"
-      (ptr t @-> t @-> int64_t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_unfold_backward =
-    foreign
-      "atg_unfold_backward"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> int64_t
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs_unfold_backward_out =
-    foreign
-      "atg_unfold_backward_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> int64_t
-       @-> int64_t
-       @-> int64_t
-       @-> returning void)
-  ;;
-
-  let stubs_unfold_copy =
-    foreign
-      "atg_unfold_copy"
-      (ptr t @-> t @-> int64_t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_unfold_copy_out =
-    foreign
-      "atg_unfold_copy_out"
-      (ptr t @-> t @-> t @-> int64_t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_uniform =
-    foreign "atg_uniform" (ptr t @-> t @-> double @-> double @-> returning void)
-  ;;
-
-  let stubs_uniform_ =
-    foreign "atg_uniform_" (ptr t @-> t @-> double @-> double @-> returning void)
-  ;;
-
-  let stubs_uniform_out =
-    foreign "atg_uniform_out" (ptr t @-> t @-> t @-> double @-> double @-> returning void)
-  ;;
-
-  let stubs_unique_consecutive =
-    foreign
-      "atg_unique_consecutive"
-      (ptr t @-> t @-> int @-> int @-> int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_unique_consecutive_out =
-    foreign
-      "atg_unique_consecutive_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int
-       @-> int
-       @-> int64_t
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_unique_dim =
-    foreign
-      "atg_unique_dim"
-      (ptr t @-> t @-> int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_unique_dim_consecutive =
-    foreign
-      "atg_unique_dim_consecutive"
-      (ptr t @-> t @-> int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_unique_dim_consecutive_out =
-    foreign
-      "atg_unique_dim_consecutive_out"
-      (ptr t @-> t @-> t @-> t @-> t @-> int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_unique_dim_out =
-    foreign
-      "atg_unique_dim_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> t
-       @-> int64_t
-       @-> int
-       @-> int
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_unsafe_chunk =
-    foreign "atg_unsafe_chunk" (t @-> int64_t @-> int64_t @-> returning (ptr t))
-  ;;
-
-  let stubs_unsafe_split =
-    foreign "atg_unsafe_split" (t @-> int64_t @-> int64_t @-> returning (ptr t))
-  ;;
-
-  let stubs_unsafe_split_tensor_out =
-    foreign
-      "atg_unsafe_split_tensor_out"
-      (ptr t @-> int @-> t @-> int64_t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_unsafe_split_with_sizes =
-    foreign
-      "atg_unsafe_split_with_sizes"
-      (t @-> ptr int64_t @-> int @-> int64_t @-> returning (ptr t))
-  ;;
-
-  let stubs_unsafe_split_with_sizes_out =
-    foreign
-      "atg_unsafe_split_with_sizes_out"
-      (ptr t @-> int @-> t @-> ptr int64_t @-> int @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_unsqueeze =
-    foreign "atg_unsqueeze" (ptr t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_unsqueeze_ =
-    foreign "atg_unsqueeze_" (ptr t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_unsqueeze_copy =
-    foreign "atg_unsqueeze_copy" (ptr t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_unsqueeze_copy_out =
-    foreign "atg_unsqueeze_copy_out" (ptr t @-> t @-> t @-> int64_t @-> returning void)
-  ;;
-
-  let stubs_upsample_bicubic2d =
-    foreign
-      "atg_upsample_bicubic2d"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_upsample_bicubic2d_backward =
-    foreign
-      "atg_upsample_bicubic2d_backward"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_upsample_bicubic2d_backward_grad_input =
-    foreign
-      "atg_upsample_bicubic2d_backward_grad_input"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_upsample_bicubic2d_out =
-    foreign
-      "atg_upsample_bicubic2d_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_upsample_bicubic2d_vec =
-    foreign
-      "atg_upsample_bicubic2d_vec"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> ptr double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_upsample_bilinear2d =
-    foreign
-      "atg_upsample_bilinear2d"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_upsample_bilinear2d_backward =
-    foreign
-      "atg_upsample_bilinear2d_backward"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_upsample_bilinear2d_backward_grad_input =
-    foreign
-      "atg_upsample_bilinear2d_backward_grad_input"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_upsample_bilinear2d_out =
-    foreign
-      "atg_upsample_bilinear2d_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_upsample_bilinear2d_vec =
-    foreign
-      "atg_upsample_bilinear2d_vec"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> ptr double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_upsample_linear1d =
-    foreign
-      "atg_upsample_linear1d"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> int @-> double @-> int @-> returning void)
-  ;;
-
-  let stubs_upsample_linear1d_backward =
-    foreign
-      "atg_upsample_linear1d_backward"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_upsample_linear1d_backward_grad_input =
-    foreign
-      "atg_upsample_linear1d_backward_grad_input"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_upsample_linear1d_out =
-    foreign
-      "atg_upsample_linear1d_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_upsample_linear1d_vec =
-    foreign
-      "atg_upsample_linear1d_vec"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> ptr double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_upsample_nearest1d =
-    foreign
-      "atg_upsample_nearest1d"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> double @-> int @-> returning void)
-  ;;
-
-  let stubs_upsample_nearest1d_backward =
-    foreign
-      "atg_upsample_nearest1d_backward"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_upsample_nearest1d_backward_grad_input =
-    foreign
-      "atg_upsample_nearest1d_backward_grad_input"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_upsample_nearest1d_out =
-    foreign
-      "atg_upsample_nearest1d_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> double @-> int @-> returning void)
-  ;;
-
-  let stubs_upsample_nearest1d_vec =
-    foreign
-      "atg_upsample_nearest1d_vec"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> ptr double @-> int @-> returning void)
-  ;;
-
-  let stubs_upsample_nearest2d =
-    foreign
-      "atg_upsample_nearest2d"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_upsample_nearest2d_backward =
-    foreign
-      "atg_upsample_nearest2d_backward"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_upsample_nearest2d_backward_grad_input =
-    foreign
-      "atg_upsample_nearest2d_backward_grad_input"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_upsample_nearest2d_out =
-    foreign
-      "atg_upsample_nearest2d_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_upsample_nearest2d_vec =
-    foreign
-      "atg_upsample_nearest2d_vec"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> ptr double @-> int @-> returning void)
-  ;;
-
-  let stubs_upsample_nearest3d =
-    foreign
-      "atg_upsample_nearest3d"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_upsample_nearest3d_backward =
-    foreign
-      "atg_upsample_nearest3d_backward"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_upsample_nearest3d_backward_grad_input =
-    foreign
-      "atg_upsample_nearest3d_backward_grad_input"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_upsample_nearest3d_out =
-    foreign
-      "atg_upsample_nearest3d_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_upsample_nearest3d_vec =
-    foreign
-      "atg_upsample_nearest3d_vec"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> ptr double @-> int @-> returning void)
-  ;;
-
-  let stubs_upsample_trilinear3d =
-    foreign
-      "atg_upsample_trilinear3d"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_upsample_trilinear3d_backward =
-    foreign
-      "atg_upsample_trilinear3d_backward"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_upsample_trilinear3d_backward_grad_input =
-    foreign
-      "atg_upsample_trilinear3d_backward_grad_input"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_upsample_trilinear3d_out =
-    foreign
-      "atg_upsample_trilinear3d_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_upsample_trilinear3d_vec =
-    foreign
-      "atg_upsample_trilinear3d_vec"
-      (ptr t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> int
-       @-> ptr double
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_value_selecting_reduction_backward =
-    foreign
-      "atg_value_selecting_reduction_backward"
-      (ptr t @-> t @-> int64_t @-> t @-> ptr int64_t @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_values = foreign "atg_values" (ptr t @-> t @-> returning void)
-  let stubs_values_copy = foreign "atg_values_copy" (ptr t @-> t @-> returning void)
-
-  let stubs_values_copy_out =
-    foreign "atg_values_copy_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_vander =
-    foreign "atg_vander" (ptr t @-> t @-> int64_t @-> int @-> int @-> returning void)
-  ;;
-end
-
-module C25 (F : Cstubs.FOREIGN) = struct
-  open F
-
-  type t = unit ptr
-
-  let t : t typ = ptr void
-
-  type scalar = unit ptr
-
-  let scalar : scalar typ = ptr void
-  let stubs_var = foreign "atg_var" (ptr t @-> t @-> int @-> returning void)
-
-  let stubs_var_correction =
-    foreign
-      "atg_var_correction"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> scalar @-> int @-> returning void)
-  ;;
-
-  let stubs_var_correction_out =
-    foreign
-      "atg_var_correction_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> scalar @-> int @-> returning void)
-  ;;
-
-  let stubs_var_dim =
-    foreign
-      "atg_var_dim"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_var_mean = foreign "atg_var_mean" (ptr t @-> t @-> int @-> returning void)
-
-  let stubs_var_mean_correction =
-    foreign
-      "atg_var_mean_correction"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> scalar @-> int @-> returning void)
-  ;;
-
-  let stubs_var_mean_correction_out =
-    foreign
-      "atg_var_mean_correction_out"
-      (ptr t
-       @-> t
-       @-> t
-       @-> t
-       @-> ptr int64_t
-       @-> int
-       @-> scalar
-       @-> int
-       @-> returning void)
-  ;;
-
-  let stubs_var_mean_dim =
-    foreign
-      "atg_var_mean_dim"
-      (ptr t @-> t @-> ptr int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_var_out =
-    foreign
-      "atg_var_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_vdot = foreign "atg_vdot" (ptr t @-> t @-> t @-> returning void)
-  let stubs_vdot_out = foreign "atg_vdot_out" (ptr t @-> t @-> t @-> t @-> returning void)
-
-  let stubs_view =
-    foreign "atg_view" (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_view_as = foreign "atg_view_as" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_view_as_complex =
-    foreign "atg_view_as_complex" (ptr t @-> t @-> returning void)
-  ;;
-
-  let stubs_view_as_complex_copy =
-    foreign "atg_view_as_complex_copy" (ptr t @-> t @-> returning void)
-  ;;
-
-  let stubs_view_as_complex_copy_out =
-    foreign "atg_view_as_complex_copy_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_view_as_real = foreign "atg_view_as_real" (ptr t @-> t @-> returning void)
-
-  let stubs_view_as_real_copy =
-    foreign "atg_view_as_real_copy" (ptr t @-> t @-> returning void)
-  ;;
-
-  let stubs_view_as_real_copy_out =
-    foreign "atg_view_as_real_copy_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_view_copy =
-    foreign "atg_view_copy" (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_view_copy_dtype =
-    foreign "atg_view_copy_dtype" (ptr t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_view_copy_dtype_out =
-    foreign "atg_view_copy_dtype_out" (ptr t @-> t @-> t @-> int @-> returning void)
-  ;;
-
-  let stubs_view_copy_out =
-    foreign
-      "atg_view_copy_out"
-      (ptr t @-> t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-
-  let stubs_view_dtype = foreign "atg_view_dtype" (ptr t @-> t @-> int @-> returning void)
-  let stubs_vsplit = foreign "atg_vsplit" (t @-> int64_t @-> returning (ptr t))
-
-  let stubs_vsplit_array =
-    foreign "atg_vsplit_array" (t @-> ptr int64_t @-> int @-> returning (ptr t))
-  ;;
-
-  let stubs_vstack = foreign "atg_vstack" (ptr t @-> ptr t @-> int @-> returning void)
-
-  let stubs_vstack_out =
-    foreign "atg_vstack_out" (ptr t @-> t @-> ptr t @-> int @-> returning void)
-  ;;
-
-  let stubs_where = foreign "atg_where" (t @-> returning (ptr t))
-
-  let stubs_where_scalar =
-    foreign "atg_where_scalar" (ptr t @-> t @-> scalar @-> scalar @-> returning void)
-  ;;
-
-  let stubs_where_scalarother =
-    foreign "atg_where_scalarother" (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_where_scalarself =
-    foreign "atg_where_scalarself" (ptr t @-> t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_where_self =
-    foreign "atg_where_self" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_where_self_out =
-    foreign "atg_where_self_out" (ptr t @-> t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_xlogy = foreign "atg_xlogy" (ptr t @-> t @-> t @-> returning void)
-  let stubs_xlogy_ = foreign "atg_xlogy_" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_xlogy_outscalar_other =
-    foreign "atg_xlogy_outscalar_other" (ptr t @-> t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_xlogy_outscalar_self =
-    foreign "atg_xlogy_outscalar_self" (ptr t @-> t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_xlogy_outtensor =
-    foreign "atg_xlogy_outtensor" (ptr t @-> t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_xlogy_scalar_other =
-    foreign "atg_xlogy_scalar_other" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_xlogy_scalar_other_ =
-    foreign "atg_xlogy_scalar_other_" (ptr t @-> t @-> scalar @-> returning void)
-  ;;
-
-  let stubs_xlogy_scalar_self =
-    foreign "atg_xlogy_scalar_self" (ptr t @-> scalar @-> t @-> returning void)
-  ;;
-
-  let stubs_zero = foreign "atg_zero" (ptr t @-> t @-> returning void)
-  let stubs_zero_ = foreign "atg_zero_" (ptr t @-> t @-> returning void)
-  let stubs_zero_out = foreign "atg_zero_out" (ptr t @-> t @-> t @-> returning void)
-
-  let stubs_zeros =
-    foreign "atg_zeros" (ptr t @-> ptr int64_t @-> int @-> int @-> int @-> returning void)
-  ;;
-
-  let stubs_zeros_like = foreign "atg_zeros_like" (ptr t @-> t @-> returning void)
-
-  let stubs_zeros_like_out =
-    foreign "atg_zeros_like_out" (ptr t @-> t @-> t @-> returning void)
-  ;;
-
-  let stubs_zeros_out =
-    foreign "atg_zeros_out" (ptr t @-> t @-> ptr int64_t @-> int @-> returning void)
-  ;;
-end
-
-module C (F : Cstubs.FOREIGN) = struct
-  include C0 (F)
-  include C1 (F)
-  include C2 (F)
-  include C3 (F)
-  include C4 (F)
-  include C5 (F)
-  include C6 (F)
-  include C7 (F)
-  include C8 (F)
-  include C9 (F)
-  include C10 (F)
-  include C11 (F)
-  include C12 (F)
-  include C13 (F)
-  include C14 (F)
-  include C15 (F)
-  include C16 (F)
-  include C17 (F)
-  include C18 (F)
-  include C19 (F)
-  include C20 (F)
-  include C21 (F)
-  include C22 (F)
-  include C23 (F)
-  include C24 (F)
-  include C25 (F)
-end
diff --git a/src/tests/jit_tests.ml b/src/tests/jit_tests.ml
index d410c96..64ce78b 100644
--- a/src/tests/jit_tests.ml
+++ b/src/tests/jit_tests.ml
@@ -37,6 +37,50 @@ let%expect_test _ =
       |}]
 ;;
 
+(*
+   test module generated with these Python code:
+
+   import torch
+   @torch.jit.script
+   def would_raise(x):
+   raise RuntimeError("Raising expcetion on purpose")
+   return x
+   torch.jit.save(would_raise, '/tmp/raise.pt')
+*)
+let%expect_test "test exception raise in torch script can be properly caught in OCaml" =
+  let model = Module.load "raise.pt" in
+  try
+    let output = Module.forward model [ Tensor.of_float0 0. ] in
+    ignore output
+  with
+  | Failure failure ->
+    Stdio.print_s [%message "Exception raised and caught" (failure : string)];
+    ();
+    [%expect
+      {|
+        ("Exception raised and caught"
+         (failure
+           "The following operation failed in the TorchScript interpreter.\
+          \nTraceback of TorchScript, serialized code (most recent call last):\
+          \n  File \"code/__torch__.py\", line 8, in forward\
+          \n    x: Tensor) -> NoneType:\
+          \n    _0 = uninitialized(NoneType)\
+          \n    ops.prim.RaiseException(\"Raising expcetion on purpose\", \"builtins.RuntimeError\")\
+          \n    ~~~~~~~~~~~~~~~~~~~~~~~ <--- HERE\
+          \n    return _0\
+          \n\
+          \nTraceback of TorchScript, original code (most recent call last):\
+          \n  File \"/tmp/ipykernel_741402/1182469162.py\", line 5, in forward\
+          \n@torch.jit.script\
+          \ndef would_raise(x):\
+          \n    raise RuntimeError(\"Raising expcetion on purpose\")\
+          \n    ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ <--- HERE\
+          \n    return x\
+          \nbuiltins.RuntimeError: Raising expcetion on purpose\
+          \n"))
+      |}]
+;;
+
 let%expect_test _ =
   (* test that we can list all the buffers in a module, modify them, and get different results *)
   (* This model just adds all the buffers and parameters together. *)
diff --git a/src/tests/raise.pt b/src/tests/raise.pt
new file mode 100644
index 0000000..9bba39d
Binary files /dev/null and b/src/tests/raise.pt differ
diff --git a/src/wrapper/dune b/src/wrapper/dune
index 9782060..7e9eea2 100644
--- a/src/wrapper/dune
+++ b/src/wrapper/dune
@@ -8,14 +8,14 @@
    (:include cxx_flags.sexp)))
  (foreign_stubs
   (language c)
-  (names torch_stubs))
+  (names torch_stubs torch_stubs_generated))
  (name torch_core)
  (public_name torch.core)
  (c_library_flags
   :standard
   -lstdc++
   (:include c_library_flags.sexp))
- (libraries ctypes.foreign ctypes.stubs ctypes)
+ (libraries ctypes.foreign torch_bindings)
  (preprocess
   (pps ppx_jane)))
 
@@ -26,19 +26,7 @@
   (bash %{deps})))
 
 (rule
- (targets torch_bindings.ml)
- (deps ../stubs/torch_bindings.ml)
- (action
-  (bash "cp ../stubs/torch_bindings.ml torch_bindings.ml")))
-
-(rule
- (targets torch_bindings_generated.ml)
- (deps ../stubs/torch_bindings_generated.ml)
- (action
-  (bash "cp ../stubs/torch_bindings_generated.ml torch_bindings_generated.ml")))
-
-(rule
- (targets torch_stubs.c torch_generated.ml)
- (deps ../stubs/torch_gen.exe)
+ (targets torch_stubs_generated.c torch_stubs_generated.ml)
+ (deps ../gen_stubs/gen_stubs.exe)
  (action
   (bash ./%{deps})))
diff --git a/src/wrapper/torch_api.cpp b/src/wrapper/torch_api.cpp
index 1bffba4..cd8a585 100644
--- a/src/wrapper/torch_api.cpp
+++ b/src/wrapper/torch_api.cpp
@@ -5,27 +5,67 @@
 #include <stdexcept>
 #include <vector>
 #include <caml/fail.h>
-#include <caml/threads.h>
+#include <caml/memory.h>
 #undef invalid_argument
 #include "torch_api.h"
+#include "ctypes_cstubs_internals.h"
 
 using namespace std;
 
+torch::Tensor *tensor_ptr_from_ocaml(gc_tensor t) { return (torch::Tensor *)t; }
+
+torch::Tensor tensor_from_ocaml(gc_tensor t) {
+  return *tensor_ptr_from_ocaml(t);
+}
+
+CAMLprim void finalize_tensor(value block) {
+  gc_tensor t = *(void **)Data_custom_val(block);
+  at_free(t);
+}
+
+static struct custom_operations ops = {"torch-tensor",
+                                       finalize_tensor,
+                                       custom_compare_default,
+                                       custom_hash_default,
+                                       custom_serialize_default,
+                                       custom_deserialize_default,
+                                       custom_compare_ext_default,
+                                       custom_fixed_length_default};
+
+CAMLprim value with_tensor_gc_internal(raw_tensor t) {
+  // See https://v2.ocaml.org/manual/intfc.html#s%3Ac-custom
+  unsigned long int off_heap_cpu_memory_bytes = 0;
+  torch::Tensor *tensor = (torch::Tensor *)t;
+  if (tensor->defined() && tensor->device() == at::kCPU) {
+    off_heap_cpu_memory_bytes = tensor->numel() * tensor->element_size();
+  }
+  value new_block = caml_alloc_custom_mem(&ops, sizeof(torch::Tensor *),
+                                          off_heap_cpu_memory_bytes);
+  *(void **)Data_custom_val(new_block) = t;
+  return new_block;
+}
+
+raw_tensor tensor_to_ocaml(torch::Tensor cpp_tensor) {
+  torch::Tensor *res = new torch::Tensor(cpp_tensor);
+  return (raw_tensor)res;
+}
+
 void at_manual_seed(int64_t seed) { torch::manual_seed(seed); }
 
-vector<torch::Tensor> of_carray_tensor(torch::Tensor **vs, int len) {
+vector<torch::Tensor> of_carray_tensor(gc_tensor *vs, int len) {
   vector<torch::Tensor> result;
   for (int i = 0; i < len; ++i)
-    result.push_back(*(vs[i]));
+    result.push_back(tensor_from_ocaml(vs[i]));
   return result;
 }
 
-c10::List<c10::optional<torch::Tensor>> of_carray_tensor_opt(torch::Tensor **vs,
+c10::List<c10::optional<torch::Tensor>> of_carray_tensor_opt(gc_tensor *vs,
                                                              int len) {
   vector<c10::optional<torch::Tensor>> result;
   for (int i = 0; i < len; ++i) {
-    result.push_back(vs[i] != nullptr ? c10::optional<torch::Tensor>(*(vs[i]))
-                                      : c10::nullopt);
+    result.push_back(
+        vs[i] ? c10::optional<torch::Tensor>(tensor_from_ocaml(vs[i]))
+              : c10::nullopt);
   }
   return c10::List<c10::optional<torch::Tensor>>(result);
 }
@@ -43,13 +83,13 @@ c10::optional<at::Device> optional_device_of_int(int d) {
 
 at::Device device_of_int(int d) { return optional_device_of_int(d).value(); }
 
-tensor at_new_tensor() {
-  PROTECT(return new torch::Tensor();)
+raw_tensor at_new_tensor() {
+  PROTECT(return tensor_to_ocaml(torch::Tensor());)
   return nullptr;
 }
 
-tensor at_tensor_of_data(void *vs, int64_t *dims, int ndims,
-                         int element_size_in_bytes, int type) {
+raw_tensor at_tensor_of_data(void *vs, int64_t *dims, int ndims,
+                             int element_size_in_bytes, int type) {
   PROTECT(
       torch::Tensor tensor = torch::zeros(torch::IntArrayRef(dims, ndims),
                                           torch::ScalarType(type));
@@ -57,13 +97,13 @@ tensor at_tensor_of_data(void *vs, int64_t *dims, int ndims,
           invalid_argument("incoherent element sizes in bytes");
       void *tensor_data = tensor.data_ptr();
       memcpy(tensor_data, vs, tensor.numel() * element_size_in_bytes);
-      return new torch::Tensor(tensor);)
+      return tensor_to_ocaml(tensor);)
   return nullptr;
 }
 
-void at_copy_data(tensor tensor, void *vs, int64_t numel,
-                  int elt_size_in_bytes) {
+void at_copy_data(gc_tensor t, void *vs, int64_t numel, int elt_size_in_bytes) {
   PROTECT(
+      torch::Tensor *tensor = tensor_ptr_from_ocaml(t);
       if ((int64_t)elt_size_in_bytes != tensor->element_size()) throw std::
           invalid_argument("incoherent element sizes in bytes");
       if ((int64_t)numel > tensor->numel()) throw std::invalid_argument(
@@ -79,45 +119,48 @@ void at_copy_data(tensor tensor, void *vs, int64_t numel,
       })
 }
 
-tensor at_float_vec(double *vs, int len, int type) {
+raw_tensor at_float_vec(double *vs, int len, int type) {
   PROTECT(torch::Tensor tensor = torch::empty({len}, torch::ScalarType(type));
           for (int i = 0; i < len; ++i) tensor[i] = vs[i];
-          return new torch::Tensor(tensor);)
+          return tensor_to_ocaml(tensor);)
   return nullptr;
 }
 
-tensor at_int_vec(int64_t *vs, int len, int type) {
+raw_tensor at_int_vec(int64_t *vs, int len, int type) {
   PROTECT(torch::Tensor tensor = torch::empty({len}, torch::ScalarType(type));
           for (int i = 0; i < len; ++i) tensor[i] = vs[i];
-          return new torch::Tensor(tensor);)
+          return tensor_to_ocaml(tensor);)
   return nullptr;
 }
 
-int at_defined(tensor t) {
-  PROTECT(return t->defined();)
+int at_defined(gc_tensor t) {
+  PROTECT(return tensor_ptr_from_ocaml(t)->defined();)
   return -1;
 }
 
-int at_is_sparse(tensor t) {
-  PROTECT(return t->is_sparse();)
+int at_is_sparse(gc_tensor t) {
+  PROTECT(return tensor_ptr_from_ocaml(t)->is_sparse();)
   return -1;
 }
 
-int at_dim(tensor t) {
-  PROTECT(return t->dim();)
+int at_dim(gc_tensor t) {
+  PROTECT(return tensor_ptr_from_ocaml(t)->dim();)
   return -1;
 }
 
-void at_shape(tensor t, int *dims) {
-  PROTECT(int i = 0; for (int dim : t->sizes()) dims[i++] = dim;)
+void at_shape(gc_tensor t, int *dims) {
+  PROTECT(int i = 0; for (int dim
+                          : tensor_ptr_from_ocaml(t)->sizes()) dims[i++] = dim;)
 }
 
-void at_stride(tensor t, int64_t *dims) {
-  PROTECT(int i = 0; for (int64_t dim : t->strides()) dims[i++] = dim;)
+void at_stride(gc_tensor t, int64_t *dims) {
+  PROTECT(int i = 0;
+          for (int64_t dim
+               : tensor_ptr_from_ocaml(t)->strides()) dims[i++] = dim;)
 }
 
-int at_scalar_type(tensor t) {
-  PROTECT(return static_cast<int>(t->scalar_type());)
+int at_scalar_type(gc_tensor t) {
+  PROTECT(return static_cast<int>(tensor_ptr_from_ocaml(t)->scalar_type());)
 }
 
 void at_autocast_clear_cache() { at::autocast::clear_cache(); }
@@ -143,17 +186,23 @@ int at_autocast_set_enabled(int b) {
   return -1;
 }
 
-int at_device(tensor tensor) {
-  PROTECT(auto device = tensor->device(); if (device.is_cpu()) return -1;
-          return device.index();)
+int at_device(gc_tensor t) {
+  PROTECT(auto device = tensor_ptr_from_ocaml(t)->device();
+          if (device.is_cpu()) return -1; return device.index();)
 }
 
-void at_backward(tensor t, int keep_graph, int create_graph) {
-  PROTECT(t->backward({}, keep_graph, create_graph);)
+void at_backward(gc_tensor t, int keep_graph, int create_graph) {
+  PROTECT(
+      caml_release_runtime_system(); try {
+        tensor_ptr_from_ocaml(t)->backward({}, keep_graph, create_graph);
+      } catch (const exception &) {
+        caml_acquire_runtime_system();
+        throw;
+      } caml_acquire_runtime_system();)
 }
 
-int at_requires_grad(tensor t) {
-  PROTECT(return t->requires_grad();)
+int at_requires_grad(gc_tensor t) {
+  PROTECT(return tensor_ptr_from_ocaml(t)->requires_grad();)
   return -1;
 }
 
@@ -163,112 +212,125 @@ int at_grad_set_enabled(int b) {
   return -1;
 }
 
-tensor at_get(tensor t, int index) {
-  PROTECT(return new torch::Tensor((*t)[index]);)
+raw_tensor at_get(gc_tensor t, int index) {
+  PROTECT(return tensor_to_ocaml(tensor_from_ocaml(t)[index]);)
   return nullptr;
 }
 
 template <typename T>
-T at_value_at_indexes(tensor t, int *indexes, int indexes_len) {
-  PROTECT(torch::Tensor tensor = *t; for (int i = 0; i < indexes_len; ++i) {
-    tensor = tensor[indexes[i]];
-  } return tensor.item<T>();)
+T at_value_at_indexes(gc_tensor t, int *indexes, int indexes_len) {
+  PROTECT(torch::Tensor tensor = tensor_from_ocaml(t);
+          for (int i = 0; i < indexes_len;
+               ++i) { tensor = tensor[indexes[i]]; } return tensor.item<T>();)
   return T();
 }
 
-double at_double_value_at_indexes(tensor t, int *indexes, int indexes_len) {
+double at_double_value_at_indexes(gc_tensor t, int *indexes, int indexes_len) {
   return at_value_at_indexes<double>(t, indexes, indexes_len);
 }
 
-int64_t at_int64_value_at_indexes(tensor t, int *indexes, int indexes_len) {
+int64_t at_int64_value_at_indexes(gc_tensor t, int *indexes, int indexes_len) {
   return at_value_at_indexes<int64_t>(t, indexes, indexes_len);
 }
 
 template <typename T>
-void at_set_value_at_indexes(tensor t, int *indexes, int indexes_len, T v) {
-  PROTECT(torch::Tensor tensor = *t; for (int i = 0; i < indexes_len; ++i) {
-    tensor = tensor[indexes[i]];
-  } tensor.fill_(v);)
+void at_set_value_at_indexes(gc_tensor t, int *indexes, int indexes_len, T v) {
+  PROTECT(torch::Tensor tensor = tensor_from_ocaml(t);
+          for (int i = 0; i < indexes_len;
+               ++i) { tensor = tensor[indexes[i]]; } tensor.fill_(v);)
 }
 
-void at_set_double_value_at_indexes(tensor t, int *indexes, int indexes_len,
+void at_set_double_value_at_indexes(gc_tensor t, int *indexes, int indexes_len,
                                     double v) {
   at_set_value_at_indexes<double>(t, indexes, indexes_len, v);
 }
 
-void at_set_int64_value_at_indexes(tensor t, int *indexes, int indexes_len,
+void at_set_int64_value_at_indexes(gc_tensor t, int *indexes, int indexes_len,
                                    int64_t v) {
   at_set_value_at_indexes<int64_t>(t, indexes, indexes_len, v);
 }
 
-void at_fill_double(tensor t, double v) { PROTECT(t->fill_(v);) }
+void at_fill_double(gc_tensor t, double v) {
+  PROTECT(tensor_ptr_from_ocaml(t)->fill_(v);)
+}
 
-void at_fill_int64(tensor t, int64_t v) { PROTECT(t->fill_(v);) }
+void at_fill_int64(gc_tensor t, int64_t v) {
+  PROTECT(tensor_ptr_from_ocaml(t)->fill_(v);)
+}
 
-void at_print(tensor t) {
+void at_print(gc_tensor t) {
   PROTECT(torch::Tensor *tensor = (torch::Tensor *)t; cout << *tensor << endl;)
 }
 
-char *at_to_string(tensor t, int line_size) {
-  PROTECT(std::ostringstream oss; torch::print(oss, *t, line_size);
+char *at_to_string(gc_tensor t, int line_size) {
+  PROTECT(std::ostringstream oss;
+          torch::print(oss, tensor_from_ocaml(t), line_size);
           return strdup(oss.str().c_str());)
   return nullptr;
 }
 
-void at_copy_(tensor dst, tensor src) { PROTECT(dst->copy_(*src);) }
-void at_set_data(tensor dst, tensor src) { PROTECT(dst->set_data(*src);) }
+void at_copy_(gc_tensor dst, gc_tensor src) {
+  PROTECT(tensor_ptr_from_ocaml(dst)->copy_(tensor_from_ocaml(src));)
+}
+void at_set_data(gc_tensor dst, gc_tensor src) {
+  PROTECT(tensor_ptr_from_ocaml(dst)->set_data(tensor_from_ocaml(src));)
+}
 
-void at_save(tensor t, char *filename) { PROTECT(torch::save(*t, filename);) }
+void at_save(gc_tensor t, char *filename) {
+  PROTECT(torch::save(tensor_from_ocaml(t), filename);)
+}
 
-void at_save_multi(tensor *tensors, char **tensor_names, int ntensors,
+void at_save_multi(gc_tensor *tensors, char **tensor_names, int ntensors,
                    char *filename) {
   PROTECT(torch::serialize::OutputArchive archive;
-          for (int i = 0; i < ntensors; ++i) archive.write(
-              std::string(tensor_names[i]), *(tensors[i]), /* buffer=*/false);
+          for (int i = 0; i < ntensors; ++i)
+              archive.write(std::string(tensor_names[i]),
+                            tensor_from_ocaml(tensors[i]), /* buffer=*/false);
           archive.save_to(filename);)
 }
 
-void at_load_multi(tensor *tensors, char **tensor_names, int ntensors,
+void at_load_multi(raw_tensor *outputs, char **tensor_names, int ntensors,
                    char *filename) {
   PROTECT(torch::serialize::InputArchive archive;
           archive.load_from(std::string(filename));
           vector<torch::Tensor> ts(ntensors);
           for (int i = 0; i < ntensors; ++i)
               archive.read(std::string(tensor_names[i]), ts[i]);
-          // Only allocate the new tensor now so that if there is an exception
+          // Only allocate the new tensors now so that if there is an exception
           // raised during [read], no memory has to be freed.
-          for (int i = 0; i < ntensors; ++i) tensors[i] =
-              new torch::Tensor(ts[i]);)
+          for (int i = 0; i < ntensors; ++i) outputs[i] =
+              tensor_to_ocaml(ts[i]);)
 }
 
-void at_load_callback(char *filename, void (*f)(char *, tensor)) {
+void at_load_callback(char *filename, void (*f)(char *, raw_tensor)) {
   PROTECT(auto module = torch::jit::load(filename);
           for (const auto &p
                : module.named_parameters()) {
             auto v = p.value;
-            f((char *)p.name.c_str(), new torch::Tensor(v));
+            f((char *)p.name.c_str(), tensor_to_ocaml(v));
           })
 }
 
-void at_load_multi_(tensor *tensors, char **tensor_names, int ntensors,
+void at_load_multi_(gc_tensor *tensors, char **tensor_names, int ntensors,
                     char *filename) {
   PROTECT(torch::NoGradGuard no_grad; torch::serialize::InputArchive archive;
           archive.load_from(std::string(filename));
           for (int i = 0; i < ntensors; ++i) {
-            if (tensors[i]->device().type() == at::kCPU)
-              archive.read(std::string(tensor_names[i]), *(tensors[i]));
+            torch::Tensor *tensor_ptr = tensor_ptr_from_ocaml(tensors[i]);
+            if (tensor_ptr->device().type() == at::kCPU)
+              archive.read(std::string(tensor_names[i]), *tensor_ptr);
             else {
               torch::Tensor tmp_tensor =
-                  torch::empty_like(*(tensors[i]), at::device(at::kCPU));
+                  torch::empty_like(*tensor_ptr, at::device(at::kCPU));
               archive.read(std::string(tensor_names[i]), tmp_tensor);
-              tensors[i]->copy_(tmp_tensor);
+              tensor_ptr->copy_(tmp_tensor);
             }
           })
 }
 
-tensor at_load(char *filename) {
+raw_tensor at_load(char *filename) {
   PROTECT(torch::Tensor tensor; torch::load(tensor, filename);
-          return new torch::Tensor(tensor);)
+          return tensor_to_ocaml(tensor);)
   return nullptr;
 }
 
@@ -290,32 +352,38 @@ void at_set_num_threads(int n_threads) {
   PROTECT(at::set_num_threads(n_threads);)
 }
 
-void at_free(tensor t) { delete (t); }
+void at_free(gc_tensor t) { delete (tensor_ptr_from_ocaml(t)); }
 
-void at_run_backward(tensor *tensors, int ntensors, tensor *inputs, int ninputs,
-                     tensor *outputs, int keep_graph, int create_graph) {
+void at_run_backward(gc_tensor *tensors, int ntensors, gc_tensor *inputs,
+                     int ninputs, raw_tensor *outputs, int keep_graph,
+                     int create_graph) {
   PROTECT(
       vector<torch::autograd::Edge> roots;
-      for (int i = 0; i < ntensors; ++i)
-          roots.push_back(torch::autograd::impl::gradient_edge(*tensors[i]));
+      for (int i = 0; i < ntensors; ++i) roots.push_back(
+          torch::autograd::impl::gradient_edge(tensor_from_ocaml(tensors[i])));
 
       vector<torch::autograd::Edge> inputs_; for (int i = 0; i < ninputs; ++i) {
-        if (!inputs[i]->requires_grad())
+        torch::Tensor *input_ = tensor_ptr_from_ocaml(inputs[i]);
+        if (!input_->requires_grad())
           throw std::invalid_argument(
               "one of the input tensor does not use set_requires_grad");
-        inputs_.push_back(torch::autograd::impl::gradient_edge(*inputs[i]));
+        inputs_.push_back(torch::autograd::impl::gradient_edge(*input_));
       }
 
       vector<torch::autograd::Variable>
           grads;
       for (int i = 0; i < ntensors; ++i)
-          grads.push_back(torch::ones_like(*tensors[i]));
+          grads.push_back(torch::ones_like(tensor_from_ocaml(tensors[i])));
 
-      auto vl = torch::autograd::Engine::get_default_engine().execute(
-          roots, grads, keep_graph, create_graph, false, inputs_);
-      for (int i = 0; i < ninputs; ++i) {
-        outputs[i] = static_cast<tensor>(new torch::autograd::Variable(vl[i]));
-      })
+      caml_release_runtime_system(); torch::autograd::variable_list vl; try {
+        vl = torch::autograd::Engine::get_default_engine().execute(
+            roots, grads, keep_graph, create_graph, false, inputs_);
+      } catch (const exception &) {
+        caml_acquire_runtime_system();
+        throw;
+      } caml_acquire_runtime_system();
+      for (int i = 0; i < ninputs;
+           ++i) { outputs[i] = tensor_to_ocaml(vl[i]); })
 }
 
 optimizer ato_adam(double learning_rate, double beta1, double beta2,
@@ -351,10 +419,10 @@ optimizer ato_sgd(double learning_rate, double momentum, double dampening,
   return nullptr;
 }
 
-void ato_add_parameters(optimizer t, tensor *tensors, int ntensors) {
+void ato_add_parameters(optimizer t, gc_tensor *tensors, int ntensors) {
   PROTECT(for (int i = 0; i < ntensors; ++i) t->param_groups()[0]
               .params()
-              .push_back(*(tensors[i]));)
+              .push_back(tensor_from_ocaml(tensors[i]));)
 }
 
 template <class T> void set_lr(optimizer t, double learning_rate) {
@@ -564,27 +632,46 @@ module atm_load_str(char *data, size_t sz, int device) {
   return nullptr;
 }
 
-tensor atm_forward(module m, tensor *tensors, int ntensors) {
-  PROTECT(std::vector<torch::jit::IValue> inputs;
-          for (int i = 0; i < ntensors; ++i) inputs.push_back(*(tensors[i]));
-          caml_release_runtime_system();
-          torch::jit::IValue output = m->forward(std::move(inputs));
-          caml_acquire_runtime_system();
-          if (!output.isTensor()) throw std::invalid_argument(
-              "forward did not return a tensor");
-          return new torch::Tensor(output.toTensor());)
+raw_tensor atm_forward(module m, gc_tensor *tensors, int ntensors) {
+  PROTECT(
+      std::vector<torch::jit::IValue> inputs;
+      for (int i = 0; i < ntensors; ++i)
+          inputs.push_back(tensor_from_ocaml(tensors[i]));
+      caml_release_runtime_system(); torch::jit::IValue output;
+
+      // In case of exception, we need to re-acquire the runtime
+      // lock before re-raising, since PROTECT re-enters ocaml.
+      try {
+        output = m->forward(std::move(inputs));
+      } catch (const exception &) {
+        caml_acquire_runtime_system();
+        throw;
+      }
+
+      caml_acquire_runtime_system();
+      if (!output.isTensor()) throw std::invalid_argument(
+          "forward did not return a tensor");
+      return tensor_to_ocaml(output.toTensor());)
   return nullptr;
 }
 
 ivalue atm_forward_(module m, ivalue *ivalues, int nivalues) {
-  PROTECT(std::vector<torch::jit::IValue> inputs;
-          for (int i = 0; i < nivalues; ++i) inputs.push_back(*(ivalues[i]));
-          caml_release_runtime_system();
-          torch::jit::IValue output = m->forward(inputs);
-          caml_acquire_runtime_system(); return new torch::jit::IValue(output);)
+  PROTECT(
+      std::vector<torch::jit::IValue> inputs;
+      for (int i = 0; i < nivalues; ++i) inputs.push_back(*(ivalues[i]));
+      caml_release_runtime_system(); torch::jit::IValue output;
+
+      // In case of exception, we need to re-acquire the runtime
+      // lock before re-raising, since PROTECT re-enters ocaml.
+      try { output = m->forward(inputs); } catch (const exception &) {
+        caml_acquire_runtime_system();
+        throw;
+      }
+
+      caml_acquire_runtime_system();
+      return new torch::jit::IValue(output);)
   return nullptr;
 }
-
 // To return this OrderedDict<string, Tensor>, we pass it a tuple
 // IValue containing
 // * list of strings IValue (names)
@@ -613,8 +700,8 @@ void atm_to(module m, int device, int dtype, bool non_blocking) {
   PROTECT(m->to(device_of_int(device), at::ScalarType(dtype), non_blocking);)
 }
 
-ivalue ati_tensor(tensor t) {
-  PROTECT(return new torch::jit::IValue(*t);)
+ivalue ati_tensor(gc_tensor t) {
+  PROTECT(return new torch::jit::IValue(tensor_from_ocaml(t));)
   return nullptr;
 }
 
@@ -694,9 +781,10 @@ ivalue ati_string_list(char **is, int nvalues) {
   return nullptr;
 }
 
-ivalue ati_tensor_list(tensor *is, int nvalues) {
+ivalue ati_tensor_list(gc_tensor *is, int nvalues) {
   PROTECT(c10::List<at::Tensor> vec;
-          for (int i = 0; i < nvalues; ++i) vec.push_back(*(is[i]));
+          for (int i = 0; i < nvalues; ++i)
+              vec.push_back(tensor_from_ocaml(is[i]));
           return new torch::jit::IValue(vec);)
   return nullptr;
 }
@@ -735,8 +823,8 @@ char *ati_to_string(ivalue i) {
   return nullptr;
 }
 
-tensor ati_to_tensor(ivalue i) {
-  PROTECT(return new torch::Tensor(i->toTensor());)
+raw_tensor ati_to_tensor(ivalue i) {
+  PROTECT(return tensor_to_ocaml(i->toTensor());)
   return nullptr;
 }
 
@@ -808,10 +896,10 @@ void ati_to_bool_list(ivalue i, char *outputs, int noutputs) {
   } for (int i = 0; i < noutputs; ++i) outputs[i] = vec[i];)
 }
 
-void ati_to_tensor_list(ivalue i, tensor *outputs, int noutputs) {
+void ati_to_tensor_list(ivalue i, raw_tensor *outputs, int noutputs) {
   PROTECT(auto vec = i->toTensorList(); if (vec.size() != noutputs) {
     throw std::invalid_argument("unexpected list<tensor> size");
-  } for (int i = 0; i < noutputs; ++i) outputs[i] = new torch::Tensor(vec[i]);)
+  } for (int i = 0; i < noutputs; ++i) outputs[i] = tensor_to_ocaml(vec[i]);)
 }
 
 ivalue ati_object_method_(ivalue i, char *method_name, ivalue *ivalues,
@@ -842,4 +930,4 @@ void at_set_graph_executor_optimize(bool o) {
   torch::jit::setGraphExecutorOptimize(o);
 }
 
-#include "torch_api_generated.cpp.h"
+#include "torch_api_generated.cpp"
diff --git a/src/wrapper/torch_api.h b/src/wrapper/torch_api.h
index 68df5a6..027e721 100644
--- a/src/wrapper/torch_api.h
+++ b/src/wrapper/torch_api.h
@@ -2,10 +2,10 @@
 #define __TORCH_API_H__
 #include <stdint.h>
 #include <stddef.h>
+#include <caml/custom.h>
 
 #ifdef __cplusplus
 extern "C" {
-typedef torch::Tensor *tensor;
 typedef torch::Scalar *scalar;
 typedef torch::optim::Optimizer *optimizer;
 typedef torch::jit::script::Module *module;
@@ -17,29 +17,33 @@ typedef torch::jit::IValue *ivalue;
     caml_failwith(strdup(e.what()));                                           \
   }
 #else
-typedef void *tensor;
 typedef void *optimizer;
 typedef void *scalar;
 typedef void *module;
 typedef void *ivalue;
 #endif
 
+typedef void *raw_tensor;
+typedef void *gc_tensor;
+
+value with_tensor_gc_internal(raw_tensor t);
+
 void at_manual_seed(int64_t);
-tensor at_new_tensor();
-tensor at_tensor_of_data(void *vs, int64_t *dims, int ndims,
-                         int element_size_in_bytes, int type);
-void at_copy_data(tensor tensor, void *vs, int64_t numel,
+raw_tensor at_new_tensor();
+raw_tensor at_tensor_of_data(void *vs, int64_t *dims, int ndims,
+                             int element_size_in_bytes, int type);
+void at_copy_data(gc_tensor t, void *vs, int64_t numel,
                   int element_size_in_bytes);
-tensor at_float_vec(double *values, int value_len, int type);
-tensor at_int_vec(int64_t *values, int value_len, int type);
+raw_tensor at_float_vec(double *values, int value_len, int type);
+raw_tensor at_int_vec(int64_t *values, int value_len, int type);
 
-int at_defined(tensor);
-int at_is_sparse(tensor);
-int at_device(tensor);
-int at_dim(tensor);
-void at_shape(tensor, int *);
-void at_stride(tensor, int *);
-int at_scalar_type(tensor);
+int at_defined(gc_tensor);
+int at_is_sparse(gc_tensor);
+int at_device(gc_tensor);
+int at_dim(gc_tensor);
+void at_shape(gc_tensor, int *);
+void at_stride(gc_tensor, int *);
+int at_scalar_type(gc_tensor);
 
 void at_autocast_clear_cache();
 int at_autocast_decrement_nesting();
@@ -47,47 +51,48 @@ int at_autocast_increment_nesting();
 int at_autocast_is_enabled();
 int at_autocast_set_enabled(int b);
 
-void at_backward(tensor, int, int);
-int at_requires_grad(tensor);
+void at_backward(gc_tensor, int, int);
+int at_requires_grad(gc_tensor);
 int at_grad_set_enabled(int);
 
-tensor at_get(tensor, int index);
-void at_fill_double(tensor, double);
-void at_fill_int64(tensor, int64_t);
+raw_tensor at_get(gc_tensor, int index);
+void at_fill_double(gc_tensor, double);
+void at_fill_int64(gc_tensor, int64_t);
 
-double at_double_value_at_indexes(tensor, int *indexes, int indexes_len);
-int64_t at_int64_value_at_indexes(tensor, int *indexes, int indexes_len);
-void at_set_double_value_at_indexes(tensor, int *indexes, int indexes_len,
+double at_double_value_at_indexes(gc_tensor, int *indexes, int indexes_len);
+int64_t at_int64_value_at_indexes(gc_tensor, int *indexes, int indexes_len);
+void at_set_double_value_at_indexes(gc_tensor, int *indexes, int indexes_len,
                                     double v);
-void at_set_int64_value_at_indexes(tensor, int *indexes, int indexes_len,
+void at_set_int64_value_at_indexes(gc_tensor, int *indexes, int indexes_len,
                                    int64_t v);
 
-void at_copy_(tensor dst, tensor src);
-void at_set_data(tensor dst, tensor src);
+void at_copy_(gc_tensor dst, gc_tensor src);
+void at_set_data(gc_tensor dst, gc_tensor src);
 
-void at_print(tensor);
-char *at_to_string(tensor, int line_size);
-void at_save(tensor, char *filename);
-tensor at_load(char *filename);
+void at_print(gc_tensor);
+char *at_to_string(gc_tensor, int line_size);
+void at_save(gc_tensor, char *filename);
+raw_tensor at_load(char *filename);
 
 int at_get_num_threads();
 void at_set_num_threads(int n_threads);
 
-void at_save_multi(tensor *tensors, char **tensor_names, int ntensors,
+void at_save_multi(gc_tensor *tensors, char **tensor_names, int ntensors,
                    char *filename);
 /* [at_load_multi] takes as input an array of nullptr for [tensors]. */
-void at_load_multi(tensor *tensors, char **tensor_names, int ntensors,
+void at_load_multi(raw_tensor *outputs, char **tensor_names, int ntensors,
                    char *filename);
 /* [at_load_multi_] takes as input an array of allocation [tensors]. */
-void at_load_multi_(tensor *tensors, char **tensor_names, int ntensors,
+void at_load_multi_(gc_tensor *tensors, char **tensor_names, int ntensors,
                     char *filename);
 
-void at_load_callback(char *filename, void (*f)(char *, tensor));
+void at_load_callback(char *filename, void (*f)(char *, raw_tensor));
 
-void at_free(tensor);
+void at_free(gc_tensor);
 
-void at_run_backward(tensor *tensors, int ntensors, tensor *inputs, int ninputs,
-                     tensor *outputs, int keep_graph, int create_graph);
+void at_run_backward(gc_tensor *tensors, int ntensors, gc_tensor *inputs,
+                     int ninputs, raw_tensor *outputs, int keep_graph,
+                     int create_graph);
 
 optimizer ato_adam(double learning_rate, double beta1, double beta2,
                    double weight_decay, double eps);
@@ -95,7 +100,7 @@ optimizer ato_rmsprop(double learning_rate, double alpha, double eps,
                       double weight_decay, double momentum, int centered);
 optimizer ato_sgd(double learning_rate, double momentum, double dampening,
                   double weight_decay, int nesterov);
-void ato_add_parameters(optimizer, tensor *, int ntensors);
+void ato_add_parameters(optimizer, gc_tensor *, int ntensors);
 void ato_set_learning_rate(optimizer, double learning_rate);
 void ato_set_momentum(optimizer, double momentum);
 void ato_zero_grad(optimizer);
@@ -113,28 +118,27 @@ void atc_set_benchmark_cudnn(int b);
 
 module atm_load(char *, int);
 module atm_load_str(char *, size_t, int);
-tensor atm_forward(module, tensor *tensors, int ntensors);
+raw_tensor atm_forward(module, gc_tensor *tensors, int ntensors);
 ivalue atm_forward_(module, ivalue *ivalues, int nivalues);
 ivalue atm_named_buffers(module);
 void atm_free(module);
 
 ivalue ati_none();
-ivalue ati_tensor(tensor);
+ivalue ati_tensor(gc_tensor);
 ivalue ati_bool(int);
 ivalue ati_int(int64_t);
 ivalue ati_double(double);
 ivalue ati_tuple(ivalue *, int);
 ivalue ati_string(char *);
-ivalue ati_tuple(ivalue *, int);
 ivalue ati_generic_list(ivalue *, int);
 ivalue ati_generic_dict(ivalue *, int);
 ivalue ati_int_list(int64_t *, int);
 ivalue ati_double_list(double *, int);
 ivalue ati_bool_list(char *, int);
 ivalue ati_string_list(char **, int);
-ivalue ati_tensor_list(tensor *, int);
+ivalue ati_tensor_list(gc_tensor *, int);
 
-tensor ati_to_tensor(ivalue);
+raw_tensor ati_to_tensor(ivalue);
 int64_t ati_to_int(ivalue);
 double ati_to_double(ivalue);
 char *ati_to_string(ivalue);
@@ -148,7 +152,7 @@ void ati_to_generic_dict(ivalue, ivalue *, int);
 void ati_to_int_list(ivalue, int64_t *, int);
 void ati_to_double_list(ivalue, double *, int);
 void ati_to_bool_list(ivalue, char *, int);
-void ati_to_tensor_list(ivalue, tensor *, int);
+void ati_to_tensor_list(ivalue, raw_tensor *, int);
 
 int ati_tag(ivalue);
 
diff --git a/src/wrapper/torch_api_generated.cpp b/src/wrapper/torch_api_generated.cpp
new file mode 100644
index 0000000..efb3208
--- /dev/null
+++ b/src/wrapper/torch_api_generated.cpp
@@ -0,0 +1,18377 @@
+// THIS FILE IS AUTOMATICALLY GENERATED, DO NOT EDIT BY HAND!
+
+raw_tensor atg___and__(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::__and__(*tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg___and__tensor_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::__and__(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg___iand__(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->__iand__(*other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg___iand__tensor_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->__iand__(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg___ilshift__(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->__ilshift__(*other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg___ilshift__tensor_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->__ilshift__(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg___ior__(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->__ior__(*other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg___ior__tensor_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->__ior__(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg___irshift__(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->__irshift__(*other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg___irshift__tensor_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->__irshift__(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg___ixor__(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->__ixor__(*other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg___ixor__tensor_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->__ixor__(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg___lshift__(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::__lshift__(*tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg___lshift__scalar_out_(gc_tensor out, gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::__lshift___out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg___lshift__tensor_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::__lshift__(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg___lshift__tensor_out_(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::__lshift___out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg___or__(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::__or__(*tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg___or__tensor_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::__or__(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg___rshift__(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::__rshift__(*tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg___rshift__scalar_out_(gc_tensor out, gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::__rshift___out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg___rshift__tensor_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::__rshift__(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg___rshift__tensor_out_(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::__rshift___out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg___xor__(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::__xor__(*tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg___xor__tensor_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::__xor__(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__adaptive_avg_pool2d(gc_tensor self, int64_t *output_size_data, int output_size_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::_adaptive_avg_pool2d(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(output_size_data, output_size_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__adaptive_avg_pool2d_backward(gc_tensor grad_output, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::_adaptive_avg_pool2d_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__adaptive_avg_pool2d_backward_out(gc_tensor out, gc_tensor grad_output, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::_adaptive_avg_pool2d_backward_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__adaptive_avg_pool2d_out(gc_tensor out, gc_tensor self, int64_t *output_size_data, int output_size_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::_adaptive_avg_pool2d_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(output_size_data, output_size_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__adaptive_avg_pool3d(gc_tensor self, int64_t *output_size_data, int output_size_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::_adaptive_avg_pool3d(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(output_size_data, output_size_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__adaptive_avg_pool3d_backward(gc_tensor grad_output, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::_adaptive_avg_pool3d_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__adaptive_avg_pool3d_backward_out(gc_tensor out, gc_tensor grad_output, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::_adaptive_avg_pool3d_backward_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__adaptive_avg_pool3d_out(gc_tensor out, gc_tensor self, int64_t *output_size_data, int output_size_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::_adaptive_avg_pool3d_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(output_size_data, output_size_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__add_batch_dim(gc_tensor self, int64_t batch_dim, int64_t level) {
+  PROTECT(
+    torch::Tensor results__ = torch::_add_batch_dim(*tensor_ptr_from_ocaml(self), batch_dim, level);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__add_relu(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::_add_relu(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__add_relu_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::_add_relu_(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__add_relu_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::_add_relu_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__add_relu_scalar(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::_add_relu(*tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__add_relu_scalar_(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::_add_relu_(*tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__add_relu_scalar_out(gc_tensor out, gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::_add_relu_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__addmm_activation(gc_tensor self, gc_tensor mat1, gc_tensor mat2, int use_gelu) {
+  PROTECT(
+    torch::Tensor results__ = torch::_addmm_activation(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(mat1), *tensor_ptr_from_ocaml(mat2), (bool)use_gelu);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__addmm_activation_out(gc_tensor out, gc_tensor self, gc_tensor mat1, gc_tensor mat2, int use_gelu) {
+  PROTECT(
+    torch::Tensor results__ = torch::_addmm_activation_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(mat1), *tensor_ptr_from_ocaml(mat2), (bool)use_gelu);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg__aminmax(raw_tensor *out__, gc_tensor self) {
+  PROTECT(
+    auto results__ = torch::_aminmax(*tensor_ptr_from_ocaml(self));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg__aminmax_dim(raw_tensor *out__, gc_tensor self, int64_t dim, int keepdim) {
+  PROTECT(
+    auto results__ = torch::_aminmax(*tensor_ptr_from_ocaml(self), dim, (bool)keepdim);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg__aminmax_dim_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor self, int64_t dim, int keepdim) {
+  PROTECT(
+    auto results__ = torch::_aminmax_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(self), dim, (bool)keepdim);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg__aminmax_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor self) {
+  PROTECT(
+    auto results__ = torch::_aminmax_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(self));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg__amp_update_scale(raw_tensor *out__, gc_tensor self, gc_tensor growth_tracker, gc_tensor found_inf, double scale_growth_factor, double scale_backoff_factor, int64_t growth_interval) {
+  PROTECT(
+    auto results__ = torch::_amp_update_scale(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(growth_tracker), *tensor_ptr_from_ocaml(found_inf), scale_growth_factor, scale_backoff_factor, growth_interval);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg__amp_update_scale_(gc_tensor self, gc_tensor growth_tracker, gc_tensor found_inf, double scale_growth_factor, double scale_backoff_factor, int64_t growth_interval) {
+  PROTECT(
+    torch::Tensor results__ = torch::_amp_update_scale_(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(growth_tracker), *tensor_ptr_from_ocaml(found_inf), scale_growth_factor, scale_backoff_factor, growth_interval);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__amp_update_scale_out(gc_tensor out, gc_tensor self, gc_tensor growth_tracker, gc_tensor found_inf, double scale_growth_factor, double scale_backoff_factor, int64_t growth_interval) {
+  PROTECT(
+    torch::Tensor results__ = torch::_amp_update_scale_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(growth_tracker), *tensor_ptr_from_ocaml(found_inf), scale_growth_factor, scale_backoff_factor, growth_interval);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg__assert_tensor_metadata(gc_tensor a, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len, int dtype) {
+  PROTECT(
+    torch::_assert_tensor_metadata(*tensor_ptr_from_ocaml(a), size_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(size_data, size_len)), stride_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(stride_data, stride_len)), torch::ScalarType(dtype));
+  )
+}
+
+raw_tensor atg__autocast_to_full_precision(gc_tensor self, int cuda_enabled, int cpu_enabled) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->_autocast_to_full_precision((bool)cuda_enabled, (bool)cpu_enabled);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__autocast_to_reduced_precision(gc_tensor self, int cuda_enabled, int cpu_enabled, int cuda_dtype, int cpu_dtype) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->_autocast_to_reduced_precision((bool)cuda_enabled, (bool)cpu_enabled, torch::ScalarType(cuda_dtype), torch::ScalarType(cpu_dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__cast_byte(gc_tensor self, int non_blocking) {
+  PROTECT(
+    torch::Tensor results__ = torch::_cast_Byte(*tensor_ptr_from_ocaml(self), (bool)non_blocking);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__cast_char(gc_tensor self, int non_blocking) {
+  PROTECT(
+    torch::Tensor results__ = torch::_cast_Char(*tensor_ptr_from_ocaml(self), (bool)non_blocking);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__cast_double(gc_tensor self, int non_blocking) {
+  PROTECT(
+    torch::Tensor results__ = torch::_cast_Double(*tensor_ptr_from_ocaml(self), (bool)non_blocking);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__cast_float(gc_tensor self, int non_blocking) {
+  PROTECT(
+    torch::Tensor results__ = torch::_cast_Float(*tensor_ptr_from_ocaml(self), (bool)non_blocking);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__cast_half(gc_tensor self, int non_blocking) {
+  PROTECT(
+    torch::Tensor results__ = torch::_cast_Half(*tensor_ptr_from_ocaml(self), (bool)non_blocking);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__cast_int(gc_tensor self, int non_blocking) {
+  PROTECT(
+    torch::Tensor results__ = torch::_cast_Int(*tensor_ptr_from_ocaml(self), (bool)non_blocking);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__cast_long(gc_tensor self, int non_blocking) {
+  PROTECT(
+    torch::Tensor results__ = torch::_cast_Long(*tensor_ptr_from_ocaml(self), (bool)non_blocking);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__cast_short(gc_tensor self, int non_blocking) {
+  PROTECT(
+    torch::Tensor results__ = torch::_cast_Short(*tensor_ptr_from_ocaml(self), (bool)non_blocking);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__cdist_backward(gc_tensor grad, gc_tensor x1, gc_tensor x2, double p, gc_tensor cdist) {
+  PROTECT(
+    torch::Tensor results__ = torch::_cdist_backward(*tensor_ptr_from_ocaml(grad), *tensor_ptr_from_ocaml(x1), *tensor_ptr_from_ocaml(x2), p, *tensor_ptr_from_ocaml(cdist));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__cdist_backward_out(gc_tensor out, gc_tensor grad, gc_tensor x1, gc_tensor x2, double p, gc_tensor cdist) {
+  PROTECT(
+    torch::Tensor results__ = torch::_cdist_backward_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(grad), *tensor_ptr_from_ocaml(x1), *tensor_ptr_from_ocaml(x2), p, *tensor_ptr_from_ocaml(cdist));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__cholesky_solve_helper(gc_tensor self, gc_tensor A, int upper) {
+  PROTECT(
+    torch::Tensor results__ = torch::_cholesky_solve_helper(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(A), (bool)upper);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__cholesky_solve_helper_out(gc_tensor out, gc_tensor self, gc_tensor A, int upper) {
+  PROTECT(
+    torch::Tensor results__ = torch::_cholesky_solve_helper_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(A), (bool)upper);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__coalesce(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::_coalesce(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__coalesce_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::_coalesce_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__coalesced(gc_tensor self, int coalesced) {
+  PROTECT(
+    torch::Tensor results__ = torch::_coalesced(*tensor_ptr_from_ocaml(self), (bool)coalesced);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__coalesced_(gc_tensor self, int coalesced) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->_coalesced_((bool)coalesced);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__coalesced_out(gc_tensor out, gc_tensor self, int coalesced) {
+  PROTECT(
+    torch::Tensor results__ = torch::_coalesced_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), (bool)coalesced);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__compute_linear_combination(gc_tensor input, gc_tensor coefficients) {
+  PROTECT(
+    torch::Tensor results__ = torch::_compute_linear_combination(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(coefficients));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__compute_linear_combination_out(gc_tensor out, gc_tensor input, gc_tensor coefficients) {
+  PROTECT(
+    torch::Tensor results__ = torch::_compute_linear_combination_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(coefficients));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__conj(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::_conj(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__conj_copy(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::_conj_copy(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__conj_copy_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::_conj_copy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__conj_physical(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::_conj_physical(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__conj_physical_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::_conj_physical_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__conv_depthwise2d(gc_tensor self, gc_tensor weight, int64_t *kernel_size_data, int kernel_size_len, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::_conv_depthwise2d(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(weight), torch::IntArrayRef(kernel_size_data, kernel_size_len), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__conv_depthwise2d_out(gc_tensor out, gc_tensor self, gc_tensor weight, int64_t *kernel_size_data, int kernel_size_len, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::_conv_depthwise2d_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(weight), torch::IntArrayRef(kernel_size_data, kernel_size_len), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__convert_indices_from_coo_to_csr(gc_tensor self, int64_t size, int out_int32) {
+  PROTECT(
+    torch::Tensor results__ = torch::_convert_indices_from_coo_to_csr(*tensor_ptr_from_ocaml(self), size, (bool)out_int32);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__convert_indices_from_coo_to_csr_out(gc_tensor out, gc_tensor self, int64_t size, int out_int32) {
+  PROTECT(
+    torch::Tensor results__ = torch::_convert_indices_from_coo_to_csr_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), size, (bool)out_int32);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__convert_indices_from_csr_to_coo(gc_tensor crow_indices, gc_tensor col_indices, int out_int32, int transpose) {
+  PROTECT(
+    torch::Tensor results__ = torch::_convert_indices_from_csr_to_coo(*tensor_ptr_from_ocaml(crow_indices), *tensor_ptr_from_ocaml(col_indices), (bool)out_int32, (bool)transpose);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__convert_indices_from_csr_to_coo_out(gc_tensor out, gc_tensor crow_indices, gc_tensor col_indices, int out_int32, int transpose) {
+  PROTECT(
+    torch::Tensor results__ = torch::_convert_indices_from_csr_to_coo_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(crow_indices), *tensor_ptr_from_ocaml(col_indices), (bool)out_int32, (bool)transpose);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__convolution(gc_tensor input, gc_tensor weight, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int transposed, int64_t *output_padding_data, int output_padding_len, int64_t groups, int benchmark, int deterministic, int cudnn_enabled, int allow_tf32) {
+  PROTECT(
+    torch::Tensor results__ = torch::_convolution(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(weight), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)transposed, torch::IntArrayRef(output_padding_data, output_padding_len), groups, (bool)benchmark, (bool)deterministic, (bool)cudnn_enabled, (bool)allow_tf32);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__convolution_deprecated(gc_tensor input, gc_tensor weight, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int transposed, int64_t *output_padding_data, int output_padding_len, int64_t groups, int benchmark, int deterministic, int cudnn_enabled) {
+  PROTECT(
+    torch::Tensor results__ = torch::_convolution(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(weight), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)transposed, torch::IntArrayRef(output_padding_data, output_padding_len), groups, (bool)benchmark, (bool)deterministic, (bool)cudnn_enabled);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__convolution_mode(gc_tensor input, gc_tensor weight, gc_tensor bias, int64_t *stride_data, int stride_len, char * padding, int64_t *dilation_data, int dilation_len, int64_t groups) {
+  PROTECT(
+    torch::Tensor results__ = torch::_convolution_mode(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(weight), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), std::string(padding), torch::IntArrayRef(dilation_data, dilation_len), groups);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__convolution_out(gc_tensor out, gc_tensor input, gc_tensor weight, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int transposed, int64_t *output_padding_data, int output_padding_len, int64_t groups, int benchmark, int deterministic, int cudnn_enabled, int allow_tf32) {
+  PROTECT(
+    torch::Tensor results__ = torch::_convolution_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(weight), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)transposed, torch::IntArrayRef(output_padding_data, output_padding_len), groups, (bool)benchmark, (bool)deterministic, (bool)cudnn_enabled, (bool)allow_tf32);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__copy_from(gc_tensor self, gc_tensor dst, int non_blocking) {
+  PROTECT(
+    torch::Tensor results__ = torch::_copy_from(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(dst), (bool)non_blocking);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__copy_from_and_resize(gc_tensor self, gc_tensor dst) {
+  PROTECT(
+    torch::Tensor results__ = torch::_copy_from_and_resize(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(dst));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__copy_from_and_resize_out(gc_tensor out, gc_tensor self, gc_tensor dst) {
+  PROTECT(
+    torch::Tensor results__ = torch::_copy_from_and_resize_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(dst));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__copy_from_out(gc_tensor out, gc_tensor self, gc_tensor dst, int non_blocking) {
+  PROTECT(
+    torch::Tensor results__ = torch::_copy_from_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(dst), (bool)non_blocking);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__cslt_compress(gc_tensor input) {
+  PROTECT(
+    torch::Tensor results__ = torch::_cslt_compress(*tensor_ptr_from_ocaml(input));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__cslt_sparse_mm(gc_tensor compressed_A, gc_tensor dense_B, gc_tensor bias, int transpose_result) {
+  PROTECT(
+    torch::Tensor results__ = torch::_cslt_sparse_mm(*tensor_ptr_from_ocaml(compressed_A), *tensor_ptr_from_ocaml(dense_B), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), (bool)transpose_result);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg__ctc_loss(raw_tensor *out__, gc_tensor log_probs, gc_tensor targets, int64_t *input_lengths_data, int input_lengths_len, int64_t *target_lengths_data, int target_lengths_len, int64_t blank, int zero_infinity) {
+  PROTECT(
+    auto results__ = torch::_ctc_loss(*tensor_ptr_from_ocaml(log_probs), *tensor_ptr_from_ocaml(targets), torch::IntArrayRef(input_lengths_data, input_lengths_len), torch::IntArrayRef(target_lengths_data, target_lengths_len), blank, (bool)zero_infinity);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg__ctc_loss_backward(gc_tensor grad, gc_tensor log_probs, gc_tensor targets, int64_t *input_lengths_data, int input_lengths_len, int64_t *target_lengths_data, int target_lengths_len, gc_tensor neg_log_likelihood, gc_tensor log_alpha, int64_t blank, int zero_infinity) {
+  PROTECT(
+    torch::Tensor results__ = torch::_ctc_loss_backward(*tensor_ptr_from_ocaml(grad), *tensor_ptr_from_ocaml(log_probs), *tensor_ptr_from_ocaml(targets), torch::IntArrayRef(input_lengths_data, input_lengths_len), torch::IntArrayRef(target_lengths_data, target_lengths_len), *tensor_ptr_from_ocaml(neg_log_likelihood), *tensor_ptr_from_ocaml(log_alpha), blank, (bool)zero_infinity);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__ctc_loss_backward_out(gc_tensor out, gc_tensor grad, gc_tensor log_probs, gc_tensor targets, int64_t *input_lengths_data, int input_lengths_len, int64_t *target_lengths_data, int target_lengths_len, gc_tensor neg_log_likelihood, gc_tensor log_alpha, int64_t blank, int zero_infinity) {
+  PROTECT(
+    torch::Tensor results__ = torch::_ctc_loss_backward_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(grad), *tensor_ptr_from_ocaml(log_probs), *tensor_ptr_from_ocaml(targets), torch::IntArrayRef(input_lengths_data, input_lengths_len), torch::IntArrayRef(target_lengths_data, target_lengths_len), *tensor_ptr_from_ocaml(neg_log_likelihood), *tensor_ptr_from_ocaml(log_alpha), blank, (bool)zero_infinity);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__ctc_loss_backward_tensor(gc_tensor grad, gc_tensor log_probs, gc_tensor targets, gc_tensor input_lengths, gc_tensor target_lengths, gc_tensor neg_log_likelihood, gc_tensor log_alpha, int64_t blank, int zero_infinity) {
+  PROTECT(
+    torch::Tensor results__ = torch::_ctc_loss_backward(*tensor_ptr_from_ocaml(grad), *tensor_ptr_from_ocaml(log_probs), *tensor_ptr_from_ocaml(targets), *tensor_ptr_from_ocaml(input_lengths), *tensor_ptr_from_ocaml(target_lengths), *tensor_ptr_from_ocaml(neg_log_likelihood), *tensor_ptr_from_ocaml(log_alpha), blank, (bool)zero_infinity);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg__ctc_loss_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor log_probs, gc_tensor targets, int64_t *input_lengths_data, int input_lengths_len, int64_t *target_lengths_data, int target_lengths_len, int64_t blank, int zero_infinity) {
+  PROTECT(
+    auto results__ = torch::_ctc_loss_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(log_probs), *tensor_ptr_from_ocaml(targets), torch::IntArrayRef(input_lengths_data, input_lengths_len), torch::IntArrayRef(target_lengths_data, target_lengths_len), blank, (bool)zero_infinity);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg__ctc_loss_tensor(raw_tensor *out__, gc_tensor log_probs, gc_tensor targets, gc_tensor input_lengths, gc_tensor target_lengths, int64_t blank, int zero_infinity) {
+  PROTECT(
+    auto results__ = torch::_ctc_loss(*tensor_ptr_from_ocaml(log_probs), *tensor_ptr_from_ocaml(targets), *tensor_ptr_from_ocaml(input_lengths), *tensor_ptr_from_ocaml(target_lengths), blank, (bool)zero_infinity);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg__ctc_loss_tensor_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor log_probs, gc_tensor targets, gc_tensor input_lengths, gc_tensor target_lengths, int64_t blank, int zero_infinity) {
+  PROTECT(
+    auto results__ = torch::_ctc_loss_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(log_probs), *tensor_ptr_from_ocaml(targets), *tensor_ptr_from_ocaml(input_lengths), *tensor_ptr_from_ocaml(target_lengths), blank, (bool)zero_infinity);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg__cudnn_ctc_loss(raw_tensor *out__, gc_tensor log_probs, gc_tensor targets, int64_t *input_lengths_data, int input_lengths_len, int64_t *target_lengths_data, int target_lengths_len, int64_t blank, int deterministic, int zero_infinity) {
+  PROTECT(
+    auto results__ = torch::_cudnn_ctc_loss(*tensor_ptr_from_ocaml(log_probs), *tensor_ptr_from_ocaml(targets), torch::IntArrayRef(input_lengths_data, input_lengths_len), torch::IntArrayRef(target_lengths_data, target_lengths_len), blank, (bool)deterministic, (bool)zero_infinity);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg__cudnn_ctc_loss_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor log_probs, gc_tensor targets, int64_t *input_lengths_data, int input_lengths_len, int64_t *target_lengths_data, int target_lengths_len, int64_t blank, int deterministic, int zero_infinity) {
+  PROTECT(
+    auto results__ = torch::_cudnn_ctc_loss_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(log_probs), *tensor_ptr_from_ocaml(targets), torch::IntArrayRef(input_lengths_data, input_lengths_len), torch::IntArrayRef(target_lengths_data, target_lengths_len), blank, (bool)deterministic, (bool)zero_infinity);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg__cudnn_ctc_loss_tensor(raw_tensor *out__, gc_tensor log_probs, gc_tensor targets, gc_tensor input_lengths, gc_tensor target_lengths, int64_t blank, int deterministic, int zero_infinity) {
+  PROTECT(
+    auto results__ = torch::_cudnn_ctc_loss(*tensor_ptr_from_ocaml(log_probs), *tensor_ptr_from_ocaml(targets), *tensor_ptr_from_ocaml(input_lengths), *tensor_ptr_from_ocaml(target_lengths), blank, (bool)deterministic, (bool)zero_infinity);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg__cudnn_init_dropout_state(double dropout, int train, int64_t dropout_seed, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::_cudnn_init_dropout_state(dropout, (bool)train, dropout_seed, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__cudnn_init_dropout_state_out(gc_tensor out, double dropout, int train, int64_t dropout_seed) {
+  PROTECT(
+    torch::Tensor results__ = torch::_cudnn_init_dropout_state_out(*tensor_ptr_from_ocaml(out), dropout, (bool)train, dropout_seed);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg__cudnn_rnn(raw_tensor *out__, gc_tensor input, gc_tensor *weight_data, int weight_len, int64_t weight_stride0, gc_tensor weight_buf, gc_tensor hx, gc_tensor cx, int64_t mode, int64_t hidden_size, int64_t proj_size, int64_t num_layers, int batch_first, double dropout, int train, int bidirectional, int64_t *batch_sizes_data, int batch_sizes_len, gc_tensor dropout_state) {
+  PROTECT(
+    auto results__ = torch::_cudnn_rnn(*tensor_ptr_from_ocaml(input), of_carray_tensor(weight_data, weight_len), weight_stride0, (weight_buf ? tensor_from_ocaml(weight_buf) : torch::Tensor()), *tensor_ptr_from_ocaml(hx), (cx ? tensor_from_ocaml(cx) : torch::Tensor()), mode, hidden_size, proj_size, num_layers, (bool)batch_first, dropout, (bool)train, (bool)bidirectional, torch::IntArrayRef(batch_sizes_data, batch_sizes_len), (dropout_state ? tensor_from_ocaml(dropout_state) : torch::Tensor()));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+    out__[3] = tensor_to_ocaml(std::get<3>(results__));
+    out__[4] = tensor_to_ocaml(std::get<4>(results__));
+  )
+}
+
+raw_tensor atg__cudnn_rnn_flatten_weight(gc_tensor *weight_arr_data, int weight_arr_len, int64_t weight_stride0, int64_t input_size, int64_t mode, int64_t hidden_size, int64_t proj_size, int64_t num_layers, int batch_first, int bidirectional) {
+  PROTECT(
+    torch::Tensor results__ = torch::_cudnn_rnn_flatten_weight(of_carray_tensor(weight_arr_data, weight_arr_len), weight_stride0, input_size, mode, hidden_size, proj_size, num_layers, (bool)batch_first, (bool)bidirectional);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__cudnn_rnn_flatten_weight_out(gc_tensor out, gc_tensor *weight_arr_data, int weight_arr_len, int64_t weight_stride0, int64_t input_size, int64_t mode, int64_t hidden_size, int64_t proj_size, int64_t num_layers, int batch_first, int bidirectional) {
+  PROTECT(
+    torch::Tensor results__ = torch::_cudnn_rnn_flatten_weight_out(*tensor_ptr_from_ocaml(out), of_carray_tensor(weight_arr_data, weight_arr_len), weight_stride0, input_size, mode, hidden_size, proj_size, num_layers, (bool)batch_first, (bool)bidirectional);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg__cudnn_rnn_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor out3, gc_tensor out4, gc_tensor input, gc_tensor *weight_data, int weight_len, int64_t weight_stride0, gc_tensor weight_buf, gc_tensor hx, gc_tensor cx, int64_t mode, int64_t hidden_size, int64_t proj_size, int64_t num_layers, int batch_first, double dropout, int train, int bidirectional, int64_t *batch_sizes_data, int batch_sizes_len, gc_tensor dropout_state) {
+  PROTECT(
+    auto results__ = torch::_cudnn_rnn_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(out2), *tensor_ptr_from_ocaml(out3), *tensor_ptr_from_ocaml(out4), *tensor_ptr_from_ocaml(input), of_carray_tensor(weight_data, weight_len), weight_stride0, (weight_buf ? tensor_from_ocaml(weight_buf) : torch::Tensor()), *tensor_ptr_from_ocaml(hx), (cx ? tensor_from_ocaml(cx) : torch::Tensor()), mode, hidden_size, proj_size, num_layers, (bool)batch_first, dropout, (bool)train, (bool)bidirectional, torch::IntArrayRef(batch_sizes_data, batch_sizes_len), (dropout_state ? tensor_from_ocaml(dropout_state) : torch::Tensor()));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+    out__[3] = tensor_to_ocaml(std::get<3>(results__));
+    out__[4] = tensor_to_ocaml(std::get<4>(results__));
+  )
+}
+
+int64_t atg__debug_has_internal_overlap(gc_tensor self) {
+  PROTECT(
+    return torch::_debug_has_internal_overlap(*tensor_ptr_from_ocaml(self));
+  )
+  return 0;
+}
+
+raw_tensor atg__dim_arange(gc_tensor like, int64_t dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::_dim_arange(*tensor_ptr_from_ocaml(like), dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+int64_t atg__dimi(gc_tensor self) {
+  PROTECT(
+    return tensor_ptr_from_ocaml(self)->_dimI();
+  )
+  return 0;
+}
+
+int64_t atg__dimv(gc_tensor self) {
+  PROTECT(
+    return tensor_ptr_from_ocaml(self)->_dimV();
+  )
+  return 0;
+}
+
+raw_tensor atg__dirichlet_grad(gc_tensor x, gc_tensor alpha, gc_tensor total) {
+  PROTECT(
+    torch::Tensor results__ = torch::_dirichlet_grad(*tensor_ptr_from_ocaml(x), *tensor_ptr_from_ocaml(alpha), *tensor_ptr_from_ocaml(total));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__dirichlet_grad_out(gc_tensor out, gc_tensor x, gc_tensor alpha, gc_tensor total) {
+  PROTECT(
+    torch::Tensor results__ = torch::_dirichlet_grad_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(x), *tensor_ptr_from_ocaml(alpha), *tensor_ptr_from_ocaml(total));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg__efficient_attention_backward(raw_tensor *out__, gc_tensor grad_out_, gc_tensor query, gc_tensor key, gc_tensor value, gc_tensor bias, gc_tensor out, gc_tensor cu_seqlens_q, gc_tensor cu_seqlens_k, int64_t max_seqlen_k, int64_t max_seqlen_q, gc_tensor logsumexp, double dropout_p, gc_tensor philox_seed, gc_tensor philox_offset, int64_t custom_mask_type, int bias_requires_grad, double scale_v, int scale_null, int64_t num_splits_key_v, int num_splits_key_null) {
+  PROTECT(
+    auto results__ = torch::_efficient_attention_backward(*tensor_ptr_from_ocaml(grad_out_), *tensor_ptr_from_ocaml(query), *tensor_ptr_from_ocaml(key), *tensor_ptr_from_ocaml(value), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), *tensor_ptr_from_ocaml(out), (cu_seqlens_q ? tensor_from_ocaml(cu_seqlens_q) : torch::Tensor()), (cu_seqlens_k ? tensor_from_ocaml(cu_seqlens_k) : torch::Tensor()), max_seqlen_k, max_seqlen_q, *tensor_ptr_from_ocaml(logsumexp), dropout_p, *tensor_ptr_from_ocaml(philox_seed), *tensor_ptr_from_ocaml(philox_offset), custom_mask_type, (bool)bias_requires_grad, scale_null ? c10::nullopt : c10::optional<double>(scale_v), num_splits_key_null ? c10::nullopt : c10::optional<int64_t>(num_splits_key_v));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+    out__[3] = tensor_to_ocaml(std::get<3>(results__));
+  )
+}
+
+raw_tensor atg__efficientzerotensor(int64_t *size_data, int size_len, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::_efficientzerotensor(torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__efficientzerotensor_out(gc_tensor out, int64_t *size_data, int size_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::_efficientzerotensor_out(*tensor_ptr_from_ocaml(out), torch::IntArrayRef(size_data, size_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg__embedding_bag(raw_tensor *out__, gc_tensor weight, gc_tensor indices, gc_tensor offsets, int scale_grad_by_freq, int64_t mode, int sparse, gc_tensor per_sample_weights, int include_last_offset, int64_t padding_idx) {
+  PROTECT(
+    auto results__ = torch::_embedding_bag(*tensor_ptr_from_ocaml(weight), *tensor_ptr_from_ocaml(indices), *tensor_ptr_from_ocaml(offsets), (bool)scale_grad_by_freq, mode, (bool)sparse, (per_sample_weights ? tensor_from_ocaml(per_sample_weights) : torch::Tensor()), (bool)include_last_offset, padding_idx);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+    out__[3] = tensor_to_ocaml(std::get<3>(results__));
+  )
+}
+
+raw_tensor atg__embedding_bag_backward(gc_tensor grad, gc_tensor indices, gc_tensor offsets, gc_tensor offset2bag, gc_tensor bag_size, gc_tensor maximum_indices, int64_t num_weights, int scale_grad_by_freq, int64_t mode, int sparse, gc_tensor per_sample_weights, int64_t padding_idx) {
+  PROTECT(
+    torch::Tensor results__ = torch::_embedding_bag_backward(*tensor_ptr_from_ocaml(grad), *tensor_ptr_from_ocaml(indices), *tensor_ptr_from_ocaml(offsets), *tensor_ptr_from_ocaml(offset2bag), *tensor_ptr_from_ocaml(bag_size), *tensor_ptr_from_ocaml(maximum_indices), num_weights, (bool)scale_grad_by_freq, mode, (bool)sparse, (per_sample_weights ? tensor_from_ocaml(per_sample_weights) : torch::Tensor()), padding_idx);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__embedding_bag_dense_backward(gc_tensor grad, gc_tensor indices, gc_tensor offset2bag, gc_tensor bag_size, gc_tensor maximum_indices, int64_t num_weights, int scale_grad_by_freq, int64_t mode, gc_tensor per_sample_weights, int64_t padding_idx) {
+  PROTECT(
+    torch::Tensor results__ = torch::_embedding_bag_dense_backward(*tensor_ptr_from_ocaml(grad), *tensor_ptr_from_ocaml(indices), *tensor_ptr_from_ocaml(offset2bag), *tensor_ptr_from_ocaml(bag_size), *tensor_ptr_from_ocaml(maximum_indices), num_weights, (bool)scale_grad_by_freq, mode, (per_sample_weights ? tensor_from_ocaml(per_sample_weights) : torch::Tensor()), padding_idx);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__embedding_bag_dense_backward_out(gc_tensor out, gc_tensor grad, gc_tensor indices, gc_tensor offset2bag, gc_tensor bag_size, gc_tensor maximum_indices, int64_t num_weights, int scale_grad_by_freq, int64_t mode, gc_tensor per_sample_weights, int64_t padding_idx) {
+  PROTECT(
+    torch::Tensor results__ = torch::_embedding_bag_dense_backward_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(grad), *tensor_ptr_from_ocaml(indices), *tensor_ptr_from_ocaml(offset2bag), *tensor_ptr_from_ocaml(bag_size), *tensor_ptr_from_ocaml(maximum_indices), num_weights, (bool)scale_grad_by_freq, mode, (per_sample_weights ? tensor_from_ocaml(per_sample_weights) : torch::Tensor()), padding_idx);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg__embedding_bag_forward_only(raw_tensor *out__, gc_tensor weight, gc_tensor indices, gc_tensor offsets, int scale_grad_by_freq, int64_t mode, int sparse, gc_tensor per_sample_weights, int include_last_offset, int64_t padding_idx) {
+  PROTECT(
+    auto results__ = torch::_embedding_bag_forward_only(*tensor_ptr_from_ocaml(weight), *tensor_ptr_from_ocaml(indices), *tensor_ptr_from_ocaml(offsets), (bool)scale_grad_by_freq, mode, (bool)sparse, (per_sample_weights ? tensor_from_ocaml(per_sample_weights) : torch::Tensor()), (bool)include_last_offset, padding_idx);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+    out__[3] = tensor_to_ocaml(std::get<3>(results__));
+  )
+}
+
+void atg__embedding_bag_forward_only_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor out3, gc_tensor weight, gc_tensor indices, gc_tensor offsets, int scale_grad_by_freq, int64_t mode, int sparse, gc_tensor per_sample_weights, int include_last_offset, int64_t padding_idx) {
+  PROTECT(
+    auto results__ = torch::_embedding_bag_forward_only_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(out2), *tensor_ptr_from_ocaml(out3), *tensor_ptr_from_ocaml(weight), *tensor_ptr_from_ocaml(indices), *tensor_ptr_from_ocaml(offsets), (bool)scale_grad_by_freq, mode, (bool)sparse, (per_sample_weights ? tensor_from_ocaml(per_sample_weights) : torch::Tensor()), (bool)include_last_offset, padding_idx);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+    out__[3] = tensor_to_ocaml(std::get<3>(results__));
+  )
+}
+
+void atg__embedding_bag_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor out3, gc_tensor weight, gc_tensor indices, gc_tensor offsets, int scale_grad_by_freq, int64_t mode, int sparse, gc_tensor per_sample_weights, int include_last_offset, int64_t padding_idx) {
+  PROTECT(
+    auto results__ = torch::_embedding_bag_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(out2), *tensor_ptr_from_ocaml(out3), *tensor_ptr_from_ocaml(weight), *tensor_ptr_from_ocaml(indices), *tensor_ptr_from_ocaml(offsets), (bool)scale_grad_by_freq, mode, (bool)sparse, (per_sample_weights ? tensor_from_ocaml(per_sample_weights) : torch::Tensor()), (bool)include_last_offset, padding_idx);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+    out__[3] = tensor_to_ocaml(std::get<3>(results__));
+  )
+}
+
+raw_tensor atg__embedding_bag_per_sample_weights_backward(gc_tensor grad, gc_tensor weight, gc_tensor indices, gc_tensor offsets, gc_tensor offset2bag, int64_t mode, int64_t padding_idx) {
+  PROTECT(
+    torch::Tensor results__ = torch::_embedding_bag_per_sample_weights_backward(*tensor_ptr_from_ocaml(grad), *tensor_ptr_from_ocaml(weight), *tensor_ptr_from_ocaml(indices), *tensor_ptr_from_ocaml(offsets), *tensor_ptr_from_ocaml(offset2bag), mode, padding_idx);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__embedding_bag_per_sample_weights_backward_out(gc_tensor out, gc_tensor grad, gc_tensor weight, gc_tensor indices, gc_tensor offsets, gc_tensor offset2bag, int64_t mode, int64_t padding_idx) {
+  PROTECT(
+    torch::Tensor results__ = torch::_embedding_bag_per_sample_weights_backward_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(grad), *tensor_ptr_from_ocaml(weight), *tensor_ptr_from_ocaml(indices), *tensor_ptr_from_ocaml(offsets), *tensor_ptr_from_ocaml(offset2bag), mode, padding_idx);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__embedding_bag_sparse_backward(gc_tensor grad, gc_tensor indices, gc_tensor offsets, gc_tensor offset2bag, gc_tensor bag_size, int64_t num_weights, int scale_grad_by_freq, int64_t mode, gc_tensor per_sample_weights, int64_t padding_idx) {
+  PROTECT(
+    torch::Tensor results__ = torch::_embedding_bag_sparse_backward(*tensor_ptr_from_ocaml(grad), *tensor_ptr_from_ocaml(indices), *tensor_ptr_from_ocaml(offsets), *tensor_ptr_from_ocaml(offset2bag), *tensor_ptr_from_ocaml(bag_size), num_weights, (bool)scale_grad_by_freq, mode, (per_sample_weights ? tensor_from_ocaml(per_sample_weights) : torch::Tensor()), padding_idx);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__empty_affine_quantized(int64_t *size_data, int size_len, int options_kind, int options_device, double scale, int64_t zero_point) {
+  PROTECT(
+    torch::Tensor results__ = torch::_empty_affine_quantized(torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)), scale, zero_point);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__empty_affine_quantized_out(gc_tensor out, int64_t *size_data, int size_len, double scale, int64_t zero_point) {
+  PROTECT(
+    torch::Tensor results__ = torch::_empty_affine_quantized_out(*tensor_ptr_from_ocaml(out), torch::IntArrayRef(size_data, size_len), scale, zero_point);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__empty_per_channel_affine_quantized(int64_t *size_data, int size_len, gc_tensor scales, gc_tensor zero_points, int64_t axis, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::_empty_per_channel_affine_quantized(torch::IntArrayRef(size_data, size_len), *tensor_ptr_from_ocaml(scales), *tensor_ptr_from_ocaml(zero_points), axis, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__empty_per_channel_affine_quantized_out(gc_tensor out, int64_t *size_data, int size_len, gc_tensor scales, gc_tensor zero_points, int64_t axis) {
+  PROTECT(
+    torch::Tensor results__ = torch::_empty_per_channel_affine_quantized_out(*tensor_ptr_from_ocaml(out), torch::IntArrayRef(size_data, size_len), *tensor_ptr_from_ocaml(scales), *tensor_ptr_from_ocaml(zero_points), axis);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__euclidean_dist(gc_tensor x1, gc_tensor x2) {
+  PROTECT(
+    torch::Tensor results__ = torch::_euclidean_dist(*tensor_ptr_from_ocaml(x1), *tensor_ptr_from_ocaml(x2));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__euclidean_dist_out(gc_tensor out, gc_tensor x1, gc_tensor x2) {
+  PROTECT(
+    torch::Tensor results__ = torch::_euclidean_dist_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(x1), *tensor_ptr_from_ocaml(x2));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__fake_quantize_learnable_per_channel_affine(gc_tensor self, gc_tensor scale, gc_tensor zero_point, int64_t axis, int64_t quant_min, int64_t quant_max, double grad_factor) {
+  PROTECT(
+    torch::Tensor results__ = torch::_fake_quantize_learnable_per_channel_affine(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(scale), *tensor_ptr_from_ocaml(zero_point), axis, quant_min, quant_max, grad_factor);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg__fake_quantize_learnable_per_channel_affine_backward(raw_tensor *out__, gc_tensor grad, gc_tensor self, gc_tensor scale, gc_tensor zero_point, int64_t axis, int64_t quant_min, int64_t quant_max, double grad_factor) {
+  PROTECT(
+    auto results__ = torch::_fake_quantize_learnable_per_channel_affine_backward(*tensor_ptr_from_ocaml(grad), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(scale), *tensor_ptr_from_ocaml(zero_point), axis, quant_min, quant_max, grad_factor);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+raw_tensor atg__fake_quantize_learnable_per_channel_affine_out(gc_tensor out, gc_tensor self, gc_tensor scale, gc_tensor zero_point, int64_t axis, int64_t quant_min, int64_t quant_max, double grad_factor) {
+  PROTECT(
+    torch::Tensor results__ = torch::_fake_quantize_learnable_per_channel_affine_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(scale), *tensor_ptr_from_ocaml(zero_point), axis, quant_min, quant_max, grad_factor);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__fake_quantize_learnable_per_tensor_affine(gc_tensor self, gc_tensor scale, gc_tensor zero_point, int64_t quant_min, int64_t quant_max, double grad_factor) {
+  PROTECT(
+    torch::Tensor results__ = torch::_fake_quantize_learnable_per_tensor_affine(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(scale), *tensor_ptr_from_ocaml(zero_point), quant_min, quant_max, grad_factor);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg__fake_quantize_learnable_per_tensor_affine_backward(raw_tensor *out__, gc_tensor grad, gc_tensor self, gc_tensor scale, gc_tensor zero_point, int64_t quant_min, int64_t quant_max, double grad_factor) {
+  PROTECT(
+    auto results__ = torch::_fake_quantize_learnable_per_tensor_affine_backward(*tensor_ptr_from_ocaml(grad), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(scale), *tensor_ptr_from_ocaml(zero_point), quant_min, quant_max, grad_factor);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+raw_tensor atg__fake_quantize_learnable_per_tensor_affine_out(gc_tensor out, gc_tensor self, gc_tensor scale, gc_tensor zero_point, int64_t quant_min, int64_t quant_max, double grad_factor) {
+  PROTECT(
+    torch::Tensor results__ = torch::_fake_quantize_learnable_per_tensor_affine_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(scale), *tensor_ptr_from_ocaml(zero_point), quant_min, quant_max, grad_factor);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg__fake_quantize_per_tensor_affine_cachemask_tensor_qparams(raw_tensor *out__, gc_tensor self, gc_tensor scale, gc_tensor zero_point, gc_tensor fake_quant_enabled, int64_t quant_min, int64_t quant_max) {
+  PROTECT(
+    auto results__ = torch::_fake_quantize_per_tensor_affine_cachemask_tensor_qparams(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(scale), *tensor_ptr_from_ocaml(zero_point), *tensor_ptr_from_ocaml(fake_quant_enabled), quant_min, quant_max);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg__fake_quantize_per_tensor_affine_cachemask_tensor_qparams_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor self, gc_tensor scale, gc_tensor zero_point, gc_tensor fake_quant_enabled, int64_t quant_min, int64_t quant_max) {
+  PROTECT(
+    auto results__ = torch::_fake_quantize_per_tensor_affine_cachemask_tensor_qparams_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(scale), *tensor_ptr_from_ocaml(zero_point), *tensor_ptr_from_ocaml(fake_quant_enabled), quant_min, quant_max);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg__fft_c2c(gc_tensor self, int64_t *dim_data, int dim_len, int64_t normalization, int forward) {
+  PROTECT(
+    torch::Tensor results__ = torch::_fft_c2c(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(dim_data, dim_len), normalization, (bool)forward);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__fft_c2c_out(gc_tensor out, gc_tensor self, int64_t *dim_data, int dim_len, int64_t normalization, int forward) {
+  PROTECT(
+    torch::Tensor results__ = torch::_fft_c2c_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(dim_data, dim_len), normalization, (bool)forward);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__fft_c2r(gc_tensor self, int64_t *dim_data, int dim_len, int64_t normalization, int64_t last_dim_size) {
+  PROTECT(
+    torch::Tensor results__ = torch::_fft_c2r(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(dim_data, dim_len), normalization, last_dim_size);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__fft_c2r_out(gc_tensor out, gc_tensor self, int64_t *dim_data, int dim_len, int64_t normalization, int64_t last_dim_size) {
+  PROTECT(
+    torch::Tensor results__ = torch::_fft_c2r_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(dim_data, dim_len), normalization, last_dim_size);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__fft_r2c(gc_tensor self, int64_t *dim_data, int dim_len, int64_t normalization, int onesided) {
+  PROTECT(
+    torch::Tensor results__ = torch::_fft_r2c(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(dim_data, dim_len), normalization, (bool)onesided);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__fft_r2c_out(gc_tensor out, gc_tensor self, int64_t *dim_data, int dim_len, int64_t normalization, int onesided) {
+  PROTECT(
+    torch::Tensor results__ = torch::_fft_r2c_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(dim_data, dim_len), normalization, (bool)onesided);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__fill_mem_eff_dropout_mask_(gc_tensor self, double dropout_p, int64_t seed, int64_t offset) {
+  PROTECT(
+    torch::Tensor results__ = torch::_fill_mem_eff_dropout_mask_(*tensor_ptr_from_ocaml(self), dropout_p, seed, offset);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg__flash_attention_backward(raw_tensor *out__, gc_tensor grad_out, gc_tensor query, gc_tensor key, gc_tensor value, gc_tensor out, gc_tensor logsumexp, gc_tensor cum_seq_q, gc_tensor cum_seq_k, int64_t max_q, int64_t max_k, double dropout_p, int is_causal, gc_tensor philox_seed, gc_tensor philox_offset, double scale_v, int scale_null) {
+  PROTECT(
+    auto results__ = torch::_flash_attention_backward(*tensor_ptr_from_ocaml(grad_out), *tensor_ptr_from_ocaml(query), *tensor_ptr_from_ocaml(key), *tensor_ptr_from_ocaml(value), *tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(logsumexp), *tensor_ptr_from_ocaml(cum_seq_q), *tensor_ptr_from_ocaml(cum_seq_k), max_q, max_k, dropout_p, (bool)is_causal, *tensor_ptr_from_ocaml(philox_seed), *tensor_ptr_from_ocaml(philox_offset), scale_null ? c10::nullopt : c10::optional<double>(scale_v));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+raw_tensor atg__foobar(gc_tensor self, int arg1, int arg2, int arg3) {
+  PROTECT(
+    torch::Tensor results__ = torch::_foobar(*tensor_ptr_from_ocaml(self), (bool)arg1, (bool)arg2, (bool)arg3);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__foobar_out(gc_tensor out, gc_tensor self, int arg1, int arg2, int arg3) {
+  PROTECT(
+    torch::Tensor results__ = torch::_foobar_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), (bool)arg1, (bool)arg2, (bool)arg3);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__functional_assert_async(gc_tensor self, char * assert_msg, gc_tensor dep_token) {
+  PROTECT(
+    torch::Tensor results__ = torch::_functional_assert_async(*tensor_ptr_from_ocaml(self), std::string(assert_msg), *tensor_ptr_from_ocaml(dep_token));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__functional_sym_constrain_range(scalar size, int64_t min_v, int min_null, int64_t max_v, int max_null, gc_tensor dep_token) {
+  PROTECT(
+    torch::Tensor results__ = torch::_functional_sym_constrain_range(*size, min_null ? c10::nullopt : c10::optional<int64_t>(min_v), max_null ? c10::nullopt : c10::optional<int64_t>(max_v), *tensor_ptr_from_ocaml(dep_token));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__functional_sym_constrain_range_for_size(scalar size, int64_t min_v, int min_null, int64_t max_v, int max_null, gc_tensor dep_token) {
+  PROTECT(
+    torch::Tensor results__ = torch::_functional_sym_constrain_range_for_size(*size, min_null ? c10::nullopt : c10::optional<int64_t>(min_v), max_null ? c10::nullopt : c10::optional<int64_t>(max_v), *tensor_ptr_from_ocaml(dep_token));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg__fused_adam(gc_tensor *out_data, int out_len, gc_tensor *self_data, int self_len, gc_tensor *grads_data, int grads_len, gc_tensor *exp_avgs_data, int exp_avgs_len, gc_tensor *exp_avg_sqs_data, int exp_avg_sqs_len, gc_tensor *max_exp_avg_sqs_data, int max_exp_avg_sqs_len, gc_tensor *state_steps_data, int state_steps_len, double lr, double beta1, double beta2, double weight_decay, double eps, int amsgrad, int maximize, gc_tensor grad_scale, gc_tensor found_inf) {
+  PROTECT(
+    torch::_fused_adam_out(of_carray_tensor(out_data, out_len), of_carray_tensor(self_data, self_len), of_carray_tensor(grads_data, grads_len), of_carray_tensor(exp_avgs_data, exp_avgs_len), of_carray_tensor(exp_avg_sqs_data, exp_avg_sqs_len), of_carray_tensor(max_exp_avg_sqs_data, max_exp_avg_sqs_len), of_carray_tensor(state_steps_data, state_steps_len), lr, beta1, beta2, weight_decay, eps, (bool)amsgrad, (bool)maximize, (grad_scale ? tensor_from_ocaml(grad_scale) : torch::Tensor()), (found_inf ? tensor_from_ocaml(found_inf) : torch::Tensor()));
+  )
+}
+
+void atg__fused_adam_(gc_tensor *self_data, int self_len, gc_tensor *grads_data, int grads_len, gc_tensor *exp_avgs_data, int exp_avgs_len, gc_tensor *exp_avg_sqs_data, int exp_avg_sqs_len, gc_tensor *max_exp_avg_sqs_data, int max_exp_avg_sqs_len, gc_tensor *state_steps_data, int state_steps_len, double lr, double beta1, double beta2, double weight_decay, double eps, int amsgrad, int maximize, gc_tensor grad_scale, gc_tensor found_inf) {
+  PROTECT(
+    torch::_fused_adam_(of_carray_tensor(self_data, self_len), of_carray_tensor(grads_data, grads_len), of_carray_tensor(exp_avgs_data, exp_avgs_len), of_carray_tensor(exp_avg_sqs_data, exp_avg_sqs_len), of_carray_tensor(max_exp_avg_sqs_data, max_exp_avg_sqs_len), of_carray_tensor(state_steps_data, state_steps_len), lr, beta1, beta2, weight_decay, eps, (bool)amsgrad, (bool)maximize, (grad_scale ? tensor_from_ocaml(grad_scale) : torch::Tensor()), (found_inf ? tensor_from_ocaml(found_inf) : torch::Tensor()));
+  )
+}
+
+void atg__fused_adam_tensor_lr_(gc_tensor *self_data, int self_len, gc_tensor *grads_data, int grads_len, gc_tensor *exp_avgs_data, int exp_avgs_len, gc_tensor *exp_avg_sqs_data, int exp_avg_sqs_len, gc_tensor *max_exp_avg_sqs_data, int max_exp_avg_sqs_len, gc_tensor *state_steps_data, int state_steps_len, gc_tensor lr, double beta1, double beta2, double weight_decay, double eps, int amsgrad, int maximize, gc_tensor grad_scale, gc_tensor found_inf) {
+  PROTECT(
+    torch::_fused_adam_(of_carray_tensor(self_data, self_len), of_carray_tensor(grads_data, grads_len), of_carray_tensor(exp_avgs_data, exp_avgs_len), of_carray_tensor(exp_avg_sqs_data, exp_avg_sqs_len), of_carray_tensor(max_exp_avg_sqs_data, max_exp_avg_sqs_len), of_carray_tensor(state_steps_data, state_steps_len), *tensor_ptr_from_ocaml(lr), beta1, beta2, weight_decay, eps, (bool)amsgrad, (bool)maximize, (grad_scale ? tensor_from_ocaml(grad_scale) : torch::Tensor()), (found_inf ? tensor_from_ocaml(found_inf) : torch::Tensor()));
+  )
+}
+
+void atg__fused_adam_tensor_lr_out(gc_tensor *out_data, int out_len, gc_tensor *self_data, int self_len, gc_tensor *grads_data, int grads_len, gc_tensor *exp_avgs_data, int exp_avgs_len, gc_tensor *exp_avg_sqs_data, int exp_avg_sqs_len, gc_tensor *max_exp_avg_sqs_data, int max_exp_avg_sqs_len, gc_tensor *state_steps_data, int state_steps_len, gc_tensor lr, double beta1, double beta2, double weight_decay, double eps, int amsgrad, int maximize, gc_tensor grad_scale, gc_tensor found_inf) {
+  PROTECT(
+    torch::_fused_adam_out(of_carray_tensor(out_data, out_len), of_carray_tensor(self_data, self_len), of_carray_tensor(grads_data, grads_len), of_carray_tensor(exp_avgs_data, exp_avgs_len), of_carray_tensor(exp_avg_sqs_data, exp_avg_sqs_len), of_carray_tensor(max_exp_avg_sqs_data, max_exp_avg_sqs_len), of_carray_tensor(state_steps_data, state_steps_len), *tensor_ptr_from_ocaml(lr), beta1, beta2, weight_decay, eps, (bool)amsgrad, (bool)maximize, (grad_scale ? tensor_from_ocaml(grad_scale) : torch::Tensor()), (found_inf ? tensor_from_ocaml(found_inf) : torch::Tensor()));
+  )
+}
+
+void atg__fused_adamw(gc_tensor *out_data, int out_len, gc_tensor *self_data, int self_len, gc_tensor *grads_data, int grads_len, gc_tensor *exp_avgs_data, int exp_avgs_len, gc_tensor *exp_avg_sqs_data, int exp_avg_sqs_len, gc_tensor *max_exp_avg_sqs_data, int max_exp_avg_sqs_len, gc_tensor *state_steps_data, int state_steps_len, double lr, double beta1, double beta2, double weight_decay, double eps, int amsgrad, int maximize, gc_tensor grad_scale, gc_tensor found_inf) {
+  PROTECT(
+    torch::_fused_adamw_out(of_carray_tensor(out_data, out_len), of_carray_tensor(self_data, self_len), of_carray_tensor(grads_data, grads_len), of_carray_tensor(exp_avgs_data, exp_avgs_len), of_carray_tensor(exp_avg_sqs_data, exp_avg_sqs_len), of_carray_tensor(max_exp_avg_sqs_data, max_exp_avg_sqs_len), of_carray_tensor(state_steps_data, state_steps_len), lr, beta1, beta2, weight_decay, eps, (bool)amsgrad, (bool)maximize, (grad_scale ? tensor_from_ocaml(grad_scale) : torch::Tensor()), (found_inf ? tensor_from_ocaml(found_inf) : torch::Tensor()));
+  )
+}
+
+void atg__fused_adamw_(gc_tensor *self_data, int self_len, gc_tensor *grads_data, int grads_len, gc_tensor *exp_avgs_data, int exp_avgs_len, gc_tensor *exp_avg_sqs_data, int exp_avg_sqs_len, gc_tensor *max_exp_avg_sqs_data, int max_exp_avg_sqs_len, gc_tensor *state_steps_data, int state_steps_len, double lr, double beta1, double beta2, double weight_decay, double eps, int amsgrad, int maximize, gc_tensor grad_scale, gc_tensor found_inf) {
+  PROTECT(
+    torch::_fused_adamw_(of_carray_tensor(self_data, self_len), of_carray_tensor(grads_data, grads_len), of_carray_tensor(exp_avgs_data, exp_avgs_len), of_carray_tensor(exp_avg_sqs_data, exp_avg_sqs_len), of_carray_tensor(max_exp_avg_sqs_data, max_exp_avg_sqs_len), of_carray_tensor(state_steps_data, state_steps_len), lr, beta1, beta2, weight_decay, eps, (bool)amsgrad, (bool)maximize, (grad_scale ? tensor_from_ocaml(grad_scale) : torch::Tensor()), (found_inf ? tensor_from_ocaml(found_inf) : torch::Tensor()));
+  )
+}
+
+void atg__fused_adamw_tensor_lr_(gc_tensor *self_data, int self_len, gc_tensor *grads_data, int grads_len, gc_tensor *exp_avgs_data, int exp_avgs_len, gc_tensor *exp_avg_sqs_data, int exp_avg_sqs_len, gc_tensor *max_exp_avg_sqs_data, int max_exp_avg_sqs_len, gc_tensor *state_steps_data, int state_steps_len, gc_tensor lr, double beta1, double beta2, double weight_decay, double eps, int amsgrad, int maximize, gc_tensor grad_scale, gc_tensor found_inf) {
+  PROTECT(
+    torch::_fused_adamw_(of_carray_tensor(self_data, self_len), of_carray_tensor(grads_data, grads_len), of_carray_tensor(exp_avgs_data, exp_avgs_len), of_carray_tensor(exp_avg_sqs_data, exp_avg_sqs_len), of_carray_tensor(max_exp_avg_sqs_data, max_exp_avg_sqs_len), of_carray_tensor(state_steps_data, state_steps_len), *tensor_ptr_from_ocaml(lr), beta1, beta2, weight_decay, eps, (bool)amsgrad, (bool)maximize, (grad_scale ? tensor_from_ocaml(grad_scale) : torch::Tensor()), (found_inf ? tensor_from_ocaml(found_inf) : torch::Tensor()));
+  )
+}
+
+void atg__fused_adamw_tensor_lr_out(gc_tensor *out_data, int out_len, gc_tensor *self_data, int self_len, gc_tensor *grads_data, int grads_len, gc_tensor *exp_avgs_data, int exp_avgs_len, gc_tensor *exp_avg_sqs_data, int exp_avg_sqs_len, gc_tensor *max_exp_avg_sqs_data, int max_exp_avg_sqs_len, gc_tensor *state_steps_data, int state_steps_len, gc_tensor lr, double beta1, double beta2, double weight_decay, double eps, int amsgrad, int maximize, gc_tensor grad_scale, gc_tensor found_inf) {
+  PROTECT(
+    torch::_fused_adamw_out(of_carray_tensor(out_data, out_len), of_carray_tensor(self_data, self_len), of_carray_tensor(grads_data, grads_len), of_carray_tensor(exp_avgs_data, exp_avgs_len), of_carray_tensor(exp_avg_sqs_data, exp_avg_sqs_len), of_carray_tensor(max_exp_avg_sqs_data, max_exp_avg_sqs_len), of_carray_tensor(state_steps_data, state_steps_len), *tensor_ptr_from_ocaml(lr), beta1, beta2, weight_decay, eps, (bool)amsgrad, (bool)maximize, (grad_scale ? tensor_from_ocaml(grad_scale) : torch::Tensor()), (found_inf ? tensor_from_ocaml(found_inf) : torch::Tensor()));
+  )
+}
+
+void atg__fused_dropout(raw_tensor *out__, gc_tensor self, double p) {
+  PROTECT(
+    auto results__ = torch::_fused_dropout(*tensor_ptr_from_ocaml(self), p);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg__fused_dropout_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor self, double p) {
+  PROTECT(
+    auto results__ = torch::_fused_dropout_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(self), p);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg__fused_moving_avg_obs_fq_helper(raw_tensor *out__, gc_tensor self, gc_tensor observer_on, gc_tensor fake_quant_on, gc_tensor running_min, gc_tensor running_max, gc_tensor scale, gc_tensor zero_point, double averaging_const, int64_t quant_min, int64_t quant_max, int64_t ch_axis, int per_row_fake_quant, int symmetric_quant) {
+  PROTECT(
+    auto results__ = torch::_fused_moving_avg_obs_fq_helper(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(observer_on), *tensor_ptr_from_ocaml(fake_quant_on), *tensor_ptr_from_ocaml(running_min), *tensor_ptr_from_ocaml(running_max), *tensor_ptr_from_ocaml(scale), *tensor_ptr_from_ocaml(zero_point), averaging_const, quant_min, quant_max, ch_axis, (bool)per_row_fake_quant, (bool)symmetric_quant);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg__fused_moving_avg_obs_fq_helper_functional(raw_tensor *out__, gc_tensor self, gc_tensor observer_on, gc_tensor fake_quant_on, gc_tensor running_min, gc_tensor running_max, gc_tensor scale, gc_tensor zero_point, double averaging_const, int64_t quant_min, int64_t quant_max, int64_t ch_axis, int per_row_fake_quant, int symmetric_quant) {
+  PROTECT(
+    auto results__ = torch::_fused_moving_avg_obs_fq_helper_functional(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(observer_on), *tensor_ptr_from_ocaml(fake_quant_on), *tensor_ptr_from_ocaml(running_min), *tensor_ptr_from_ocaml(running_max), *tensor_ptr_from_ocaml(scale), *tensor_ptr_from_ocaml(zero_point), averaging_const, quant_min, quant_max, ch_axis, (bool)per_row_fake_quant, (bool)symmetric_quant);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+    out__[3] = tensor_to_ocaml(std::get<3>(results__));
+    out__[4] = tensor_to_ocaml(std::get<4>(results__));
+    out__[5] = tensor_to_ocaml(std::get<5>(results__));
+  )
+}
+
+void atg__fused_moving_avg_obs_fq_helper_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor self, gc_tensor observer_on, gc_tensor fake_quant_on, gc_tensor running_min, gc_tensor running_max, gc_tensor scale, gc_tensor zero_point, double averaging_const, int64_t quant_min, int64_t quant_max, int64_t ch_axis, int per_row_fake_quant, int symmetric_quant) {
+  PROTECT(
+    auto results__ = torch::_fused_moving_avg_obs_fq_helper_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(observer_on), *tensor_ptr_from_ocaml(fake_quant_on), *tensor_ptr_from_ocaml(running_min), *tensor_ptr_from_ocaml(running_max), *tensor_ptr_from_ocaml(scale), *tensor_ptr_from_ocaml(zero_point), averaging_const, quant_min, quant_max, ch_axis, (bool)per_row_fake_quant, (bool)symmetric_quant);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+int64_t atg__fused_sdp_choice(gc_tensor query, gc_tensor key, gc_tensor value, gc_tensor attn_mask, double dropout_p, int is_causal, double scale_v, int scale_null) {
+  PROTECT(
+    return torch::_fused_sdp_choice(*tensor_ptr_from_ocaml(query), *tensor_ptr_from_ocaml(key), *tensor_ptr_from_ocaml(value), (attn_mask ? tensor_from_ocaml(attn_mask) : torch::Tensor()), dropout_p, (bool)is_causal, scale_null ? c10::nullopt : c10::optional<double>(scale_v));
+  )
+  return 0;
+}
+
+raw_tensor atg__fw_primal(gc_tensor self, int64_t level) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->_fw_primal(level);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__fw_primal_copy(gc_tensor self, int64_t level) {
+  PROTECT(
+    torch::Tensor results__ = torch::_fw_primal_copy(*tensor_ptr_from_ocaml(self), level);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__fw_primal_copy_out(gc_tensor out, gc_tensor self, int64_t level) {
+  PROTECT(
+    torch::Tensor results__ = torch::_fw_primal_copy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), level);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__gather_sparse_backward(gc_tensor self, int64_t dim, gc_tensor index, gc_tensor grad) {
+  PROTECT(
+    torch::Tensor results__ = torch::_gather_sparse_backward(*tensor_ptr_from_ocaml(self), dim, *tensor_ptr_from_ocaml(index), *tensor_ptr_from_ocaml(grad));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__grid_sampler_2d_cpu_fallback(gc_tensor input, gc_tensor grid, int64_t interpolation_mode, int64_t padding_mode, int align_corners) {
+  PROTECT(
+    torch::Tensor results__ = torch::_grid_sampler_2d_cpu_fallback(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(grid), interpolation_mode, padding_mode, (bool)align_corners);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg__grid_sampler_2d_cpu_fallback_backward(raw_tensor *out__, gc_tensor grad_output, gc_tensor input, gc_tensor grid, int64_t interpolation_mode, int64_t padding_mode, int align_corners) {
+  PROTECT(
+    auto results__ = torch::_grid_sampler_2d_cpu_fallback_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(grid), interpolation_mode, padding_mode, (bool)align_corners);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg__grid_sampler_2d_cpu_fallback_out(gc_tensor out, gc_tensor input, gc_tensor grid, int64_t interpolation_mode, int64_t padding_mode, int align_corners) {
+  PROTECT(
+    torch::Tensor results__ = torch::_grid_sampler_2d_cpu_fallback_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(grid), interpolation_mode, padding_mode, (bool)align_corners);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+int atg__has_compatible_shallow_copy_type(gc_tensor self, gc_tensor from) {
+  PROTECT(
+    return torch::_has_compatible_shallow_copy_type(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(from));
+  )
+  return 0;
+}
+
+int atg__has_same_storage_numel(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    return torch::_has_same_storage_numel(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+  )
+  return 0;
+}
+
+raw_tensor *atg__histogramdd_bin_edges(gc_tensor self, int64_t *bins_data, int bins_len, double *range_data, int range_len, gc_tensor weight, int density) {
+  PROTECT(
+    auto results__ = torch::_histogramdd_bin_edges(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(bins_data, bins_len), at::ArrayRef<double>(range_data, range_len), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), (bool)density);
+    int sz = results__.size();
+    raw_tensor *out__ = (raw_tensor*)malloc((sz + 1) * sizeof(raw_tensor));
+    for (int i = 0; i < sz; ++i)
+      out__[i] = tensor_to_ocaml(results__[i]);
+    out__[sz] = nullptr;
+    return out__;
+  )
+}
+
+void atg__histogramdd_bin_edges_out(gc_tensor *out_data, int out_len, gc_tensor self, int64_t *bins_data, int bins_len, double *range_data, int range_len, gc_tensor weight, int density) {
+  PROTECT(
+    torch::_histogramdd_bin_edges_out(of_carray_tensor(out_data, out_len), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(bins_data, bins_len), at::ArrayRef<double>(range_data, range_len), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), (bool)density);
+  )
+}
+
+raw_tensor atg__histogramdd_from_bin_cts(gc_tensor self, int64_t *bins_data, int bins_len, double *range_data, int range_len, gc_tensor weight, int density) {
+  PROTECT(
+    torch::Tensor results__ = torch::_histogramdd_from_bin_cts(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(bins_data, bins_len), at::ArrayRef<double>(range_data, range_len), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), (bool)density);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__histogramdd_from_bin_cts_out(gc_tensor out, gc_tensor self, int64_t *bins_data, int bins_len, double *range_data, int range_len, gc_tensor weight, int density) {
+  PROTECT(
+    torch::Tensor results__ = torch::_histogramdd_from_bin_cts_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(bins_data, bins_len), at::ArrayRef<double>(range_data, range_len), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), (bool)density);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__histogramdd_from_bin_tensors(gc_tensor self, gc_tensor *bins_data, int bins_len, gc_tensor weight, int density) {
+  PROTECT(
+    torch::Tensor results__ = torch::_histogramdd_from_bin_tensors(*tensor_ptr_from_ocaml(self), of_carray_tensor(bins_data, bins_len), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), (bool)density);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__histogramdd_from_bin_tensors_out(gc_tensor out, gc_tensor self, gc_tensor *bins_data, int bins_len, gc_tensor weight, int density) {
+  PROTECT(
+    torch::Tensor results__ = torch::_histogramdd_from_bin_tensors_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), of_carray_tensor(bins_data, bins_len), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), (bool)density);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__index_put_impl(gc_tensor self, gc_tensor *indices_data, int indices_len, gc_tensor values, int accumulate, int unsafe) {
+  PROTECT(
+    torch::Tensor results__ = torch::_index_put_impl(*tensor_ptr_from_ocaml(self), of_carray_tensor_opt(indices_data, indices_len), *tensor_ptr_from_ocaml(values), (bool)accumulate, (bool)unsafe);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__index_put_impl_(gc_tensor self, gc_tensor *indices_data, int indices_len, gc_tensor values, int accumulate, int unsafe) {
+  PROTECT(
+    torch::Tensor results__ = torch::_index_put_impl_(*tensor_ptr_from_ocaml(self), of_carray_tensor_opt(indices_data, indices_len), *tensor_ptr_from_ocaml(values), (bool)accumulate, (bool)unsafe);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__index_put_impl_out(gc_tensor out, gc_tensor self, gc_tensor *indices_data, int indices_len, gc_tensor values, int accumulate, int unsafe) {
+  PROTECT(
+    torch::Tensor results__ = torch::_index_put_impl_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), of_carray_tensor_opt(indices_data, indices_len), *tensor_ptr_from_ocaml(values), (bool)accumulate, (bool)unsafe);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__indices(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->_indices();
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__indices_copy(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::_indices_copy(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__indices_copy_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::_indices_copy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__int_mm(gc_tensor self, gc_tensor mat2) {
+  PROTECT(
+    torch::Tensor results__ = torch::_int_mm(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(mat2));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__int_mm_out(gc_tensor out, gc_tensor self, gc_tensor mat2) {
+  PROTECT(
+    torch::Tensor results__ = torch::_int_mm_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(mat2));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__is_all_true(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::_is_all_true(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__is_any_true(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::_is_any_true(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+int atg__is_zerotensor(gc_tensor self) {
+  PROTECT(
+    return torch::_is_zerotensor(*tensor_ptr_from_ocaml(self));
+  )
+  return 0;
+}
+
+void atg__linalg_check_errors(gc_tensor info, char * api_name, int is_matrix) {
+  PROTECT(
+    torch::_linalg_check_errors(*tensor_ptr_from_ocaml(info), std::string(api_name), (bool)is_matrix);
+  )
+}
+
+void atg__linalg_det(raw_tensor *out__, gc_tensor A) {
+  PROTECT(
+    auto results__ = torch::_linalg_det(*tensor_ptr_from_ocaml(A));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+void atg__linalg_det_result(raw_tensor *out__, gc_tensor result, gc_tensor LU, gc_tensor pivots, gc_tensor A) {
+  PROTECT(
+    auto results__ = torch::_linalg_det_out(*tensor_ptr_from_ocaml(result), *tensor_ptr_from_ocaml(LU), *tensor_ptr_from_ocaml(pivots), *tensor_ptr_from_ocaml(A));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+void atg__linalg_eigh(raw_tensor *out__, gc_tensor A, char * UPLO, int compute_v) {
+  PROTECT(
+    auto results__ = torch::_linalg_eigh(*tensor_ptr_from_ocaml(A), std::string(UPLO), (bool)compute_v);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg__linalg_eigh_eigenvalues(raw_tensor *out__, gc_tensor eigenvalues, gc_tensor eigenvectors, gc_tensor A, char * UPLO, int compute_v) {
+  PROTECT(
+    auto results__ = torch::_linalg_eigh_out(*tensor_ptr_from_ocaml(eigenvalues), *tensor_ptr_from_ocaml(eigenvectors), *tensor_ptr_from_ocaml(A), std::string(UPLO), (bool)compute_v);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg__linalg_slogdet(raw_tensor *out__, gc_tensor A) {
+  PROTECT(
+    auto results__ = torch::_linalg_slogdet(*tensor_ptr_from_ocaml(A));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+    out__[3] = tensor_to_ocaml(std::get<3>(results__));
+  )
+}
+
+void atg__linalg_slogdet_sign(raw_tensor *out__, gc_tensor sign, gc_tensor logabsdet, gc_tensor LU, gc_tensor pivots, gc_tensor A) {
+  PROTECT(
+    auto results__ = torch::_linalg_slogdet_out(*tensor_ptr_from_ocaml(sign), *tensor_ptr_from_ocaml(logabsdet), *tensor_ptr_from_ocaml(LU), *tensor_ptr_from_ocaml(pivots), *tensor_ptr_from_ocaml(A));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+    out__[3] = tensor_to_ocaml(std::get<3>(results__));
+  )
+}
+
+void atg__linalg_solve_ex(raw_tensor *out__, gc_tensor A, gc_tensor B, int left, int check_errors) {
+  PROTECT(
+    auto results__ = torch::_linalg_solve_ex(*tensor_ptr_from_ocaml(A), *tensor_ptr_from_ocaml(B), (bool)left, (bool)check_errors);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+    out__[3] = tensor_to_ocaml(std::get<3>(results__));
+  )
+}
+
+void atg__linalg_solve_ex_result(raw_tensor *out__, gc_tensor result, gc_tensor LU, gc_tensor pivots, gc_tensor info, gc_tensor A, gc_tensor B, int left, int check_errors) {
+  PROTECT(
+    auto results__ = torch::_linalg_solve_ex_out(*tensor_ptr_from_ocaml(result), *tensor_ptr_from_ocaml(LU), *tensor_ptr_from_ocaml(pivots), *tensor_ptr_from_ocaml(info), *tensor_ptr_from_ocaml(A), *tensor_ptr_from_ocaml(B), (bool)left, (bool)check_errors);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+    out__[3] = tensor_to_ocaml(std::get<3>(results__));
+  )
+}
+
+void atg__linalg_svd(raw_tensor *out__, gc_tensor A, int full_matrices, int compute_uv, char * driver) {
+  PROTECT(
+    auto results__ = torch::_linalg_svd(*tensor_ptr_from_ocaml(A), (bool)full_matrices, (bool)compute_uv, std::string(driver));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+void atg__linalg_svd_u(raw_tensor *out__, gc_tensor U, gc_tensor S, gc_tensor Vh, gc_tensor A, int full_matrices, int compute_uv, char * driver) {
+  PROTECT(
+    auto results__ = torch::_linalg_svd_out(*tensor_ptr_from_ocaml(U), *tensor_ptr_from_ocaml(S), *tensor_ptr_from_ocaml(Vh), *tensor_ptr_from_ocaml(A), (bool)full_matrices, (bool)compute_uv, std::string(driver));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+raw_tensor atg__log_softmax(gc_tensor self, int64_t dim, int half_to_float) {
+  PROTECT(
+    torch::Tensor results__ = torch::_log_softmax(*tensor_ptr_from_ocaml(self), dim, (bool)half_to_float);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__log_softmax_backward_data(gc_tensor grad_output, gc_tensor output, int64_t dim, int input_dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::_log_softmax_backward_data(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(output), dim, torch::ScalarType(input_dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__log_softmax_backward_data_out(gc_tensor out, gc_tensor grad_output, gc_tensor output, int64_t dim, int input_dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::_log_softmax_backward_data_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(output), dim, torch::ScalarType(input_dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__log_softmax_out(gc_tensor out, gc_tensor self, int64_t dim, int half_to_float) {
+  PROTECT(
+    torch::Tensor results__ = torch::_log_softmax_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dim, (bool)half_to_float);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__logcumsumexp(gc_tensor self, int64_t dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::_logcumsumexp(*tensor_ptr_from_ocaml(self), dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__logcumsumexp_out(gc_tensor out, gc_tensor self, int64_t dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::_logcumsumexp_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg__lstm_mps(raw_tensor *out__, gc_tensor input, gc_tensor *hx_data, int hx_len, gc_tensor *params_data, int params_len, int has_biases, int64_t num_layers, double dropout, int train, int bidirectional, int batch_first) {
+  PROTECT(
+    auto results__ = torch::_lstm_mps(*tensor_ptr_from_ocaml(input), of_carray_tensor(hx_data, hx_len), of_carray_tensor(params_data, params_len), (bool)has_biases, num_layers, dropout, (bool)train, (bool)bidirectional, (bool)batch_first);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+    out__[3] = tensor_to_ocaml(std::get<3>(results__));
+    out__[4] = tensor_to_ocaml(std::get<4>(results__));
+    out__[5] = tensor_to_ocaml(std::get<5>(results__));
+  )
+}
+
+void atg__lstm_mps_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor out3, gc_tensor out4, gc_tensor out5, gc_tensor input, gc_tensor *hx_data, int hx_len, gc_tensor *params_data, int params_len, int has_biases, int64_t num_layers, double dropout, int train, int bidirectional, int batch_first) {
+  PROTECT(
+    auto results__ = torch::_lstm_mps_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(out2), *tensor_ptr_from_ocaml(out3), *tensor_ptr_from_ocaml(out4), *tensor_ptr_from_ocaml(out5), *tensor_ptr_from_ocaml(input), of_carray_tensor(hx_data, hx_len), of_carray_tensor(params_data, params_len), (bool)has_biases, num_layers, dropout, (bool)train, (bool)bidirectional, (bool)batch_first);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+    out__[3] = tensor_to_ocaml(std::get<3>(results__));
+    out__[4] = tensor_to_ocaml(std::get<4>(results__));
+    out__[5] = tensor_to_ocaml(std::get<5>(results__));
+  )
+}
+
+void atg__lu_with_info(raw_tensor *out__, gc_tensor self, int pivot, int check_errors) {
+  PROTECT(
+    auto results__ = torch::_lu_with_info(*tensor_ptr_from_ocaml(self), (bool)pivot, (bool)check_errors);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+raw_tensor atg__make_dep_token(int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::_make_dep_token(at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__make_dual(gc_tensor primal, gc_tensor tangent, int64_t level) {
+  PROTECT(
+    torch::Tensor results__ = torch::_make_dual(*tensor_ptr_from_ocaml(primal), *tensor_ptr_from_ocaml(tangent), level);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__make_dual_copy(gc_tensor primal, gc_tensor tangent, int64_t level) {
+  PROTECT(
+    torch::Tensor results__ = torch::_make_dual_copy(*tensor_ptr_from_ocaml(primal), *tensor_ptr_from_ocaml(tangent), level);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__make_dual_copy_out(gc_tensor out, gc_tensor primal, gc_tensor tangent, int64_t level) {
+  PROTECT(
+    torch::Tensor results__ = torch::_make_dual_copy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(primal), *tensor_ptr_from_ocaml(tangent), level);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__make_per_channel_quantized_tensor(gc_tensor self, gc_tensor scale, gc_tensor zero_point, int64_t axis) {
+  PROTECT(
+    torch::Tensor results__ = torch::_make_per_channel_quantized_tensor(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(scale), *tensor_ptr_from_ocaml(zero_point), axis);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__make_per_channel_quantized_tensor_out(gc_tensor out, gc_tensor self, gc_tensor scale, gc_tensor zero_point, int64_t axis) {
+  PROTECT(
+    torch::Tensor results__ = torch::_make_per_channel_quantized_tensor_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(scale), *tensor_ptr_from_ocaml(zero_point), axis);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__make_per_tensor_quantized_tensor(gc_tensor self, double scale, int64_t zero_point) {
+  PROTECT(
+    torch::Tensor results__ = torch::_make_per_tensor_quantized_tensor(*tensor_ptr_from_ocaml(self), scale, zero_point);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__make_per_tensor_quantized_tensor_out(gc_tensor out, gc_tensor self, double scale, int64_t zero_point) {
+  PROTECT(
+    torch::Tensor results__ = torch::_make_per_tensor_quantized_tensor_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), scale, zero_point);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__masked_scale(gc_tensor self, gc_tensor mask, double scale) {
+  PROTECT(
+    torch::Tensor results__ = torch::_masked_scale(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(mask), scale);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__masked_scale_out(gc_tensor out, gc_tensor self, gc_tensor mask, double scale) {
+  PROTECT(
+    torch::Tensor results__ = torch::_masked_scale_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(mask), scale);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__masked_softmax(gc_tensor self, gc_tensor mask, int64_t dim_v, int dim_null, int64_t mask_type_v, int mask_type_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::_masked_softmax(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(mask), dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v), mask_type_null ? c10::nullopt : c10::optional<int64_t>(mask_type_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__masked_softmax_backward(gc_tensor grad_output, gc_tensor output, gc_tensor mask, int64_t dim_v, int dim_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::_masked_softmax_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(output), *tensor_ptr_from_ocaml(mask), dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__masked_softmax_backward_out(gc_tensor out, gc_tensor grad_output, gc_tensor output, gc_tensor mask, int64_t dim_v, int dim_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::_masked_softmax_backward_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(output), *tensor_ptr_from_ocaml(mask), dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__masked_softmax_out(gc_tensor out, gc_tensor self, gc_tensor mask, int64_t dim_v, int dim_null, int64_t mask_type_v, int mask_type_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::_masked_softmax_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(mask), dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v), mask_type_null ? c10::nullopt : c10::optional<int64_t>(mask_type_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__mkldnn_reshape(gc_tensor self, int64_t *shape_data, int shape_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::_mkldnn_reshape(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(shape_data, shape_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__mkldnn_reshape_out(gc_tensor out, gc_tensor self, int64_t *shape_data, int shape_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::_mkldnn_reshape_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(shape_data, shape_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__mkldnn_transpose(gc_tensor self, int64_t dim0, int64_t dim1) {
+  PROTECT(
+    torch::Tensor results__ = torch::_mkldnn_transpose(*tensor_ptr_from_ocaml(self), dim0, dim1);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__mkldnn_transpose_(gc_tensor self, int64_t dim0, int64_t dim1) {
+  PROTECT(
+    torch::Tensor results__ = torch::_mkldnn_transpose_(*tensor_ptr_from_ocaml(self), dim0, dim1);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__mkldnn_transpose_out(gc_tensor out, gc_tensor self, int64_t dim0, int64_t dim1) {
+  PROTECT(
+    torch::Tensor results__ = torch::_mkldnn_transpose_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dim0, dim1);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__mps_convolution(gc_tensor self, gc_tensor weight, gc_tensor bias, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups) {
+  PROTECT(
+    torch::Tensor results__ = torch::_mps_convolution(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(weight), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(dilation_data, dilation_len), groups);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__mps_convolution_out(gc_tensor out, gc_tensor self, gc_tensor weight, gc_tensor bias, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups) {
+  PROTECT(
+    torch::Tensor results__ = torch::_mps_convolution_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(weight), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(dilation_data, dilation_len), groups);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__mps_convolution_transpose(gc_tensor self, gc_tensor weight, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups) {
+  PROTECT(
+    torch::Tensor results__ = torch::_mps_convolution_transpose(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(weight), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(output_padding_data, output_padding_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(dilation_data, dilation_len), groups);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__mps_convolution_transpose_out(gc_tensor out, gc_tensor self, gc_tensor weight, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups) {
+  PROTECT(
+    torch::Tensor results__ = torch::_mps_convolution_transpose_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(weight), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(output_padding_data, output_padding_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(dilation_data, dilation_len), groups);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg__native_batch_norm_legit(raw_tensor *out__, gc_tensor input, gc_tensor weight, gc_tensor bias, gc_tensor running_mean, gc_tensor running_var, int training, double momentum, double eps) {
+  PROTECT(
+    auto results__ = torch::_native_batch_norm_legit(*tensor_ptr_from_ocaml(input), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), *tensor_ptr_from_ocaml(running_mean), *tensor_ptr_from_ocaml(running_var), (bool)training, momentum, eps);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+void atg__native_batch_norm_legit_functional(raw_tensor *out__, gc_tensor input, gc_tensor weight, gc_tensor bias, gc_tensor running_mean, gc_tensor running_var, int training, double momentum, double eps) {
+  PROTECT(
+    auto results__ = torch::_native_batch_norm_legit_functional(*tensor_ptr_from_ocaml(input), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), *tensor_ptr_from_ocaml(running_mean), *tensor_ptr_from_ocaml(running_var), (bool)training, momentum, eps);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+    out__[3] = tensor_to_ocaml(std::get<3>(results__));
+    out__[4] = tensor_to_ocaml(std::get<4>(results__));
+  )
+}
+
+void atg__native_batch_norm_legit_no_stats(raw_tensor *out__, gc_tensor input, gc_tensor weight, gc_tensor bias, int training, double momentum, double eps) {
+  PROTECT(
+    auto results__ = torch::_native_batch_norm_legit(*tensor_ptr_from_ocaml(input), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), (bool)training, momentum, eps);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+void atg__native_batch_norm_legit_no_stats_out(raw_tensor *out__, gc_tensor out, gc_tensor save_mean, gc_tensor save_invstd, gc_tensor input, gc_tensor weight, gc_tensor bias, int training, double momentum, double eps) {
+  PROTECT(
+    auto results__ = torch::_native_batch_norm_legit_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(save_mean), *tensor_ptr_from_ocaml(save_invstd), *tensor_ptr_from_ocaml(input), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), (bool)training, momentum, eps);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+void atg__native_batch_norm_legit_no_training(raw_tensor *out__, gc_tensor input, gc_tensor weight, gc_tensor bias, gc_tensor running_mean, gc_tensor running_var, double momentum, double eps) {
+  PROTECT(
+    auto results__ = torch::_native_batch_norm_legit_no_training(*tensor_ptr_from_ocaml(input), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), *tensor_ptr_from_ocaml(running_mean), *tensor_ptr_from_ocaml(running_var), momentum, eps);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+void atg__native_batch_norm_legit_no_training_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor input, gc_tensor weight, gc_tensor bias, gc_tensor running_mean, gc_tensor running_var, double momentum, double eps) {
+  PROTECT(
+    auto results__ = torch::_native_batch_norm_legit_no_training_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(out2), *tensor_ptr_from_ocaml(input), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), *tensor_ptr_from_ocaml(running_mean), *tensor_ptr_from_ocaml(running_var), momentum, eps);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+void atg__native_batch_norm_legit_out(raw_tensor *out__, gc_tensor out, gc_tensor save_mean, gc_tensor save_invstd, gc_tensor input, gc_tensor weight, gc_tensor bias, gc_tensor running_mean, gc_tensor running_var, int training, double momentum, double eps) {
+  PROTECT(
+    auto results__ = torch::_native_batch_norm_legit_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(save_mean), *tensor_ptr_from_ocaml(save_invstd), *tensor_ptr_from_ocaml(input), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), *tensor_ptr_from_ocaml(running_mean), *tensor_ptr_from_ocaml(running_var), (bool)training, momentum, eps);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+void atg__native_multi_head_attention(raw_tensor *out__, gc_tensor query, gc_tensor key, gc_tensor value, int64_t embed_dim, int64_t num_head, gc_tensor qkv_weight, gc_tensor qkv_bias, gc_tensor proj_weight, gc_tensor proj_bias, gc_tensor mask, int need_weights, int average_attn_weights, int64_t mask_type_v, int mask_type_null) {
+  PROTECT(
+    auto results__ = torch::_native_multi_head_attention(*tensor_ptr_from_ocaml(query), *tensor_ptr_from_ocaml(key), *tensor_ptr_from_ocaml(value), embed_dim, num_head, *tensor_ptr_from_ocaml(qkv_weight), *tensor_ptr_from_ocaml(qkv_bias), *tensor_ptr_from_ocaml(proj_weight), *tensor_ptr_from_ocaml(proj_bias), (mask ? tensor_from_ocaml(mask) : torch::Tensor()), (bool)need_weights, (bool)average_attn_weights, mask_type_null ? c10::nullopt : c10::optional<int64_t>(mask_type_v));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg__native_multi_head_attention_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor query, gc_tensor key, gc_tensor value, int64_t embed_dim, int64_t num_head, gc_tensor qkv_weight, gc_tensor qkv_bias, gc_tensor proj_weight, gc_tensor proj_bias, gc_tensor mask, int need_weights, int average_attn_weights, int64_t mask_type_v, int mask_type_null) {
+  PROTECT(
+    auto results__ = torch::_native_multi_head_attention_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(query), *tensor_ptr_from_ocaml(key), *tensor_ptr_from_ocaml(value), embed_dim, num_head, *tensor_ptr_from_ocaml(qkv_weight), *tensor_ptr_from_ocaml(qkv_bias), *tensor_ptr_from_ocaml(proj_weight), *tensor_ptr_from_ocaml(proj_bias), (mask ? tensor_from_ocaml(mask) : torch::Tensor()), (bool)need_weights, (bool)average_attn_weights, mask_type_null ? c10::nullopt : c10::optional<int64_t>(mask_type_v));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg__neg_view(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::_neg_view(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__neg_view_copy(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::_neg_view_copy(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__neg_view_copy_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::_neg_view_copy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__nested_from_padded(gc_tensor padded, gc_tensor cpu_nested_shape_example, int fuse_transform_0213) {
+  PROTECT(
+    torch::Tensor results__ = torch::_nested_from_padded(*tensor_ptr_from_ocaml(padded), *tensor_ptr_from_ocaml(cpu_nested_shape_example), (bool)fuse_transform_0213);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__nested_from_padded_and_nested_example(gc_tensor padded, gc_tensor nt_example) {
+  PROTECT(
+    torch::Tensor results__ = torch::_nested_from_padded_and_nested_example(*tensor_ptr_from_ocaml(padded), *tensor_ptr_from_ocaml(nt_example));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__nested_from_padded_and_nested_example_out(gc_tensor out, gc_tensor padded, gc_tensor nt_example) {
+  PROTECT(
+    torch::Tensor results__ = torch::_nested_from_padded_and_nested_example_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(padded), *tensor_ptr_from_ocaml(nt_example));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__nested_from_padded_out(gc_tensor out, gc_tensor padded, gc_tensor cpu_nested_shape_example, int fuse_transform_0213) {
+  PROTECT(
+    torch::Tensor results__ = torch::_nested_from_padded_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(padded), *tensor_ptr_from_ocaml(cpu_nested_shape_example), (bool)fuse_transform_0213);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__nested_select_backward(gc_tensor grad_output, gc_tensor self, int64_t dim, int64_t index) {
+  PROTECT(
+    torch::Tensor results__ = torch::_nested_select_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), dim, index);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__nested_sum_backward(gc_tensor grad, gc_tensor self, int64_t *dim_data, int dim_len, int keepdim) {
+  PROTECT(
+    torch::Tensor results__ = torch::_nested_sum_backward(*tensor_ptr_from_ocaml(grad), *tensor_ptr_from_ocaml(self), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), (bool)keepdim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__nested_view_from_buffer(gc_tensor self, gc_tensor nested_size, gc_tensor nested_strides, gc_tensor offsets) {
+  PROTECT(
+    torch::Tensor results__ = torch::_nested_view_from_buffer(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(nested_size), *tensor_ptr_from_ocaml(nested_strides), *tensor_ptr_from_ocaml(offsets));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__nested_view_from_buffer_copy(gc_tensor self, gc_tensor nested_size, gc_tensor nested_strides, gc_tensor offsets) {
+  PROTECT(
+    torch::Tensor results__ = torch::_nested_view_from_buffer_copy(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(nested_size), *tensor_ptr_from_ocaml(nested_strides), *tensor_ptr_from_ocaml(offsets));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__nested_view_from_buffer_copy_out(gc_tensor out, gc_tensor self, gc_tensor nested_size, gc_tensor nested_strides, gc_tensor offsets) {
+  PROTECT(
+    torch::Tensor results__ = torch::_nested_view_from_buffer_copy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(nested_size), *tensor_ptr_from_ocaml(nested_strides), *tensor_ptr_from_ocaml(offsets));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__new_zeros_with_same_feature_meta(gc_tensor self, gc_tensor other, int64_t self_num_batch_dims) {
+  PROTECT(
+    torch::Tensor results__ = torch::_new_zeros_with_same_feature_meta(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other), self_num_batch_dims);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__new_zeros_with_same_feature_meta_out(gc_tensor out, gc_tensor self, gc_tensor other, int64_t self_num_batch_dims) {
+  PROTECT(
+    torch::Tensor results__ = torch::_new_zeros_with_same_feature_meta_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other), self_num_batch_dims);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+int atg__nnpack_available() {
+  PROTECT(
+    return torch::_nnpack_available();
+  )
+  return 0;
+}
+
+raw_tensor atg__nnpack_spatial_convolution(gc_tensor input, gc_tensor weight, gc_tensor bias, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::_nnpack_spatial_convolution(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(weight), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(stride_data, stride_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__nnpack_spatial_convolution_out(gc_tensor out, gc_tensor input, gc_tensor weight, gc_tensor bias, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::_nnpack_spatial_convolution_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(weight), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(stride_data, stride_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+int64_t atg__nnz(gc_tensor self) {
+  PROTECT(
+    return tensor_ptr_from_ocaml(self)->_nnz();
+  )
+  return 0;
+}
+
+void atg__pack_padded_sequence(raw_tensor *out__, gc_tensor input, gc_tensor lengths, int batch_first) {
+  PROTECT(
+    auto results__ = torch::_pack_padded_sequence(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(lengths), (bool)batch_first);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg__pack_padded_sequence_backward(gc_tensor grad, int64_t *input_size_data, int input_size_len, gc_tensor batch_sizes, int batch_first) {
+  PROTECT(
+    torch::Tensor results__ = torch::_pack_padded_sequence_backward(*tensor_ptr_from_ocaml(grad), torch::IntArrayRef(input_size_data, input_size_len), *tensor_ptr_from_ocaml(batch_sizes), (bool)batch_first);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg__pack_padded_sequence_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor input, gc_tensor lengths, int batch_first) {
+  PROTECT(
+    auto results__ = torch::_pack_padded_sequence_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(lengths), (bool)batch_first);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg__pad_circular(gc_tensor self, int64_t *pad_data, int pad_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::_pad_circular(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(pad_data, pad_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__pad_enum(gc_tensor self, int64_t *pad_data, int pad_len, int64_t mode, double value_v, int value_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::_pad_enum(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(pad_data, pad_len), mode, value_null ? c10::nullopt : c10::optional<double>(value_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg__pad_packed_sequence(raw_tensor *out__, gc_tensor data, gc_tensor batch_sizes, int batch_first, scalar padding_value, int64_t total_length) {
+  PROTECT(
+    auto results__ = torch::_pad_packed_sequence(*tensor_ptr_from_ocaml(data), *tensor_ptr_from_ocaml(batch_sizes), (bool)batch_first, *padding_value, total_length);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg__pdist_backward(gc_tensor grad, gc_tensor self, double p, gc_tensor pdist) {
+  PROTECT(
+    torch::Tensor results__ = torch::_pdist_backward(*tensor_ptr_from_ocaml(grad), *tensor_ptr_from_ocaml(self), p, *tensor_ptr_from_ocaml(pdist));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__pdist_backward_out(gc_tensor out, gc_tensor grad, gc_tensor self, double p, gc_tensor pdist) {
+  PROTECT(
+    torch::Tensor results__ = torch::_pdist_backward_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(grad), *tensor_ptr_from_ocaml(self), p, *tensor_ptr_from_ocaml(pdist));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__pin_memory(gc_tensor self, int device) {
+  PROTECT(
+    torch::Tensor results__ = torch::_pin_memory(*tensor_ptr_from_ocaml(self), device_of_int(device));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__pin_memory_out(gc_tensor out, gc_tensor self, int device) {
+  PROTECT(
+    torch::Tensor results__ = torch::_pin_memory_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), device_of_int(device));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__prelu_kernel(gc_tensor self, gc_tensor weight) {
+  PROTECT(
+    torch::Tensor results__ = torch::_prelu_kernel(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(weight));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg__prelu_kernel_backward(raw_tensor *out__, gc_tensor grad_output, gc_tensor self, gc_tensor weight) {
+  PROTECT(
+    auto results__ = torch::_prelu_kernel_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(weight));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg__propagate_xla_data(gc_tensor input, gc_tensor output) {
+  PROTECT(
+    torch::_propagate_xla_data(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(output));
+  )
+}
+
+raw_tensor atg__remove_batch_dim(gc_tensor self, int64_t level, int64_t batch_size, int64_t out_dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::_remove_batch_dim(*tensor_ptr_from_ocaml(self), level, batch_size, out_dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__reshape_alias(gc_tensor self, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::_reshape_alias(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(size_data, size_len), torch::IntArrayRef(stride_data, stride_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__reshape_alias_copy(gc_tensor self, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::_reshape_alias_copy(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(size_data, size_len), torch::IntArrayRef(stride_data, stride_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__reshape_alias_copy_out(gc_tensor out, gc_tensor self, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::_reshape_alias_copy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(size_data, size_len), torch::IntArrayRef(stride_data, stride_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__reshape_copy(gc_tensor self, int64_t *size_data, int size_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::_reshape_copy(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(size_data, size_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__reshape_from_tensor(gc_tensor self, gc_tensor shape) {
+  PROTECT(
+    torch::Tensor results__ = torch::_reshape_from_tensor(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(shape));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__resize_output(gc_tensor self, int64_t *size_data, int size_len, int device) {
+  PROTECT(
+    torch::Tensor results__ = torch::_resize_output(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(size_data, size_len), device_of_int(device));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__resize_output_(gc_tensor self, int64_t *size_data, int size_len, int device) {
+  PROTECT(
+    torch::Tensor results__ = torch::_resize_output_(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(size_data, size_len), device_of_int(device));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__resize_output_out(gc_tensor out, gc_tensor self, int64_t *size_data, int size_len, int device) {
+  PROTECT(
+    torch::Tensor results__ = torch::_resize_output_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(size_data, size_len), device_of_int(device));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg__rowwise_prune(raw_tensor *out__, gc_tensor weight, gc_tensor mask, int compressed_indices_dtype) {
+  PROTECT(
+    auto results__ = torch::_rowwise_prune(*tensor_ptr_from_ocaml(weight), *tensor_ptr_from_ocaml(mask), torch::ScalarType(compressed_indices_dtype));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg__sample_dirichlet(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sample_dirichlet(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__sample_dirichlet_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sample_dirichlet_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__saturate_weight_to_fp16(gc_tensor weight) {
+  PROTECT(
+    torch::Tensor results__ = torch::_saturate_weight_to_fp16(*tensor_ptr_from_ocaml(weight));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg__scaled_dot_product_attention_math(raw_tensor *out__, gc_tensor query, gc_tensor key, gc_tensor value, gc_tensor attn_mask, double dropout_p, int is_causal, gc_tensor dropout_mask, double scale_v, int scale_null) {
+  PROTECT(
+    auto results__ = torch::_scaled_dot_product_attention_math(*tensor_ptr_from_ocaml(query), *tensor_ptr_from_ocaml(key), *tensor_ptr_from_ocaml(value), (attn_mask ? tensor_from_ocaml(attn_mask) : torch::Tensor()), dropout_p, (bool)is_causal, (dropout_mask ? tensor_from_ocaml(dropout_mask) : torch::Tensor()), scale_null ? c10::nullopt : c10::optional<double>(scale_v));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg__scaled_dot_product_efficient_attention(raw_tensor *out__, gc_tensor query, gc_tensor key, gc_tensor value, gc_tensor attn_bias, int compute_log_sumexp, double dropout_p, int is_causal, double scale_v, int scale_null) {
+  PROTECT(
+    auto results__ = torch::_scaled_dot_product_efficient_attention(*tensor_ptr_from_ocaml(query), *tensor_ptr_from_ocaml(key), *tensor_ptr_from_ocaml(value), (attn_bias ? tensor_from_ocaml(attn_bias) : torch::Tensor()), (bool)compute_log_sumexp, dropout_p, (bool)is_causal, scale_null ? c10::nullopt : c10::optional<double>(scale_v));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+    out__[3] = tensor_to_ocaml(std::get<3>(results__));
+  )
+}
+
+void atg__scaled_dot_product_flash_attention_backward(raw_tensor *out__, gc_tensor grad_out, gc_tensor query, gc_tensor key, gc_tensor value, gc_tensor out, gc_tensor logsumexp, gc_tensor cum_seq_q, gc_tensor cum_seq_k, int64_t max_q, int64_t max_k, double dropout_p, int is_causal, gc_tensor philox_seed, gc_tensor philox_offset, double scale_v, int scale_null) {
+  PROTECT(
+    auto results__ = torch::_scaled_dot_product_flash_attention_backward(*tensor_ptr_from_ocaml(grad_out), *tensor_ptr_from_ocaml(query), *tensor_ptr_from_ocaml(key), *tensor_ptr_from_ocaml(value), *tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(logsumexp), *tensor_ptr_from_ocaml(cum_seq_q), *tensor_ptr_from_ocaml(cum_seq_k), max_q, max_k, dropout_p, (bool)is_causal, *tensor_ptr_from_ocaml(philox_seed), *tensor_ptr_from_ocaml(philox_offset), scale_null ? c10::nullopt : c10::optional<double>(scale_v));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+void atg__scaled_mm(raw_tensor *out__, gc_tensor self, gc_tensor mat2, gc_tensor bias, int out_dtype, gc_tensor scale_a, gc_tensor scale_b, gc_tensor scale_result) {
+  PROTECT(
+    auto results__ = torch::_scaled_mm(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(mat2), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::ScalarType(out_dtype), (scale_a ? tensor_from_ocaml(scale_a) : torch::Tensor()), (scale_b ? tensor_from_ocaml(scale_b) : torch::Tensor()), (scale_result ? tensor_from_ocaml(scale_result) : torch::Tensor()));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg__scaled_mm_out(raw_tensor *out__, gc_tensor out, gc_tensor out_amax, gc_tensor self, gc_tensor mat2, gc_tensor bias, int out_dtype, gc_tensor scale_a, gc_tensor scale_b, gc_tensor scale_result) {
+  PROTECT(
+    auto results__ = torch::_scaled_mm_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(out_amax), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(mat2), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::ScalarType(out_dtype), (scale_a ? tensor_from_ocaml(scale_a) : torch::Tensor()), (scale_b ? tensor_from_ocaml(scale_b) : torch::Tensor()), (scale_result ? tensor_from_ocaml(scale_result) : torch::Tensor()));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg__scatter_reduce(gc_tensor self, int64_t dim, gc_tensor index, gc_tensor src, char * reduce, int include_self) {
+  PROTECT(
+    torch::Tensor results__ = torch::scatter_reduce(*tensor_ptr_from_ocaml(self), dim, *tensor_ptr_from_ocaml(index), *tensor_ptr_from_ocaml(src), std::string(reduce), (bool)include_self);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__scatter_reduce_(gc_tensor self, int64_t dim, gc_tensor index, gc_tensor src, char * reduce, int include_self) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->scatter_reduce_(dim, *tensor_ptr_from_ocaml(index), *tensor_ptr_from_ocaml(src), std::string(reduce), (bool)include_self);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__scatter_reduce_two_out(gc_tensor out, gc_tensor self, int64_t dim, gc_tensor index, gc_tensor src, char * reduce, int include_self) {
+  PROTECT(
+    torch::Tensor results__ = torch::scatter_reduce_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dim, *tensor_ptr_from_ocaml(index), *tensor_ptr_from_ocaml(src), std::string(reduce), (bool)include_self);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__segment_reduce_backward(gc_tensor grad, gc_tensor output, gc_tensor data, char * reduce, gc_tensor lengths, gc_tensor offsets, int64_t axis, scalar initial) {
+  PROTECT(
+    torch::Tensor results__ = torch::_segment_reduce_backward(*tensor_ptr_from_ocaml(grad), *tensor_ptr_from_ocaml(output), *tensor_ptr_from_ocaml(data), std::string(reduce), (lengths ? tensor_from_ocaml(lengths) : torch::Tensor()), (offsets ? tensor_from_ocaml(offsets) : torch::Tensor()), axis, *initial);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__segment_reduce_backward_out(gc_tensor out, gc_tensor grad, gc_tensor output, gc_tensor data, char * reduce, gc_tensor lengths, gc_tensor offsets, int64_t axis, scalar initial) {
+  PROTECT(
+    torch::Tensor results__ = torch::_segment_reduce_backward_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(grad), *tensor_ptr_from_ocaml(output), *tensor_ptr_from_ocaml(data), std::string(reduce), (lengths ? tensor_from_ocaml(lengths) : torch::Tensor()), (offsets ? tensor_from_ocaml(offsets) : torch::Tensor()), axis, *initial);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__shape_as_tensor(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::_shape_as_tensor(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg__slow_conv2d_backward(raw_tensor *out__, gc_tensor grad_input, gc_tensor grad_weight, gc_tensor grad_bias, gc_tensor grad_output, gc_tensor self, gc_tensor weight, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len) {
+  PROTECT(
+    auto results__ = torch::_slow_conv2d_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_weight), *tensor_ptr_from_ocaml(grad_bias), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(weight), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+void atg__sobol_engine_draw(raw_tensor *out__, gc_tensor quasi, int64_t n, gc_tensor sobolstate, int64_t dimension, int64_t num_generated, int dtype) {
+  PROTECT(
+    auto results__ = torch::_sobol_engine_draw(*tensor_ptr_from_ocaml(quasi), n, *tensor_ptr_from_ocaml(sobolstate), dimension, num_generated, torch::ScalarType(dtype));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg__sobol_engine_ff_(gc_tensor self, int64_t n, gc_tensor sobolstate, int64_t dimension, int64_t num_generated) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sobol_engine_ff_(*tensor_ptr_from_ocaml(self), n, *tensor_ptr_from_ocaml(sobolstate), dimension, num_generated);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__sobol_engine_initialize_state_(gc_tensor self, int64_t dimension) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sobol_engine_initialize_state_(*tensor_ptr_from_ocaml(self), dimension);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__sobol_engine_scramble_(gc_tensor self, gc_tensor ltm, int64_t dimension) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sobol_engine_scramble_(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(ltm), dimension);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__softmax(gc_tensor self, int64_t dim, int half_to_float) {
+  PROTECT(
+    torch::Tensor results__ = torch::_softmax(*tensor_ptr_from_ocaml(self), dim, (bool)half_to_float);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__softmax_backward_data(gc_tensor grad_output, gc_tensor output, int64_t dim, int input_dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::_softmax_backward_data(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(output), dim, torch::ScalarType(input_dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__softmax_backward_data_out(gc_tensor grad_input, gc_tensor grad_output, gc_tensor output, int64_t dim, int input_dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::_softmax_backward_data_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(output), dim, torch::ScalarType(input_dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__softmax_out(gc_tensor out, gc_tensor self, int64_t dim, int half_to_float) {
+  PROTECT(
+    torch::Tensor results__ = torch::_softmax_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dim, (bool)half_to_float);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__sparse_addmm(gc_tensor self, gc_tensor mat1, gc_tensor mat2) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sparse_addmm(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(mat1), *tensor_ptr_from_ocaml(mat2));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__sparse_addmm_out(gc_tensor out, gc_tensor self, gc_tensor mat1, gc_tensor mat2) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sparse_addmm_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(mat1), *tensor_ptr_from_ocaml(mat2));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__sparse_broadcast_to(gc_tensor self, int64_t *size_data, int size_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sparse_broadcast_to(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(size_data, size_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__sparse_broadcast_to_copy(gc_tensor self, int64_t *size_data, int size_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sparse_broadcast_to_copy(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(size_data, size_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__sparse_broadcast_to_copy_out(gc_tensor out, gc_tensor self, int64_t *size_data, int size_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sparse_broadcast_to_copy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(size_data, size_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__sparse_bsc_tensor_unsafe(gc_tensor ccol_indices, gc_tensor row_indices, gc_tensor values, int64_t *size_data, int size_len, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sparse_bsc_tensor_unsafe(*tensor_ptr_from_ocaml(ccol_indices), *tensor_ptr_from_ocaml(row_indices), *tensor_ptr_from_ocaml(values), torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__sparse_bsr_tensor_unsafe(gc_tensor crow_indices, gc_tensor col_indices, gc_tensor values, int64_t *size_data, int size_len, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sparse_bsr_tensor_unsafe(*tensor_ptr_from_ocaml(crow_indices), *tensor_ptr_from_ocaml(col_indices), *tensor_ptr_from_ocaml(values), torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__sparse_compressed_tensor_unsafe(gc_tensor compressed_indices, gc_tensor plain_indices, gc_tensor values, int64_t *size_data, int size_len, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sparse_compressed_tensor_unsafe(*tensor_ptr_from_ocaml(compressed_indices), *tensor_ptr_from_ocaml(plain_indices), *tensor_ptr_from_ocaml(values), torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__sparse_coo_tensor_unsafe(gc_tensor indices, gc_tensor values, int64_t *size_data, int size_len, int options_kind, int options_device, int is_coalesced) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sparse_coo_tensor_unsafe(*tensor_ptr_from_ocaml(indices), *tensor_ptr_from_ocaml(values), torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)), (bool)is_coalesced);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__sparse_coo_tensor_with_dims(int64_t sparse_dim, int64_t dense_dim, int64_t *size_data, int size_len, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sparse_coo_tensor_with_dims(sparse_dim, dense_dim, torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__sparse_coo_tensor_with_dims_and_tensors(int64_t sparse_dim, int64_t dense_dim, int64_t *size_data, int size_len, gc_tensor indices, gc_tensor values, int options_kind, int options_device, int is_coalesced) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sparse_coo_tensor_with_dims_and_tensors(sparse_dim, dense_dim, torch::IntArrayRef(size_data, size_len), *tensor_ptr_from_ocaml(indices), *tensor_ptr_from_ocaml(values), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)), (bool)is_coalesced);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__sparse_coo_tensor_with_dims_and_tensors_out(gc_tensor out, int64_t sparse_dim, int64_t dense_dim, int64_t *size_data, int size_len, gc_tensor indices, gc_tensor values, int is_coalesced) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sparse_coo_tensor_with_dims_and_tensors_out(*tensor_ptr_from_ocaml(out), sparse_dim, dense_dim, torch::IntArrayRef(size_data, size_len), *tensor_ptr_from_ocaml(indices), *tensor_ptr_from_ocaml(values), (bool)is_coalesced);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__sparse_coo_tensor_with_dims_out(gc_tensor out, int64_t sparse_dim, int64_t dense_dim, int64_t *size_data, int size_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sparse_coo_tensor_with_dims_out(*tensor_ptr_from_ocaml(out), sparse_dim, dense_dim, torch::IntArrayRef(size_data, size_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__sparse_csc_tensor_unsafe(gc_tensor ccol_indices, gc_tensor row_indices, gc_tensor values, int64_t *size_data, int size_len, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sparse_csc_tensor_unsafe(*tensor_ptr_from_ocaml(ccol_indices), *tensor_ptr_from_ocaml(row_indices), *tensor_ptr_from_ocaml(values), torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__sparse_csr_prod(gc_tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sparse_csr_prod(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(dim_data, dim_len), (bool)keepdim, torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__sparse_csr_prod_dim_dtype_out(gc_tensor out, gc_tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sparse_csr_prod_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(dim_data, dim_len), (bool)keepdim, torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__sparse_csr_sum(gc_tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sparse_csr_sum(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(dim_data, dim_len), (bool)keepdim, torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__sparse_csr_sum_dim_dtype_out(gc_tensor out, gc_tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sparse_csr_sum_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(dim_data, dim_len), (bool)keepdim, torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__sparse_csr_tensor_unsafe(gc_tensor crow_indices, gc_tensor col_indices, gc_tensor values, int64_t *size_data, int size_len, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sparse_csr_tensor_unsafe(*tensor_ptr_from_ocaml(crow_indices), *tensor_ptr_from_ocaml(col_indices), *tensor_ptr_from_ocaml(values), torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__sparse_log_softmax(gc_tensor self, int64_t dim, int half_to_float) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sparse_log_softmax(*tensor_ptr_from_ocaml(self), dim, (bool)half_to_float);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__sparse_log_softmax_backward_data(gc_tensor grad_output, gc_tensor output, int64_t dim, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sparse_log_softmax_backward_data(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(output), dim, *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__sparse_log_softmax_backward_data_out(gc_tensor out, gc_tensor grad_output, gc_tensor output, int64_t dim, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sparse_log_softmax_backward_data_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(output), dim, *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__sparse_log_softmax_int(gc_tensor self, int64_t dim, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sparse_log_softmax(*tensor_ptr_from_ocaml(self), dim, torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__sparse_log_softmax_out(gc_tensor out, gc_tensor self, int64_t dim, int half_to_float) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sparse_log_softmax_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dim, (bool)half_to_float);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__sparse_mask_projection(gc_tensor self, gc_tensor mask, int accumulate_matches) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->_sparse_mask_projection(*tensor_ptr_from_ocaml(mask), (bool)accumulate_matches);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__sparse_mask_projection_out(gc_tensor out, gc_tensor self, gc_tensor mask, int accumulate_matches) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sparse_mask_projection_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(mask), (bool)accumulate_matches);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__sparse_mm(gc_tensor sparse, gc_tensor dense) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sparse_mm(*tensor_ptr_from_ocaml(sparse), *tensor_ptr_from_ocaml(dense));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__sparse_mm_reduce(gc_tensor sparse, gc_tensor dense, char * reduce) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sparse_mm(*tensor_ptr_from_ocaml(sparse), *tensor_ptr_from_ocaml(dense), std::string(reduce));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg__sparse_mm_reduce_impl(raw_tensor *out__, gc_tensor self, gc_tensor other, char * reduce) {
+  PROTECT(
+    auto results__ = torch::_sparse_mm_reduce_impl(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other), std::string(reduce));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg__sparse_semi_structured_linear(gc_tensor input, gc_tensor weight, gc_tensor meta, gc_tensor bias, char * activation) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sparse_semi_structured_linear(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(weight), *tensor_ptr_from_ocaml(meta), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), std::string(activation));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__sparse_softmax(gc_tensor self, int64_t dim, int half_to_float) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sparse_softmax(*tensor_ptr_from_ocaml(self), dim, (bool)half_to_float);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__sparse_softmax_backward_data(gc_tensor grad_output, gc_tensor output, int64_t dim, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sparse_softmax_backward_data(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(output), dim, *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__sparse_softmax_backward_data_out(gc_tensor out, gc_tensor grad_output, gc_tensor output, int64_t dim, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sparse_softmax_backward_data_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(output), dim, *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__sparse_softmax_int(gc_tensor self, int64_t dim, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sparse_softmax(*tensor_ptr_from_ocaml(self), dim, torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__sparse_softmax_out(gc_tensor out, gc_tensor self, int64_t dim, int half_to_float) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sparse_softmax_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dim, (bool)half_to_float);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__sparse_sparse_matmul(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sparse_sparse_matmul(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__sparse_sparse_matmul_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sparse_sparse_matmul_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__sparse_sum(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sparse_sum(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__sparse_sum_backward(gc_tensor grad, gc_tensor self, int64_t *dim_data, int dim_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sparse_sum_backward(*tensor_ptr_from_ocaml(grad), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(dim_data, dim_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__sparse_sum_backward_out(gc_tensor out, gc_tensor grad, gc_tensor self, int64_t *dim_data, int dim_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sparse_sum_backward_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(grad), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(dim_data, dim_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__sparse_sum_dim(gc_tensor self, int64_t *dim_data, int dim_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sparse_sum(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(dim_data, dim_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__sparse_sum_dim_dtype(gc_tensor self, int64_t *dim_data, int dim_len, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sparse_sum(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(dim_data, dim_len), torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__sparse_sum_dim_out(gc_tensor out, gc_tensor self, int64_t *dim_data, int dim_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sparse_sum_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(dim_data, dim_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__sparse_sum_dtype(gc_tensor self, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::_sparse_sum(*tensor_ptr_from_ocaml(self), torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__spdiags(gc_tensor diagonals, gc_tensor offsets, int64_t *shape_data, int shape_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::_spdiags(*tensor_ptr_from_ocaml(diagonals), *tensor_ptr_from_ocaml(offsets), torch::IntArrayRef(shape_data, shape_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__spdiags_out(gc_tensor out, gc_tensor diagonals, gc_tensor offsets, int64_t *shape_data, int shape_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::_spdiags_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(diagonals), *tensor_ptr_from_ocaml(offsets), torch::IntArrayRef(shape_data, shape_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__stack(gc_tensor *tensors_data, int tensors_len, int64_t dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::_stack(of_carray_tensor(tensors_data, tensors_len), dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__stack_out(gc_tensor out, gc_tensor *tensors_data, int tensors_len, int64_t dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::_stack_out(*tensor_ptr_from_ocaml(out), of_carray_tensor(tensors_data, tensors_len), dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__standard_gamma(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::_standard_gamma(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__standard_gamma_grad(gc_tensor self, gc_tensor output) {
+  PROTECT(
+    torch::Tensor results__ = torch::_standard_gamma_grad(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(output));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__standard_gamma_grad_out(gc_tensor out, gc_tensor self, gc_tensor output) {
+  PROTECT(
+    torch::Tensor results__ = torch::_standard_gamma_grad_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(output));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__standard_gamma_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::_standard_gamma_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__test_ambiguous_defaults(gc_tensor dummy, int64_t a, int64_t b) {
+  PROTECT(
+    torch::Tensor results__ = torch::_test_ambiguous_defaults(*tensor_ptr_from_ocaml(dummy), a, b);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__test_ambiguous_defaults_b(gc_tensor dummy, int64_t a, char * b) {
+  PROTECT(
+    torch::Tensor results__ = torch::_test_ambiguous_defaults(*tensor_ptr_from_ocaml(dummy), a, std::string(b));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__test_autograd_multiple_dispatch(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::_test_autograd_multiple_dispatch(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__test_autograd_multiple_dispatch_fullcoverage_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::_test_autograd_multiple_dispatch_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__test_autograd_multiple_dispatch_ntonly(gc_tensor self, int b) {
+  PROTECT(
+    torch::Tensor results__ = torch::_test_autograd_multiple_dispatch(*tensor_ptr_from_ocaml(self), (bool)b);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__test_autograd_multiple_dispatch_view(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::_test_autograd_multiple_dispatch_view(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__test_autograd_multiple_dispatch_view_copy(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::_test_autograd_multiple_dispatch_view_copy(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__test_autograd_multiple_dispatch_view_copy_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::_test_autograd_multiple_dispatch_view_copy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__test_check_tensor(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::_test_check_tensor(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__test_functorch_fallback(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::_test_functorch_fallback(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__test_functorch_fallback_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::_test_functorch_fallback_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__test_optional_filled_intlist(gc_tensor values, int64_t *addends_data, int addends_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::_test_optional_filled_intlist(*tensor_ptr_from_ocaml(values), addends_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(addends_data, addends_len)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__test_optional_filled_intlist_out(gc_tensor out, gc_tensor values, int64_t *addends_data, int addends_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::_test_optional_filled_intlist_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(values), addends_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(addends_data, addends_len)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__test_optional_floatlist(gc_tensor values, double *addends_data, int addends_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::_test_optional_floatlist(*tensor_ptr_from_ocaml(values), at::ArrayRef<double>(addends_data, addends_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__test_optional_floatlist_out(gc_tensor out, gc_tensor values, double *addends_data, int addends_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::_test_optional_floatlist_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(values), at::ArrayRef<double>(addends_data, addends_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__test_optional_intlist(gc_tensor values, int64_t *addends_data, int addends_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::_test_optional_intlist(*tensor_ptr_from_ocaml(values), addends_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(addends_data, addends_len)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__test_optional_intlist_out(gc_tensor out, gc_tensor values, int64_t *addends_data, int addends_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::_test_optional_intlist_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(values), addends_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(addends_data, addends_len)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__test_serialization_subcmul(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::_test_serialization_subcmul(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__test_string_default(gc_tensor dummy, char * a, char * b) {
+  PROTECT(
+    torch::Tensor results__ = torch::_test_string_default(*tensor_ptr_from_ocaml(dummy), std::string(a), std::string(b));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__test_warn_in_autograd(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::_test_warn_in_autograd(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__test_warn_in_autograd_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::_test_warn_in_autograd_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg__thnn_differentiable_gru_cell_backward(raw_tensor *out__, gc_tensor grad_hy, gc_tensor input_gates, gc_tensor hidden_gates, gc_tensor hx, gc_tensor input_bias, gc_tensor hidden_bias) {
+  PROTECT(
+    auto results__ = torch::_thnn_differentiable_gru_cell_backward(*tensor_ptr_from_ocaml(grad_hy), *tensor_ptr_from_ocaml(input_gates), *tensor_ptr_from_ocaml(hidden_gates), *tensor_ptr_from_ocaml(hx), (input_bias ? tensor_from_ocaml(input_bias) : torch::Tensor()), (hidden_bias ? tensor_from_ocaml(hidden_bias) : torch::Tensor()));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+    out__[3] = tensor_to_ocaml(std::get<3>(results__));
+    out__[4] = tensor_to_ocaml(std::get<4>(results__));
+  )
+}
+
+void atg__thnn_differentiable_lstm_cell_backward(raw_tensor *out__, gc_tensor grad_hy, gc_tensor grad_cy, gc_tensor input_gates, gc_tensor hidden_gates, gc_tensor input_bias, gc_tensor hidden_bias, gc_tensor cx, gc_tensor cy) {
+  PROTECT(
+    auto results__ = torch::_thnn_differentiable_lstm_cell_backward((grad_hy ? tensor_from_ocaml(grad_hy) : torch::Tensor()), (grad_cy ? tensor_from_ocaml(grad_cy) : torch::Tensor()), *tensor_ptr_from_ocaml(input_gates), *tensor_ptr_from_ocaml(hidden_gates), (input_bias ? tensor_from_ocaml(input_bias) : torch::Tensor()), (hidden_bias ? tensor_from_ocaml(hidden_bias) : torch::Tensor()), *tensor_ptr_from_ocaml(cx), *tensor_ptr_from_ocaml(cy));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+    out__[3] = tensor_to_ocaml(std::get<3>(results__));
+    out__[4] = tensor_to_ocaml(std::get<4>(results__));
+  )
+}
+
+void atg__thnn_fused_gru_cell(raw_tensor *out__, gc_tensor input_gates, gc_tensor hidden_gates, gc_tensor hx, gc_tensor input_bias, gc_tensor hidden_bias) {
+  PROTECT(
+    auto results__ = torch::_thnn_fused_gru_cell(*tensor_ptr_from_ocaml(input_gates), *tensor_ptr_from_ocaml(hidden_gates), *tensor_ptr_from_ocaml(hx), (input_bias ? tensor_from_ocaml(input_bias) : torch::Tensor()), (hidden_bias ? tensor_from_ocaml(hidden_bias) : torch::Tensor()));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg__thnn_fused_gru_cell_backward(raw_tensor *out__, gc_tensor grad_hy, gc_tensor workspace, int has_bias) {
+  PROTECT(
+    auto results__ = torch::_thnn_fused_gru_cell_backward(*tensor_ptr_from_ocaml(grad_hy), *tensor_ptr_from_ocaml(workspace), (bool)has_bias);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+    out__[3] = tensor_to_ocaml(std::get<3>(results__));
+    out__[4] = tensor_to_ocaml(std::get<4>(results__));
+  )
+}
+
+void atg__thnn_fused_gru_cell_backward_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor out3, gc_tensor out4, gc_tensor grad_hy, gc_tensor workspace, int has_bias) {
+  PROTECT(
+    auto results__ = torch::_thnn_fused_gru_cell_backward_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(out2), *tensor_ptr_from_ocaml(out3), *tensor_ptr_from_ocaml(out4), *tensor_ptr_from_ocaml(grad_hy), *tensor_ptr_from_ocaml(workspace), (bool)has_bias);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+    out__[3] = tensor_to_ocaml(std::get<3>(results__));
+    out__[4] = tensor_to_ocaml(std::get<4>(results__));
+  )
+}
+
+void atg__thnn_fused_gru_cell_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor input_gates, gc_tensor hidden_gates, gc_tensor hx, gc_tensor input_bias, gc_tensor hidden_bias) {
+  PROTECT(
+    auto results__ = torch::_thnn_fused_gru_cell_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(input_gates), *tensor_ptr_from_ocaml(hidden_gates), *tensor_ptr_from_ocaml(hx), (input_bias ? tensor_from_ocaml(input_bias) : torch::Tensor()), (hidden_bias ? tensor_from_ocaml(hidden_bias) : torch::Tensor()));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg__thnn_fused_lstm_cell(raw_tensor *out__, gc_tensor input_gates, gc_tensor hidden_gates, gc_tensor cx, gc_tensor input_bias, gc_tensor hidden_bias) {
+  PROTECT(
+    auto results__ = torch::_thnn_fused_lstm_cell(*tensor_ptr_from_ocaml(input_gates), *tensor_ptr_from_ocaml(hidden_gates), *tensor_ptr_from_ocaml(cx), (input_bias ? tensor_from_ocaml(input_bias) : torch::Tensor()), (hidden_bias ? tensor_from_ocaml(hidden_bias) : torch::Tensor()));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+void atg__thnn_fused_lstm_cell_backward(raw_tensor *out__, gc_tensor grad_hy, gc_tensor grad_cy, gc_tensor cx, gc_tensor cy, gc_tensor workspace, int has_bias) {
+  PROTECT(
+    auto results__ = torch::_thnn_fused_lstm_cell_backward((grad_hy ? tensor_from_ocaml(grad_hy) : torch::Tensor()), (grad_cy ? tensor_from_ocaml(grad_cy) : torch::Tensor()), *tensor_ptr_from_ocaml(cx), *tensor_ptr_from_ocaml(cy), *tensor_ptr_from_ocaml(workspace), (bool)has_bias);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+    out__[3] = tensor_to_ocaml(std::get<3>(results__));
+    out__[4] = tensor_to_ocaml(std::get<4>(results__));
+  )
+}
+
+void atg__thnn_fused_lstm_cell_backward_impl(raw_tensor *out__, gc_tensor grad_hy, gc_tensor grad_cy, gc_tensor cx, gc_tensor cy, gc_tensor workspace, int has_bias) {
+  PROTECT(
+    auto results__ = torch::_thnn_fused_lstm_cell_backward_impl((grad_hy ? tensor_from_ocaml(grad_hy) : torch::Tensor()), (grad_cy ? tensor_from_ocaml(grad_cy) : torch::Tensor()), *tensor_ptr_from_ocaml(cx), *tensor_ptr_from_ocaml(cy), *tensor_ptr_from_ocaml(workspace), (bool)has_bias);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+void atg__thnn_fused_lstm_cell_backward_impl_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor grad_hy, gc_tensor grad_cy, gc_tensor cx, gc_tensor cy, gc_tensor workspace, int has_bias) {
+  PROTECT(
+    auto results__ = torch::_thnn_fused_lstm_cell_backward_impl_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(out2), (grad_hy ? tensor_from_ocaml(grad_hy) : torch::Tensor()), (grad_cy ? tensor_from_ocaml(grad_cy) : torch::Tensor()), *tensor_ptr_from_ocaml(cx), *tensor_ptr_from_ocaml(cy), *tensor_ptr_from_ocaml(workspace), (bool)has_bias);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+void atg__thnn_fused_lstm_cell_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor input_gates, gc_tensor hidden_gates, gc_tensor cx, gc_tensor input_bias, gc_tensor hidden_bias) {
+  PROTECT(
+    auto results__ = torch::_thnn_fused_lstm_cell_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(out2), *tensor_ptr_from_ocaml(input_gates), *tensor_ptr_from_ocaml(hidden_gates), *tensor_ptr_from_ocaml(cx), (input_bias ? tensor_from_ocaml(input_bias) : torch::Tensor()), (hidden_bias ? tensor_from_ocaml(hidden_bias) : torch::Tensor()));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+raw_tensor atg__to_copy(gc_tensor self, int options_kind, int options_device, int non_blocking) {
+  PROTECT(
+    torch::Tensor results__ = torch::_to_copy(*tensor_ptr_from_ocaml(self), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)), (bool)non_blocking);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__to_copy_out(gc_tensor out, gc_tensor self, int non_blocking) {
+  PROTECT(
+    torch::Tensor results__ = torch::_to_copy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), (bool)non_blocking);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor *atg__to_cpu(gc_tensor *tensors_data, int tensors_len) {
+  PROTECT(
+    auto results__ = torch::_to_cpu(of_carray_tensor(tensors_data, tensors_len));
+    int sz = results__.size();
+    raw_tensor *out__ = (raw_tensor*)malloc((sz + 1) * sizeof(raw_tensor));
+    for (int i = 0; i < sz; ++i)
+      out__[i] = tensor_to_ocaml(results__[i]);
+    out__[sz] = nullptr;
+    return out__;
+  )
+}
+
+raw_tensor atg__to_dense(gc_tensor self, int dtype, int masked_grad) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->_to_dense(torch::ScalarType(dtype), (bool)masked_grad);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__to_dense_out(gc_tensor out, gc_tensor self, int dtype, int masked_grad) {
+  PROTECT(
+    torch::Tensor results__ = torch::_to_dense_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::ScalarType(dtype), (bool)masked_grad);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__to_sparse_bsc(gc_tensor self, int64_t *blocksize_data, int blocksize_len, int64_t dense_dim_v, int dense_dim_null) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->_to_sparse_bsc(torch::IntArrayRef(blocksize_data, blocksize_len), dense_dim_null ? c10::nullopt : c10::optional<int64_t>(dense_dim_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__to_sparse_bsc_out(gc_tensor out, gc_tensor self, int64_t *blocksize_data, int blocksize_len, int64_t dense_dim_v, int dense_dim_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::_to_sparse_bsc_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(blocksize_data, blocksize_len), dense_dim_null ? c10::nullopt : c10::optional<int64_t>(dense_dim_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__to_sparse_bsr(gc_tensor self, int64_t *blocksize_data, int blocksize_len, int64_t dense_dim_v, int dense_dim_null) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->_to_sparse_bsr(torch::IntArrayRef(blocksize_data, blocksize_len), dense_dim_null ? c10::nullopt : c10::optional<int64_t>(dense_dim_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__to_sparse_bsr_out(gc_tensor out, gc_tensor self, int64_t *blocksize_data, int blocksize_len, int64_t dense_dim_v, int dense_dim_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::_to_sparse_bsr_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(blocksize_data, blocksize_len), dense_dim_null ? c10::nullopt : c10::optional<int64_t>(dense_dim_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__to_sparse_csc(gc_tensor self, int64_t dense_dim_v, int dense_dim_null) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->_to_sparse_csc(dense_dim_null ? c10::nullopt : c10::optional<int64_t>(dense_dim_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__to_sparse_csc_out(gc_tensor out, gc_tensor self, int64_t dense_dim_v, int dense_dim_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::_to_sparse_csc_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dense_dim_null ? c10::nullopt : c10::optional<int64_t>(dense_dim_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__to_sparse_csr(gc_tensor self, int64_t dense_dim_v, int dense_dim_null) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->_to_sparse_csr(dense_dim_null ? c10::nullopt : c10::optional<int64_t>(dense_dim_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__to_sparse_csr_out(gc_tensor out, gc_tensor self, int64_t dense_dim_v, int dense_dim_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::_to_sparse_csr_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dense_dim_null ? c10::nullopt : c10::optional<int64_t>(dense_dim_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg__to_sparse_semi_structured(raw_tensor *out__, gc_tensor dense) {
+  PROTECT(
+    auto results__ = torch::_to_sparse_semi_structured(*tensor_ptr_from_ocaml(dense));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg__transform_bias_rescale_qkv(raw_tensor *out__, gc_tensor qkv, gc_tensor qkv_bias, int64_t num_heads) {
+  PROTECT(
+    auto results__ = torch::_transform_bias_rescale_qkv(*tensor_ptr_from_ocaml(qkv), *tensor_ptr_from_ocaml(qkv_bias), num_heads);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+void atg__transform_bias_rescale_qkv_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor qkv, gc_tensor qkv_bias, int64_t num_heads) {
+  PROTECT(
+    auto results__ = torch::_transform_bias_rescale_qkv_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(out2), *tensor_ptr_from_ocaml(qkv), *tensor_ptr_from_ocaml(qkv_bias), num_heads);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+raw_tensor atg__transformer_encoder_layer_fwd(gc_tensor src, int64_t embed_dim, int64_t num_heads, gc_tensor qkv_weight, gc_tensor qkv_bias, gc_tensor proj_weight, gc_tensor proj_bias, int use_gelu, int norm_first, double eps, gc_tensor norm_weight_1, gc_tensor norm_bias_1, gc_tensor norm_weight_2, gc_tensor norm_bias_2, gc_tensor ffn_weight_1, gc_tensor ffn_bias_1, gc_tensor ffn_weight_2, gc_tensor ffn_bias_2, gc_tensor mask, int64_t mask_type_v, int mask_type_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::_transformer_encoder_layer_fwd(*tensor_ptr_from_ocaml(src), embed_dim, num_heads, *tensor_ptr_from_ocaml(qkv_weight), *tensor_ptr_from_ocaml(qkv_bias), *tensor_ptr_from_ocaml(proj_weight), *tensor_ptr_from_ocaml(proj_bias), (bool)use_gelu, (bool)norm_first, eps, *tensor_ptr_from_ocaml(norm_weight_1), *tensor_ptr_from_ocaml(norm_bias_1), *tensor_ptr_from_ocaml(norm_weight_2), *tensor_ptr_from_ocaml(norm_bias_2), *tensor_ptr_from_ocaml(ffn_weight_1), *tensor_ptr_from_ocaml(ffn_bias_1), *tensor_ptr_from_ocaml(ffn_weight_2), *tensor_ptr_from_ocaml(ffn_bias_2), (mask ? tensor_from_ocaml(mask) : torch::Tensor()), mask_type_null ? c10::nullopt : c10::optional<int64_t>(mask_type_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__transformer_encoder_layer_fwd_out(gc_tensor out, gc_tensor src, int64_t embed_dim, int64_t num_heads, gc_tensor qkv_weight, gc_tensor qkv_bias, gc_tensor proj_weight, gc_tensor proj_bias, int use_gelu, int norm_first, double eps, gc_tensor norm_weight_1, gc_tensor norm_bias_1, gc_tensor norm_weight_2, gc_tensor norm_bias_2, gc_tensor ffn_weight_1, gc_tensor ffn_bias_1, gc_tensor ffn_weight_2, gc_tensor ffn_bias_2, gc_tensor mask, int64_t mask_type_v, int mask_type_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::_transformer_encoder_layer_fwd_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(src), embed_dim, num_heads, *tensor_ptr_from_ocaml(qkv_weight), *tensor_ptr_from_ocaml(qkv_bias), *tensor_ptr_from_ocaml(proj_weight), *tensor_ptr_from_ocaml(proj_bias), (bool)use_gelu, (bool)norm_first, eps, *tensor_ptr_from_ocaml(norm_weight_1), *tensor_ptr_from_ocaml(norm_bias_1), *tensor_ptr_from_ocaml(norm_weight_2), *tensor_ptr_from_ocaml(norm_bias_2), *tensor_ptr_from_ocaml(ffn_weight_1), *tensor_ptr_from_ocaml(ffn_bias_1), *tensor_ptr_from_ocaml(ffn_weight_2), *tensor_ptr_from_ocaml(ffn_bias_2), (mask ? tensor_from_ocaml(mask) : torch::Tensor()), mask_type_null ? c10::nullopt : c10::optional<int64_t>(mask_type_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__trilinear(gc_tensor i1, gc_tensor i2, gc_tensor i3, int64_t *expand1_data, int expand1_len, int64_t *expand2_data, int expand2_len, int64_t *expand3_data, int expand3_len, int64_t *sumdim_data, int sumdim_len, int64_t unroll_dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::_trilinear(*tensor_ptr_from_ocaml(i1), *tensor_ptr_from_ocaml(i2), *tensor_ptr_from_ocaml(i3), torch::IntArrayRef(expand1_data, expand1_len), torch::IntArrayRef(expand2_data, expand2_len), torch::IntArrayRef(expand3_data, expand3_len), torch::IntArrayRef(sumdim_data, sumdim_len), unroll_dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__trilinear_out(gc_tensor out, gc_tensor i1, gc_tensor i2, gc_tensor i3, int64_t *expand1_data, int expand1_len, int64_t *expand2_data, int expand2_len, int64_t *expand3_data, int expand3_len, int64_t *sumdim_data, int sumdim_len, int64_t unroll_dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::_trilinear_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(i1), *tensor_ptr_from_ocaml(i2), *tensor_ptr_from_ocaml(i3), torch::IntArrayRef(expand1_data, expand1_len), torch::IntArrayRef(expand2_data, expand2_len), torch::IntArrayRef(expand3_data, expand3_len), torch::IntArrayRef(sumdim_data, sumdim_len), unroll_dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__triton_multi_head_attention(gc_tensor query, gc_tensor key, gc_tensor value, int64_t embed_dim, int64_t num_head, gc_tensor qkv_weight, gc_tensor qkv_bias, gc_tensor proj_weight, gc_tensor proj_bias, gc_tensor mask) {
+  PROTECT(
+    torch::Tensor results__ = torch::_triton_multi_head_attention(*tensor_ptr_from_ocaml(query), *tensor_ptr_from_ocaml(key), *tensor_ptr_from_ocaml(value), embed_dim, num_head, *tensor_ptr_from_ocaml(qkv_weight), *tensor_ptr_from_ocaml(qkv_bias), *tensor_ptr_from_ocaml(proj_weight), *tensor_ptr_from_ocaml(proj_bias), (mask ? tensor_from_ocaml(mask) : torch::Tensor()));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__triton_multi_head_attention_out(gc_tensor out, gc_tensor query, gc_tensor key, gc_tensor value, int64_t embed_dim, int64_t num_head, gc_tensor qkv_weight, gc_tensor qkv_bias, gc_tensor proj_weight, gc_tensor proj_bias, gc_tensor mask) {
+  PROTECT(
+    torch::Tensor results__ = torch::_triton_multi_head_attention_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(query), *tensor_ptr_from_ocaml(key), *tensor_ptr_from_ocaml(value), embed_dim, num_head, *tensor_ptr_from_ocaml(qkv_weight), *tensor_ptr_from_ocaml(qkv_bias), *tensor_ptr_from_ocaml(proj_weight), *tensor_ptr_from_ocaml(proj_bias), (mask ? tensor_from_ocaml(mask) : torch::Tensor()));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__triton_scaled_dot_attention(gc_tensor q, gc_tensor k, gc_tensor v, double dropout_p) {
+  PROTECT(
+    torch::Tensor results__ = torch::_triton_scaled_dot_attention(*tensor_ptr_from_ocaml(q), *tensor_ptr_from_ocaml(k), *tensor_ptr_from_ocaml(v), dropout_p);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__triton_scaled_dot_attention_out(gc_tensor out, gc_tensor q, gc_tensor k, gc_tensor v, double dropout_p) {
+  PROTECT(
+    torch::Tensor results__ = torch::_triton_scaled_dot_attention_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(q), *tensor_ptr_from_ocaml(k), *tensor_ptr_from_ocaml(v), dropout_p);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg__unique(raw_tensor *out__, gc_tensor self, int sorted, int return_inverse) {
+  PROTECT(
+    auto results__ = torch::_unique(*tensor_ptr_from_ocaml(self), (bool)sorted, (bool)return_inverse);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg__unique2(raw_tensor *out__, gc_tensor self, int sorted, int return_inverse, int return_counts) {
+  PROTECT(
+    auto results__ = torch::_unique2(*tensor_ptr_from_ocaml(self), (bool)sorted, (bool)return_inverse, (bool)return_counts);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+void atg__unique2_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor self, int sorted, int return_inverse, int return_counts) {
+  PROTECT(
+    auto results__ = torch::_unique2_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(out2), *tensor_ptr_from_ocaml(self), (bool)sorted, (bool)return_inverse, (bool)return_counts);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+void atg__unique_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor self, int sorted, int return_inverse) {
+  PROTECT(
+    auto results__ = torch::_unique_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(self), (bool)sorted, (bool)return_inverse);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg__unpack_dual(raw_tensor *out__, gc_tensor dual, int64_t level) {
+  PROTECT(
+    auto results__ = torch::_unpack_dual(*tensor_ptr_from_ocaml(dual), level);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg__unsafe_index(gc_tensor self, gc_tensor *indices_data, int indices_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::_unsafe_index(*tensor_ptr_from_ocaml(self), of_carray_tensor_opt(indices_data, indices_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__unsafe_index_put(gc_tensor self, gc_tensor *indices_data, int indices_len, gc_tensor values, int accumulate) {
+  PROTECT(
+    torch::Tensor results__ = torch::_unsafe_index_put(*tensor_ptr_from_ocaml(self), of_carray_tensor_opt(indices_data, indices_len), *tensor_ptr_from_ocaml(values), (bool)accumulate);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__unsafe_view(gc_tensor self, int64_t *size_data, int size_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::_unsafe_view(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(size_data, size_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__unsafe_view_out(gc_tensor out, gc_tensor self, int64_t *size_data, int size_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::_unsafe_view_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(size_data, size_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__upsample_bicubic2d_aa(gc_tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::_upsample_bicubic2d_aa(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(output_size_data, output_size_len), (bool)align_corners, scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__upsample_bicubic2d_aa_backward(gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::_upsample_bicubic2d_aa_backward(*tensor_ptr_from_ocaml(grad_output), torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), (bool)align_corners, scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__upsample_bicubic2d_aa_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::_upsample_bicubic2d_aa_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), (bool)align_corners, scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__upsample_bicubic2d_aa_out(gc_tensor out, gc_tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::_upsample_bicubic2d_aa_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(output_size_data, output_size_len), (bool)align_corners, scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__upsample_bicubic2d_aa_vec(gc_tensor input, int64_t *output_size_data, int output_size_len, int align_corners, double *scale_factors_data, int scale_factors_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::_upsample_bicubic2d_aa(*tensor_ptr_from_ocaml(input), output_size_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(output_size_data, output_size_len)), (bool)align_corners, at::ArrayRef<double>(scale_factors_data, scale_factors_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__upsample_bilinear2d_aa(gc_tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::_upsample_bilinear2d_aa(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(output_size_data, output_size_len), (bool)align_corners, scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__upsample_bilinear2d_aa_backward(gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::_upsample_bilinear2d_aa_backward(*tensor_ptr_from_ocaml(grad_output), torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), (bool)align_corners, scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__upsample_bilinear2d_aa_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::_upsample_bilinear2d_aa_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), (bool)align_corners, scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__upsample_bilinear2d_aa_out(gc_tensor out, gc_tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::_upsample_bilinear2d_aa_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(output_size_data, output_size_len), (bool)align_corners, scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__upsample_bilinear2d_aa_vec(gc_tensor input, int64_t *output_size_data, int output_size_len, int align_corners, double *scale_factors_data, int scale_factors_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::_upsample_bilinear2d_aa(*tensor_ptr_from_ocaml(input), output_size_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(output_size_data, output_size_len)), (bool)align_corners, at::ArrayRef<double>(scale_factors_data, scale_factors_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__upsample_nearest_exact1d(gc_tensor self, int64_t *output_size_data, int output_size_len, double scales_v, int scales_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::_upsample_nearest_exact1d(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(output_size_data, output_size_len), scales_null ? c10::nullopt : c10::optional<double>(scales_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__upsample_nearest_exact1d_backward(gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_v, int scales_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::_upsample_nearest_exact1d_backward(*tensor_ptr_from_ocaml(grad_output), torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), scales_null ? c10::nullopt : c10::optional<double>(scales_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__upsample_nearest_exact1d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_v, int scales_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::_upsample_nearest_exact1d_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), scales_null ? c10::nullopt : c10::optional<double>(scales_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__upsample_nearest_exact1d_out(gc_tensor out, gc_tensor self, int64_t *output_size_data, int output_size_len, double scales_v, int scales_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::_upsample_nearest_exact1d_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(output_size_data, output_size_len), scales_null ? c10::nullopt : c10::optional<double>(scales_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__upsample_nearest_exact1d_vec(gc_tensor input, int64_t *output_size_data, int output_size_len, double *scale_factors_data, int scale_factors_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::_upsample_nearest_exact1d(*tensor_ptr_from_ocaml(input), output_size_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(output_size_data, output_size_len)), at::ArrayRef<double>(scale_factors_data, scale_factors_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__upsample_nearest_exact2d(gc_tensor self, int64_t *output_size_data, int output_size_len, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::_upsample_nearest_exact2d(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(output_size_data, output_size_len), scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__upsample_nearest_exact2d_backward(gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::_upsample_nearest_exact2d_backward(*tensor_ptr_from_ocaml(grad_output), torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__upsample_nearest_exact2d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::_upsample_nearest_exact2d_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__upsample_nearest_exact2d_out(gc_tensor out, gc_tensor self, int64_t *output_size_data, int output_size_len, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::_upsample_nearest_exact2d_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(output_size_data, output_size_len), scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__upsample_nearest_exact2d_vec(gc_tensor input, int64_t *output_size_data, int output_size_len, double *scale_factors_data, int scale_factors_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::_upsample_nearest_exact2d(*tensor_ptr_from_ocaml(input), output_size_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(output_size_data, output_size_len)), at::ArrayRef<double>(scale_factors_data, scale_factors_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__upsample_nearest_exact3d(gc_tensor self, int64_t *output_size_data, int output_size_len, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::_upsample_nearest_exact3d(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(output_size_data, output_size_len), scales_d_null ? c10::nullopt : c10::optional<double>(scales_d_v), scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__upsample_nearest_exact3d_backward(gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::_upsample_nearest_exact3d_backward(*tensor_ptr_from_ocaml(grad_output), torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), scales_d_null ? c10::nullopt : c10::optional<double>(scales_d_v), scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__upsample_nearest_exact3d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::_upsample_nearest_exact3d_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), scales_d_null ? c10::nullopt : c10::optional<double>(scales_d_v), scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__upsample_nearest_exact3d_out(gc_tensor out, gc_tensor self, int64_t *output_size_data, int output_size_len, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::_upsample_nearest_exact3d_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(output_size_data, output_size_len), scales_d_null ? c10::nullopt : c10::optional<double>(scales_d_v), scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__upsample_nearest_exact3d_vec(gc_tensor input, int64_t *output_size_data, int output_size_len, double *scale_factors_data, int scale_factors_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::_upsample_nearest_exact3d(*tensor_ptr_from_ocaml(input), output_size_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(output_size_data, output_size_len)), at::ArrayRef<double>(scale_factors_data, scale_factors_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+int atg__use_cudnn_ctc_loss(gc_tensor log_probs, gc_tensor targets, int64_t *input_lengths_data, int input_lengths_len, int64_t *target_lengths_data, int target_lengths_len, int64_t blank) {
+  PROTECT(
+    return torch::_use_cudnn_ctc_loss(*tensor_ptr_from_ocaml(log_probs), *tensor_ptr_from_ocaml(targets), torch::IntArrayRef(input_lengths_data, input_lengths_len), torch::IntArrayRef(target_lengths_data, target_lengths_len), blank);
+  )
+  return 0;
+}
+
+int atg__use_cudnn_ctc_loss_tensor(gc_tensor log_probs, gc_tensor targets, gc_tensor input_lengths, gc_tensor target_lengths, int64_t blank) {
+  PROTECT(
+    return torch::_use_cudnn_ctc_loss(*tensor_ptr_from_ocaml(log_probs), *tensor_ptr_from_ocaml(targets), *tensor_ptr_from_ocaml(input_lengths), *tensor_ptr_from_ocaml(target_lengths), blank);
+  )
+  return 0;
+}
+
+int atg__use_cudnn_rnn_flatten_weight() {
+  PROTECT(
+    return torch::_use_cudnn_rnn_flatten_weight();
+  )
+  return 0;
+}
+
+void atg__validate_compressed_sparse_indices(int is_crow, gc_tensor compressed_idx, gc_tensor plain_idx, int64_t cdim, int64_t dim, int64_t nnz) {
+  PROTECT(
+    torch::_validate_compressed_sparse_indices((bool)is_crow, *tensor_ptr_from_ocaml(compressed_idx), *tensor_ptr_from_ocaml(plain_idx), cdim, dim, nnz);
+  )
+}
+
+void atg__validate_sparse_bsc_tensor_args(gc_tensor ccol_indices, gc_tensor row_indices, gc_tensor values, int64_t *size_data, int size_len) {
+  PROTECT(
+    torch::_validate_sparse_bsc_tensor_args(*tensor_ptr_from_ocaml(ccol_indices), *tensor_ptr_from_ocaml(row_indices), *tensor_ptr_from_ocaml(values), torch::IntArrayRef(size_data, size_len));
+  )
+}
+
+void atg__validate_sparse_bsr_tensor_args(gc_tensor crow_indices, gc_tensor col_indices, gc_tensor values, int64_t *size_data, int size_len) {
+  PROTECT(
+    torch::_validate_sparse_bsr_tensor_args(*tensor_ptr_from_ocaml(crow_indices), *tensor_ptr_from_ocaml(col_indices), *tensor_ptr_from_ocaml(values), torch::IntArrayRef(size_data, size_len));
+  )
+}
+
+void atg__validate_sparse_csc_tensor_args(gc_tensor ccol_indices, gc_tensor row_indices, gc_tensor values, int64_t *size_data, int size_len) {
+  PROTECT(
+    torch::_validate_sparse_csc_tensor_args(*tensor_ptr_from_ocaml(ccol_indices), *tensor_ptr_from_ocaml(row_indices), *tensor_ptr_from_ocaml(values), torch::IntArrayRef(size_data, size_len));
+  )
+}
+
+raw_tensor atg__values(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->_values();
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__values_copy(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::_values_copy(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg__values_copy_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::_values_copy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+int64_t atg__version(gc_tensor self) {
+  PROTECT(
+    return tensor_ptr_from_ocaml(self)->_version();
+  )
+  return 0;
+}
+
+raw_tensor atg__weight_norm(gc_tensor v, gc_tensor g, int64_t dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::_weight_norm(*tensor_ptr_from_ocaml(v), *tensor_ptr_from_ocaml(g), dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg__weight_norm_differentiable_backward(raw_tensor *out__, gc_tensor grad_w, gc_tensor saved_v, gc_tensor saved_g, gc_tensor saved_norms, int64_t dim) {
+  PROTECT(
+    auto results__ = torch::_weight_norm_differentiable_backward(*tensor_ptr_from_ocaml(grad_w), *tensor_ptr_from_ocaml(saved_v), *tensor_ptr_from_ocaml(saved_g), *tensor_ptr_from_ocaml(saved_norms), dim);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg__weight_norm_interface(raw_tensor *out__, gc_tensor v, gc_tensor g, int64_t dim) {
+  PROTECT(
+    auto results__ = torch::_weight_norm_interface(*tensor_ptr_from_ocaml(v), *tensor_ptr_from_ocaml(g), dim);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg__weight_norm_interface_backward(raw_tensor *out__, gc_tensor grad_w, gc_tensor saved_v, gc_tensor saved_g, gc_tensor saved_norms, int64_t dim) {
+  PROTECT(
+    auto results__ = torch::_weight_norm_interface_backward(*tensor_ptr_from_ocaml(grad_w), *tensor_ptr_from_ocaml(saved_v), *tensor_ptr_from_ocaml(saved_g), *tensor_ptr_from_ocaml(saved_norms), dim);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg__weight_norm_interface_backward_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor grad_w, gc_tensor saved_v, gc_tensor saved_g, gc_tensor saved_norms, int64_t dim) {
+  PROTECT(
+    auto results__ = torch::_weight_norm_interface_backward_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(grad_w), *tensor_ptr_from_ocaml(saved_v), *tensor_ptr_from_ocaml(saved_g), *tensor_ptr_from_ocaml(saved_norms), dim);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg__weight_norm_interface_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor v, gc_tensor g, int64_t dim) {
+  PROTECT(
+    auto results__ = torch::_weight_norm_interface_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(v), *tensor_ptr_from_ocaml(g), dim);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_abs(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::abs(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_abs_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::abs_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_abs_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::abs_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_absolute(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::absolute(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_absolute_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->absolute_();
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_absolute_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::absolute_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_acos(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::acos(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_acos_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::acos_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_acos_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::acos_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_acosh(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::acosh(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_acosh_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::acosh_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_acosh_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::acosh_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_adaptive_avg_pool1d(gc_tensor self, int64_t *output_size_data, int output_size_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::adaptive_avg_pool1d(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(output_size_data, output_size_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_adaptive_avg_pool2d(gc_tensor self, int64_t *output_size_data, int output_size_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::adaptive_avg_pool2d(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(output_size_data, output_size_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_adaptive_avg_pool2d_out(gc_tensor out, gc_tensor self, int64_t *output_size_data, int output_size_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::adaptive_avg_pool2d_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(output_size_data, output_size_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_adaptive_avg_pool3d(gc_tensor self, int64_t *output_size_data, int output_size_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::adaptive_avg_pool3d(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(output_size_data, output_size_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_adaptive_avg_pool3d_backward(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::adaptive_avg_pool3d_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_adaptive_avg_pool3d_out(gc_tensor out, gc_tensor self, int64_t *output_size_data, int output_size_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::adaptive_avg_pool3d_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(output_size_data, output_size_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_adaptive_max_pool1d(raw_tensor *out__, gc_tensor self, int64_t *output_size_data, int output_size_len) {
+  PROTECT(
+    auto results__ = torch::adaptive_max_pool1d(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(output_size_data, output_size_len));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_adaptive_max_pool2d(raw_tensor *out__, gc_tensor self, int64_t *output_size_data, int output_size_len) {
+  PROTECT(
+    auto results__ = torch::adaptive_max_pool2d(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(output_size_data, output_size_len));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_adaptive_max_pool2d_backward(gc_tensor grad_output, gc_tensor self, gc_tensor indices) {
+  PROTECT(
+    torch::Tensor results__ = torch::adaptive_max_pool2d_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(indices));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_adaptive_max_pool2d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, gc_tensor indices) {
+  PROTECT(
+    torch::Tensor results__ = torch::adaptive_max_pool2d_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(indices));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_adaptive_max_pool2d_out(raw_tensor *out__, gc_tensor out, gc_tensor indices, gc_tensor self, int64_t *output_size_data, int output_size_len) {
+  PROTECT(
+    auto results__ = torch::adaptive_max_pool2d_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(indices), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(output_size_data, output_size_len));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_adaptive_max_pool3d(raw_tensor *out__, gc_tensor self, int64_t *output_size_data, int output_size_len) {
+  PROTECT(
+    auto results__ = torch::adaptive_max_pool3d(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(output_size_data, output_size_len));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_adaptive_max_pool3d_backward(gc_tensor grad_output, gc_tensor self, gc_tensor indices) {
+  PROTECT(
+    torch::Tensor results__ = torch::adaptive_max_pool3d_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(indices));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_adaptive_max_pool3d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, gc_tensor indices) {
+  PROTECT(
+    torch::Tensor results__ = torch::adaptive_max_pool3d_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(indices));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_adaptive_max_pool3d_out(raw_tensor *out__, gc_tensor out, gc_tensor indices, gc_tensor self, int64_t *output_size_data, int output_size_len) {
+  PROTECT(
+    auto results__ = torch::adaptive_max_pool3d_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(indices), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(output_size_data, output_size_len));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_add(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::add(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_add_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->add_(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_add_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::add_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_add_scalar(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::add(*tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_add_scalar_(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->add_(*other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_add_scalar_out(gc_tensor out, gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::add_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_addbmm(gc_tensor self, gc_tensor batch1, gc_tensor batch2) {
+  PROTECT(
+    torch::Tensor results__ = torch::addbmm(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(batch1), *tensor_ptr_from_ocaml(batch2));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_addbmm_(gc_tensor self, gc_tensor batch1, gc_tensor batch2) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->addbmm_(*tensor_ptr_from_ocaml(batch1), *tensor_ptr_from_ocaml(batch2));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_addbmm_out(gc_tensor out, gc_tensor self, gc_tensor batch1, gc_tensor batch2) {
+  PROTECT(
+    torch::Tensor results__ = torch::addbmm_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(batch1), *tensor_ptr_from_ocaml(batch2));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_addcdiv(gc_tensor self, gc_tensor tensor1, gc_tensor tensor2) {
+  PROTECT(
+    torch::Tensor results__ = torch::addcdiv(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(tensor1), *tensor_ptr_from_ocaml(tensor2));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_addcdiv_(gc_tensor self, gc_tensor tensor1, gc_tensor tensor2) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->addcdiv_(*tensor_ptr_from_ocaml(tensor1), *tensor_ptr_from_ocaml(tensor2));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_addcdiv_out(gc_tensor out, gc_tensor self, gc_tensor tensor1, gc_tensor tensor2) {
+  PROTECT(
+    torch::Tensor results__ = torch::addcdiv_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(tensor1), *tensor_ptr_from_ocaml(tensor2));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_addcmul(gc_tensor self, gc_tensor tensor1, gc_tensor tensor2) {
+  PROTECT(
+    torch::Tensor results__ = torch::addcmul(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(tensor1), *tensor_ptr_from_ocaml(tensor2));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_addcmul_(gc_tensor self, gc_tensor tensor1, gc_tensor tensor2) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->addcmul_(*tensor_ptr_from_ocaml(tensor1), *tensor_ptr_from_ocaml(tensor2));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_addcmul_out(gc_tensor out, gc_tensor self, gc_tensor tensor1, gc_tensor tensor2) {
+  PROTECT(
+    torch::Tensor results__ = torch::addcmul_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(tensor1), *tensor_ptr_from_ocaml(tensor2));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_addmm(gc_tensor self, gc_tensor mat1, gc_tensor mat2) {
+  PROTECT(
+    torch::Tensor results__ = torch::addmm(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(mat1), *tensor_ptr_from_ocaml(mat2));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_addmm_(gc_tensor self, gc_tensor mat1, gc_tensor mat2) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->addmm_(*tensor_ptr_from_ocaml(mat1), *tensor_ptr_from_ocaml(mat2));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_addmm_out(gc_tensor out, gc_tensor self, gc_tensor mat1, gc_tensor mat2) {
+  PROTECT(
+    torch::Tensor results__ = torch::addmm_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(mat1), *tensor_ptr_from_ocaml(mat2));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_addmv(gc_tensor self, gc_tensor mat, gc_tensor vec) {
+  PROTECT(
+    torch::Tensor results__ = torch::addmv(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(mat), *tensor_ptr_from_ocaml(vec));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_addmv_(gc_tensor self, gc_tensor mat, gc_tensor vec) {
+  PROTECT(
+    torch::Tensor results__ = torch::addmv_(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(mat), *tensor_ptr_from_ocaml(vec));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_addmv_out(gc_tensor out, gc_tensor self, gc_tensor mat, gc_tensor vec) {
+  PROTECT(
+    torch::Tensor results__ = torch::addmv_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(mat), *tensor_ptr_from_ocaml(vec));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_addr(gc_tensor self, gc_tensor vec1, gc_tensor vec2) {
+  PROTECT(
+    torch::Tensor results__ = torch::addr(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(vec1), *tensor_ptr_from_ocaml(vec2));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_addr_(gc_tensor self, gc_tensor vec1, gc_tensor vec2) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->addr_(*tensor_ptr_from_ocaml(vec1), *tensor_ptr_from_ocaml(vec2));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_addr_out(gc_tensor out, gc_tensor self, gc_tensor vec1, gc_tensor vec2) {
+  PROTECT(
+    torch::Tensor results__ = torch::addr_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(vec1), *tensor_ptr_from_ocaml(vec2));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_adjoint(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::adjoint(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_affine_grid_generator(gc_tensor theta, int64_t *size_data, int size_len, int align_corners) {
+  PROTECT(
+    torch::Tensor results__ = torch::affine_grid_generator(*tensor_ptr_from_ocaml(theta), torch::IntArrayRef(size_data, size_len), (bool)align_corners);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_affine_grid_generator_backward(gc_tensor grad, int64_t *size_data, int size_len, int align_corners) {
+  PROTECT(
+    torch::Tensor results__ = torch::affine_grid_generator_backward(*tensor_ptr_from_ocaml(grad), torch::IntArrayRef(size_data, size_len), (bool)align_corners);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_affine_grid_generator_out(gc_tensor out, gc_tensor theta, int64_t *size_data, int size_len, int align_corners) {
+  PROTECT(
+    torch::Tensor results__ = torch::affine_grid_generator_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(theta), torch::IntArrayRef(size_data, size_len), (bool)align_corners);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_alias(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::alias(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_alias_copy(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::alias_copy(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_alias_copy_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::alias_copy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_align_as(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->align_as(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor *atg_align_tensors(gc_tensor *tensors_data, int tensors_len) {
+  PROTECT(
+    auto results__ = torch::align_tensors(of_carray_tensor(tensors_data, tensors_len));
+    int sz = results__.size();
+    raw_tensor *out__ = (raw_tensor*)malloc((sz + 1) * sizeof(raw_tensor));
+    for (int i = 0; i < sz; ++i)
+      out__[i] = tensor_to_ocaml(results__[i]);
+    out__[sz] = nullptr;
+    return out__;
+  )
+}
+
+raw_tensor atg_all(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::all(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_all_all_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::all_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_all_dim(gc_tensor self, int64_t dim, int keepdim) {
+  PROTECT(
+    torch::Tensor results__ = torch::all(*tensor_ptr_from_ocaml(self), dim, (bool)keepdim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_all_out(gc_tensor out, gc_tensor self, int64_t dim, int keepdim) {
+  PROTECT(
+    torch::Tensor results__ = torch::all_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dim, (bool)keepdim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+int atg_allclose(gc_tensor self, gc_tensor other, double rtol, double atol, int equal_nan) {
+  PROTECT(
+    return torch::allclose(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other), rtol, atol, (bool)equal_nan);
+  )
+  return 0;
+}
+
+raw_tensor atg_alpha_dropout(gc_tensor input, double p, int train) {
+  PROTECT(
+    torch::Tensor results__ = torch::alpha_dropout(*tensor_ptr_from_ocaml(input), p, (bool)train);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_alpha_dropout_(gc_tensor self, double p, int train) {
+  PROTECT(
+    torch::Tensor results__ = torch::alpha_dropout_(*tensor_ptr_from_ocaml(self), p, (bool)train);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_amax(gc_tensor self, int64_t *dim_data, int dim_len, int keepdim) {
+  PROTECT(
+    torch::Tensor results__ = torch::amax(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(dim_data, dim_len), (bool)keepdim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_amax_out(gc_tensor out, gc_tensor self, int64_t *dim_data, int dim_len, int keepdim) {
+  PROTECT(
+    torch::Tensor results__ = torch::amax_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(dim_data, dim_len), (bool)keepdim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_amin(gc_tensor self, int64_t *dim_data, int dim_len, int keepdim) {
+  PROTECT(
+    torch::Tensor results__ = torch::amin(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(dim_data, dim_len), (bool)keepdim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_amin_out(gc_tensor out, gc_tensor self, int64_t *dim_data, int dim_len, int keepdim) {
+  PROTECT(
+    torch::Tensor results__ = torch::amin_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(dim_data, dim_len), (bool)keepdim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_aminmax(raw_tensor *out__, gc_tensor self, int64_t dim_v, int dim_null, int keepdim) {
+  PROTECT(
+    auto results__ = torch::aminmax(*tensor_ptr_from_ocaml(self), dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v), (bool)keepdim);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_aminmax_out(raw_tensor *out__, gc_tensor min, gc_tensor max, gc_tensor self, int64_t dim_v, int dim_null, int keepdim) {
+  PROTECT(
+    auto results__ = torch::aminmax_out(*tensor_ptr_from_ocaml(min), *tensor_ptr_from_ocaml(max), *tensor_ptr_from_ocaml(self), dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v), (bool)keepdim);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_angle(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::angle(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_angle_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::angle_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_any(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::any(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_any_all_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::any_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_any_dim(gc_tensor self, int64_t dim, int keepdim) {
+  PROTECT(
+    torch::Tensor results__ = torch::any(*tensor_ptr_from_ocaml(self), dim, (bool)keepdim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_any_out(gc_tensor out, gc_tensor self, int64_t dim, int keepdim) {
+  PROTECT(
+    torch::Tensor results__ = torch::any_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dim, (bool)keepdim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_arange(scalar end, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::arange(*end, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_arange_start(scalar start, scalar end, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::arange(*start, *end, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_arange_start_step(scalar start, scalar end, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::arange(*start, *end, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_arccos(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::arccos(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_arccos_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::arccos_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_arccos_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::arccos_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_arccosh(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::arccosh(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_arccosh_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::arccosh_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_arccosh_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::arccosh_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_arcsin(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::arcsin(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_arcsin_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::arcsin_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_arcsin_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::arcsin_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_arcsinh(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::arcsinh(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_arcsinh_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::arcsinh_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_arcsinh_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::arcsinh_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_arctan(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::arctan(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_arctan2(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::arctan2(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_arctan2_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->arctan2_(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_arctan2_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::arctan2_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_arctan_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::arctan_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_arctan_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::arctan_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_arctanh(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::arctanh(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_arctanh_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::arctanh_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_arctanh_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::arctanh_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_argmax(gc_tensor self, int64_t dim_v, int dim_null, int keepdim) {
+  PROTECT(
+    torch::Tensor results__ = torch::argmax(*tensor_ptr_from_ocaml(self), dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v), (bool)keepdim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_argmax_out(gc_tensor out, gc_tensor self, int64_t dim_v, int dim_null, int keepdim) {
+  PROTECT(
+    torch::Tensor results__ = torch::argmax_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v), (bool)keepdim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_argmin(gc_tensor self, int64_t dim_v, int dim_null, int keepdim) {
+  PROTECT(
+    torch::Tensor results__ = torch::argmin(*tensor_ptr_from_ocaml(self), dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v), (bool)keepdim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_argmin_out(gc_tensor out, gc_tensor self, int64_t dim_v, int dim_null, int keepdim) {
+  PROTECT(
+    torch::Tensor results__ = torch::argmin_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v), (bool)keepdim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_argsort(gc_tensor self, int64_t dim, int descending) {
+  PROTECT(
+    torch::Tensor results__ = torch::argsort(*tensor_ptr_from_ocaml(self), dim, (bool)descending);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_argsort_stable(gc_tensor self, int stable, int64_t dim, int descending) {
+  PROTECT(
+    torch::Tensor results__ = torch::argsort(*tensor_ptr_from_ocaml(self), (bool)stable, dim, (bool)descending);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_argsort_stable_out(gc_tensor out, gc_tensor self, int stable, int64_t dim, int descending) {
+  PROTECT(
+    torch::Tensor results__ = torch::argsort_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), (bool)stable, dim, (bool)descending);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_argwhere(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::argwhere(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_as_strided(gc_tensor self, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len, int64_t storage_offset_v, int storage_offset_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::as_strided(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(size_data, size_len), torch::IntArrayRef(stride_data, stride_len), storage_offset_null ? c10::nullopt : c10::optional<int64_t>(storage_offset_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_as_strided_(gc_tensor self, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len, int64_t storage_offset_v, int storage_offset_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::as_strided_(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(size_data, size_len), torch::IntArrayRef(stride_data, stride_len), storage_offset_null ? c10::nullopt : c10::optional<int64_t>(storage_offset_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_as_strided_copy(gc_tensor self, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len, int64_t storage_offset_v, int storage_offset_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::as_strided_copy(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(size_data, size_len), torch::IntArrayRef(stride_data, stride_len), storage_offset_null ? c10::nullopt : c10::optional<int64_t>(storage_offset_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_as_strided_copy_out(gc_tensor out, gc_tensor self, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len, int64_t storage_offset_v, int storage_offset_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::as_strided_copy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(size_data, size_len), torch::IntArrayRef(stride_data, stride_len), storage_offset_null ? c10::nullopt : c10::optional<int64_t>(storage_offset_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_as_strided_scatter(gc_tensor self, gc_tensor src, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len, int64_t storage_offset_v, int storage_offset_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::as_strided_scatter(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(src), torch::IntArrayRef(size_data, size_len), torch::IntArrayRef(stride_data, stride_len), storage_offset_null ? c10::nullopt : c10::optional<int64_t>(storage_offset_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_as_strided_scatter_out(gc_tensor out, gc_tensor self, gc_tensor src, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len, int64_t storage_offset_v, int storage_offset_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::as_strided_scatter_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(src), torch::IntArrayRef(size_data, size_len), torch::IntArrayRef(stride_data, stride_len), storage_offset_null ? c10::nullopt : c10::optional<int64_t>(storage_offset_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_asin(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::asin(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_asin_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::asin_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_asin_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::asin_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_asinh(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::asinh(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_asinh_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::asinh_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_asinh_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::asinh_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_atan(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::atan(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_atan2(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::atan2(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_atan2_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->atan2_(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_atan2_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::atan2_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_atan_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::atan_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_atan_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::atan_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_atanh(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::atanh(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_atanh_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::atanh_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_atanh_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::atanh_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_atleast_1d(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::atleast_1d(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor *atg_atleast_1d_sequence(gc_tensor *tensors_data, int tensors_len) {
+  PROTECT(
+    auto results__ = torch::atleast_1d(of_carray_tensor(tensors_data, tensors_len));
+    int sz = results__.size();
+    raw_tensor *out__ = (raw_tensor*)malloc((sz + 1) * sizeof(raw_tensor));
+    for (int i = 0; i < sz; ++i)
+      out__[i] = tensor_to_ocaml(results__[i]);
+    out__[sz] = nullptr;
+    return out__;
+  )
+}
+
+raw_tensor atg_atleast_2d(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::atleast_2d(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor *atg_atleast_2d_sequence(gc_tensor *tensors_data, int tensors_len) {
+  PROTECT(
+    auto results__ = torch::atleast_2d(of_carray_tensor(tensors_data, tensors_len));
+    int sz = results__.size();
+    raw_tensor *out__ = (raw_tensor*)malloc((sz + 1) * sizeof(raw_tensor));
+    for (int i = 0; i < sz; ++i)
+      out__[i] = tensor_to_ocaml(results__[i]);
+    out__[sz] = nullptr;
+    return out__;
+  )
+}
+
+raw_tensor atg_atleast_3d(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::atleast_3d(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor *atg_atleast_3d_sequence(gc_tensor *tensors_data, int tensors_len) {
+  PROTECT(
+    auto results__ = torch::atleast_3d(of_carray_tensor(tensors_data, tensors_len));
+    int sz = results__.size();
+    raw_tensor *out__ = (raw_tensor*)malloc((sz + 1) * sizeof(raw_tensor));
+    for (int i = 0; i < sz; ++i)
+      out__[i] = tensor_to_ocaml(results__[i]);
+    out__[sz] = nullptr;
+    return out__;
+  )
+}
+
+raw_tensor atg_avg_pool1d(gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int ceil_mode, int count_include_pad) {
+  PROTECT(
+    torch::Tensor results__ = torch::avg_pool1d(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), (bool)ceil_mode, (bool)count_include_pad);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_avg_pool2d(gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int ceil_mode, int count_include_pad, int64_t divisor_override_v, int divisor_override_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::avg_pool2d(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), (bool)ceil_mode, (bool)count_include_pad, divisor_override_null ? c10::nullopt : c10::optional<int64_t>(divisor_override_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_avg_pool2d_backward(gc_tensor grad_output, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int ceil_mode, int count_include_pad, int64_t divisor_override_v, int divisor_override_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::avg_pool2d_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), (bool)ceil_mode, (bool)count_include_pad, divisor_override_null ? c10::nullopt : c10::optional<int64_t>(divisor_override_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_avg_pool2d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int ceil_mode, int count_include_pad, int64_t divisor_override_v, int divisor_override_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::avg_pool2d_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), (bool)ceil_mode, (bool)count_include_pad, divisor_override_null ? c10::nullopt : c10::optional<int64_t>(divisor_override_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_avg_pool2d_out(gc_tensor out, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int ceil_mode, int count_include_pad, int64_t divisor_override_v, int divisor_override_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::avg_pool2d_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), (bool)ceil_mode, (bool)count_include_pad, divisor_override_null ? c10::nullopt : c10::optional<int64_t>(divisor_override_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_avg_pool3d(gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int ceil_mode, int count_include_pad, int64_t divisor_override_v, int divisor_override_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::avg_pool3d(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), (bool)ceil_mode, (bool)count_include_pad, divisor_override_null ? c10::nullopt : c10::optional<int64_t>(divisor_override_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_avg_pool3d_backward(gc_tensor grad_output, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int ceil_mode, int count_include_pad, int64_t divisor_override_v, int divisor_override_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::avg_pool3d_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), (bool)ceil_mode, (bool)count_include_pad, divisor_override_null ? c10::nullopt : c10::optional<int64_t>(divisor_override_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_avg_pool3d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int ceil_mode, int count_include_pad, int64_t divisor_override_v, int divisor_override_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::avg_pool3d_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), (bool)ceil_mode, (bool)count_include_pad, divisor_override_null ? c10::nullopt : c10::optional<int64_t>(divisor_override_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_avg_pool3d_out(gc_tensor out, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int ceil_mode, int count_include_pad, int64_t divisor_override_v, int divisor_override_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::avg_pool3d_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), (bool)ceil_mode, (bool)count_include_pad, divisor_override_null ? c10::nullopt : c10::optional<int64_t>(divisor_override_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_baddbmm(gc_tensor self, gc_tensor batch1, gc_tensor batch2) {
+  PROTECT(
+    torch::Tensor results__ = torch::baddbmm(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(batch1), *tensor_ptr_from_ocaml(batch2));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_baddbmm_(gc_tensor self, gc_tensor batch1, gc_tensor batch2) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->baddbmm_(*tensor_ptr_from_ocaml(batch1), *tensor_ptr_from_ocaml(batch2));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_baddbmm_out(gc_tensor out, gc_tensor self, gc_tensor batch1, gc_tensor batch2) {
+  PROTECT(
+    torch::Tensor results__ = torch::baddbmm_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(batch1), *tensor_ptr_from_ocaml(batch2));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bartlett_window(int64_t window_length, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::bartlett_window(window_length, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bartlett_window_out(gc_tensor out, int64_t window_length) {
+  PROTECT(
+    torch::Tensor results__ = torch::bartlett_window_out(*tensor_ptr_from_ocaml(out), window_length);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bartlett_window_periodic(int64_t window_length, int periodic, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::bartlett_window(window_length, (bool)periodic, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bartlett_window_periodic_out(gc_tensor out, int64_t window_length, int periodic) {
+  PROTECT(
+    torch::Tensor results__ = torch::bartlett_window_out(*tensor_ptr_from_ocaml(out), window_length, (bool)periodic);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_batch_norm(gc_tensor input, gc_tensor weight, gc_tensor bias, gc_tensor running_mean, gc_tensor running_var, int training, double momentum, double eps, int cudnn_enabled) {
+  PROTECT(
+    torch::Tensor results__ = torch::batch_norm(*tensor_ptr_from_ocaml(input), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), (running_mean ? tensor_from_ocaml(running_mean) : torch::Tensor()), (running_var ? tensor_from_ocaml(running_var) : torch::Tensor()), (bool)training, momentum, eps, (bool)cudnn_enabled);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_batch_norm_backward_elemt(gc_tensor grad_out, gc_tensor input, gc_tensor mean, gc_tensor invstd, gc_tensor weight, gc_tensor sum_dy, gc_tensor sum_dy_xmu, gc_tensor count) {
+  PROTECT(
+    torch::Tensor results__ = torch::batch_norm_backward_elemt(*tensor_ptr_from_ocaml(grad_out), *tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(mean), *tensor_ptr_from_ocaml(invstd), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), *tensor_ptr_from_ocaml(sum_dy), *tensor_ptr_from_ocaml(sum_dy_xmu), *tensor_ptr_from_ocaml(count));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_batch_norm_backward_elemt_out(gc_tensor out, gc_tensor grad_out, gc_tensor input, gc_tensor mean, gc_tensor invstd, gc_tensor weight, gc_tensor sum_dy, gc_tensor sum_dy_xmu, gc_tensor count) {
+  PROTECT(
+    torch::Tensor results__ = torch::batch_norm_backward_elemt_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(grad_out), *tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(mean), *tensor_ptr_from_ocaml(invstd), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), *tensor_ptr_from_ocaml(sum_dy), *tensor_ptr_from_ocaml(sum_dy_xmu), *tensor_ptr_from_ocaml(count));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_batch_norm_backward_reduce(raw_tensor *out__, gc_tensor grad_out, gc_tensor input, gc_tensor mean, gc_tensor invstd, gc_tensor weight, int input_g, int weight_g, int bias_g) {
+  PROTECT(
+    auto results__ = torch::batch_norm_backward_reduce(*tensor_ptr_from_ocaml(grad_out), *tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(mean), *tensor_ptr_from_ocaml(invstd), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), (bool)input_g, (bool)weight_g, (bool)bias_g);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+    out__[3] = tensor_to_ocaml(std::get<3>(results__));
+  )
+}
+
+void atg_batch_norm_backward_reduce_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor out3, gc_tensor grad_out, gc_tensor input, gc_tensor mean, gc_tensor invstd, gc_tensor weight, int input_g, int weight_g, int bias_g) {
+  PROTECT(
+    auto results__ = torch::batch_norm_backward_reduce_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(out2), *tensor_ptr_from_ocaml(out3), *tensor_ptr_from_ocaml(grad_out), *tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(mean), *tensor_ptr_from_ocaml(invstd), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), (bool)input_g, (bool)weight_g, (bool)bias_g);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+    out__[3] = tensor_to_ocaml(std::get<3>(results__));
+  )
+}
+
+raw_tensor atg_batch_norm_elemt(gc_tensor input, gc_tensor weight, gc_tensor bias, gc_tensor mean, gc_tensor invstd, double eps) {
+  PROTECT(
+    torch::Tensor results__ = torch::batch_norm_elemt(*tensor_ptr_from_ocaml(input), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), *tensor_ptr_from_ocaml(mean), *tensor_ptr_from_ocaml(invstd), eps);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_batch_norm_elemt_out(gc_tensor out, gc_tensor input, gc_tensor weight, gc_tensor bias, gc_tensor mean, gc_tensor invstd, double eps) {
+  PROTECT(
+    torch::Tensor results__ = torch::batch_norm_elemt_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(input), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), *tensor_ptr_from_ocaml(mean), *tensor_ptr_from_ocaml(invstd), eps);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_batch_norm_gather_stats(raw_tensor *out__, gc_tensor input, gc_tensor mean, gc_tensor invstd, gc_tensor running_mean, gc_tensor running_var, double momentum, double eps, int64_t count) {
+  PROTECT(
+    auto results__ = torch::batch_norm_gather_stats(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(mean), *tensor_ptr_from_ocaml(invstd), (running_mean ? tensor_from_ocaml(running_mean) : torch::Tensor()), (running_var ? tensor_from_ocaml(running_var) : torch::Tensor()), momentum, eps, count);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_batch_norm_gather_stats_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor input, gc_tensor mean, gc_tensor invstd, gc_tensor running_mean, gc_tensor running_var, double momentum, double eps, int64_t count) {
+  PROTECT(
+    auto results__ = torch::batch_norm_gather_stats_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(mean), *tensor_ptr_from_ocaml(invstd), (running_mean ? tensor_from_ocaml(running_mean) : torch::Tensor()), (running_var ? tensor_from_ocaml(running_var) : torch::Tensor()), momentum, eps, count);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_batch_norm_gather_stats_with_counts(raw_tensor *out__, gc_tensor input, gc_tensor mean, gc_tensor invstd, gc_tensor running_mean, gc_tensor running_var, double momentum, double eps, gc_tensor counts) {
+  PROTECT(
+    auto results__ = torch::batch_norm_gather_stats_with_counts(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(mean), *tensor_ptr_from_ocaml(invstd), (running_mean ? tensor_from_ocaml(running_mean) : torch::Tensor()), (running_var ? tensor_from_ocaml(running_var) : torch::Tensor()), momentum, eps, *tensor_ptr_from_ocaml(counts));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_batch_norm_gather_stats_with_counts_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor input, gc_tensor mean, gc_tensor invstd, gc_tensor running_mean, gc_tensor running_var, double momentum, double eps, gc_tensor counts) {
+  PROTECT(
+    auto results__ = torch::batch_norm_gather_stats_with_counts_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(mean), *tensor_ptr_from_ocaml(invstd), (running_mean ? tensor_from_ocaml(running_mean) : torch::Tensor()), (running_var ? tensor_from_ocaml(running_var) : torch::Tensor()), momentum, eps, *tensor_ptr_from_ocaml(counts));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_batch_norm_stats(raw_tensor *out__, gc_tensor input, double eps) {
+  PROTECT(
+    auto results__ = torch::batch_norm_stats(*tensor_ptr_from_ocaml(input), eps);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_batch_norm_stats_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor input, double eps) {
+  PROTECT(
+    auto results__ = torch::batch_norm_stats_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(input), eps);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_batch_norm_update_stats(raw_tensor *out__, gc_tensor input, gc_tensor running_mean, gc_tensor running_var, double momentum) {
+  PROTECT(
+    auto results__ = torch::batch_norm_update_stats(*tensor_ptr_from_ocaml(input), (running_mean ? tensor_from_ocaml(running_mean) : torch::Tensor()), (running_var ? tensor_from_ocaml(running_var) : torch::Tensor()), momentum);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_batch_norm_update_stats_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor input, gc_tensor running_mean, gc_tensor running_var, double momentum) {
+  PROTECT(
+    auto results__ = torch::batch_norm_update_stats_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(input), (running_mean ? tensor_from_ocaml(running_mean) : torch::Tensor()), (running_var ? tensor_from_ocaml(running_var) : torch::Tensor()), momentum);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_bernoulli(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::bernoulli(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bernoulli_(gc_tensor self, gc_tensor p) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->bernoulli_(*tensor_ptr_from_ocaml(p));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bernoulli_float_(gc_tensor self, double p) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->bernoulli_(p);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bernoulli_p(gc_tensor self, double p) {
+  PROTECT(
+    torch::Tensor results__ = torch::bernoulli(*tensor_ptr_from_ocaml(self), p);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bernoulli_tensor(gc_tensor self, gc_tensor p) {
+  PROTECT(
+    torch::Tensor results__ = torch::bernoulli(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(p));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bilinear(gc_tensor input1, gc_tensor input2, gc_tensor weight, gc_tensor bias) {
+  PROTECT(
+    torch::Tensor results__ = torch::bilinear(*tensor_ptr_from_ocaml(input1), *tensor_ptr_from_ocaml(input2), *tensor_ptr_from_ocaml(weight), (bias ? tensor_from_ocaml(bias) : torch::Tensor()));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_binary_cross_entropy(gc_tensor self, gc_tensor target, gc_tensor weight, int64_t reduction) {
+  PROTECT(
+    torch::Tensor results__ = torch::binary_cross_entropy(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(target), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), reduction);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_binary_cross_entropy_backward(gc_tensor grad_output, gc_tensor self, gc_tensor target, gc_tensor weight, int64_t reduction) {
+  PROTECT(
+    torch::Tensor results__ = torch::binary_cross_entropy_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(target), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), reduction);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_binary_cross_entropy_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, gc_tensor target, gc_tensor weight, int64_t reduction) {
+  PROTECT(
+    torch::Tensor results__ = torch::binary_cross_entropy_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(target), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), reduction);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_binary_cross_entropy_out(gc_tensor out, gc_tensor self, gc_tensor target, gc_tensor weight, int64_t reduction) {
+  PROTECT(
+    torch::Tensor results__ = torch::binary_cross_entropy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(target), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), reduction);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_binary_cross_entropy_with_logits(gc_tensor self, gc_tensor target, gc_tensor weight, gc_tensor pos_weight, int64_t reduction) {
+  PROTECT(
+    torch::Tensor results__ = torch::binary_cross_entropy_with_logits(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(target), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), (pos_weight ? tensor_from_ocaml(pos_weight) : torch::Tensor()), reduction);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_binary_cross_entropy_with_logits_out(gc_tensor out, gc_tensor self, gc_tensor target, gc_tensor weight, gc_tensor pos_weight, int64_t reduction) {
+  PROTECT(
+    torch::Tensor results__ = torch::binary_cross_entropy_with_logits_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(target), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), (pos_weight ? tensor_from_ocaml(pos_weight) : torch::Tensor()), reduction);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bincount(gc_tensor self, gc_tensor weights, int64_t minlength) {
+  PROTECT(
+    torch::Tensor results__ = torch::bincount(*tensor_ptr_from_ocaml(self), (weights ? tensor_from_ocaml(weights) : torch::Tensor()), minlength);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bincount_out(gc_tensor out, gc_tensor self, gc_tensor weights, int64_t minlength) {
+  PROTECT(
+    torch::Tensor results__ = torch::bincount_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), (weights ? tensor_from_ocaml(weights) : torch::Tensor()), minlength);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_binomial(gc_tensor count, gc_tensor prob) {
+  PROTECT(
+    torch::Tensor results__ = torch::binomial(*tensor_ptr_from_ocaml(count), *tensor_ptr_from_ocaml(prob));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_binomial_out(gc_tensor out, gc_tensor count, gc_tensor prob) {
+  PROTECT(
+    torch::Tensor results__ = torch::binomial_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(count), *tensor_ptr_from_ocaml(prob));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bitwise_and(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::bitwise_and(*tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bitwise_and_(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->bitwise_and_(*other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bitwise_and_scalar_out(gc_tensor out, gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::bitwise_and_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bitwise_and_scalar_tensor(scalar self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::bitwise_and(*self, *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bitwise_and_scalar_tensor_out(gc_tensor out, scalar self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::bitwise_and_out(*tensor_ptr_from_ocaml(out), *self, *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bitwise_and_tensor(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::bitwise_and(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bitwise_and_tensor_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->bitwise_and_(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bitwise_and_tensor_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::bitwise_and_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bitwise_left_shift(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::bitwise_left_shift(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bitwise_left_shift_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->bitwise_left_shift_(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bitwise_left_shift_scalar_tensor(scalar self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::bitwise_left_shift(*self, *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bitwise_left_shift_scalar_tensor_out(gc_tensor out, scalar self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::bitwise_left_shift_out(*tensor_ptr_from_ocaml(out), *self, *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bitwise_left_shift_tensor_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::bitwise_left_shift_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bitwise_left_shift_tensor_scalar(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::bitwise_left_shift(*tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bitwise_left_shift_tensor_scalar_(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->bitwise_left_shift_(*other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bitwise_left_shift_tensor_scalar_out(gc_tensor out, gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::bitwise_left_shift_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bitwise_not(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::bitwise_not(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bitwise_not_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->bitwise_not_();
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bitwise_not_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::bitwise_not_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bitwise_or(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::bitwise_or(*tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bitwise_or_(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->bitwise_or_(*other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bitwise_or_scalar_out(gc_tensor out, gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::bitwise_or_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bitwise_or_scalar_tensor(scalar self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::bitwise_or(*self, *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bitwise_or_scalar_tensor_out(gc_tensor out, scalar self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::bitwise_or_out(*tensor_ptr_from_ocaml(out), *self, *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bitwise_or_tensor(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::bitwise_or(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bitwise_or_tensor_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->bitwise_or_(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bitwise_or_tensor_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::bitwise_or_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bitwise_right_shift(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::bitwise_right_shift(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bitwise_right_shift_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->bitwise_right_shift_(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bitwise_right_shift_scalar_tensor(scalar self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::bitwise_right_shift(*self, *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bitwise_right_shift_scalar_tensor_out(gc_tensor out, scalar self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::bitwise_right_shift_out(*tensor_ptr_from_ocaml(out), *self, *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bitwise_right_shift_tensor_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::bitwise_right_shift_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bitwise_right_shift_tensor_scalar(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::bitwise_right_shift(*tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bitwise_right_shift_tensor_scalar_(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->bitwise_right_shift_(*other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bitwise_right_shift_tensor_scalar_out(gc_tensor out, gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::bitwise_right_shift_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bitwise_xor(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::bitwise_xor(*tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bitwise_xor_(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->bitwise_xor_(*other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bitwise_xor_scalar_out(gc_tensor out, gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::bitwise_xor_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bitwise_xor_scalar_tensor(scalar self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::bitwise_xor(*self, *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bitwise_xor_scalar_tensor_out(gc_tensor out, scalar self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::bitwise_xor_out(*tensor_ptr_from_ocaml(out), *self, *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bitwise_xor_tensor(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::bitwise_xor(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bitwise_xor_tensor_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->bitwise_xor_(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bitwise_xor_tensor_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::bitwise_xor_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_blackman_window(int64_t window_length, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::blackman_window(window_length, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_blackman_window_out(gc_tensor out, int64_t window_length) {
+  PROTECT(
+    torch::Tensor results__ = torch::blackman_window_out(*tensor_ptr_from_ocaml(out), window_length);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_blackman_window_periodic(int64_t window_length, int periodic, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::blackman_window(window_length, (bool)periodic, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_blackman_window_periodic_out(gc_tensor out, int64_t window_length, int periodic) {
+  PROTECT(
+    torch::Tensor results__ = torch::blackman_window_out(*tensor_ptr_from_ocaml(out), window_length, (bool)periodic);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_block_diag(gc_tensor *tensors_data, int tensors_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::block_diag(of_carray_tensor(tensors_data, tensors_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_block_diag_out(gc_tensor out, gc_tensor *tensors_data, int tensors_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::block_diag_out(*tensor_ptr_from_ocaml(out), of_carray_tensor(tensors_data, tensors_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bmm(gc_tensor self, gc_tensor mat2) {
+  PROTECT(
+    torch::Tensor results__ = torch::bmm(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(mat2));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bmm_out(gc_tensor out, gc_tensor self, gc_tensor mat2) {
+  PROTECT(
+    torch::Tensor results__ = torch::bmm_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(mat2));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor *atg_broadcast_tensors(gc_tensor *tensors_data, int tensors_len) {
+  PROTECT(
+    auto results__ = torch::broadcast_tensors(of_carray_tensor(tensors_data, tensors_len));
+    int sz = results__.size();
+    raw_tensor *out__ = (raw_tensor*)malloc((sz + 1) * sizeof(raw_tensor));
+    for (int i = 0; i < sz; ++i)
+      out__[i] = tensor_to_ocaml(results__[i]);
+    out__[sz] = nullptr;
+    return out__;
+  )
+}
+
+raw_tensor atg_broadcast_to(gc_tensor self, int64_t *size_data, int size_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::broadcast_to(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(size_data, size_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bucketize(gc_tensor self, gc_tensor boundaries, int out_int32, int right) {
+  PROTECT(
+    torch::Tensor results__ = torch::bucketize(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(boundaries), (bool)out_int32, (bool)right);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bucketize_scalar(scalar self, gc_tensor boundaries, int out_int32, int right) {
+  PROTECT(
+    torch::Tensor results__ = torch::bucketize(*self, *tensor_ptr_from_ocaml(boundaries), (bool)out_int32, (bool)right);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bucketize_scalar_out(gc_tensor out, scalar self, gc_tensor boundaries, int out_int32, int right) {
+  PROTECT(
+    torch::Tensor results__ = torch::bucketize_out(*tensor_ptr_from_ocaml(out), *self, *tensor_ptr_from_ocaml(boundaries), (bool)out_int32, (bool)right);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_bucketize_tensor_out(gc_tensor out, gc_tensor self, gc_tensor boundaries, int out_int32, int right) {
+  PROTECT(
+    torch::Tensor results__ = torch::bucketize_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(boundaries), (bool)out_int32, (bool)right);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+int atg_can_cast(int from, int to) {
+  PROTECT(
+    return torch::can_cast(torch::ScalarType(from), torch::ScalarType(to));
+  )
+  return 0;
+}
+
+raw_tensor atg_cartesian_prod(gc_tensor *tensors_data, int tensors_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::cartesian_prod(of_carray_tensor(tensors_data, tensors_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_cat(gc_tensor *tensors_data, int tensors_len, int64_t dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::cat(of_carray_tensor(tensors_data, tensors_len), dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_cat_out(gc_tensor out, gc_tensor *tensors_data, int tensors_len, int64_t dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::cat_out(*tensor_ptr_from_ocaml(out), of_carray_tensor(tensors_data, tensors_len), dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_cauchy(gc_tensor self, double median, double sigma) {
+  PROTECT(
+    torch::Tensor results__ = torch::cauchy(*tensor_ptr_from_ocaml(self), median, sigma);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_cauchy_(gc_tensor self, double median, double sigma) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->cauchy_(median, sigma);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_cauchy_out(gc_tensor out, gc_tensor self, double median, double sigma) {
+  PROTECT(
+    torch::Tensor results__ = torch::cauchy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), median, sigma);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_ccol_indices(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->ccol_indices();
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_ccol_indices_copy(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::ccol_indices_copy(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_ccol_indices_copy_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::ccol_indices_copy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_cdist(gc_tensor x1, gc_tensor x2, double p, int64_t compute_mode_v, int compute_mode_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::cdist(*tensor_ptr_from_ocaml(x1), *tensor_ptr_from_ocaml(x2), p, compute_mode_null ? c10::nullopt : c10::optional<int64_t>(compute_mode_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_ceil(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::ceil(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_ceil_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::ceil_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_ceil_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::ceil_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_celu(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::celu(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_celu_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::celu_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_celu_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::celu_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_chain_matmul(gc_tensor *matrices_data, int matrices_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::chain_matmul(of_carray_tensor(matrices_data, matrices_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_chain_matmul_out(gc_tensor out, gc_tensor *matrices_data, int matrices_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::chain_matmul_out(*tensor_ptr_from_ocaml(out), of_carray_tensor(matrices_data, matrices_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_chalf(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->chalf();
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_channel_shuffle(gc_tensor self, int64_t groups) {
+  PROTECT(
+    torch::Tensor results__ = torch::channel_shuffle(*tensor_ptr_from_ocaml(self), groups);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_channel_shuffle_out(gc_tensor out, gc_tensor self, int64_t groups) {
+  PROTECT(
+    torch::Tensor results__ = torch::channel_shuffle_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), groups);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_cholesky(gc_tensor self, int upper) {
+  PROTECT(
+    torch::Tensor results__ = torch::cholesky(*tensor_ptr_from_ocaml(self), (bool)upper);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_cholesky_inverse(gc_tensor self, int upper) {
+  PROTECT(
+    torch::Tensor results__ = torch::cholesky_inverse(*tensor_ptr_from_ocaml(self), (bool)upper);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_cholesky_inverse_out(gc_tensor out, gc_tensor self, int upper) {
+  PROTECT(
+    torch::Tensor results__ = torch::cholesky_inverse_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), (bool)upper);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_cholesky_out(gc_tensor out, gc_tensor self, int upper) {
+  PROTECT(
+    torch::Tensor results__ = torch::cholesky_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), (bool)upper);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_cholesky_solve(gc_tensor self, gc_tensor input2, int upper) {
+  PROTECT(
+    torch::Tensor results__ = torch::cholesky_solve(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(input2), (bool)upper);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_cholesky_solve_out(gc_tensor out, gc_tensor self, gc_tensor input2, int upper) {
+  PROTECT(
+    torch::Tensor results__ = torch::cholesky_solve_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(input2), (bool)upper);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_choose_qparams_optimized(raw_tensor *out__, gc_tensor input, int64_t numel, int64_t n_bins, double ratio, int64_t bit_width) {
+  PROTECT(
+    auto results__ = torch::choose_qparams_optimized(*tensor_ptr_from_ocaml(input), numel, n_bins, ratio, bit_width);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor *atg_chunk(gc_tensor self, int64_t chunks, int64_t dim) {
+  PROTECT(
+    auto results__ = torch::chunk(*tensor_ptr_from_ocaml(self), chunks, dim);
+    int sz = results__.size();
+    raw_tensor *out__ = (raw_tensor*)malloc((sz + 1) * sizeof(raw_tensor));
+    for (int i = 0; i < sz; ++i)
+      out__[i] = tensor_to_ocaml(results__[i]);
+    out__[sz] = nullptr;
+    return out__;
+  )
+}
+
+raw_tensor atg_clamp(gc_tensor self, scalar min, scalar max) {
+  PROTECT(
+    torch::Tensor results__ = torch::clamp(*tensor_ptr_from_ocaml(self), *min, *max);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_clamp_(gc_tensor self, scalar min, scalar max) {
+  PROTECT(
+    torch::Tensor results__ = torch::clamp_(*tensor_ptr_from_ocaml(self), *min, *max);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_clamp_max(gc_tensor self, scalar max) {
+  PROTECT(
+    torch::Tensor results__ = torch::clamp_max(*tensor_ptr_from_ocaml(self), *max);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_clamp_max_(gc_tensor self, scalar max) {
+  PROTECT(
+    torch::Tensor results__ = torch::clamp_max_(*tensor_ptr_from_ocaml(self), *max);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_clamp_max_out(gc_tensor out, gc_tensor self, scalar max) {
+  PROTECT(
+    torch::Tensor results__ = torch::clamp_max_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *max);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_clamp_max_tensor(gc_tensor self, gc_tensor max) {
+  PROTECT(
+    torch::Tensor results__ = torch::clamp_max(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(max));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_clamp_max_tensor_(gc_tensor self, gc_tensor max) {
+  PROTECT(
+    torch::Tensor results__ = torch::clamp_max_(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(max));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_clamp_max_tensor_out(gc_tensor out, gc_tensor self, gc_tensor max) {
+  PROTECT(
+    torch::Tensor results__ = torch::clamp_max_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(max));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_clamp_min(gc_tensor self, scalar min) {
+  PROTECT(
+    torch::Tensor results__ = torch::clamp_min(*tensor_ptr_from_ocaml(self), *min);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_clamp_min_(gc_tensor self, scalar min) {
+  PROTECT(
+    torch::Tensor results__ = torch::clamp_min_(*tensor_ptr_from_ocaml(self), *min);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_clamp_min_out(gc_tensor out, gc_tensor self, scalar min) {
+  PROTECT(
+    torch::Tensor results__ = torch::clamp_min_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *min);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_clamp_min_tensor(gc_tensor self, gc_tensor min) {
+  PROTECT(
+    torch::Tensor results__ = torch::clamp_min(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(min));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_clamp_min_tensor_(gc_tensor self, gc_tensor min) {
+  PROTECT(
+    torch::Tensor results__ = torch::clamp_min_(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(min));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_clamp_min_tensor_out(gc_tensor out, gc_tensor self, gc_tensor min) {
+  PROTECT(
+    torch::Tensor results__ = torch::clamp_min_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(min));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_clamp_out(gc_tensor out, gc_tensor self, scalar min, scalar max) {
+  PROTECT(
+    torch::Tensor results__ = torch::clamp_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *min, *max);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_clamp_tensor(gc_tensor self, gc_tensor min, gc_tensor max) {
+  PROTECT(
+    torch::Tensor results__ = torch::clamp(*tensor_ptr_from_ocaml(self), (min ? tensor_from_ocaml(min) : torch::Tensor()), (max ? tensor_from_ocaml(max) : torch::Tensor()));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_clamp_tensor_(gc_tensor self, gc_tensor min, gc_tensor max) {
+  PROTECT(
+    torch::Tensor results__ = torch::clamp_(*tensor_ptr_from_ocaml(self), (min ? tensor_from_ocaml(min) : torch::Tensor()), (max ? tensor_from_ocaml(max) : torch::Tensor()));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_clamp_tensor_out(gc_tensor out, gc_tensor self, gc_tensor min, gc_tensor max) {
+  PROTECT(
+    torch::Tensor results__ = torch::clamp_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), (min ? tensor_from_ocaml(min) : torch::Tensor()), (max ? tensor_from_ocaml(max) : torch::Tensor()));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_clip(gc_tensor self, scalar min, scalar max) {
+  PROTECT(
+    torch::Tensor results__ = torch::clip(*tensor_ptr_from_ocaml(self), *min, *max);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_clip_(gc_tensor self, scalar min, scalar max) {
+  PROTECT(
+    torch::Tensor results__ = torch::clip_(*tensor_ptr_from_ocaml(self), *min, *max);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_clip_out(gc_tensor out, gc_tensor self, scalar min, scalar max) {
+  PROTECT(
+    torch::Tensor results__ = torch::clip_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *min, *max);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_clip_tensor(gc_tensor self, gc_tensor min, gc_tensor max) {
+  PROTECT(
+    torch::Tensor results__ = torch::clip(*tensor_ptr_from_ocaml(self), (min ? tensor_from_ocaml(min) : torch::Tensor()), (max ? tensor_from_ocaml(max) : torch::Tensor()));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_clip_tensor_(gc_tensor self, gc_tensor min, gc_tensor max) {
+  PROTECT(
+    torch::Tensor results__ = torch::clip_(*tensor_ptr_from_ocaml(self), (min ? tensor_from_ocaml(min) : torch::Tensor()), (max ? tensor_from_ocaml(max) : torch::Tensor()));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_clip_tensor_out(gc_tensor out, gc_tensor self, gc_tensor min, gc_tensor max) {
+  PROTECT(
+    torch::Tensor results__ = torch::clip_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), (min ? tensor_from_ocaml(min) : torch::Tensor()), (max ? tensor_from_ocaml(max) : torch::Tensor()));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_clone(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::clone(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_clone_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::clone_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_coalesce(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->coalesce();
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_col2im(gc_tensor self, int64_t *output_size_data, int output_size_len, int64_t *kernel_size_data, int kernel_size_len, int64_t *dilation_data, int dilation_len, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::col2im(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(dilation_data, dilation_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(stride_data, stride_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_col2im_out(gc_tensor out, gc_tensor self, int64_t *output_size_data, int output_size_len, int64_t *kernel_size_data, int kernel_size_len, int64_t *dilation_data, int dilation_len, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::col2im_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(dilation_data, dilation_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(stride_data, stride_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_col_indices(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->col_indices();
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_col_indices_copy(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::col_indices_copy(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_col_indices_copy_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::col_indices_copy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_column_stack(gc_tensor *tensors_data, int tensors_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::column_stack(of_carray_tensor(tensors_data, tensors_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_column_stack_out(gc_tensor out, gc_tensor *tensors_data, int tensors_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::column_stack_out(*tensor_ptr_from_ocaml(out), of_carray_tensor(tensors_data, tensors_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_combinations(gc_tensor self, int64_t r, int with_replacement) {
+  PROTECT(
+    torch::Tensor results__ = torch::combinations(*tensor_ptr_from_ocaml(self), r, (bool)with_replacement);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_complex(gc_tensor real, gc_tensor imag) {
+  PROTECT(
+    torch::Tensor results__ = torch::complex(*tensor_ptr_from_ocaml(real), *tensor_ptr_from_ocaml(imag));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_complex_out(gc_tensor out, gc_tensor real, gc_tensor imag) {
+  PROTECT(
+    torch::Tensor results__ = torch::complex_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(real), *tensor_ptr_from_ocaml(imag));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_concat(gc_tensor *tensors_data, int tensors_len, int64_t dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::concat(of_carray_tensor(tensors_data, tensors_len), dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_concat_out(gc_tensor out, gc_tensor *tensors_data, int tensors_len, int64_t dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::concat_out(*tensor_ptr_from_ocaml(out), of_carray_tensor(tensors_data, tensors_len), dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_concatenate(gc_tensor *tensors_data, int tensors_len, int64_t dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::concatenate(of_carray_tensor(tensors_data, tensors_len), dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_concatenate_out(gc_tensor out, gc_tensor *tensors_data, int tensors_len, int64_t dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::concatenate_out(*tensor_ptr_from_ocaml(out), of_carray_tensor(tensors_data, tensors_len), dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_conj(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::conj(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_conj_physical(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::conj_physical(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_conj_physical_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::conj_physical_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_conj_physical_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::conj_physical_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_constant_pad_nd(gc_tensor self, int64_t *pad_data, int pad_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::constant_pad_nd(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(pad_data, pad_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_constant_pad_nd_out(gc_tensor out, gc_tensor self, int64_t *pad_data, int pad_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::constant_pad_nd_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(pad_data, pad_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_contiguous(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->contiguous();
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_conv1d(gc_tensor input, gc_tensor weight, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int64_t groups) {
+  PROTECT(
+    torch::Tensor results__ = torch::conv1d(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(weight), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), groups);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_conv1d_padding(gc_tensor input, gc_tensor weight, gc_tensor bias, int64_t *stride_data, int stride_len, char * padding, int64_t *dilation_data, int dilation_len, int64_t groups) {
+  PROTECT(
+    torch::Tensor results__ = torch::conv1d(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(weight), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), std::string(padding), torch::IntArrayRef(dilation_data, dilation_len), groups);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_conv2d(gc_tensor input, gc_tensor weight, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int64_t groups) {
+  PROTECT(
+    torch::Tensor results__ = torch::conv2d(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(weight), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), groups);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_conv2d_padding(gc_tensor input, gc_tensor weight, gc_tensor bias, int64_t *stride_data, int stride_len, char * padding, int64_t *dilation_data, int dilation_len, int64_t groups) {
+  PROTECT(
+    torch::Tensor results__ = torch::conv2d(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(weight), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), std::string(padding), torch::IntArrayRef(dilation_data, dilation_len), groups);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_conv3d(gc_tensor input, gc_tensor weight, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int64_t groups) {
+  PROTECT(
+    torch::Tensor results__ = torch::conv3d(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(weight), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), groups);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_conv3d_padding(gc_tensor input, gc_tensor weight, gc_tensor bias, int64_t *stride_data, int stride_len, char * padding, int64_t *dilation_data, int dilation_len, int64_t groups) {
+  PROTECT(
+    torch::Tensor results__ = torch::conv3d(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(weight), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), std::string(padding), torch::IntArrayRef(dilation_data, dilation_len), groups);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_conv_depthwise3d(gc_tensor self, gc_tensor weight, int64_t *kernel_size_data, int kernel_size_len, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::conv_depthwise3d(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(weight), torch::IntArrayRef(kernel_size_data, kernel_size_len), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_conv_depthwise3d_out(gc_tensor out, gc_tensor self, gc_tensor weight, int64_t *kernel_size_data, int kernel_size_len, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::conv_depthwise3d_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(weight), torch::IntArrayRef(kernel_size_data, kernel_size_len), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_conv_tbc(gc_tensor self, gc_tensor weight, gc_tensor bias, int64_t pad) {
+  PROTECT(
+    torch::Tensor results__ = torch::conv_tbc(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(weight), *tensor_ptr_from_ocaml(bias), pad);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_conv_tbc_backward(raw_tensor *out__, gc_tensor self, gc_tensor input, gc_tensor weight, gc_tensor bias, int64_t pad) {
+  PROTECT(
+    auto results__ = torch::conv_tbc_backward(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(weight), *tensor_ptr_from_ocaml(bias), pad);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+raw_tensor atg_conv_tbc_out(gc_tensor out, gc_tensor self, gc_tensor weight, gc_tensor bias, int64_t pad) {
+  PROTECT(
+    torch::Tensor results__ = torch::conv_tbc_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(weight), *tensor_ptr_from_ocaml(bias), pad);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_conv_transpose1d(gc_tensor input, gc_tensor weight, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t groups, int64_t *dilation_data, int dilation_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::conv_transpose1d(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(weight), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(output_padding_data, output_padding_len), groups, torch::IntArrayRef(dilation_data, dilation_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_conv_transpose2d(gc_tensor input, gc_tensor weight, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t groups, int64_t *dilation_data, int dilation_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::conv_transpose2d(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(weight), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(output_padding_data, output_padding_len), groups, torch::IntArrayRef(dilation_data, dilation_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_conv_transpose3d(gc_tensor input, gc_tensor weight, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t groups, int64_t *dilation_data, int dilation_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::conv_transpose3d(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(weight), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(output_padding_data, output_padding_len), groups, torch::IntArrayRef(dilation_data, dilation_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_convolution(gc_tensor input, gc_tensor weight, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int transposed, int64_t *output_padding_data, int output_padding_len, int64_t groups) {
+  PROTECT(
+    torch::Tensor results__ = torch::convolution(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(weight), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)transposed, torch::IntArrayRef(output_padding_data, output_padding_len), groups);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_convolution_out(gc_tensor out, gc_tensor input, gc_tensor weight, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int transposed, int64_t *output_padding_data, int output_padding_len, int64_t groups) {
+  PROTECT(
+    torch::Tensor results__ = torch::convolution_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(weight), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)transposed, torch::IntArrayRef(output_padding_data, output_padding_len), groups);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_convolution_overrideable(gc_tensor input, gc_tensor weight, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int transposed, int64_t *output_padding_data, int output_padding_len, int64_t groups) {
+  PROTECT(
+    torch::Tensor results__ = torch::convolution_overrideable(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(weight), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)transposed, torch::IntArrayRef(output_padding_data, output_padding_len), groups);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_convolution_overrideable_out(gc_tensor out, gc_tensor input, gc_tensor weight, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int transposed, int64_t *output_padding_data, int output_padding_len, int64_t groups) {
+  PROTECT(
+    torch::Tensor results__ = torch::convolution_overrideable_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(weight), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)transposed, torch::IntArrayRef(output_padding_data, output_padding_len), groups);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_copy(gc_tensor self, gc_tensor src, int non_blocking) {
+  PROTECT(
+    torch::Tensor results__ = torch::copy(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(src), (bool)non_blocking);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_copy_out(gc_tensor out, gc_tensor self, gc_tensor src, int non_blocking) {
+  PROTECT(
+    torch::Tensor results__ = torch::copy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(src), (bool)non_blocking);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_copy_sparse_to_sparse(gc_tensor self, gc_tensor src, int non_blocking) {
+  PROTECT(
+    torch::Tensor results__ = torch::copy_sparse_to_sparse(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(src), (bool)non_blocking);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_copy_sparse_to_sparse_(gc_tensor self, gc_tensor src, int non_blocking) {
+  PROTECT(
+    torch::Tensor results__ = torch::copy_sparse_to_sparse_(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(src), (bool)non_blocking);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_copy_sparse_to_sparse_out(gc_tensor out, gc_tensor self, gc_tensor src, int non_blocking) {
+  PROTECT(
+    torch::Tensor results__ = torch::copy_sparse_to_sparse_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(src), (bool)non_blocking);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_copysign(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::copysign(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_copysign_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->copysign_(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_copysign_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::copysign_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_copysign_scalar(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::copysign(*tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_copysign_scalar_(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->copysign_(*other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_copysign_scalar_out(gc_tensor out, gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::copysign_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_corrcoef(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::corrcoef(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_cos(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::cos(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_cos_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::cos_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_cos_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::cos_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_cosh(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::cosh(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_cosh_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::cosh_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_cosh_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::cosh_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_cosine_embedding_loss(gc_tensor input1, gc_tensor input2, gc_tensor target, double margin, int64_t reduction) {
+  PROTECT(
+    torch::Tensor results__ = torch::cosine_embedding_loss(*tensor_ptr_from_ocaml(input1), *tensor_ptr_from_ocaml(input2), *tensor_ptr_from_ocaml(target), margin, reduction);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_cosine_similarity(gc_tensor x1, gc_tensor x2, int64_t dim, double eps) {
+  PROTECT(
+    torch::Tensor results__ = torch::cosine_similarity(*tensor_ptr_from_ocaml(x1), *tensor_ptr_from_ocaml(x2), dim, eps);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_count_nonzero(gc_tensor out, gc_tensor self, int64_t *dim_data, int dim_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::count_nonzero_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(dim_data, dim_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_count_nonzero_out(gc_tensor out, gc_tensor self, int64_t dim_v, int dim_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::count_nonzero_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_cov(gc_tensor self, int64_t correction, gc_tensor fweights, gc_tensor aweights) {
+  PROTECT(
+    torch::Tensor results__ = torch::cov(*tensor_ptr_from_ocaml(self), correction, (fweights ? tensor_from_ocaml(fweights) : torch::Tensor()), (aweights ? tensor_from_ocaml(aweights) : torch::Tensor()));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_cross(gc_tensor self, gc_tensor other, int64_t dim_v, int dim_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::cross(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other), dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_cross_entropy_loss(gc_tensor self, gc_tensor target, gc_tensor weight, int64_t reduction, int64_t ignore_index, double label_smoothing) {
+  PROTECT(
+    torch::Tensor results__ = torch::cross_entropy_loss(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(target), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), reduction, ignore_index, label_smoothing);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_cross_out(gc_tensor out, gc_tensor self, gc_tensor other, int64_t dim_v, int dim_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::cross_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other), dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_crow_indices(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->crow_indices();
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_crow_indices_copy(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::crow_indices_copy(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_crow_indices_copy_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::crow_indices_copy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_ctc_loss(gc_tensor log_probs, gc_tensor targets, int64_t *input_lengths_data, int input_lengths_len, int64_t *target_lengths_data, int target_lengths_len, int64_t blank, int64_t reduction, int zero_infinity) {
+  PROTECT(
+    torch::Tensor results__ = torch::ctc_loss(*tensor_ptr_from_ocaml(log_probs), *tensor_ptr_from_ocaml(targets), torch::IntArrayRef(input_lengths_data, input_lengths_len), torch::IntArrayRef(target_lengths_data, target_lengths_len), blank, reduction, (bool)zero_infinity);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_ctc_loss_tensor(gc_tensor log_probs, gc_tensor targets, gc_tensor input_lengths, gc_tensor target_lengths, int64_t blank, int64_t reduction, int zero_infinity) {
+  PROTECT(
+    torch::Tensor results__ = torch::ctc_loss(*tensor_ptr_from_ocaml(log_probs), *tensor_ptr_from_ocaml(targets), *tensor_ptr_from_ocaml(input_lengths), *tensor_ptr_from_ocaml(target_lengths), blank, reduction, (bool)zero_infinity);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_cudnn_affine_grid_generator(gc_tensor theta, int64_t n, int64_t C, int64_t H, int64_t W) {
+  PROTECT(
+    torch::Tensor results__ = torch::cudnn_affine_grid_generator(*tensor_ptr_from_ocaml(theta), n, C, H, W);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_cudnn_affine_grid_generator_backward(gc_tensor grad, int64_t n, int64_t C, int64_t H, int64_t W) {
+  PROTECT(
+    torch::Tensor results__ = torch::cudnn_affine_grid_generator_backward(*tensor_ptr_from_ocaml(grad), n, C, H, W);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_cudnn_affine_grid_generator_backward_out(gc_tensor out, gc_tensor grad, int64_t n, int64_t C, int64_t H, int64_t W) {
+  PROTECT(
+    torch::Tensor results__ = torch::cudnn_affine_grid_generator_backward_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(grad), n, C, H, W);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_cudnn_affine_grid_generator_out(gc_tensor out, gc_tensor theta, int64_t n, int64_t C, int64_t H, int64_t W) {
+  PROTECT(
+    torch::Tensor results__ = torch::cudnn_affine_grid_generator_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(theta), n, C, H, W);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_cudnn_batch_norm(raw_tensor *out__, gc_tensor input, gc_tensor weight, gc_tensor bias, gc_tensor running_mean, gc_tensor running_var, int training, double exponential_average_factor, double epsilon) {
+  PROTECT(
+    auto results__ = torch::cudnn_batch_norm(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(weight), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), (running_mean ? tensor_from_ocaml(running_mean) : torch::Tensor()), (running_var ? tensor_from_ocaml(running_var) : torch::Tensor()), (bool)training, exponential_average_factor, epsilon);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+    out__[3] = tensor_to_ocaml(std::get<3>(results__));
+  )
+}
+
+void atg_cudnn_batch_norm_backward(raw_tensor *out__, gc_tensor input, gc_tensor grad_output, gc_tensor weight, gc_tensor running_mean, gc_tensor running_var, gc_tensor save_mean, gc_tensor save_var, double epsilon, gc_tensor reserveSpace) {
+  PROTECT(
+    auto results__ = torch::cudnn_batch_norm_backward(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(weight), (running_mean ? tensor_from_ocaml(running_mean) : torch::Tensor()), (running_var ? tensor_from_ocaml(running_var) : torch::Tensor()), (save_mean ? tensor_from_ocaml(save_mean) : torch::Tensor()), (save_var ? tensor_from_ocaml(save_var) : torch::Tensor()), epsilon, *tensor_ptr_from_ocaml(reserveSpace));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+void atg_cudnn_batch_norm_backward_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor input, gc_tensor grad_output, gc_tensor weight, gc_tensor running_mean, gc_tensor running_var, gc_tensor save_mean, gc_tensor save_var, double epsilon, gc_tensor reserveSpace) {
+  PROTECT(
+    auto results__ = torch::cudnn_batch_norm_backward_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(out2), *tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(weight), (running_mean ? tensor_from_ocaml(running_mean) : torch::Tensor()), (running_var ? tensor_from_ocaml(running_var) : torch::Tensor()), (save_mean ? tensor_from_ocaml(save_mean) : torch::Tensor()), (save_var ? tensor_from_ocaml(save_var) : torch::Tensor()), epsilon, *tensor_ptr_from_ocaml(reserveSpace));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+void atg_cudnn_batch_norm_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor out3, gc_tensor input, gc_tensor weight, gc_tensor bias, gc_tensor running_mean, gc_tensor running_var, int training, double exponential_average_factor, double epsilon) {
+  PROTECT(
+    auto results__ = torch::cudnn_batch_norm_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(out2), *tensor_ptr_from_ocaml(out3), *tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(weight), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), (running_mean ? tensor_from_ocaml(running_mean) : torch::Tensor()), (running_var ? tensor_from_ocaml(running_var) : torch::Tensor()), (bool)training, exponential_average_factor, epsilon);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+    out__[3] = tensor_to_ocaml(std::get<3>(results__));
+  )
+}
+
+raw_tensor atg_cudnn_convolution(gc_tensor self, gc_tensor weight, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int benchmark, int deterministic, int allow_tf32) {
+  PROTECT(
+    torch::Tensor results__ = torch::cudnn_convolution(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(weight), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(dilation_data, dilation_len), groups, (bool)benchmark, (bool)deterministic, (bool)allow_tf32);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_cudnn_convolution_add_relu(gc_tensor self, gc_tensor weight, gc_tensor z, scalar alpha, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int64_t groups) {
+  PROTECT(
+    torch::Tensor results__ = torch::cudnn_convolution_add_relu(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(weight), *tensor_ptr_from_ocaml(z), *alpha, (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), groups);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_cudnn_convolution_add_relu_out(gc_tensor out, gc_tensor self, gc_tensor weight, gc_tensor z, scalar alpha, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int64_t groups) {
+  PROTECT(
+    torch::Tensor results__ = torch::cudnn_convolution_add_relu_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(weight), *tensor_ptr_from_ocaml(z), *alpha, (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), groups);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_cudnn_convolution_out(gc_tensor out, gc_tensor self, gc_tensor weight, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int benchmark, int deterministic, int allow_tf32) {
+  PROTECT(
+    torch::Tensor results__ = torch::cudnn_convolution_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(weight), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(dilation_data, dilation_len), groups, (bool)benchmark, (bool)deterministic, (bool)allow_tf32);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_cudnn_convolution_relu(gc_tensor self, gc_tensor weight, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int64_t groups) {
+  PROTECT(
+    torch::Tensor results__ = torch::cudnn_convolution_relu(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(weight), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), groups);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_cudnn_convolution_relu_out(gc_tensor out, gc_tensor self, gc_tensor weight, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int64_t groups) {
+  PROTECT(
+    torch::Tensor results__ = torch::cudnn_convolution_relu_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(weight), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), groups);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_cudnn_convolution_transpose(gc_tensor self, gc_tensor weight, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int benchmark, int deterministic, int allow_tf32) {
+  PROTECT(
+    torch::Tensor results__ = torch::cudnn_convolution_transpose(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(weight), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(output_padding_data, output_padding_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(dilation_data, dilation_len), groups, (bool)benchmark, (bool)deterministic, (bool)allow_tf32);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_cudnn_convolution_transpose_out(gc_tensor out, gc_tensor self, gc_tensor weight, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int benchmark, int deterministic, int allow_tf32) {
+  PROTECT(
+    torch::Tensor results__ = torch::cudnn_convolution_transpose_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(weight), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(output_padding_data, output_padding_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(dilation_data, dilation_len), groups, (bool)benchmark, (bool)deterministic, (bool)allow_tf32);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_cudnn_grid_sampler(gc_tensor self, gc_tensor grid) {
+  PROTECT(
+    torch::Tensor results__ = torch::cudnn_grid_sampler(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(grid));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_cudnn_grid_sampler_backward(raw_tensor *out__, gc_tensor self, gc_tensor grid, gc_tensor grad_output) {
+  PROTECT(
+    auto results__ = torch::cudnn_grid_sampler_backward(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(grid), *tensor_ptr_from_ocaml(grad_output));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_cudnn_grid_sampler_backward_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor self, gc_tensor grid, gc_tensor grad_output) {
+  PROTECT(
+    auto results__ = torch::cudnn_grid_sampler_backward_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(grid), *tensor_ptr_from_ocaml(grad_output));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_cudnn_grid_sampler_out(gc_tensor out, gc_tensor self, gc_tensor grid) {
+  PROTECT(
+    torch::Tensor results__ = torch::cudnn_grid_sampler_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(grid));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+int atg_cudnn_is_acceptable(gc_tensor self) {
+  PROTECT(
+    return torch::cudnn_is_acceptable(*tensor_ptr_from_ocaml(self));
+  )
+  return 0;
+}
+
+void atg_cummax(raw_tensor *out__, gc_tensor self, int64_t dim) {
+  PROTECT(
+    auto results__ = torch::cummax(*tensor_ptr_from_ocaml(self), dim);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_cummax_out(raw_tensor *out__, gc_tensor values, gc_tensor indices, gc_tensor self, int64_t dim) {
+  PROTECT(
+    auto results__ = torch::cummax_out(*tensor_ptr_from_ocaml(values), *tensor_ptr_from_ocaml(indices), *tensor_ptr_from_ocaml(self), dim);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_cummaxmin_backward(gc_tensor grad, gc_tensor input, gc_tensor indices, int64_t dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::cummaxmin_backward(*tensor_ptr_from_ocaml(grad), *tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(indices), dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_cummin(raw_tensor *out__, gc_tensor self, int64_t dim) {
+  PROTECT(
+    auto results__ = torch::cummin(*tensor_ptr_from_ocaml(self), dim);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_cummin_out(raw_tensor *out__, gc_tensor values, gc_tensor indices, gc_tensor self, int64_t dim) {
+  PROTECT(
+    auto results__ = torch::cummin_out(*tensor_ptr_from_ocaml(values), *tensor_ptr_from_ocaml(indices), *tensor_ptr_from_ocaml(self), dim);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_cumprod(gc_tensor self, int64_t dim, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::cumprod(*tensor_ptr_from_ocaml(self), dim, torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_cumprod_(gc_tensor self, int64_t dim, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->cumprod_(dim, torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_cumprod_backward(gc_tensor grad, gc_tensor input, int64_t dim, gc_tensor output) {
+  PROTECT(
+    torch::Tensor results__ = torch::cumprod_backward(*tensor_ptr_from_ocaml(grad), *tensor_ptr_from_ocaml(input), dim, *tensor_ptr_from_ocaml(output));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_cumprod_out(gc_tensor out, gc_tensor self, int64_t dim, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::cumprod_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dim, torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_cumsum(gc_tensor self, int64_t dim, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::cumsum(*tensor_ptr_from_ocaml(self), dim, torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_cumsum_(gc_tensor self, int64_t dim, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->cumsum_(dim, torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_cumsum_out(gc_tensor out, gc_tensor self, int64_t dim, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::cumsum_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dim, torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_cumulative_trapezoid(gc_tensor y, int64_t dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::cumulative_trapezoid(*tensor_ptr_from_ocaml(y), dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_cumulative_trapezoid_x(gc_tensor y, gc_tensor x, int64_t dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::cumulative_trapezoid(*tensor_ptr_from_ocaml(y), *tensor_ptr_from_ocaml(x), dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_data(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->data();
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_deg2rad(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::deg2rad(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_deg2rad_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::deg2rad_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_deg2rad_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::deg2rad_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+int64_t atg_dense_dim(gc_tensor self) {
+  PROTECT(
+    return tensor_ptr_from_ocaml(self)->dense_dim();
+  )
+  return 0;
+}
+
+raw_tensor atg_dequantize(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::dequantize(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_dequantize_self_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::dequantize_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor *atg_dequantize_tensors(gc_tensor *tensors_data, int tensors_len) {
+  PROTECT(
+    auto results__ = torch::dequantize(of_carray_tensor(tensors_data, tensors_len));
+    int sz = results__.size();
+    raw_tensor *out__ = (raw_tensor*)malloc((sz + 1) * sizeof(raw_tensor));
+    for (int i = 0; i < sz; ++i)
+      out__[i] = tensor_to_ocaml(results__[i]);
+    out__[sz] = nullptr;
+    return out__;
+  )
+}
+
+void atg_dequantize_tensors_out(gc_tensor *out_data, int out_len, gc_tensor *tensors_data, int tensors_len) {
+  PROTECT(
+    torch::dequantize_out(of_carray_tensor(out_data, out_len), of_carray_tensor(tensors_data, tensors_len));
+  )
+}
+
+raw_tensor atg_det(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::det(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_detach(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::detach(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_detach_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::detach_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_detach_copy(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::detach_copy(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_detach_copy_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::detach_copy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_diag(gc_tensor self, int64_t diagonal) {
+  PROTECT(
+    torch::Tensor results__ = torch::diag(*tensor_ptr_from_ocaml(self), diagonal);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_diag_embed(gc_tensor self, int64_t offset, int64_t dim1, int64_t dim2) {
+  PROTECT(
+    torch::Tensor results__ = torch::diag_embed(*tensor_ptr_from_ocaml(self), offset, dim1, dim2);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_diag_embed_out(gc_tensor out, gc_tensor self, int64_t offset, int64_t dim1, int64_t dim2) {
+  PROTECT(
+    torch::Tensor results__ = torch::diag_embed_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), offset, dim1, dim2);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_diag_out(gc_tensor out, gc_tensor self, int64_t diagonal) {
+  PROTECT(
+    torch::Tensor results__ = torch::diag_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), diagonal);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_diagflat(gc_tensor self, int64_t offset) {
+  PROTECT(
+    torch::Tensor results__ = torch::diagflat(*tensor_ptr_from_ocaml(self), offset);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_diagonal(gc_tensor self, int64_t offset, int64_t dim1, int64_t dim2) {
+  PROTECT(
+    torch::Tensor results__ = torch::diagonal(*tensor_ptr_from_ocaml(self), offset, dim1, dim2);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_diagonal_backward(gc_tensor grad_output, int64_t *input_sizes_data, int input_sizes_len, int64_t offset, int64_t dim1, int64_t dim2) {
+  PROTECT(
+    torch::Tensor results__ = torch::diagonal_backward(*tensor_ptr_from_ocaml(grad_output), torch::IntArrayRef(input_sizes_data, input_sizes_len), offset, dim1, dim2);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_diagonal_backward_out(gc_tensor out, gc_tensor grad_output, int64_t *input_sizes_data, int input_sizes_len, int64_t offset, int64_t dim1, int64_t dim2) {
+  PROTECT(
+    torch::Tensor results__ = torch::diagonal_backward_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(grad_output), torch::IntArrayRef(input_sizes_data, input_sizes_len), offset, dim1, dim2);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_diagonal_copy(gc_tensor self, int64_t offset, int64_t dim1, int64_t dim2) {
+  PROTECT(
+    torch::Tensor results__ = torch::diagonal_copy(*tensor_ptr_from_ocaml(self), offset, dim1, dim2);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_diagonal_copy_out(gc_tensor out, gc_tensor self, int64_t offset, int64_t dim1, int64_t dim2) {
+  PROTECT(
+    torch::Tensor results__ = torch::diagonal_copy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), offset, dim1, dim2);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_diagonal_scatter(gc_tensor self, gc_tensor src, int64_t offset, int64_t dim1, int64_t dim2) {
+  PROTECT(
+    torch::Tensor results__ = torch::diagonal_scatter(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(src), offset, dim1, dim2);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_diagonal_scatter_out(gc_tensor out, gc_tensor self, gc_tensor src, int64_t offset, int64_t dim1, int64_t dim2) {
+  PROTECT(
+    torch::Tensor results__ = torch::diagonal_scatter_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(src), offset, dim1, dim2);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_diff(gc_tensor self, int64_t n, int64_t dim, gc_tensor prepend, gc_tensor append) {
+  PROTECT(
+    torch::Tensor results__ = torch::diff(*tensor_ptr_from_ocaml(self), n, dim, (prepend ? tensor_from_ocaml(prepend) : torch::Tensor()), (append ? tensor_from_ocaml(append) : torch::Tensor()));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_diff_out(gc_tensor out, gc_tensor self, int64_t n, int64_t dim, gc_tensor prepend, gc_tensor append) {
+  PROTECT(
+    torch::Tensor results__ = torch::diff_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), n, dim, (prepend ? tensor_from_ocaml(prepend) : torch::Tensor()), (append ? tensor_from_ocaml(append) : torch::Tensor()));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_digamma(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::digamma(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_digamma_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->digamma_();
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_digamma_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::digamma_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_dist(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::dist(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_dist_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::dist_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_div(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::div(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_div_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->div_(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_div_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::div_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_div_out_mode(gc_tensor out, gc_tensor self, gc_tensor other, char * rounding_mode) {
+  PROTECT(
+    torch::Tensor results__ = torch::div_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other), std::string(rounding_mode));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_div_scalar(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::div(*tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_div_scalar_(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->div_(*other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_div_scalar_mode(gc_tensor self, scalar other, char * rounding_mode) {
+  PROTECT(
+    torch::Tensor results__ = torch::div(*tensor_ptr_from_ocaml(self), *other, std::string(rounding_mode));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_div_scalar_mode_(gc_tensor self, scalar other, char * rounding_mode) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->div_(*other, std::string(rounding_mode));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_div_scalar_mode_out(gc_tensor out, gc_tensor self, scalar other, char * rounding_mode) {
+  PROTECT(
+    torch::Tensor results__ = torch::div_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *other, std::string(rounding_mode));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_div_scalar_out(gc_tensor out, gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::div_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_div_tensor_mode(gc_tensor self, gc_tensor other, char * rounding_mode) {
+  PROTECT(
+    torch::Tensor results__ = torch::div(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other), std::string(rounding_mode));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_div_tensor_mode_(gc_tensor self, gc_tensor other, char * rounding_mode) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->div_(*tensor_ptr_from_ocaml(other), std::string(rounding_mode));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_divide(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::divide(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_divide_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->divide_(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_divide_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::divide_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_divide_out_mode(gc_tensor out, gc_tensor self, gc_tensor other, char * rounding_mode) {
+  PROTECT(
+    torch::Tensor results__ = torch::divide_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other), std::string(rounding_mode));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_divide_scalar(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::divide(*tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_divide_scalar_(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->divide_(*other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_divide_scalar_mode(gc_tensor self, scalar other, char * rounding_mode) {
+  PROTECT(
+    torch::Tensor results__ = torch::divide(*tensor_ptr_from_ocaml(self), *other, std::string(rounding_mode));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_divide_scalar_mode_(gc_tensor self, scalar other, char * rounding_mode) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->divide_(*other, std::string(rounding_mode));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_divide_tensor_mode(gc_tensor self, gc_tensor other, char * rounding_mode) {
+  PROTECT(
+    torch::Tensor results__ = torch::divide(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other), std::string(rounding_mode));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_divide_tensor_mode_(gc_tensor self, gc_tensor other, char * rounding_mode) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->divide_(*tensor_ptr_from_ocaml(other), std::string(rounding_mode));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_dot(gc_tensor self, gc_tensor tensor) {
+  PROTECT(
+    torch::Tensor results__ = torch::dot(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(tensor));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_dot_out(gc_tensor out, gc_tensor self, gc_tensor tensor) {
+  PROTECT(
+    torch::Tensor results__ = torch::dot_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(tensor));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_dropout(gc_tensor input, double p, int train) {
+  PROTECT(
+    torch::Tensor results__ = torch::dropout(*tensor_ptr_from_ocaml(input), p, (bool)train);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_dropout_(gc_tensor self, double p, int train) {
+  PROTECT(
+    torch::Tensor results__ = torch::dropout_(*tensor_ptr_from_ocaml(self), p, (bool)train);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor *atg_dsplit(gc_tensor self, int64_t sections) {
+  PROTECT(
+    auto results__ = torch::dsplit(*tensor_ptr_from_ocaml(self), sections);
+    int sz = results__.size();
+    raw_tensor *out__ = (raw_tensor*)malloc((sz + 1) * sizeof(raw_tensor));
+    for (int i = 0; i < sz; ++i)
+      out__[i] = tensor_to_ocaml(results__[i]);
+    out__[sz] = nullptr;
+    return out__;
+  )
+}
+
+raw_tensor *atg_dsplit_array(gc_tensor self, int64_t *indices_data, int indices_len) {
+  PROTECT(
+    auto results__ = torch::dsplit(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(indices_data, indices_len));
+    int sz = results__.size();
+    raw_tensor *out__ = (raw_tensor*)malloc((sz + 1) * sizeof(raw_tensor));
+    for (int i = 0; i < sz; ++i)
+      out__[i] = tensor_to_ocaml(results__[i]);
+    out__[sz] = nullptr;
+    return out__;
+  )
+}
+
+raw_tensor atg_dstack(gc_tensor *tensors_data, int tensors_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::dstack(of_carray_tensor(tensors_data, tensors_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_dstack_out(gc_tensor out, gc_tensor *tensors_data, int tensors_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::dstack_out(*tensor_ptr_from_ocaml(out), of_carray_tensor(tensors_data, tensors_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_einsum(char * equation, gc_tensor *tensors_data, int tensors_len, int64_t *path_data, int path_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::einsum(std::string(equation), of_carray_tensor(tensors_data, tensors_len), path_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(path_data, path_len)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_elu(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::elu(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_elu_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::elu_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_elu_backward(gc_tensor grad_output, scalar alpha, scalar scale, scalar input_scale, int is_result, gc_tensor self_or_result) {
+  PROTECT(
+    torch::Tensor results__ = torch::elu_backward(*tensor_ptr_from_ocaml(grad_output), *alpha, *scale, *input_scale, (bool)is_result, *tensor_ptr_from_ocaml(self_or_result));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_elu_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, scalar alpha, scalar scale, scalar input_scale, int is_result, gc_tensor self_or_result) {
+  PROTECT(
+    torch::Tensor results__ = torch::elu_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), *alpha, *scale, *input_scale, (bool)is_result, *tensor_ptr_from_ocaml(self_or_result));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_elu_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::elu_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_embedding(gc_tensor weight, gc_tensor indices, int64_t padding_idx, int scale_grad_by_freq, int sparse) {
+  PROTECT(
+    torch::Tensor results__ = torch::embedding(*tensor_ptr_from_ocaml(weight), *tensor_ptr_from_ocaml(indices), padding_idx, (bool)scale_grad_by_freq, (bool)sparse);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_embedding_backward(gc_tensor grad, gc_tensor indices, int64_t num_weights, int64_t padding_idx, int scale_grad_by_freq, int sparse) {
+  PROTECT(
+    torch::Tensor results__ = torch::embedding_backward(*tensor_ptr_from_ocaml(grad), *tensor_ptr_from_ocaml(indices), num_weights, padding_idx, (bool)scale_grad_by_freq, (bool)sparse);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_embedding_bag(raw_tensor *out__, gc_tensor weight, gc_tensor indices, gc_tensor offsets, int scale_grad_by_freq, int64_t mode, int sparse, gc_tensor per_sample_weights, int include_last_offset) {
+  PROTECT(
+    auto results__ = torch::embedding_bag(*tensor_ptr_from_ocaml(weight), *tensor_ptr_from_ocaml(indices), *tensor_ptr_from_ocaml(offsets), (bool)scale_grad_by_freq, mode, (bool)sparse, (per_sample_weights ? tensor_from_ocaml(per_sample_weights) : torch::Tensor()), (bool)include_last_offset);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+    out__[3] = tensor_to_ocaml(std::get<3>(results__));
+  )
+}
+
+void atg_embedding_bag_padding_idx(raw_tensor *out__, gc_tensor weight, gc_tensor indices, gc_tensor offsets, int scale_grad_by_freq, int64_t mode, int sparse, gc_tensor per_sample_weights, int include_last_offset, int64_t padding_idx_v, int padding_idx_null) {
+  PROTECT(
+    auto results__ = torch::embedding_bag(*tensor_ptr_from_ocaml(weight), *tensor_ptr_from_ocaml(indices), *tensor_ptr_from_ocaml(offsets), (bool)scale_grad_by_freq, mode, (bool)sparse, (per_sample_weights ? tensor_from_ocaml(per_sample_weights) : torch::Tensor()), (bool)include_last_offset, padding_idx_null ? c10::nullopt : c10::optional<int64_t>(padding_idx_v));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+    out__[3] = tensor_to_ocaml(std::get<3>(results__));
+  )
+}
+
+raw_tensor atg_embedding_dense_backward(gc_tensor grad_output, gc_tensor indices, int64_t num_weights, int64_t padding_idx, int scale_grad_by_freq) {
+  PROTECT(
+    torch::Tensor results__ = torch::embedding_dense_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(indices), num_weights, padding_idx, (bool)scale_grad_by_freq);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_embedding_dense_backward_out(gc_tensor out, gc_tensor grad_output, gc_tensor indices, int64_t num_weights, int64_t padding_idx, int scale_grad_by_freq) {
+  PROTECT(
+    torch::Tensor results__ = torch::embedding_dense_backward_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(indices), num_weights, padding_idx, (bool)scale_grad_by_freq);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_embedding_out(gc_tensor out, gc_tensor weight, gc_tensor indices, int64_t padding_idx, int scale_grad_by_freq, int sparse) {
+  PROTECT(
+    torch::Tensor results__ = torch::embedding_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(weight), *tensor_ptr_from_ocaml(indices), padding_idx, (bool)scale_grad_by_freq, (bool)sparse);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_embedding_renorm(gc_tensor self, gc_tensor indices, double max_norm, double norm_type) {
+  PROTECT(
+    torch::Tensor results__ = torch::embedding_renorm(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(indices), max_norm, norm_type);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_embedding_renorm_(gc_tensor self, gc_tensor indices, double max_norm, double norm_type) {
+  PROTECT(
+    torch::Tensor results__ = torch::embedding_renorm_(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(indices), max_norm, norm_type);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_embedding_renorm_out(gc_tensor out, gc_tensor self, gc_tensor indices, double max_norm, double norm_type) {
+  PROTECT(
+    torch::Tensor results__ = torch::embedding_renorm_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(indices), max_norm, norm_type);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_embedding_sparse_backward(gc_tensor grad, gc_tensor indices, int64_t num_weights, int64_t padding_idx, int scale_grad_by_freq) {
+  PROTECT(
+    torch::Tensor results__ = torch::embedding_sparse_backward(*tensor_ptr_from_ocaml(grad), *tensor_ptr_from_ocaml(indices), num_weights, padding_idx, (bool)scale_grad_by_freq);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_empty(int64_t *size_data, int size_len, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::empty(torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_empty_like(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::empty_like(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_empty_like_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::empty_like_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_empty_out(gc_tensor out, int64_t *size_data, int size_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::empty_out(*tensor_ptr_from_ocaml(out), torch::IntArrayRef(size_data, size_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_empty_permuted(int64_t *size_data, int size_len, int64_t *physical_layout_data, int physical_layout_len, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::empty_permuted(torch::IntArrayRef(size_data, size_len), torch::IntArrayRef(physical_layout_data, physical_layout_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_empty_permuted_out(gc_tensor out, int64_t *size_data, int size_len, int64_t *physical_layout_data, int physical_layout_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::empty_permuted_out(*tensor_ptr_from_ocaml(out), torch::IntArrayRef(size_data, size_len), torch::IntArrayRef(physical_layout_data, physical_layout_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_empty_quantized(int64_t *size_data, int size_len, gc_tensor qtensor, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::empty_quantized(torch::IntArrayRef(size_data, size_len), *tensor_ptr_from_ocaml(qtensor), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_empty_quantized_out(gc_tensor out, int64_t *size_data, int size_len, gc_tensor qtensor) {
+  PROTECT(
+    torch::Tensor results__ = torch::empty_quantized_out(*tensor_ptr_from_ocaml(out), torch::IntArrayRef(size_data, size_len), *tensor_ptr_from_ocaml(qtensor));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_empty_strided(int64_t *size_data, int size_len, int64_t *stride_data, int stride_len, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::empty_strided(torch::IntArrayRef(size_data, size_len), torch::IntArrayRef(stride_data, stride_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_empty_strided_out(gc_tensor out, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::empty_strided_out(*tensor_ptr_from_ocaml(out), torch::IntArrayRef(size_data, size_len), torch::IntArrayRef(stride_data, stride_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_eq(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::eq(*tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_eq_(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->eq_(*other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_eq_scalar_out(gc_tensor out, gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::eq_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_eq_tensor(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::eq(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_eq_tensor_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->eq_(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_eq_tensor_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::eq_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+int atg_equal(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    return torch::equal(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+  )
+  return 0;
+}
+
+raw_tensor atg_erf(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::erf(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_erf_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::erf_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_erf_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::erf_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_erfc(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::erfc(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_erfc_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::erfc_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_erfc_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::erfc_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_erfinv(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::erfinv(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_erfinv_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->erfinv_();
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_erfinv_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::erfinv_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_exp(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::exp(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_exp2(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::exp2(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_exp2_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::exp2_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_exp2_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::exp2_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_exp_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::exp_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_exp_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::exp_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_expand(gc_tensor self, int64_t *size_data, int size_len, int implicit) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->expand(torch::IntArrayRef(size_data, size_len), (bool)implicit);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_expand_as(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->expand_as(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_expand_copy(gc_tensor self, int64_t *size_data, int size_len, int implicit) {
+  PROTECT(
+    torch::Tensor results__ = torch::expand_copy(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(size_data, size_len), (bool)implicit);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_expand_copy_out(gc_tensor out, gc_tensor self, int64_t *size_data, int size_len, int implicit) {
+  PROTECT(
+    torch::Tensor results__ = torch::expand_copy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(size_data, size_len), (bool)implicit);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_expm1(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::expm1(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_expm1_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::expm1_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_expm1_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::expm1_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_exponential(gc_tensor self, double lambd) {
+  PROTECT(
+    torch::Tensor results__ = torch::exponential(*tensor_ptr_from_ocaml(self), lambd);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_exponential_(gc_tensor self, double lambd) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->exponential_(lambd);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_exponential_out(gc_tensor out, gc_tensor self, double lambd) {
+  PROTECT(
+    torch::Tensor results__ = torch::exponential_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), lambd);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_eye(int64_t n, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::eye(n, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_eye_m(int64_t n, int64_t m, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::eye(n, m, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_eye_m_out(gc_tensor out, int64_t n, int64_t m) {
+  PROTECT(
+    torch::Tensor results__ = torch::eye_out(*tensor_ptr_from_ocaml(out), n, m);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_eye_out(gc_tensor out, int64_t n) {
+  PROTECT(
+    torch::Tensor results__ = torch::eye_out(*tensor_ptr_from_ocaml(out), n);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fake_quantize_per_channel_affine(gc_tensor self, gc_tensor scale, gc_tensor zero_point, int64_t axis, int64_t quant_min, int64_t quant_max) {
+  PROTECT(
+    torch::Tensor results__ = torch::fake_quantize_per_channel_affine(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(scale), *tensor_ptr_from_ocaml(zero_point), axis, quant_min, quant_max);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_fake_quantize_per_channel_affine_cachemask(raw_tensor *out__, gc_tensor self, gc_tensor scale, gc_tensor zero_point, int64_t axis, int64_t quant_min, int64_t quant_max) {
+  PROTECT(
+    auto results__ = torch::fake_quantize_per_channel_affine_cachemask(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(scale), *tensor_ptr_from_ocaml(zero_point), axis, quant_min, quant_max);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_fake_quantize_per_channel_affine_cachemask_backward(gc_tensor grad, gc_tensor mask) {
+  PROTECT(
+    torch::Tensor results__ = torch::fake_quantize_per_channel_affine_cachemask_backward(*tensor_ptr_from_ocaml(grad), *tensor_ptr_from_ocaml(mask));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_fake_quantize_per_channel_affine_cachemask_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor self, gc_tensor scale, gc_tensor zero_point, int64_t axis, int64_t quant_min, int64_t quant_max) {
+  PROTECT(
+    auto results__ = torch::fake_quantize_per_channel_affine_cachemask_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(scale), *tensor_ptr_from_ocaml(zero_point), axis, quant_min, quant_max);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_fake_quantize_per_tensor_affine(gc_tensor self, double scale, int64_t zero_point, int64_t quant_min, int64_t quant_max) {
+  PROTECT(
+    torch::Tensor results__ = torch::fake_quantize_per_tensor_affine(*tensor_ptr_from_ocaml(self), scale, zero_point, quant_min, quant_max);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_fake_quantize_per_tensor_affine_cachemask(raw_tensor *out__, gc_tensor self, double scale, int64_t zero_point, int64_t quant_min, int64_t quant_max) {
+  PROTECT(
+    auto results__ = torch::fake_quantize_per_tensor_affine_cachemask(*tensor_ptr_from_ocaml(self), scale, zero_point, quant_min, quant_max);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_fake_quantize_per_tensor_affine_cachemask_backward(gc_tensor grad, gc_tensor mask) {
+  PROTECT(
+    torch::Tensor results__ = torch::fake_quantize_per_tensor_affine_cachemask_backward(*tensor_ptr_from_ocaml(grad), *tensor_ptr_from_ocaml(mask));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_fake_quantize_per_tensor_affine_cachemask_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor self, double scale, int64_t zero_point, int64_t quant_min, int64_t quant_max) {
+  PROTECT(
+    auto results__ = torch::fake_quantize_per_tensor_affine_cachemask_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(self), scale, zero_point, quant_min, quant_max);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_fake_quantize_per_tensor_affine_tensor_qparams(gc_tensor self, gc_tensor scale, gc_tensor zero_point, int64_t quant_min, int64_t quant_max) {
+  PROTECT(
+    torch::Tensor results__ = torch::fake_quantize_per_tensor_affine(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(scale), *tensor_ptr_from_ocaml(zero_point), quant_min, quant_max);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fbgemm_linear_fp16_weight(gc_tensor input, gc_tensor packed_weight, gc_tensor bias) {
+  PROTECT(
+    torch::Tensor results__ = torch::fbgemm_linear_fp16_weight(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(packed_weight), *tensor_ptr_from_ocaml(bias));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fbgemm_linear_fp16_weight_fp32_activation(gc_tensor input, gc_tensor packed_weight, gc_tensor bias) {
+  PROTECT(
+    torch::Tensor results__ = torch::fbgemm_linear_fp16_weight_fp32_activation(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(packed_weight), *tensor_ptr_from_ocaml(bias));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fbgemm_linear_int8_weight(gc_tensor input, gc_tensor weight, gc_tensor packed, gc_tensor col_offsets, scalar weight_scale, scalar weight_zero_point, gc_tensor bias) {
+  PROTECT(
+    torch::Tensor results__ = torch::fbgemm_linear_int8_weight(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(weight), *tensor_ptr_from_ocaml(packed), *tensor_ptr_from_ocaml(col_offsets), *weight_scale, *weight_zero_point, *tensor_ptr_from_ocaml(bias));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fbgemm_linear_int8_weight_fp32_activation(gc_tensor input, gc_tensor weight, gc_tensor packed, gc_tensor col_offsets, scalar weight_scale, scalar weight_zero_point, gc_tensor bias) {
+  PROTECT(
+    torch::Tensor results__ = torch::fbgemm_linear_int8_weight_fp32_activation(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(weight), *tensor_ptr_from_ocaml(packed), *tensor_ptr_from_ocaml(col_offsets), *weight_scale, *weight_zero_point, *tensor_ptr_from_ocaml(bias));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fbgemm_pack_gemm_matrix_fp16(gc_tensor input) {
+  PROTECT(
+    torch::Tensor results__ = torch::fbgemm_pack_gemm_matrix_fp16(*tensor_ptr_from_ocaml(input));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fbgemm_pack_quantized_matrix(gc_tensor input) {
+  PROTECT(
+    torch::Tensor results__ = torch::fbgemm_pack_quantized_matrix(*tensor_ptr_from_ocaml(input));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fbgemm_pack_quantized_matrix_kn(gc_tensor input, int64_t K, int64_t n) {
+  PROTECT(
+    torch::Tensor results__ = torch::fbgemm_pack_quantized_matrix(*tensor_ptr_from_ocaml(input), K, n);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_feature_alpha_dropout(gc_tensor input, double p, int train) {
+  PROTECT(
+    torch::Tensor results__ = torch::feature_alpha_dropout(*tensor_ptr_from_ocaml(input), p, (bool)train);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_feature_alpha_dropout_(gc_tensor self, double p, int train) {
+  PROTECT(
+    torch::Tensor results__ = torch::feature_alpha_dropout_(*tensor_ptr_from_ocaml(self), p, (bool)train);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_feature_dropout(gc_tensor input, double p, int train) {
+  PROTECT(
+    torch::Tensor results__ = torch::feature_dropout(*tensor_ptr_from_ocaml(input), p, (bool)train);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_feature_dropout_(gc_tensor self, double p, int train) {
+  PROTECT(
+    torch::Tensor results__ = torch::feature_dropout_(*tensor_ptr_from_ocaml(self), p, (bool)train);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fft_fft(gc_tensor self, int64_t n_v, int n_null, int64_t dim, char * norm) {
+  PROTECT(
+    torch::Tensor results__ = torch::fft_fft(*tensor_ptr_from_ocaml(self), n_null ? c10::nullopt : c10::optional<int64_t>(n_v), dim, std::string(norm));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fft_fft2(gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
+  PROTECT(
+    torch::Tensor results__ = torch::fft_fft2(*tensor_ptr_from_ocaml(self), s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), torch::IntArrayRef(dim_data, dim_len), std::string(norm));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fft_fft2_out(gc_tensor out, gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
+  PROTECT(
+    torch::Tensor results__ = torch::fft_fft2_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), torch::IntArrayRef(dim_data, dim_len), std::string(norm));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fft_fft_out(gc_tensor out, gc_tensor self, int64_t n_v, int n_null, int64_t dim, char * norm) {
+  PROTECT(
+    torch::Tensor results__ = torch::fft_fft_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), n_null ? c10::nullopt : c10::optional<int64_t>(n_v), dim, std::string(norm));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fft_fftfreq(int64_t n, double d, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::fft_fftfreq(n, d, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fft_fftfreq_out(gc_tensor out, int64_t n, double d) {
+  PROTECT(
+    torch::Tensor results__ = torch::fft_fftfreq_out(*tensor_ptr_from_ocaml(out), n, d);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fft_fftn(gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
+  PROTECT(
+    torch::Tensor results__ = torch::fft_fftn(*tensor_ptr_from_ocaml(self), s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), std::string(norm));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fft_fftn_out(gc_tensor out, gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
+  PROTECT(
+    torch::Tensor results__ = torch::fft_fftn_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), std::string(norm));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fft_fftshift(gc_tensor self, int64_t *dim_data, int dim_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::fft_fftshift(*tensor_ptr_from_ocaml(self), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fft_hfft(gc_tensor self, int64_t n_v, int n_null, int64_t dim, char * norm) {
+  PROTECT(
+    torch::Tensor results__ = torch::fft_hfft(*tensor_ptr_from_ocaml(self), n_null ? c10::nullopt : c10::optional<int64_t>(n_v), dim, std::string(norm));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fft_hfft2(gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
+  PROTECT(
+    torch::Tensor results__ = torch::fft_hfft2(*tensor_ptr_from_ocaml(self), s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), torch::IntArrayRef(dim_data, dim_len), std::string(norm));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fft_hfft2_out(gc_tensor out, gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
+  PROTECT(
+    torch::Tensor results__ = torch::fft_hfft2_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), torch::IntArrayRef(dim_data, dim_len), std::string(norm));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fft_hfft_out(gc_tensor out, gc_tensor self, int64_t n_v, int n_null, int64_t dim, char * norm) {
+  PROTECT(
+    torch::Tensor results__ = torch::fft_hfft_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), n_null ? c10::nullopt : c10::optional<int64_t>(n_v), dim, std::string(norm));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fft_hfftn(gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
+  PROTECT(
+    torch::Tensor results__ = torch::fft_hfftn(*tensor_ptr_from_ocaml(self), s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), std::string(norm));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fft_hfftn_out(gc_tensor out, gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
+  PROTECT(
+    torch::Tensor results__ = torch::fft_hfftn_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), std::string(norm));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fft_ifft(gc_tensor self, int64_t n_v, int n_null, int64_t dim, char * norm) {
+  PROTECT(
+    torch::Tensor results__ = torch::fft_ifft(*tensor_ptr_from_ocaml(self), n_null ? c10::nullopt : c10::optional<int64_t>(n_v), dim, std::string(norm));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fft_ifft2(gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
+  PROTECT(
+    torch::Tensor results__ = torch::fft_ifft2(*tensor_ptr_from_ocaml(self), s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), torch::IntArrayRef(dim_data, dim_len), std::string(norm));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fft_ifft2_out(gc_tensor out, gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
+  PROTECT(
+    torch::Tensor results__ = torch::fft_ifft2_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), torch::IntArrayRef(dim_data, dim_len), std::string(norm));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fft_ifft_out(gc_tensor out, gc_tensor self, int64_t n_v, int n_null, int64_t dim, char * norm) {
+  PROTECT(
+    torch::Tensor results__ = torch::fft_ifft_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), n_null ? c10::nullopt : c10::optional<int64_t>(n_v), dim, std::string(norm));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fft_ifftn(gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
+  PROTECT(
+    torch::Tensor results__ = torch::fft_ifftn(*tensor_ptr_from_ocaml(self), s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), std::string(norm));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fft_ifftn_out(gc_tensor out, gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
+  PROTECT(
+    torch::Tensor results__ = torch::fft_ifftn_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), std::string(norm));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fft_ifftshift(gc_tensor self, int64_t *dim_data, int dim_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::fft_ifftshift(*tensor_ptr_from_ocaml(self), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fft_ihfft(gc_tensor self, int64_t n_v, int n_null, int64_t dim, char * norm) {
+  PROTECT(
+    torch::Tensor results__ = torch::fft_ihfft(*tensor_ptr_from_ocaml(self), n_null ? c10::nullopt : c10::optional<int64_t>(n_v), dim, std::string(norm));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fft_ihfft2(gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
+  PROTECT(
+    torch::Tensor results__ = torch::fft_ihfft2(*tensor_ptr_from_ocaml(self), s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), torch::IntArrayRef(dim_data, dim_len), std::string(norm));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fft_ihfft2_out(gc_tensor out, gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
+  PROTECT(
+    torch::Tensor results__ = torch::fft_ihfft2_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), torch::IntArrayRef(dim_data, dim_len), std::string(norm));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fft_ihfft_out(gc_tensor out, gc_tensor self, int64_t n_v, int n_null, int64_t dim, char * norm) {
+  PROTECT(
+    torch::Tensor results__ = torch::fft_ihfft_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), n_null ? c10::nullopt : c10::optional<int64_t>(n_v), dim, std::string(norm));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fft_ihfftn(gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
+  PROTECT(
+    torch::Tensor results__ = torch::fft_ihfftn(*tensor_ptr_from_ocaml(self), s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), std::string(norm));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fft_ihfftn_out(gc_tensor out, gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
+  PROTECT(
+    torch::Tensor results__ = torch::fft_ihfftn_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), std::string(norm));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fft_irfft(gc_tensor self, int64_t n_v, int n_null, int64_t dim, char * norm) {
+  PROTECT(
+    torch::Tensor results__ = torch::fft_irfft(*tensor_ptr_from_ocaml(self), n_null ? c10::nullopt : c10::optional<int64_t>(n_v), dim, std::string(norm));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fft_irfft2(gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
+  PROTECT(
+    torch::Tensor results__ = torch::fft_irfft2(*tensor_ptr_from_ocaml(self), s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), torch::IntArrayRef(dim_data, dim_len), std::string(norm));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fft_irfft2_out(gc_tensor out, gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
+  PROTECT(
+    torch::Tensor results__ = torch::fft_irfft2_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), torch::IntArrayRef(dim_data, dim_len), std::string(norm));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fft_irfft_out(gc_tensor out, gc_tensor self, int64_t n_v, int n_null, int64_t dim, char * norm) {
+  PROTECT(
+    torch::Tensor results__ = torch::fft_irfft_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), n_null ? c10::nullopt : c10::optional<int64_t>(n_v), dim, std::string(norm));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fft_irfftn(gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
+  PROTECT(
+    torch::Tensor results__ = torch::fft_irfftn(*tensor_ptr_from_ocaml(self), s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), std::string(norm));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fft_irfftn_out(gc_tensor out, gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
+  PROTECT(
+    torch::Tensor results__ = torch::fft_irfftn_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), std::string(norm));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fft_rfft(gc_tensor self, int64_t n_v, int n_null, int64_t dim, char * norm) {
+  PROTECT(
+    torch::Tensor results__ = torch::fft_rfft(*tensor_ptr_from_ocaml(self), n_null ? c10::nullopt : c10::optional<int64_t>(n_v), dim, std::string(norm));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fft_rfft2(gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
+  PROTECT(
+    torch::Tensor results__ = torch::fft_rfft2(*tensor_ptr_from_ocaml(self), s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), torch::IntArrayRef(dim_data, dim_len), std::string(norm));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fft_rfft2_out(gc_tensor out, gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
+  PROTECT(
+    torch::Tensor results__ = torch::fft_rfft2_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), torch::IntArrayRef(dim_data, dim_len), std::string(norm));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fft_rfft_out(gc_tensor out, gc_tensor self, int64_t n_v, int n_null, int64_t dim, char * norm) {
+  PROTECT(
+    torch::Tensor results__ = torch::fft_rfft_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), n_null ? c10::nullopt : c10::optional<int64_t>(n_v), dim, std::string(norm));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fft_rfftfreq(int64_t n, double d, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::fft_rfftfreq(n, d, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fft_rfftfreq_out(gc_tensor out, int64_t n, double d) {
+  PROTECT(
+    torch::Tensor results__ = torch::fft_rfftfreq_out(*tensor_ptr_from_ocaml(out), n, d);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fft_rfftn(gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
+  PROTECT(
+    torch::Tensor results__ = torch::fft_rfftn(*tensor_ptr_from_ocaml(self), s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), std::string(norm));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fft_rfftn_out(gc_tensor out, gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
+  PROTECT(
+    torch::Tensor results__ = torch::fft_rfftn_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), std::string(norm));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fill(gc_tensor self, scalar value) {
+  PROTECT(
+    torch::Tensor results__ = torch::fill(*tensor_ptr_from_ocaml(self), *value);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fill_(gc_tensor self, scalar value) {
+  PROTECT(
+    torch::Tensor results__ = torch::fill_(*tensor_ptr_from_ocaml(self), *value);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fill_diagonal_(gc_tensor self, scalar fill_value, int wrap) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->fill_diagonal_(*fill_value, (bool)wrap);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fill_scalar_out(gc_tensor out, gc_tensor self, scalar value) {
+  PROTECT(
+    torch::Tensor results__ = torch::fill_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *value);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fill_tensor(gc_tensor self, gc_tensor value) {
+  PROTECT(
+    torch::Tensor results__ = torch::fill(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(value));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fill_tensor_(gc_tensor self, gc_tensor value) {
+  PROTECT(
+    torch::Tensor results__ = torch::fill_(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(value));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fill_tensor_out(gc_tensor out, gc_tensor self, gc_tensor value) {
+  PROTECT(
+    torch::Tensor results__ = torch::fill_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(value));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fix(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::fix(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fix_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::fix_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fix_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::fix_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_flatten(gc_tensor self, int64_t start_dim, int64_t end_dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::flatten(*tensor_ptr_from_ocaml(self), start_dim, end_dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_flatten_dense_tensors(gc_tensor *tensors_data, int tensors_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::flatten_dense_tensors(of_carray_tensor(tensors_data, tensors_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_flip(gc_tensor self, int64_t *dims_data, int dims_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::flip(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(dims_data, dims_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_flip_out(gc_tensor out, gc_tensor self, int64_t *dims_data, int dims_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::flip_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(dims_data, dims_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fliplr(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::fliplr(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_flipud(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::flipud(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_float_power(gc_tensor self, gc_tensor exponent) {
+  PROTECT(
+    torch::Tensor results__ = torch::float_power(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(exponent));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_float_power_(gc_tensor self, scalar exponent) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->float_power_(*exponent);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_float_power_scalar(scalar self, gc_tensor exponent) {
+  PROTECT(
+    torch::Tensor results__ = torch::float_power(*self, *tensor_ptr_from_ocaml(exponent));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_float_power_scalar_out(gc_tensor out, scalar self, gc_tensor exponent) {
+  PROTECT(
+    torch::Tensor results__ = torch::float_power_out(*tensor_ptr_from_ocaml(out), *self, *tensor_ptr_from_ocaml(exponent));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_float_power_tensor_(gc_tensor self, gc_tensor exponent) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->float_power_(*tensor_ptr_from_ocaml(exponent));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_float_power_tensor_scalar(gc_tensor self, scalar exponent) {
+  PROTECT(
+    torch::Tensor results__ = torch::float_power(*tensor_ptr_from_ocaml(self), *exponent);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_float_power_tensor_scalar_out(gc_tensor out, gc_tensor self, scalar exponent) {
+  PROTECT(
+    torch::Tensor results__ = torch::float_power_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *exponent);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_float_power_tensor_tensor_out(gc_tensor out, gc_tensor self, gc_tensor exponent) {
+  PROTECT(
+    torch::Tensor results__ = torch::float_power_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(exponent));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_floor(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::floor(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_floor_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::floor_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_floor_divide(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::floor_divide(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_floor_divide_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->floor_divide_(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_floor_divide_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::floor_divide_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_floor_divide_scalar(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::floor_divide(*tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_floor_divide_scalar_(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->floor_divide_(*other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_floor_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::floor_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fmax(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::fmax(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fmax_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::fmax_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fmin(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::fmin(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fmin_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::fmin_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fmod(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::fmod(*tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fmod_(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->fmod_(*other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fmod_scalar_out(gc_tensor out, gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::fmod_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fmod_tensor(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::fmod(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fmod_tensor_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->fmod_(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fmod_tensor_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::fmod_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_frac(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::frac(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_frac_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::frac_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_frac_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::frac_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_fractional_max_pool2d(raw_tensor *out__, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *output_size_data, int output_size_len, gc_tensor random_samples) {
+  PROTECT(
+    auto results__ = torch::fractional_max_pool2d(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(output_size_data, output_size_len), *tensor_ptr_from_ocaml(random_samples));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_fractional_max_pool2d_backward(gc_tensor grad_output, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *output_size_data, int output_size_len, gc_tensor indices) {
+  PROTECT(
+    torch::Tensor results__ = torch::fractional_max_pool2d_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(output_size_data, output_size_len), *tensor_ptr_from_ocaml(indices));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fractional_max_pool2d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *output_size_data, int output_size_len, gc_tensor indices) {
+  PROTECT(
+    torch::Tensor results__ = torch::fractional_max_pool2d_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(output_size_data, output_size_len), *tensor_ptr_from_ocaml(indices));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_fractional_max_pool2d_output(raw_tensor *out__, gc_tensor output, gc_tensor indices, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *output_size_data, int output_size_len, gc_tensor random_samples) {
+  PROTECT(
+    auto results__ = torch::fractional_max_pool2d_out(*tensor_ptr_from_ocaml(output), *tensor_ptr_from_ocaml(indices), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(output_size_data, output_size_len), *tensor_ptr_from_ocaml(random_samples));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_fractional_max_pool3d(raw_tensor *out__, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *output_size_data, int output_size_len, gc_tensor random_samples) {
+  PROTECT(
+    auto results__ = torch::fractional_max_pool3d(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(output_size_data, output_size_len), *tensor_ptr_from_ocaml(random_samples));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_fractional_max_pool3d_backward(gc_tensor grad_output, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *output_size_data, int output_size_len, gc_tensor indices) {
+  PROTECT(
+    torch::Tensor results__ = torch::fractional_max_pool3d_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(output_size_data, output_size_len), *tensor_ptr_from_ocaml(indices));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fractional_max_pool3d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *output_size_data, int output_size_len, gc_tensor indices) {
+  PROTECT(
+    torch::Tensor results__ = torch::fractional_max_pool3d_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(output_size_data, output_size_len), *tensor_ptr_from_ocaml(indices));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_fractional_max_pool3d_output(raw_tensor *out__, gc_tensor output, gc_tensor indices, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *output_size_data, int output_size_len, gc_tensor random_samples) {
+  PROTECT(
+    auto results__ = torch::fractional_max_pool3d_out(*tensor_ptr_from_ocaml(output), *tensor_ptr_from_ocaml(indices), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(output_size_data, output_size_len), *tensor_ptr_from_ocaml(random_samples));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_frexp(raw_tensor *out__, gc_tensor self) {
+  PROTECT(
+    auto results__ = torch::frexp(*tensor_ptr_from_ocaml(self));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_frexp_tensor_out(raw_tensor *out__, gc_tensor mantissa, gc_tensor exponent, gc_tensor self) {
+  PROTECT(
+    auto results__ = torch::frexp_out(*tensor_ptr_from_ocaml(mantissa), *tensor_ptr_from_ocaml(exponent), *tensor_ptr_from_ocaml(self));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_frobenius_norm(gc_tensor self, int64_t *dim_data, int dim_len, int keepdim) {
+  PROTECT(
+    torch::Tensor results__ = torch::frobenius_norm(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(dim_data, dim_len), (bool)keepdim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_frobenius_norm_out(gc_tensor out, gc_tensor self, int64_t *dim_data, int dim_len, int keepdim) {
+  PROTECT(
+    torch::Tensor results__ = torch::frobenius_norm_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(dim_data, dim_len), (bool)keepdim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_from_file(char * filename, int shared, int64_t size_v, int size_null, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::from_file(std::string(filename), (bool)shared, size_null ? c10::nullopt : c10::optional<int64_t>(size_v), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_from_file_out(gc_tensor out, char * filename, int shared, int64_t size_v, int size_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::from_file_out(*tensor_ptr_from_ocaml(out), std::string(filename), (bool)shared, size_null ? c10::nullopt : c10::optional<int64_t>(size_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_full(int64_t *size_data, int size_len, scalar fill_value, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::full(torch::IntArrayRef(size_data, size_len), *fill_value, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_full_like(gc_tensor self, scalar fill_value) {
+  PROTECT(
+    torch::Tensor results__ = torch::full_like(*tensor_ptr_from_ocaml(self), *fill_value);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_full_like_out(gc_tensor out, gc_tensor self, scalar fill_value) {
+  PROTECT(
+    torch::Tensor results__ = torch::full_like_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *fill_value);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_full_out(gc_tensor out, int64_t *size_data, int size_len, scalar fill_value) {
+  PROTECT(
+    torch::Tensor results__ = torch::full_out(*tensor_ptr_from_ocaml(out), torch::IntArrayRef(size_data, size_len), *fill_value);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_fused_moving_avg_obs_fake_quant(gc_tensor self, gc_tensor observer_on, gc_tensor fake_quant_on, gc_tensor running_min, gc_tensor running_max, gc_tensor scale, gc_tensor zero_point, double averaging_const, int64_t quant_min, int64_t quant_max, int64_t ch_axis, int per_row_fake_quant, int symmetric_quant) {
+  PROTECT(
+    torch::Tensor results__ = torch::fused_moving_avg_obs_fake_quant(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(observer_on), *tensor_ptr_from_ocaml(fake_quant_on), *tensor_ptr_from_ocaml(running_min), *tensor_ptr_from_ocaml(running_max), *tensor_ptr_from_ocaml(scale), *tensor_ptr_from_ocaml(zero_point), averaging_const, quant_min, quant_max, ch_axis, (bool)per_row_fake_quant, (bool)symmetric_quant);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_gather(gc_tensor self, int64_t dim, gc_tensor index, int sparse_grad) {
+  PROTECT(
+    torch::Tensor results__ = torch::gather(*tensor_ptr_from_ocaml(self), dim, *tensor_ptr_from_ocaml(index), (bool)sparse_grad);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_gather_backward(gc_tensor grad, gc_tensor self, int64_t dim, gc_tensor index, int sparse_grad) {
+  PROTECT(
+    torch::Tensor results__ = torch::gather_backward(*tensor_ptr_from_ocaml(grad), *tensor_ptr_from_ocaml(self), dim, *tensor_ptr_from_ocaml(index), (bool)sparse_grad);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_gather_out(gc_tensor out, gc_tensor self, int64_t dim, gc_tensor index, int sparse_grad) {
+  PROTECT(
+    torch::Tensor results__ = torch::gather_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dim, *tensor_ptr_from_ocaml(index), (bool)sparse_grad);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_gcd(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::gcd(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_gcd_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::gcd_(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_gcd_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::gcd_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_ge(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::ge(*tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_ge_(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->ge_(*other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_ge_scalar_out(gc_tensor out, gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::ge_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_ge_tensor(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::ge(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_ge_tensor_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->ge_(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_ge_tensor_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::ge_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_gelu(gc_tensor self, char * approximate) {
+  PROTECT(
+    torch::Tensor results__ = torch::gelu(*tensor_ptr_from_ocaml(self), std::string(approximate));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_gelu_(gc_tensor self, char * approximate) {
+  PROTECT(
+    torch::Tensor results__ = torch::gelu_(*tensor_ptr_from_ocaml(self), std::string(approximate));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_gelu_backward(gc_tensor grad_output, gc_tensor self, char * approximate) {
+  PROTECT(
+    torch::Tensor results__ = torch::gelu_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), std::string(approximate));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_gelu_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, char * approximate) {
+  PROTECT(
+    torch::Tensor results__ = torch::gelu_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), std::string(approximate));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_gelu_out(gc_tensor out, gc_tensor self, char * approximate) {
+  PROTECT(
+    torch::Tensor results__ = torch::gelu_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), std::string(approximate));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_geometric(gc_tensor self, double p) {
+  PROTECT(
+    torch::Tensor results__ = torch::geometric(*tensor_ptr_from_ocaml(self), p);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_geometric_(gc_tensor self, double p) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->geometric_(p);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_geometric_out(gc_tensor out, gc_tensor self, double p) {
+  PROTECT(
+    torch::Tensor results__ = torch::geometric_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), p);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_geqrf(raw_tensor *out__, gc_tensor self) {
+  PROTECT(
+    auto results__ = torch::geqrf(*tensor_ptr_from_ocaml(self));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_geqrf_a(raw_tensor *out__, gc_tensor a, gc_tensor tau, gc_tensor self) {
+  PROTECT(
+    auto results__ = torch::geqrf_out(*tensor_ptr_from_ocaml(a), *tensor_ptr_from_ocaml(tau), *tensor_ptr_from_ocaml(self));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_ger(gc_tensor self, gc_tensor vec2) {
+  PROTECT(
+    torch::Tensor results__ = torch::ger(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(vec2));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_ger_out(gc_tensor out, gc_tensor self, gc_tensor vec2) {
+  PROTECT(
+    torch::Tensor results__ = torch::ger_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(vec2));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_glu(gc_tensor self, int64_t dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::glu(*tensor_ptr_from_ocaml(self), dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_glu_backward(gc_tensor grad_output, gc_tensor self, int64_t dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::glu_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_glu_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, int64_t dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::glu_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_glu_backward_jvp(gc_tensor grad_x, gc_tensor grad_glu, gc_tensor x, gc_tensor dgrad_glu, gc_tensor dx, int64_t dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::glu_backward_jvp(*tensor_ptr_from_ocaml(grad_x), *tensor_ptr_from_ocaml(grad_glu), *tensor_ptr_from_ocaml(x), *tensor_ptr_from_ocaml(dgrad_glu), *tensor_ptr_from_ocaml(dx), dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_glu_backward_jvp_out(gc_tensor out, gc_tensor grad_x, gc_tensor grad_glu, gc_tensor x, gc_tensor dgrad_glu, gc_tensor dx, int64_t dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::glu_backward_jvp_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(grad_x), *tensor_ptr_from_ocaml(grad_glu), *tensor_ptr_from_ocaml(x), *tensor_ptr_from_ocaml(dgrad_glu), *tensor_ptr_from_ocaml(dx), dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_glu_jvp(gc_tensor glu, gc_tensor x, gc_tensor dx, int64_t dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::glu_jvp(*tensor_ptr_from_ocaml(glu), *tensor_ptr_from_ocaml(x), *tensor_ptr_from_ocaml(dx), dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_glu_jvp_out(gc_tensor out, gc_tensor glu, gc_tensor x, gc_tensor dx, int64_t dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::glu_jvp_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(glu), *tensor_ptr_from_ocaml(x), *tensor_ptr_from_ocaml(dx), dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_glu_out(gc_tensor out, gc_tensor self, int64_t dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::glu_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_grad(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->grad();
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_greater(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::greater(*tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_greater_(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->greater_(*other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_greater_equal(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::greater_equal(*tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_greater_equal_(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->greater_equal_(*other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_greater_equal_scalar_out(gc_tensor out, gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::greater_equal_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_greater_equal_tensor(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::greater_equal(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_greater_equal_tensor_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->greater_equal_(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_greater_equal_tensor_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::greater_equal_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_greater_scalar_out(gc_tensor out, gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::greater_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_greater_tensor(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::greater(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_greater_tensor_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->greater_(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_greater_tensor_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::greater_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_grid_sampler(gc_tensor input, gc_tensor grid, int64_t interpolation_mode, int64_t padding_mode, int align_corners) {
+  PROTECT(
+    torch::Tensor results__ = torch::grid_sampler(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(grid), interpolation_mode, padding_mode, (bool)align_corners);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_grid_sampler_2d(gc_tensor input, gc_tensor grid, int64_t interpolation_mode, int64_t padding_mode, int align_corners) {
+  PROTECT(
+    torch::Tensor results__ = torch::grid_sampler_2d(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(grid), interpolation_mode, padding_mode, (bool)align_corners);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_grid_sampler_2d_out(gc_tensor out, gc_tensor input, gc_tensor grid, int64_t interpolation_mode, int64_t padding_mode, int align_corners) {
+  PROTECT(
+    torch::Tensor results__ = torch::grid_sampler_2d_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(grid), interpolation_mode, padding_mode, (bool)align_corners);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_grid_sampler_3d(gc_tensor input, gc_tensor grid, int64_t interpolation_mode, int64_t padding_mode, int align_corners) {
+  PROTECT(
+    torch::Tensor results__ = torch::grid_sampler_3d(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(grid), interpolation_mode, padding_mode, (bool)align_corners);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_grid_sampler_3d_out(gc_tensor out, gc_tensor input, gc_tensor grid, int64_t interpolation_mode, int64_t padding_mode, int align_corners) {
+  PROTECT(
+    torch::Tensor results__ = torch::grid_sampler_3d_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(grid), interpolation_mode, padding_mode, (bool)align_corners);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_group_norm(gc_tensor input, int64_t num_groups, gc_tensor weight, gc_tensor bias, double eps, int cudnn_enabled) {
+  PROTECT(
+    torch::Tensor results__ = torch::group_norm(*tensor_ptr_from_ocaml(input), num_groups, (weight ? tensor_from_ocaml(weight) : torch::Tensor()), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), eps, (bool)cudnn_enabled);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_gru(raw_tensor *out__, gc_tensor input, gc_tensor hx, gc_tensor *params_data, int params_len, int has_biases, int64_t num_layers, double dropout, int train, int bidirectional, int batch_first) {
+  PROTECT(
+    auto results__ = torch::gru(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(hx), of_carray_tensor(params_data, params_len), (bool)has_biases, num_layers, dropout, (bool)train, (bool)bidirectional, (bool)batch_first);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_gru_cell(gc_tensor input, gc_tensor hx, gc_tensor w_ih, gc_tensor w_hh, gc_tensor b_ih, gc_tensor b_hh) {
+  PROTECT(
+    torch::Tensor results__ = torch::gru_cell(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(hx), *tensor_ptr_from_ocaml(w_ih), *tensor_ptr_from_ocaml(w_hh), (b_ih ? tensor_from_ocaml(b_ih) : torch::Tensor()), (b_hh ? tensor_from_ocaml(b_hh) : torch::Tensor()));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_gru_data(raw_tensor *out__, gc_tensor data, gc_tensor batch_sizes, gc_tensor hx, gc_tensor *params_data, int params_len, int has_biases, int64_t num_layers, double dropout, int train, int bidirectional) {
+  PROTECT(
+    auto results__ = torch::gru(*tensor_ptr_from_ocaml(data), *tensor_ptr_from_ocaml(batch_sizes), *tensor_ptr_from_ocaml(hx), of_carray_tensor(params_data, params_len), (bool)has_biases, num_layers, dropout, (bool)train, (bool)bidirectional);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_gt(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::gt(*tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_gt_(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->gt_(*other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_gt_scalar_out(gc_tensor out, gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::gt_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_gt_tensor(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::gt(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_gt_tensor_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->gt_(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_gt_tensor_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::gt_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_hamming_window(int64_t window_length, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::hamming_window(window_length, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_hamming_window_out(gc_tensor out, int64_t window_length) {
+  PROTECT(
+    torch::Tensor results__ = torch::hamming_window_out(*tensor_ptr_from_ocaml(out), window_length);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_hamming_window_periodic(int64_t window_length, int periodic, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::hamming_window(window_length, (bool)periodic, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_hamming_window_periodic_alpha(int64_t window_length, int periodic, double alpha, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::hamming_window(window_length, (bool)periodic, alpha, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_hamming_window_periodic_alpha_beta(int64_t window_length, int periodic, double alpha, double beta, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::hamming_window(window_length, (bool)periodic, alpha, beta, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_hamming_window_periodic_alpha_beta_out(gc_tensor out, int64_t window_length, int periodic, double alpha, double beta) {
+  PROTECT(
+    torch::Tensor results__ = torch::hamming_window_out(*tensor_ptr_from_ocaml(out), window_length, (bool)periodic, alpha, beta);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_hamming_window_periodic_alpha_out(gc_tensor out, int64_t window_length, int periodic, double alpha) {
+  PROTECT(
+    torch::Tensor results__ = torch::hamming_window_out(*tensor_ptr_from_ocaml(out), window_length, (bool)periodic, alpha);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_hamming_window_periodic_out(gc_tensor out, int64_t window_length, int periodic) {
+  PROTECT(
+    torch::Tensor results__ = torch::hamming_window_out(*tensor_ptr_from_ocaml(out), window_length, (bool)periodic);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_hann_window(int64_t window_length, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::hann_window(window_length, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_hann_window_out(gc_tensor out, int64_t window_length) {
+  PROTECT(
+    torch::Tensor results__ = torch::hann_window_out(*tensor_ptr_from_ocaml(out), window_length);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_hann_window_periodic(int64_t window_length, int periodic, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::hann_window(window_length, (bool)periodic, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_hann_window_periodic_out(gc_tensor out, int64_t window_length, int periodic) {
+  PROTECT(
+    torch::Tensor results__ = torch::hann_window_out(*tensor_ptr_from_ocaml(out), window_length, (bool)periodic);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_hardshrink(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::hardshrink(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_hardshrink_backward(gc_tensor grad_out, gc_tensor self, scalar lambd) {
+  PROTECT(
+    torch::Tensor results__ = torch::hardshrink_backward(*tensor_ptr_from_ocaml(grad_out), *tensor_ptr_from_ocaml(self), *lambd);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_hardshrink_backward_grad_input(gc_tensor grad_input, gc_tensor grad_out, gc_tensor self, scalar lambd) {
+  PROTECT(
+    torch::Tensor results__ = torch::hardshrink_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_out), *tensor_ptr_from_ocaml(self), *lambd);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_hardshrink_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::hardshrink_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_hardsigmoid(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::hardsigmoid(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_hardsigmoid_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::hardsigmoid_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_hardsigmoid_backward(gc_tensor grad_output, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::hardsigmoid_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_hardsigmoid_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::hardsigmoid_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_hardsigmoid_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::hardsigmoid_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_hardswish(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::hardswish(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_hardswish_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::hardswish_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_hardswish_backward(gc_tensor grad_output, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::hardswish_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_hardswish_backward_out(gc_tensor out, gc_tensor grad_output, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::hardswish_backward_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_hardswish_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::hardswish_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_hardtanh(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::hardtanh(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_hardtanh_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::hardtanh_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_hardtanh_backward(gc_tensor grad_output, gc_tensor self, scalar min_val, scalar max_val) {
+  PROTECT(
+    torch::Tensor results__ = torch::hardtanh_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), *min_val, *max_val);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_hardtanh_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, scalar min_val, scalar max_val) {
+  PROTECT(
+    torch::Tensor results__ = torch::hardtanh_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), *min_val, *max_val);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_hardtanh_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::hardtanh_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_heaviside(gc_tensor self, gc_tensor values) {
+  PROTECT(
+    torch::Tensor results__ = torch::heaviside(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(values));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_heaviside_(gc_tensor self, gc_tensor values) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->heaviside_(*tensor_ptr_from_ocaml(values));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_heaviside_out(gc_tensor out, gc_tensor self, gc_tensor values) {
+  PROTECT(
+    torch::Tensor results__ = torch::heaviside_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(values));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_hinge_embedding_loss(gc_tensor self, gc_tensor target, double margin, int64_t reduction) {
+  PROTECT(
+    torch::Tensor results__ = torch::hinge_embedding_loss(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(target), margin, reduction);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_histc(gc_tensor self, int64_t bins) {
+  PROTECT(
+    torch::Tensor results__ = torch::histc(*tensor_ptr_from_ocaml(self), bins);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_histc_out(gc_tensor out, gc_tensor self, int64_t bins) {
+  PROTECT(
+    torch::Tensor results__ = torch::histc_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), bins);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor *atg_hsplit(gc_tensor self, int64_t sections) {
+  PROTECT(
+    auto results__ = torch::hsplit(*tensor_ptr_from_ocaml(self), sections);
+    int sz = results__.size();
+    raw_tensor *out__ = (raw_tensor*)malloc((sz + 1) * sizeof(raw_tensor));
+    for (int i = 0; i < sz; ++i)
+      out__[i] = tensor_to_ocaml(results__[i]);
+    out__[sz] = nullptr;
+    return out__;
+  )
+}
+
+raw_tensor *atg_hsplit_array(gc_tensor self, int64_t *indices_data, int indices_len) {
+  PROTECT(
+    auto results__ = torch::hsplit(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(indices_data, indices_len));
+    int sz = results__.size();
+    raw_tensor *out__ = (raw_tensor*)malloc((sz + 1) * sizeof(raw_tensor));
+    for (int i = 0; i < sz; ++i)
+      out__[i] = tensor_to_ocaml(results__[i]);
+    out__[sz] = nullptr;
+    return out__;
+  )
+}
+
+raw_tensor atg_hspmm(gc_tensor mat1, gc_tensor mat2) {
+  PROTECT(
+    torch::Tensor results__ = torch::hspmm(*tensor_ptr_from_ocaml(mat1), *tensor_ptr_from_ocaml(mat2));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_hspmm_out(gc_tensor out, gc_tensor mat1, gc_tensor mat2) {
+  PROTECT(
+    torch::Tensor results__ = torch::hspmm_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(mat1), *tensor_ptr_from_ocaml(mat2));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_hstack(gc_tensor *tensors_data, int tensors_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::hstack(of_carray_tensor(tensors_data, tensors_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_hstack_out(gc_tensor out, gc_tensor *tensors_data, int tensors_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::hstack_out(*tensor_ptr_from_ocaml(out), of_carray_tensor(tensors_data, tensors_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_huber_loss(gc_tensor self, gc_tensor target, int64_t reduction, double delta) {
+  PROTECT(
+    torch::Tensor results__ = torch::huber_loss(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(target), reduction, delta);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_huber_loss_backward(gc_tensor grad_output, gc_tensor self, gc_tensor target, int64_t reduction, double delta) {
+  PROTECT(
+    torch::Tensor results__ = torch::huber_loss_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(target), reduction, delta);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_huber_loss_backward_out(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, gc_tensor target, int64_t reduction, double delta) {
+  PROTECT(
+    torch::Tensor results__ = torch::huber_loss_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(target), reduction, delta);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_huber_loss_out(gc_tensor out, gc_tensor self, gc_tensor target, int64_t reduction, double delta) {
+  PROTECT(
+    torch::Tensor results__ = torch::huber_loss_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(target), reduction, delta);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_hypot(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::hypot(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_hypot_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->hypot_(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_hypot_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::hypot_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_i0(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::i0(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_i0_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::i0_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_i0_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::i0_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_igamma(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::igamma(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_igamma_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->igamma_(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_igamma_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::igamma_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_igammac(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::igammac(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_igammac_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->igammac_(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_igammac_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::igammac_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_im2col(gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *dilation_data, int dilation_len, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::im2col(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(dilation_data, dilation_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(stride_data, stride_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_im2col_out(gc_tensor out, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *dilation_data, int dilation_len, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::im2col_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(dilation_data, dilation_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(stride_data, stride_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_imag(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::imag(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_index(gc_tensor self, gc_tensor *indices_data, int indices_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::index(*tensor_ptr_from_ocaml(self), of_carray_tensor_opt(indices_data, indices_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_index_add(gc_tensor self, int64_t dim, gc_tensor index, gc_tensor source) {
+  PROTECT(
+    torch::Tensor results__ = torch::index_add(*tensor_ptr_from_ocaml(self), dim, *tensor_ptr_from_ocaml(index), *tensor_ptr_from_ocaml(source));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_index_add_(gc_tensor self, int64_t dim, gc_tensor index, gc_tensor source) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->index_add_(dim, *tensor_ptr_from_ocaml(index), *tensor_ptr_from_ocaml(source));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_index_add_out(gc_tensor out, gc_tensor self, int64_t dim, gc_tensor index, gc_tensor source) {
+  PROTECT(
+    torch::Tensor results__ = torch::index_add_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dim, *tensor_ptr_from_ocaml(index), *tensor_ptr_from_ocaml(source));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_index_copy(gc_tensor self, int64_t dim, gc_tensor index, gc_tensor source) {
+  PROTECT(
+    torch::Tensor results__ = torch::index_copy(*tensor_ptr_from_ocaml(self), dim, *tensor_ptr_from_ocaml(index), *tensor_ptr_from_ocaml(source));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_index_copy_(gc_tensor self, int64_t dim, gc_tensor index, gc_tensor source) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->index_copy_(dim, *tensor_ptr_from_ocaml(index), *tensor_ptr_from_ocaml(source));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_index_copy_out(gc_tensor out, gc_tensor self, int64_t dim, gc_tensor index, gc_tensor source) {
+  PROTECT(
+    torch::Tensor results__ = torch::index_copy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dim, *tensor_ptr_from_ocaml(index), *tensor_ptr_from_ocaml(source));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_index_fill(gc_tensor self, int64_t dim, gc_tensor index, scalar value) {
+  PROTECT(
+    torch::Tensor results__ = torch::index_fill(*tensor_ptr_from_ocaml(self), dim, *tensor_ptr_from_ocaml(index), *value);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_index_fill_(gc_tensor self, int64_t dim, gc_tensor index, scalar value) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->index_fill_(dim, *tensor_ptr_from_ocaml(index), *value);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_index_fill_int_scalar_out(gc_tensor out, gc_tensor self, int64_t dim, gc_tensor index, scalar value) {
+  PROTECT(
+    torch::Tensor results__ = torch::index_fill_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dim, *tensor_ptr_from_ocaml(index), *value);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_index_fill_int_tensor(gc_tensor self, int64_t dim, gc_tensor index, gc_tensor value) {
+  PROTECT(
+    torch::Tensor results__ = torch::index_fill(*tensor_ptr_from_ocaml(self), dim, *tensor_ptr_from_ocaml(index), *tensor_ptr_from_ocaml(value));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_index_fill_int_tensor_(gc_tensor self, int64_t dim, gc_tensor index, gc_tensor value) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->index_fill_(dim, *tensor_ptr_from_ocaml(index), *tensor_ptr_from_ocaml(value));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_index_fill_int_tensor_out(gc_tensor out, gc_tensor self, int64_t dim, gc_tensor index, gc_tensor value) {
+  PROTECT(
+    torch::Tensor results__ = torch::index_fill_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dim, *tensor_ptr_from_ocaml(index), *tensor_ptr_from_ocaml(value));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_index_put(gc_tensor self, gc_tensor *indices_data, int indices_len, gc_tensor values, int accumulate) {
+  PROTECT(
+    torch::Tensor results__ = torch::index_put(*tensor_ptr_from_ocaml(self), of_carray_tensor_opt(indices_data, indices_len), *tensor_ptr_from_ocaml(values), (bool)accumulate);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_index_put_(gc_tensor self, gc_tensor *indices_data, int indices_len, gc_tensor values, int accumulate) {
+  PROTECT(
+    torch::Tensor results__ = torch::index_put_(*tensor_ptr_from_ocaml(self), of_carray_tensor_opt(indices_data, indices_len), *tensor_ptr_from_ocaml(values), (bool)accumulate);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_index_put_out(gc_tensor out, gc_tensor self, gc_tensor *indices_data, int indices_len, gc_tensor values, int accumulate) {
+  PROTECT(
+    torch::Tensor results__ = torch::index_put_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), of_carray_tensor_opt(indices_data, indices_len), *tensor_ptr_from_ocaml(values), (bool)accumulate);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_index_reduce(gc_tensor self, int64_t dim, gc_tensor index, gc_tensor source, char * reduce, int include_self) {
+  PROTECT(
+    torch::Tensor results__ = torch::index_reduce(*tensor_ptr_from_ocaml(self), dim, *tensor_ptr_from_ocaml(index), *tensor_ptr_from_ocaml(source), std::string(reduce), (bool)include_self);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_index_reduce_(gc_tensor self, int64_t dim, gc_tensor index, gc_tensor source, char * reduce, int include_self) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->index_reduce_(dim, *tensor_ptr_from_ocaml(index), *tensor_ptr_from_ocaml(source), std::string(reduce), (bool)include_self);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_index_reduce_out(gc_tensor out, gc_tensor self, int64_t dim, gc_tensor index, gc_tensor source, char * reduce, int include_self) {
+  PROTECT(
+    torch::Tensor results__ = torch::index_reduce_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dim, *tensor_ptr_from_ocaml(index), *tensor_ptr_from_ocaml(source), std::string(reduce), (bool)include_self);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_index_select(gc_tensor self, int64_t dim, gc_tensor index) {
+  PROTECT(
+    torch::Tensor results__ = torch::index_select(*tensor_ptr_from_ocaml(self), dim, *tensor_ptr_from_ocaml(index));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_index_select_backward(gc_tensor grad, int64_t *self_sizes_data, int self_sizes_len, int64_t dim, gc_tensor index) {
+  PROTECT(
+    torch::Tensor results__ = torch::index_select_backward(*tensor_ptr_from_ocaml(grad), torch::IntArrayRef(self_sizes_data, self_sizes_len), dim, *tensor_ptr_from_ocaml(index));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_index_select_out(gc_tensor out, gc_tensor self, int64_t dim, gc_tensor index) {
+  PROTECT(
+    torch::Tensor results__ = torch::index_select_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dim, *tensor_ptr_from_ocaml(index));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_index_tensor_out(gc_tensor out, gc_tensor self, gc_tensor *indices_data, int indices_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::index_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), of_carray_tensor_opt(indices_data, indices_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_indices(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->indices();
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_indices_copy(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::indices_copy(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_indices_copy_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::indices_copy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_infinitely_differentiable_gelu_backward(gc_tensor grad, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::infinitely_differentiable_gelu_backward(*tensor_ptr_from_ocaml(grad), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_inner(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::inner(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_inner_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::inner_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_instance_norm(gc_tensor input, gc_tensor weight, gc_tensor bias, gc_tensor running_mean, gc_tensor running_var, int use_input_stats, double momentum, double eps, int cudnn_enabled) {
+  PROTECT(
+    torch::Tensor results__ = torch::instance_norm(*tensor_ptr_from_ocaml(input), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), (running_mean ? tensor_from_ocaml(running_mean) : torch::Tensor()), (running_var ? tensor_from_ocaml(running_var) : torch::Tensor()), (bool)use_input_stats, momentum, eps, (bool)cudnn_enabled);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_int_repr(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::int_repr(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_int_repr_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::int_repr_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_inverse(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::inverse(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_inverse_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::inverse_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+int atg_is_coalesced(gc_tensor self) {
+  PROTECT(
+    return tensor_ptr_from_ocaml(self)->is_coalesced();
+  )
+  return 0;
+}
+
+int atg_is_complex(gc_tensor self) {
+  PROTECT(
+    return torch::is_complex(*tensor_ptr_from_ocaml(self));
+  )
+  return 0;
+}
+
+int atg_is_conj(gc_tensor self) {
+  PROTECT(
+    return torch::is_conj(*tensor_ptr_from_ocaml(self));
+  )
+  return 0;
+}
+
+int atg_is_distributed(gc_tensor self) {
+  PROTECT(
+    return torch::is_distributed(*tensor_ptr_from_ocaml(self));
+  )
+  return 0;
+}
+
+int atg_is_floating_point(gc_tensor self) {
+  PROTECT(
+    return torch::is_floating_point(*tensor_ptr_from_ocaml(self));
+  )
+  return 0;
+}
+
+int atg_is_inference(gc_tensor self) {
+  PROTECT(
+    return torch::is_inference(*tensor_ptr_from_ocaml(self));
+  )
+  return 0;
+}
+
+int atg_is_leaf(gc_tensor self) {
+  PROTECT(
+    return tensor_ptr_from_ocaml(self)->is_leaf();
+  )
+  return 0;
+}
+
+int atg_is_neg(gc_tensor self) {
+  PROTECT(
+    return torch::is_neg(*tensor_ptr_from_ocaml(self));
+  )
+  return 0;
+}
+
+int atg_is_nonzero(gc_tensor self) {
+  PROTECT(
+    return torch::is_nonzero(*tensor_ptr_from_ocaml(self));
+  )
+  return 0;
+}
+
+int atg_is_pinned(gc_tensor self, int device) {
+  PROTECT(
+    return tensor_ptr_from_ocaml(self)->is_pinned(device_of_int(device));
+  )
+  return 0;
+}
+
+int atg_is_same_size(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    return torch::is_same_size(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+  )
+  return 0;
+}
+
+int atg_is_set_to(gc_tensor self, gc_tensor tensor) {
+  PROTECT(
+    return tensor_ptr_from_ocaml(self)->is_set_to(*tensor_ptr_from_ocaml(tensor));
+  )
+  return 0;
+}
+
+int atg_is_signed(gc_tensor self) {
+  PROTECT(
+    return torch::is_signed(*tensor_ptr_from_ocaml(self));
+  )
+  return 0;
+}
+
+int atg_is_vulkan_available() {
+  PROTECT(
+    return torch::is_vulkan_available();
+  )
+  return 0;
+}
+
+raw_tensor atg_isclose(gc_tensor self, gc_tensor other, double rtol, double atol, int equal_nan) {
+  PROTECT(
+    torch::Tensor results__ = torch::isclose(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other), rtol, atol, (bool)equal_nan);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_isfinite(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::isfinite(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_isin(gc_tensor elements, gc_tensor test_elements, int assume_unique, int invert) {
+  PROTECT(
+    torch::Tensor results__ = torch::isin(*tensor_ptr_from_ocaml(elements), *tensor_ptr_from_ocaml(test_elements), (bool)assume_unique, (bool)invert);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_isin_scalar_tensor(scalar element, gc_tensor test_elements, int assume_unique, int invert) {
+  PROTECT(
+    torch::Tensor results__ = torch::isin(*element, *tensor_ptr_from_ocaml(test_elements), (bool)assume_unique, (bool)invert);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_isin_scalar_tensor_out(gc_tensor out, scalar element, gc_tensor test_elements, int assume_unique, int invert) {
+  PROTECT(
+    torch::Tensor results__ = torch::isin_out(*tensor_ptr_from_ocaml(out), *element, *tensor_ptr_from_ocaml(test_elements), (bool)assume_unique, (bool)invert);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_isin_tensor_scalar(gc_tensor elements, scalar test_element, int assume_unique, int invert) {
+  PROTECT(
+    torch::Tensor results__ = torch::isin(*tensor_ptr_from_ocaml(elements), *test_element, (bool)assume_unique, (bool)invert);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_isin_tensor_scalar_out(gc_tensor out, gc_tensor elements, scalar test_element, int assume_unique, int invert) {
+  PROTECT(
+    torch::Tensor results__ = torch::isin_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(elements), *test_element, (bool)assume_unique, (bool)invert);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_isin_tensor_tensor_out(gc_tensor out, gc_tensor elements, gc_tensor test_elements, int assume_unique, int invert) {
+  PROTECT(
+    torch::Tensor results__ = torch::isin_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(elements), *tensor_ptr_from_ocaml(test_elements), (bool)assume_unique, (bool)invert);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_isinf(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::isinf(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_isinf_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::isinf_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_isnan(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::isnan(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_isnan_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::isnan_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_isneginf(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::isneginf(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_isneginf_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::isneginf_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_isposinf(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::isposinf(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_isposinf_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::isposinf_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_isreal(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::isreal(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_istft(gc_tensor self, int64_t n_fft, int64_t hop_length_v, int hop_length_null, int64_t win_length_v, int win_length_null, gc_tensor window, int center, int normalized, int onesided, int64_t length_v, int length_null, int return_complex) {
+  PROTECT(
+    torch::Tensor results__ = torch::istft(*tensor_ptr_from_ocaml(self), n_fft, hop_length_null ? c10::nullopt : c10::optional<int64_t>(hop_length_v), win_length_null ? c10::nullopt : c10::optional<int64_t>(win_length_v), (window ? tensor_from_ocaml(window) : torch::Tensor()), (bool)center, (bool)normalized, (bool)onesided, length_null ? c10::nullopt : c10::optional<int64_t>(length_v), (bool)return_complex);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_kaiser_window(int64_t window_length, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::kaiser_window(window_length, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_kaiser_window_beta(int64_t window_length, int periodic, double beta, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::kaiser_window(window_length, (bool)periodic, beta, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_kaiser_window_beta_out(gc_tensor out, int64_t window_length, int periodic, double beta) {
+  PROTECT(
+    torch::Tensor results__ = torch::kaiser_window_out(*tensor_ptr_from_ocaml(out), window_length, (bool)periodic, beta);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_kaiser_window_out(gc_tensor out, int64_t window_length) {
+  PROTECT(
+    torch::Tensor results__ = torch::kaiser_window_out(*tensor_ptr_from_ocaml(out), window_length);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_kaiser_window_periodic(int64_t window_length, int periodic, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::kaiser_window(window_length, (bool)periodic, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_kaiser_window_periodic_out(gc_tensor out, int64_t window_length, int periodic) {
+  PROTECT(
+    torch::Tensor results__ = torch::kaiser_window_out(*tensor_ptr_from_ocaml(out), window_length, (bool)periodic);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_kl_div(gc_tensor self, gc_tensor target, int64_t reduction, int log_target) {
+  PROTECT(
+    torch::Tensor results__ = torch::kl_div(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(target), reduction, (bool)log_target);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_kron(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::kron(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_kron_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::kron_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_kthvalue(raw_tensor *out__, gc_tensor self, int64_t k, int64_t dim, int keepdim) {
+  PROTECT(
+    auto results__ = torch::kthvalue(*tensor_ptr_from_ocaml(self), k, dim, (bool)keepdim);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_kthvalue_values(raw_tensor *out__, gc_tensor values, gc_tensor indices, gc_tensor self, int64_t k, int64_t dim, int keepdim) {
+  PROTECT(
+    auto results__ = torch::kthvalue_out(*tensor_ptr_from_ocaml(values), *tensor_ptr_from_ocaml(indices), *tensor_ptr_from_ocaml(self), k, dim, (bool)keepdim);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_l1_loss(gc_tensor self, gc_tensor target, int64_t reduction) {
+  PROTECT(
+    torch::Tensor results__ = torch::l1_loss(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(target), reduction);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_layer_norm(gc_tensor input, int64_t *normalized_shape_data, int normalized_shape_len, gc_tensor weight, gc_tensor bias, double eps, int cudnn_enable) {
+  PROTECT(
+    torch::Tensor results__ = torch::layer_norm(*tensor_ptr_from_ocaml(input), torch::IntArrayRef(normalized_shape_data, normalized_shape_len), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), eps, (bool)cudnn_enable);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_lcm(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::lcm(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_lcm_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::lcm_(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_lcm_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::lcm_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_ldexp(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::ldexp(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_ldexp_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::ldexp_(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_ldexp_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::ldexp_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_le(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::le(*tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_le_(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->le_(*other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_le_scalar_out(gc_tensor out, gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::le_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_le_tensor(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::le(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_le_tensor_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->le_(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_le_tensor_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::le_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_leaky_relu(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::leaky_relu(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_leaky_relu_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::leaky_relu_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_leaky_relu_backward(gc_tensor grad_output, gc_tensor self, scalar negative_slope, int self_is_result) {
+  PROTECT(
+    torch::Tensor results__ = torch::leaky_relu_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), *negative_slope, (bool)self_is_result);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_leaky_relu_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, scalar negative_slope, int self_is_result) {
+  PROTECT(
+    torch::Tensor results__ = torch::leaky_relu_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), *negative_slope, (bool)self_is_result);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_leaky_relu_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::leaky_relu_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_lerp(gc_tensor self, gc_tensor end, scalar weight) {
+  PROTECT(
+    torch::Tensor results__ = torch::lerp(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(end), *weight);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_lerp_(gc_tensor self, gc_tensor end, scalar weight) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->lerp_(*tensor_ptr_from_ocaml(end), *weight);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_lerp_scalar_out(gc_tensor out, gc_tensor self, gc_tensor end, scalar weight) {
+  PROTECT(
+    torch::Tensor results__ = torch::lerp_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(end), *weight);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_lerp_tensor(gc_tensor self, gc_tensor end, gc_tensor weight) {
+  PROTECT(
+    torch::Tensor results__ = torch::lerp(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(end), *tensor_ptr_from_ocaml(weight));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_lerp_tensor_(gc_tensor self, gc_tensor end, gc_tensor weight) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->lerp_(*tensor_ptr_from_ocaml(end), *tensor_ptr_from_ocaml(weight));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_lerp_tensor_out(gc_tensor out, gc_tensor self, gc_tensor end, gc_tensor weight) {
+  PROTECT(
+    torch::Tensor results__ = torch::lerp_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(end), *tensor_ptr_from_ocaml(weight));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_less(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::less(*tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_less_(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->less_(*other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_less_equal(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::less_equal(*tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_less_equal_(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->less_equal_(*other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_less_equal_scalar_out(gc_tensor out, gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::less_equal_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_less_equal_tensor(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::less_equal(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_less_equal_tensor_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->less_equal_(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_less_equal_tensor_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::less_equal_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_less_scalar_out(gc_tensor out, gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::less_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_less_tensor(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::less(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_less_tensor_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->less_(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_less_tensor_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::less_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_lgamma(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::lgamma(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_lgamma_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->lgamma_();
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_lgamma_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::lgamma_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_lift(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::lift(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_lift_fresh(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::lift_fresh(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_lift_fresh_copy(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::lift_fresh_copy(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_lift_fresh_copy_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::lift_fresh_copy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_lift_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::lift_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_cholesky(gc_tensor self, int upper) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_cholesky(*tensor_ptr_from_ocaml(self), (bool)upper);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_linalg_cholesky_ex(raw_tensor *out__, gc_tensor self, int upper, int check_errors) {
+  PROTECT(
+    auto results__ = torch::linalg_cholesky_ex(*tensor_ptr_from_ocaml(self), (bool)upper, (bool)check_errors);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_linalg_cholesky_ex_l(raw_tensor *out__, gc_tensor L, gc_tensor info, gc_tensor self, int upper, int check_errors) {
+  PROTECT(
+    auto results__ = torch::linalg_cholesky_ex_out(*tensor_ptr_from_ocaml(L), *tensor_ptr_from_ocaml(info), *tensor_ptr_from_ocaml(self), (bool)upper, (bool)check_errors);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_linalg_cholesky_out(gc_tensor out, gc_tensor self, int upper) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_cholesky_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), (bool)upper);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_cond(gc_tensor self, scalar p) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_cond(*tensor_ptr_from_ocaml(self), *p);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_cond_out(gc_tensor out, gc_tensor self, scalar p) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_cond_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *p);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_cond_p_str(gc_tensor self, char * p) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_cond(*tensor_ptr_from_ocaml(self), std::string(p));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_cond_p_str_out(gc_tensor out, gc_tensor self, char * p) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_cond_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), std::string(p));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_cross(gc_tensor self, gc_tensor other, int64_t dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_cross(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other), dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_cross_out(gc_tensor out, gc_tensor self, gc_tensor other, int64_t dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_cross_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other), dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_det(gc_tensor A) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_det(*tensor_ptr_from_ocaml(A));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_det_out(gc_tensor out, gc_tensor A) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_det_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(A));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_diagonal(gc_tensor A, int64_t offset, int64_t dim1, int64_t dim2) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_diagonal(*tensor_ptr_from_ocaml(A), offset, dim1, dim2);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_linalg_eig(raw_tensor *out__, gc_tensor self) {
+  PROTECT(
+    auto results__ = torch::linalg_eig(*tensor_ptr_from_ocaml(self));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_linalg_eig_out(raw_tensor *out__, gc_tensor eigenvalues, gc_tensor eigenvectors, gc_tensor self) {
+  PROTECT(
+    auto results__ = torch::linalg_eig_out(*tensor_ptr_from_ocaml(eigenvalues), *tensor_ptr_from_ocaml(eigenvectors), *tensor_ptr_from_ocaml(self));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_linalg_eigh(raw_tensor *out__, gc_tensor self, char * UPLO) {
+  PROTECT(
+    auto results__ = torch::linalg_eigh(*tensor_ptr_from_ocaml(self), std::string(UPLO));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_linalg_eigh_eigvals(raw_tensor *out__, gc_tensor eigvals, gc_tensor eigvecs, gc_tensor self, char * UPLO) {
+  PROTECT(
+    auto results__ = torch::linalg_eigh_out(*tensor_ptr_from_ocaml(eigvals), *tensor_ptr_from_ocaml(eigvecs), *tensor_ptr_from_ocaml(self), std::string(UPLO));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_linalg_eigvals(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_eigvals(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_eigvals_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_eigvals_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_eigvalsh(gc_tensor self, char * UPLO) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_eigvalsh(*tensor_ptr_from_ocaml(self), std::string(UPLO));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_eigvalsh_out(gc_tensor out, gc_tensor self, char * UPLO) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_eigvalsh_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), std::string(UPLO));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_householder_product(gc_tensor input, gc_tensor tau) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_householder_product(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(tau));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_householder_product_out(gc_tensor out, gc_tensor input, gc_tensor tau) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_householder_product_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(tau));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_inv(gc_tensor A) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_inv(*tensor_ptr_from_ocaml(A));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_linalg_inv_ex(raw_tensor *out__, gc_tensor A, int check_errors) {
+  PROTECT(
+    auto results__ = torch::linalg_inv_ex(*tensor_ptr_from_ocaml(A), (bool)check_errors);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_linalg_inv_ex_inverse(raw_tensor *out__, gc_tensor inverse, gc_tensor info, gc_tensor A, int check_errors) {
+  PROTECT(
+    auto results__ = torch::linalg_inv_ex_out(*tensor_ptr_from_ocaml(inverse), *tensor_ptr_from_ocaml(info), *tensor_ptr_from_ocaml(A), (bool)check_errors);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_linalg_inv_out(gc_tensor out, gc_tensor A) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_inv_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(A));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_linalg_ldl_factor(raw_tensor *out__, gc_tensor self, int hermitian) {
+  PROTECT(
+    auto results__ = torch::linalg_ldl_factor(*tensor_ptr_from_ocaml(self), (bool)hermitian);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_linalg_ldl_factor_ex(raw_tensor *out__, gc_tensor self, int hermitian, int check_errors) {
+  PROTECT(
+    auto results__ = torch::linalg_ldl_factor_ex(*tensor_ptr_from_ocaml(self), (bool)hermitian, (bool)check_errors);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+void atg_linalg_ldl_factor_ex_out(raw_tensor *out__, gc_tensor LD, gc_tensor pivots, gc_tensor info, gc_tensor self, int hermitian, int check_errors) {
+  PROTECT(
+    auto results__ = torch::linalg_ldl_factor_ex_out(*tensor_ptr_from_ocaml(LD), *tensor_ptr_from_ocaml(pivots), *tensor_ptr_from_ocaml(info), *tensor_ptr_from_ocaml(self), (bool)hermitian, (bool)check_errors);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+void atg_linalg_ldl_factor_out(raw_tensor *out__, gc_tensor LD, gc_tensor pivots, gc_tensor self, int hermitian) {
+  PROTECT(
+    auto results__ = torch::linalg_ldl_factor_out(*tensor_ptr_from_ocaml(LD), *tensor_ptr_from_ocaml(pivots), *tensor_ptr_from_ocaml(self), (bool)hermitian);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_linalg_ldl_solve(gc_tensor LD, gc_tensor pivots, gc_tensor B, int hermitian) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_ldl_solve(*tensor_ptr_from_ocaml(LD), *tensor_ptr_from_ocaml(pivots), *tensor_ptr_from_ocaml(B), (bool)hermitian);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_ldl_solve_out(gc_tensor out, gc_tensor LD, gc_tensor pivots, gc_tensor B, int hermitian) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_ldl_solve_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(LD), *tensor_ptr_from_ocaml(pivots), *tensor_ptr_from_ocaml(B), (bool)hermitian);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_linalg_lstsq(raw_tensor *out__, gc_tensor self, gc_tensor b, double rcond_v, int rcond_null, char * driver) {
+  PROTECT(
+    auto results__ = torch::linalg_lstsq(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(b), rcond_null ? c10::nullopt : c10::optional<double>(rcond_v), std::string(driver));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+    out__[3] = tensor_to_ocaml(std::get<3>(results__));
+  )
+}
+
+void atg_linalg_lstsq_out(raw_tensor *out__, gc_tensor solution, gc_tensor residuals, gc_tensor rank, gc_tensor singular_values, gc_tensor self, gc_tensor b, double rcond_v, int rcond_null, char * driver) {
+  PROTECT(
+    auto results__ = torch::linalg_lstsq_out(*tensor_ptr_from_ocaml(solution), *tensor_ptr_from_ocaml(residuals), *tensor_ptr_from_ocaml(rank), *tensor_ptr_from_ocaml(singular_values), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(b), rcond_null ? c10::nullopt : c10::optional<double>(rcond_v), std::string(driver));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+    out__[3] = tensor_to_ocaml(std::get<3>(results__));
+  )
+}
+
+void atg_linalg_lu(raw_tensor *out__, gc_tensor A, int pivot) {
+  PROTECT(
+    auto results__ = torch::linalg_lu(*tensor_ptr_from_ocaml(A), (bool)pivot);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+void atg_linalg_lu_factor(raw_tensor *out__, gc_tensor A, int pivot) {
+  PROTECT(
+    auto results__ = torch::linalg_lu_factor(*tensor_ptr_from_ocaml(A), (bool)pivot);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_linalg_lu_factor_ex(raw_tensor *out__, gc_tensor A, int pivot, int check_errors) {
+  PROTECT(
+    auto results__ = torch::linalg_lu_factor_ex(*tensor_ptr_from_ocaml(A), (bool)pivot, (bool)check_errors);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+void atg_linalg_lu_factor_ex_out(raw_tensor *out__, gc_tensor LU, gc_tensor pivots, gc_tensor info, gc_tensor A, int pivot, int check_errors) {
+  PROTECT(
+    auto results__ = torch::linalg_lu_factor_ex_out(*tensor_ptr_from_ocaml(LU), *tensor_ptr_from_ocaml(pivots), *tensor_ptr_from_ocaml(info), *tensor_ptr_from_ocaml(A), (bool)pivot, (bool)check_errors);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+void atg_linalg_lu_factor_out(raw_tensor *out__, gc_tensor LU, gc_tensor pivots, gc_tensor A, int pivot) {
+  PROTECT(
+    auto results__ = torch::linalg_lu_factor_out(*tensor_ptr_from_ocaml(LU), *tensor_ptr_from_ocaml(pivots), *tensor_ptr_from_ocaml(A), (bool)pivot);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_linalg_lu_out(raw_tensor *out__, gc_tensor P, gc_tensor L, gc_tensor U, gc_tensor A, int pivot) {
+  PROTECT(
+    auto results__ = torch::linalg_lu_out(*tensor_ptr_from_ocaml(P), *tensor_ptr_from_ocaml(L), *tensor_ptr_from_ocaml(U), *tensor_ptr_from_ocaml(A), (bool)pivot);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+raw_tensor atg_linalg_lu_solve(gc_tensor LU, gc_tensor pivots, gc_tensor B, int left, int adjoint) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_lu_solve(*tensor_ptr_from_ocaml(LU), *tensor_ptr_from_ocaml(pivots), *tensor_ptr_from_ocaml(B), (bool)left, (bool)adjoint);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_lu_solve_out(gc_tensor out, gc_tensor LU, gc_tensor pivots, gc_tensor B, int left, int adjoint) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_lu_solve_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(LU), *tensor_ptr_from_ocaml(pivots), *tensor_ptr_from_ocaml(B), (bool)left, (bool)adjoint);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_matmul(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_matmul(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_matmul_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_matmul_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_matrix_exp(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_matrix_exp(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_matrix_exp_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_matrix_exp_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_matrix_power(gc_tensor self, int64_t n) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_matrix_power(*tensor_ptr_from_ocaml(self), n);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_matrix_power_out(gc_tensor out, gc_tensor self, int64_t n) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_matrix_power_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), n);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_matrix_rank(gc_tensor self, double tol, int hermitian) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_matrix_rank(*tensor_ptr_from_ocaml(self), tol, (bool)hermitian);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_matrix_rank_atol_rtol_float(gc_tensor self, double atol_v, int atol_null, double rtol_v, int rtol_null, int hermitian) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_matrix_rank(*tensor_ptr_from_ocaml(self), atol_null ? c10::nullopt : c10::optional<double>(atol_v), rtol_null ? c10::nullopt : c10::optional<double>(rtol_v), (bool)hermitian);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_matrix_rank_atol_rtol_float_out(gc_tensor out, gc_tensor self, double atol_v, int atol_null, double rtol_v, int rtol_null, int hermitian) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_matrix_rank_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), atol_null ? c10::nullopt : c10::optional<double>(atol_v), rtol_null ? c10::nullopt : c10::optional<double>(rtol_v), (bool)hermitian);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_matrix_rank_atol_rtol_tensor(gc_tensor input, gc_tensor atol, gc_tensor rtol, int hermitian) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_matrix_rank(*tensor_ptr_from_ocaml(input), (atol ? tensor_from_ocaml(atol) : torch::Tensor()), (rtol ? tensor_from_ocaml(rtol) : torch::Tensor()), (bool)hermitian);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_matrix_rank_atol_rtol_tensor_out(gc_tensor out, gc_tensor input, gc_tensor atol, gc_tensor rtol, int hermitian) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_matrix_rank_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(input), (atol ? tensor_from_ocaml(atol) : torch::Tensor()), (rtol ? tensor_from_ocaml(rtol) : torch::Tensor()), (bool)hermitian);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_matrix_rank_out(gc_tensor out, gc_tensor self, double tol, int hermitian) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_matrix_rank_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), tol, (bool)hermitian);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_matrix_rank_out_tol_tensor(gc_tensor out, gc_tensor input, gc_tensor tol, int hermitian) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_matrix_rank_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(tol), (bool)hermitian);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_matrix_rank_tol_tensor(gc_tensor input, gc_tensor tol, int hermitian) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_matrix_rank(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(tol), (bool)hermitian);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_multi_dot(gc_tensor *tensors_data, int tensors_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_multi_dot(of_carray_tensor(tensors_data, tensors_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_multi_dot_out(gc_tensor out, gc_tensor *tensors_data, int tensors_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_multi_dot_out(*tensor_ptr_from_ocaml(out), of_carray_tensor(tensors_data, tensors_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_pinv(gc_tensor self, double rcond, int hermitian) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_pinv(*tensor_ptr_from_ocaml(self), rcond, (bool)hermitian);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_pinv_atol_rtol_float(gc_tensor self, double atol_v, int atol_null, double rtol_v, int rtol_null, int hermitian) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_pinv(*tensor_ptr_from_ocaml(self), atol_null ? c10::nullopt : c10::optional<double>(atol_v), rtol_null ? c10::nullopt : c10::optional<double>(rtol_v), (bool)hermitian);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_pinv_atol_rtol_float_out(gc_tensor out, gc_tensor self, double atol_v, int atol_null, double rtol_v, int rtol_null, int hermitian) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_pinv_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), atol_null ? c10::nullopt : c10::optional<double>(atol_v), rtol_null ? c10::nullopt : c10::optional<double>(rtol_v), (bool)hermitian);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_pinv_atol_rtol_tensor(gc_tensor self, gc_tensor atol, gc_tensor rtol, int hermitian) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_pinv(*tensor_ptr_from_ocaml(self), (atol ? tensor_from_ocaml(atol) : torch::Tensor()), (rtol ? tensor_from_ocaml(rtol) : torch::Tensor()), (bool)hermitian);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_pinv_atol_rtol_tensor_out(gc_tensor out, gc_tensor self, gc_tensor atol, gc_tensor rtol, int hermitian) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_pinv_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), (atol ? tensor_from_ocaml(atol) : torch::Tensor()), (rtol ? tensor_from_ocaml(rtol) : torch::Tensor()), (bool)hermitian);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_pinv_out(gc_tensor out, gc_tensor self, double rcond, int hermitian) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_pinv_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), rcond, (bool)hermitian);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_pinv_out_rcond_tensor(gc_tensor out, gc_tensor self, gc_tensor rcond, int hermitian) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_pinv_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(rcond), (bool)hermitian);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_pinv_rcond_tensor(gc_tensor self, gc_tensor rcond, int hermitian) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_pinv(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(rcond), (bool)hermitian);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_linalg_qr(raw_tensor *out__, gc_tensor A, char * mode) {
+  PROTECT(
+    auto results__ = torch::linalg_qr(*tensor_ptr_from_ocaml(A), std::string(mode));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_linalg_qr_out(raw_tensor *out__, gc_tensor Q, gc_tensor R, gc_tensor A, char * mode) {
+  PROTECT(
+    auto results__ = torch::linalg_qr_out(*tensor_ptr_from_ocaml(Q), *tensor_ptr_from_ocaml(R), *tensor_ptr_from_ocaml(A), std::string(mode));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_linalg_slogdet(raw_tensor *out__, gc_tensor A) {
+  PROTECT(
+    auto results__ = torch::linalg_slogdet(*tensor_ptr_from_ocaml(A));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_linalg_slogdet_out(raw_tensor *out__, gc_tensor sign, gc_tensor logabsdet, gc_tensor A) {
+  PROTECT(
+    auto results__ = torch::linalg_slogdet_out(*tensor_ptr_from_ocaml(sign), *tensor_ptr_from_ocaml(logabsdet), *tensor_ptr_from_ocaml(A));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_linalg_solve(gc_tensor A, gc_tensor B, int left) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_solve(*tensor_ptr_from_ocaml(A), *tensor_ptr_from_ocaml(B), (bool)left);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_linalg_solve_ex(raw_tensor *out__, gc_tensor A, gc_tensor B, int left, int check_errors) {
+  PROTECT(
+    auto results__ = torch::linalg_solve_ex(*tensor_ptr_from_ocaml(A), *tensor_ptr_from_ocaml(B), (bool)left, (bool)check_errors);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_linalg_solve_ex_out(raw_tensor *out__, gc_tensor result, gc_tensor info, gc_tensor A, gc_tensor B, int left, int check_errors) {
+  PROTECT(
+    auto results__ = torch::linalg_solve_ex_out(*tensor_ptr_from_ocaml(result), *tensor_ptr_from_ocaml(info), *tensor_ptr_from_ocaml(A), *tensor_ptr_from_ocaml(B), (bool)left, (bool)check_errors);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_linalg_solve_out(gc_tensor out, gc_tensor A, gc_tensor B, int left) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_solve_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(A), *tensor_ptr_from_ocaml(B), (bool)left);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_solve_triangular(gc_tensor self, gc_tensor B, int upper, int left, int unitriangular) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_solve_triangular(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(B), (bool)upper, (bool)left, (bool)unitriangular);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_solve_triangular_out(gc_tensor out, gc_tensor self, gc_tensor B, int upper, int left, int unitriangular) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_solve_triangular_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(B), (bool)upper, (bool)left, (bool)unitriangular);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_linalg_svd(raw_tensor *out__, gc_tensor A, int full_matrices, char * driver) {
+  PROTECT(
+    auto results__ = torch::linalg_svd(*tensor_ptr_from_ocaml(A), (bool)full_matrices, std::string(driver));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+void atg_linalg_svd_u(raw_tensor *out__, gc_tensor U, gc_tensor S, gc_tensor Vh, gc_tensor A, int full_matrices, char * driver) {
+  PROTECT(
+    auto results__ = torch::linalg_svd_out(*tensor_ptr_from_ocaml(U), *tensor_ptr_from_ocaml(S), *tensor_ptr_from_ocaml(Vh), *tensor_ptr_from_ocaml(A), (bool)full_matrices, std::string(driver));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+raw_tensor atg_linalg_svdvals(gc_tensor A, char * driver) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_svdvals(*tensor_ptr_from_ocaml(A), std::string(driver));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_svdvals_out(gc_tensor out, gc_tensor A, char * driver) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_svdvals_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(A), std::string(driver));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_tensorinv(gc_tensor self, int64_t ind) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_tensorinv(*tensor_ptr_from_ocaml(self), ind);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_tensorinv_out(gc_tensor out, gc_tensor self, int64_t ind) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_tensorinv_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), ind);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_tensorsolve(gc_tensor self, gc_tensor other, int64_t *dims_data, int dims_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_tensorsolve(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other), dims_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dims_data, dims_len)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_tensorsolve_out(gc_tensor out, gc_tensor self, gc_tensor other, int64_t *dims_data, int dims_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_tensorsolve_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other), dims_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dims_data, dims_len)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_vander(gc_tensor x, int64_t n_v, int n_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_vander(*tensor_ptr_from_ocaml(x), n_null ? c10::nullopt : c10::optional<int64_t>(n_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_vecdot(gc_tensor x, gc_tensor y, int64_t dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_vecdot(*tensor_ptr_from_ocaml(x), *tensor_ptr_from_ocaml(y), dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linalg_vecdot_out(gc_tensor out, gc_tensor x, gc_tensor y, int64_t dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::linalg_vecdot_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(x), *tensor_ptr_from_ocaml(y), dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linear(gc_tensor input, gc_tensor weight, gc_tensor bias) {
+  PROTECT(
+    torch::Tensor results__ = torch::linear(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(weight), (bias ? tensor_from_ocaml(bias) : torch::Tensor()));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linear_out(gc_tensor out, gc_tensor input, gc_tensor weight, gc_tensor bias) {
+  PROTECT(
+    torch::Tensor results__ = torch::linear_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(weight), (bias ? tensor_from_ocaml(bias) : torch::Tensor()));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linspace(scalar start, scalar end, int64_t steps, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::linspace(*start, *end, steps, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_linspace_out(gc_tensor out, scalar start, scalar end, int64_t steps) {
+  PROTECT(
+    torch::Tensor results__ = torch::linspace_out(*tensor_ptr_from_ocaml(out), *start, *end, steps);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_log(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::log(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_log10(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::log10(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_log10_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::log10_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_log10_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::log10_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_log1p(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::log1p(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_log1p_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::log1p_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_log1p_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::log1p_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_log2(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::log2(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_log2_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::log2_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_log2_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::log2_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_log_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::log_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_log_normal(gc_tensor self, double mean, double std) {
+  PROTECT(
+    torch::Tensor results__ = torch::log_normal(*tensor_ptr_from_ocaml(self), mean, std);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_log_normal_(gc_tensor self, double mean, double std) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->log_normal_(mean, std);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_log_normal_out(gc_tensor out, gc_tensor self, double mean, double std) {
+  PROTECT(
+    torch::Tensor results__ = torch::log_normal_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), mean, std);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_log_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::log_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_log_sigmoid(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::log_sigmoid(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_log_sigmoid_backward(gc_tensor grad_output, gc_tensor self, gc_tensor buffer) {
+  PROTECT(
+    torch::Tensor results__ = torch::log_sigmoid_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(buffer));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_log_sigmoid_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, gc_tensor buffer) {
+  PROTECT(
+    torch::Tensor results__ = torch::log_sigmoid_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(buffer));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_log_sigmoid_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::log_sigmoid_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_log_softmax(gc_tensor self, int64_t dim, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::log_softmax(*tensor_ptr_from_ocaml(self), dim, torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_log_softmax_int_out(gc_tensor out, gc_tensor self, int64_t dim, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::log_softmax_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dim, torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_logaddexp(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::logaddexp(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_logaddexp2(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::logaddexp2(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_logaddexp2_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::logaddexp2_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_logaddexp_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::logaddexp_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_logcumsumexp(gc_tensor self, int64_t dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::logcumsumexp(*tensor_ptr_from_ocaml(self), dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_logcumsumexp_out(gc_tensor out, gc_tensor self, int64_t dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::logcumsumexp_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_logdet(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::logdet(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_logical_and(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::logical_and(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_logical_and_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->logical_and_(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_logical_and_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::logical_and_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_logical_not(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::logical_not(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_logical_not_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->logical_not_();
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_logical_not_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::logical_not_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_logical_or(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::logical_or(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_logical_or_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->logical_or_(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_logical_or_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::logical_or_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_logical_xor(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::logical_xor(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_logical_xor_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->logical_xor_(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_logical_xor_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::logical_xor_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_logit(gc_tensor self, double eps_v, int eps_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::logit(*tensor_ptr_from_ocaml(self), eps_null ? c10::nullopt : c10::optional<double>(eps_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_logit_(gc_tensor self, double eps_v, int eps_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::logit_(*tensor_ptr_from_ocaml(self), eps_null ? c10::nullopt : c10::optional<double>(eps_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_logit_backward(gc_tensor grad_output, gc_tensor self, double eps_v, int eps_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::logit_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), eps_null ? c10::nullopt : c10::optional<double>(eps_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_logit_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, double eps_v, int eps_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::logit_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), eps_null ? c10::nullopt : c10::optional<double>(eps_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_logit_out(gc_tensor out, gc_tensor self, double eps_v, int eps_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::logit_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), eps_null ? c10::nullopt : c10::optional<double>(eps_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_logspace(scalar start, scalar end, int64_t steps, double base, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::logspace(*start, *end, steps, base, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_logspace_out(gc_tensor out, scalar start, scalar end, int64_t steps, double base) {
+  PROTECT(
+    torch::Tensor results__ = torch::logspace_out(*tensor_ptr_from_ocaml(out), *start, *end, steps, base);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_logsumexp(gc_tensor self, int64_t *dim_data, int dim_len, int keepdim) {
+  PROTECT(
+    torch::Tensor results__ = torch::logsumexp(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(dim_data, dim_len), (bool)keepdim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_logsumexp_out(gc_tensor out, gc_tensor self, int64_t *dim_data, int dim_len, int keepdim) {
+  PROTECT(
+    torch::Tensor results__ = torch::logsumexp_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(dim_data, dim_len), (bool)keepdim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_lstm(raw_tensor *out__, gc_tensor input, gc_tensor *hx_data, int hx_len, gc_tensor *params_data, int params_len, int has_biases, int64_t num_layers, double dropout, int train, int bidirectional, int batch_first) {
+  PROTECT(
+    auto results__ = torch::lstm(*tensor_ptr_from_ocaml(input), of_carray_tensor(hx_data, hx_len), of_carray_tensor(params_data, params_len), (bool)has_biases, num_layers, dropout, (bool)train, (bool)bidirectional, (bool)batch_first);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+void atg_lstm_cell(raw_tensor *out__, gc_tensor input, gc_tensor *hx_data, int hx_len, gc_tensor w_ih, gc_tensor w_hh, gc_tensor b_ih, gc_tensor b_hh) {
+  PROTECT(
+    auto results__ = torch::lstm_cell(*tensor_ptr_from_ocaml(input), of_carray_tensor(hx_data, hx_len), *tensor_ptr_from_ocaml(w_ih), *tensor_ptr_from_ocaml(w_hh), (b_ih ? tensor_from_ocaml(b_ih) : torch::Tensor()), (b_hh ? tensor_from_ocaml(b_hh) : torch::Tensor()));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_lstm_data(raw_tensor *out__, gc_tensor data, gc_tensor batch_sizes, gc_tensor *hx_data, int hx_len, gc_tensor *params_data, int params_len, int has_biases, int64_t num_layers, double dropout, int train, int bidirectional) {
+  PROTECT(
+    auto results__ = torch::lstm(*tensor_ptr_from_ocaml(data), *tensor_ptr_from_ocaml(batch_sizes), of_carray_tensor(hx_data, hx_len), of_carray_tensor(params_data, params_len), (bool)has_biases, num_layers, dropout, (bool)train, (bool)bidirectional);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+void atg_lstm_mps_backward(gc_tensor out0, gc_tensor *out1_data, int out1_len, gc_tensor *out2_data, int out2_len, gc_tensor grad_y, gc_tensor grad_hy, gc_tensor grad_cy, gc_tensor z_state, gc_tensor cell_state_fwd, gc_tensor input, gc_tensor layersOutputs, gc_tensor *hx_data, int hx_len, gc_tensor *params_data, int params_len, int has_biases, int64_t num_layers, double dropout, int train, int bidirectional, int batch_first) {
+  PROTECT(
+    torch::lstm_mps_backward_out(*tensor_ptr_from_ocaml(out0), of_carray_tensor(out1_data, out1_len), of_carray_tensor(out2_data, out2_len), (grad_y ? tensor_from_ocaml(grad_y) : torch::Tensor()), (grad_hy ? tensor_from_ocaml(grad_hy) : torch::Tensor()), (grad_cy ? tensor_from_ocaml(grad_cy) : torch::Tensor()), *tensor_ptr_from_ocaml(z_state), *tensor_ptr_from_ocaml(cell_state_fwd), *tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(layersOutputs), of_carray_tensor(hx_data, hx_len), of_carray_tensor(params_data, params_len), (bool)has_biases, num_layers, dropout, (bool)train, (bool)bidirectional, (bool)batch_first);
+  )
+}
+
+raw_tensor atg_lt(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::lt(*tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_lt_(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->lt_(*other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_lt_scalar_out(gc_tensor out, gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::lt_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_lt_tensor(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::lt(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_lt_tensor_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->lt_(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_lt_tensor_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::lt_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_lu_solve(gc_tensor self, gc_tensor LU_data, gc_tensor LU_pivots) {
+  PROTECT(
+    torch::Tensor results__ = torch::lu_solve(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(LU_data), *tensor_ptr_from_ocaml(LU_pivots));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_lu_solve_out(gc_tensor out, gc_tensor self, gc_tensor LU_data, gc_tensor LU_pivots) {
+  PROTECT(
+    torch::Tensor results__ = torch::lu_solve_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(LU_data), *tensor_ptr_from_ocaml(LU_pivots));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_lu_unpack(raw_tensor *out__, gc_tensor LU_data, gc_tensor LU_pivots, int unpack_data, int unpack_pivots) {
+  PROTECT(
+    auto results__ = torch::lu_unpack(*tensor_ptr_from_ocaml(LU_data), *tensor_ptr_from_ocaml(LU_pivots), (bool)unpack_data, (bool)unpack_pivots);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+void atg_lu_unpack_out(raw_tensor *out__, gc_tensor P, gc_tensor L, gc_tensor U, gc_tensor LU_data, gc_tensor LU_pivots, int unpack_data, int unpack_pivots) {
+  PROTECT(
+    auto results__ = torch::lu_unpack_out(*tensor_ptr_from_ocaml(P), *tensor_ptr_from_ocaml(L), *tensor_ptr_from_ocaml(U), *tensor_ptr_from_ocaml(LU_data), *tensor_ptr_from_ocaml(LU_pivots), (bool)unpack_data, (bool)unpack_pivots);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+raw_tensor atg_margin_ranking_loss(gc_tensor input1, gc_tensor input2, gc_tensor target, double margin, int64_t reduction) {
+  PROTECT(
+    torch::Tensor results__ = torch::margin_ranking_loss(*tensor_ptr_from_ocaml(input1), *tensor_ptr_from_ocaml(input2), *tensor_ptr_from_ocaml(target), margin, reduction);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_masked_fill(gc_tensor self, gc_tensor mask, scalar value) {
+  PROTECT(
+    torch::Tensor results__ = torch::masked_fill(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(mask), *value);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_masked_fill_(gc_tensor self, gc_tensor mask, scalar value) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->masked_fill_(*tensor_ptr_from_ocaml(mask), *value);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_masked_fill_scalar_out(gc_tensor out, gc_tensor self, gc_tensor mask, scalar value) {
+  PROTECT(
+    torch::Tensor results__ = torch::masked_fill_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(mask), *value);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_masked_fill_tensor(gc_tensor self, gc_tensor mask, gc_tensor value) {
+  PROTECT(
+    torch::Tensor results__ = torch::masked_fill(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(mask), *tensor_ptr_from_ocaml(value));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_masked_fill_tensor_(gc_tensor self, gc_tensor mask, gc_tensor value) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->masked_fill_(*tensor_ptr_from_ocaml(mask), *tensor_ptr_from_ocaml(value));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_masked_fill_tensor_out(gc_tensor out, gc_tensor self, gc_tensor mask, gc_tensor value) {
+  PROTECT(
+    torch::Tensor results__ = torch::masked_fill_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(mask), *tensor_ptr_from_ocaml(value));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_masked_scatter(gc_tensor self, gc_tensor mask, gc_tensor source) {
+  PROTECT(
+    torch::Tensor results__ = torch::masked_scatter(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(mask), *tensor_ptr_from_ocaml(source));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_masked_scatter_(gc_tensor self, gc_tensor mask, gc_tensor source) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->masked_scatter_(*tensor_ptr_from_ocaml(mask), *tensor_ptr_from_ocaml(source));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_masked_scatter_out(gc_tensor out, gc_tensor self, gc_tensor mask, gc_tensor source) {
+  PROTECT(
+    torch::Tensor results__ = torch::masked_scatter_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(mask), *tensor_ptr_from_ocaml(source));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_masked_select(gc_tensor self, gc_tensor mask) {
+  PROTECT(
+    torch::Tensor results__ = torch::masked_select(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(mask));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_masked_select_backward(gc_tensor grad, gc_tensor input, gc_tensor mask) {
+  PROTECT(
+    torch::Tensor results__ = torch::masked_select_backward(*tensor_ptr_from_ocaml(grad), *tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(mask));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_masked_select_out(gc_tensor out, gc_tensor self, gc_tensor mask) {
+  PROTECT(
+    torch::Tensor results__ = torch::masked_select_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(mask));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_matmul(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::matmul(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_matmul_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::matmul_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_matrix_exp(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::matrix_exp(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_matrix_exp_backward(gc_tensor self, gc_tensor grad) {
+  PROTECT(
+    torch::Tensor results__ = torch::matrix_exp_backward(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(grad));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_matrix_h(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->matrix_H();
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_matrix_power(gc_tensor self, int64_t n) {
+  PROTECT(
+    torch::Tensor results__ = torch::matrix_power(*tensor_ptr_from_ocaml(self), n);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_matrix_power_out(gc_tensor out, gc_tensor self, int64_t n) {
+  PROTECT(
+    torch::Tensor results__ = torch::matrix_power_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), n);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_max(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::max(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_max_dim(raw_tensor *out__, gc_tensor self, int64_t dim, int keepdim) {
+  PROTECT(
+    auto results__ = torch::max(*tensor_ptr_from_ocaml(self), dim, (bool)keepdim);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_max_dim_max(raw_tensor *out__, gc_tensor max, gc_tensor max_values, gc_tensor self, int64_t dim, int keepdim) {
+  PROTECT(
+    auto results__ = torch::max_out(*tensor_ptr_from_ocaml(max), *tensor_ptr_from_ocaml(max_values), *tensor_ptr_from_ocaml(self), dim, (bool)keepdim);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_max_other(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::max(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_max_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::max_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_max_pool1d(gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
+  PROTECT(
+    torch::Tensor results__ = torch::max_pool1d(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_max_pool1d_with_indices(raw_tensor *out__, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
+  PROTECT(
+    auto results__ = torch::max_pool1d_with_indices(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_max_pool2d(gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
+  PROTECT(
+    torch::Tensor results__ = torch::max_pool2d(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_max_pool2d_backward(gc_tensor grad_output, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
+  PROTECT(
+    torch::Tensor results__ = torch::max_pool2d_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_max_pool2d_backward_out(gc_tensor out, gc_tensor grad_output, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
+  PROTECT(
+    torch::Tensor results__ = torch::max_pool2d_backward_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_max_pool2d_with_indices(raw_tensor *out__, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
+  PROTECT(
+    auto results__ = torch::max_pool2d_with_indices(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_max_pool2d_with_indices_backward(gc_tensor grad_output, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode, gc_tensor indices) {
+  PROTECT(
+    torch::Tensor results__ = torch::max_pool2d_with_indices_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode, *tensor_ptr_from_ocaml(indices));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_max_pool2d_with_indices_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode, gc_tensor indices) {
+  PROTECT(
+    torch::Tensor results__ = torch::max_pool2d_with_indices_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode, *tensor_ptr_from_ocaml(indices));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_max_pool2d_with_indices_out(raw_tensor *out__, gc_tensor out, gc_tensor indices, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
+  PROTECT(
+    auto results__ = torch::max_pool2d_with_indices_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(indices), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_max_pool3d(gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
+  PROTECT(
+    torch::Tensor results__ = torch::max_pool3d(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_max_pool3d_with_indices(raw_tensor *out__, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
+  PROTECT(
+    auto results__ = torch::max_pool3d_with_indices(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_max_pool3d_with_indices_backward(gc_tensor grad_output, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode, gc_tensor indices) {
+  PROTECT(
+    torch::Tensor results__ = torch::max_pool3d_with_indices_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode, *tensor_ptr_from_ocaml(indices));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_max_pool3d_with_indices_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode, gc_tensor indices) {
+  PROTECT(
+    torch::Tensor results__ = torch::max_pool3d_with_indices_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode, *tensor_ptr_from_ocaml(indices));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_max_pool3d_with_indices_out(raw_tensor *out__, gc_tensor out, gc_tensor indices, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
+  PROTECT(
+    auto results__ = torch::max_pool3d_with_indices_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(indices), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_max_unary_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::max_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_max_unpool2d(gc_tensor self, gc_tensor indices, int64_t *output_size_data, int output_size_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::max_unpool2d(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(indices), torch::IntArrayRef(output_size_data, output_size_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_max_unpool2d_out(gc_tensor out, gc_tensor self, gc_tensor indices, int64_t *output_size_data, int output_size_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::max_unpool2d_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(indices), torch::IntArrayRef(output_size_data, output_size_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_max_unpool3d(gc_tensor self, gc_tensor indices, int64_t *output_size_data, int output_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::max_unpool3d(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(indices), torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_max_unpool3d_out(gc_tensor out, gc_tensor self, gc_tensor indices, int64_t *output_size_data, int output_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::max_unpool3d_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(indices), torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_maximum(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::maximum(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_maximum_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::maximum_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_mean(gc_tensor self, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::mean(*tensor_ptr_from_ocaml(self), torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_mean_dim(gc_tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::mean(*tensor_ptr_from_ocaml(self), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), (bool)keepdim, torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_mean_out(gc_tensor out, gc_tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::mean_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), (bool)keepdim, torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_median(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::median(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_median_dim(raw_tensor *out__, gc_tensor self, int64_t dim, int keepdim) {
+  PROTECT(
+    auto results__ = torch::median(*tensor_ptr_from_ocaml(self), dim, (bool)keepdim);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_median_dim_values(raw_tensor *out__, gc_tensor values, gc_tensor indices, gc_tensor self, int64_t dim, int keepdim) {
+  PROTECT(
+    auto results__ = torch::median_out(*tensor_ptr_from_ocaml(values), *tensor_ptr_from_ocaml(indices), *tensor_ptr_from_ocaml(self), dim, (bool)keepdim);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_median_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::median_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor *atg_meshgrid(gc_tensor *tensors_data, int tensors_len) {
+  PROTECT(
+    auto results__ = torch::meshgrid(of_carray_tensor(tensors_data, tensors_len));
+    int sz = results__.size();
+    raw_tensor *out__ = (raw_tensor*)malloc((sz + 1) * sizeof(raw_tensor));
+    for (int i = 0; i < sz; ++i)
+      out__[i] = tensor_to_ocaml(results__[i]);
+    out__[sz] = nullptr;
+    return out__;
+  )
+}
+
+raw_tensor *atg_meshgrid_indexing(gc_tensor *tensors_data, int tensors_len, char * indexing) {
+  PROTECT(
+    auto results__ = torch::meshgrid(of_carray_tensor(tensors_data, tensors_len), std::string(indexing));
+    int sz = results__.size();
+    raw_tensor *out__ = (raw_tensor*)malloc((sz + 1) * sizeof(raw_tensor));
+    for (int i = 0; i < sz; ++i)
+      out__[i] = tensor_to_ocaml(results__[i]);
+    out__[sz] = nullptr;
+    return out__;
+  )
+}
+
+raw_tensor atg_mh(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->mH();
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_min(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::min(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_min_dim(raw_tensor *out__, gc_tensor self, int64_t dim, int keepdim) {
+  PROTECT(
+    auto results__ = torch::min(*tensor_ptr_from_ocaml(self), dim, (bool)keepdim);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_min_dim_min(raw_tensor *out__, gc_tensor min, gc_tensor min_indices, gc_tensor self, int64_t dim, int keepdim) {
+  PROTECT(
+    auto results__ = torch::min_out(*tensor_ptr_from_ocaml(min), *tensor_ptr_from_ocaml(min_indices), *tensor_ptr_from_ocaml(self), dim, (bool)keepdim);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_min_other(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::min(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_min_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::min_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_min_unary_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::min_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_minimum(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::minimum(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_minimum_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::minimum_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_miopen_batch_norm(raw_tensor *out__, gc_tensor input, gc_tensor weight, gc_tensor bias, gc_tensor running_mean, gc_tensor running_var, int training, double exponential_average_factor, double epsilon) {
+  PROTECT(
+    auto results__ = torch::miopen_batch_norm(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(weight), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), (running_mean ? tensor_from_ocaml(running_mean) : torch::Tensor()), (running_var ? tensor_from_ocaml(running_var) : torch::Tensor()), (bool)training, exponential_average_factor, epsilon);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+void atg_miopen_batch_norm_backward(raw_tensor *out__, gc_tensor input, gc_tensor grad_output, gc_tensor weight, gc_tensor running_mean, gc_tensor running_var, gc_tensor save_mean, gc_tensor save_var, double epsilon) {
+  PROTECT(
+    auto results__ = torch::miopen_batch_norm_backward(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(weight), (running_mean ? tensor_from_ocaml(running_mean) : torch::Tensor()), (running_var ? tensor_from_ocaml(running_var) : torch::Tensor()), (save_mean ? tensor_from_ocaml(save_mean) : torch::Tensor()), (save_var ? tensor_from_ocaml(save_var) : torch::Tensor()), epsilon);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+void atg_miopen_batch_norm_backward_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor input, gc_tensor grad_output, gc_tensor weight, gc_tensor running_mean, gc_tensor running_var, gc_tensor save_mean, gc_tensor save_var, double epsilon) {
+  PROTECT(
+    auto results__ = torch::miopen_batch_norm_backward_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(out2), *tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(weight), (running_mean ? tensor_from_ocaml(running_mean) : torch::Tensor()), (running_var ? tensor_from_ocaml(running_var) : torch::Tensor()), (save_mean ? tensor_from_ocaml(save_mean) : torch::Tensor()), (save_var ? tensor_from_ocaml(save_var) : torch::Tensor()), epsilon);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+void atg_miopen_batch_norm_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor input, gc_tensor weight, gc_tensor bias, gc_tensor running_mean, gc_tensor running_var, int training, double exponential_average_factor, double epsilon) {
+  PROTECT(
+    auto results__ = torch::miopen_batch_norm_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(out2), *tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(weight), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), (running_mean ? tensor_from_ocaml(running_mean) : torch::Tensor()), (running_var ? tensor_from_ocaml(running_var) : torch::Tensor()), (bool)training, exponential_average_factor, epsilon);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+raw_tensor atg_miopen_convolution(gc_tensor self, gc_tensor weight, gc_tensor bias, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int benchmark, int deterministic) {
+  PROTECT(
+    torch::Tensor results__ = torch::miopen_convolution(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(weight), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(dilation_data, dilation_len), groups, (bool)benchmark, (bool)deterministic);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_miopen_convolution_add_relu(gc_tensor self, gc_tensor weight, gc_tensor z, scalar alpha, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int64_t groups) {
+  PROTECT(
+    torch::Tensor results__ = torch::miopen_convolution_add_relu(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(weight), *tensor_ptr_from_ocaml(z), *alpha, (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), groups);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_miopen_convolution_out(gc_tensor out, gc_tensor self, gc_tensor weight, gc_tensor bias, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int benchmark, int deterministic) {
+  PROTECT(
+    torch::Tensor results__ = torch::miopen_convolution_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(weight), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(dilation_data, dilation_len), groups, (bool)benchmark, (bool)deterministic);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_miopen_convolution_relu(gc_tensor self, gc_tensor weight, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int64_t groups) {
+  PROTECT(
+    torch::Tensor results__ = torch::miopen_convolution_relu(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(weight), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), groups);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_miopen_convolution_transpose(gc_tensor self, gc_tensor weight, gc_tensor bias, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int benchmark, int deterministic) {
+  PROTECT(
+    torch::Tensor results__ = torch::miopen_convolution_transpose(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(weight), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(output_padding_data, output_padding_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(dilation_data, dilation_len), groups, (bool)benchmark, (bool)deterministic);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_miopen_convolution_transpose_out(gc_tensor out, gc_tensor self, gc_tensor weight, gc_tensor bias, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int benchmark, int deterministic) {
+  PROTECT(
+    torch::Tensor results__ = torch::miopen_convolution_transpose_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(weight), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(output_padding_data, output_padding_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(dilation_data, dilation_len), groups, (bool)benchmark, (bool)deterministic);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_miopen_depthwise_convolution(gc_tensor self, gc_tensor weight, gc_tensor bias, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int benchmark, int deterministic) {
+  PROTECT(
+    torch::Tensor results__ = torch::miopen_depthwise_convolution(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(weight), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(dilation_data, dilation_len), groups, (bool)benchmark, (bool)deterministic);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_miopen_depthwise_convolution_out(gc_tensor out, gc_tensor self, gc_tensor weight, gc_tensor bias, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int benchmark, int deterministic) {
+  PROTECT(
+    torch::Tensor results__ = torch::miopen_depthwise_convolution_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(weight), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(dilation_data, dilation_len), groups, (bool)benchmark, (bool)deterministic);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_miopen_rnn(raw_tensor *out__, gc_tensor input, gc_tensor *weight_data, int weight_len, int64_t weight_stride0, gc_tensor hx, gc_tensor cx, int64_t mode, int64_t hidden_size, int64_t num_layers, int batch_first, double dropout, int train, int bidirectional, int64_t *batch_sizes_data, int batch_sizes_len, gc_tensor dropout_state) {
+  PROTECT(
+    auto results__ = torch::miopen_rnn(*tensor_ptr_from_ocaml(input), of_carray_tensor(weight_data, weight_len), weight_stride0, *tensor_ptr_from_ocaml(hx), (cx ? tensor_from_ocaml(cx) : torch::Tensor()), mode, hidden_size, num_layers, (bool)batch_first, dropout, (bool)train, (bool)bidirectional, torch::IntArrayRef(batch_sizes_data, batch_sizes_len), (dropout_state ? tensor_from_ocaml(dropout_state) : torch::Tensor()));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+    out__[3] = tensor_to_ocaml(std::get<3>(results__));
+    out__[4] = tensor_to_ocaml(std::get<4>(results__));
+  )
+}
+
+void atg_miopen_rnn_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor out3, gc_tensor out4, gc_tensor input, gc_tensor *weight_data, int weight_len, int64_t weight_stride0, gc_tensor hx, gc_tensor cx, int64_t mode, int64_t hidden_size, int64_t num_layers, int batch_first, double dropout, int train, int bidirectional, int64_t *batch_sizes_data, int batch_sizes_len, gc_tensor dropout_state) {
+  PROTECT(
+    auto results__ = torch::miopen_rnn_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(out2), *tensor_ptr_from_ocaml(out3), *tensor_ptr_from_ocaml(out4), *tensor_ptr_from_ocaml(input), of_carray_tensor(weight_data, weight_len), weight_stride0, *tensor_ptr_from_ocaml(hx), (cx ? tensor_from_ocaml(cx) : torch::Tensor()), mode, hidden_size, num_layers, (bool)batch_first, dropout, (bool)train, (bool)bidirectional, torch::IntArrayRef(batch_sizes_data, batch_sizes_len), (dropout_state ? tensor_from_ocaml(dropout_state) : torch::Tensor()));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+    out__[3] = tensor_to_ocaml(std::get<3>(results__));
+    out__[4] = tensor_to_ocaml(std::get<4>(results__));
+  )
+}
+
+raw_tensor atg_mish(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::mish(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_mish_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::mish_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_mish_backward(gc_tensor grad_output, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::mish_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_mish_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::mish_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_mkldnn_adaptive_avg_pool2d(gc_tensor self, int64_t *output_size_data, int output_size_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::mkldnn_adaptive_avg_pool2d(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(output_size_data, output_size_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_mkldnn_adaptive_avg_pool2d_backward(gc_tensor grad_output, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::mkldnn_adaptive_avg_pool2d_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_mkldnn_adaptive_avg_pool2d_backward_out(gc_tensor out, gc_tensor grad_output, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::mkldnn_adaptive_avg_pool2d_backward_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_mkldnn_adaptive_avg_pool2d_out(gc_tensor out, gc_tensor self, int64_t *output_size_data, int output_size_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::mkldnn_adaptive_avg_pool2d_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(output_size_data, output_size_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_mkldnn_convolution(gc_tensor self, gc_tensor weight, gc_tensor bias, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups) {
+  PROTECT(
+    torch::Tensor results__ = torch::mkldnn_convolution(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(weight), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(dilation_data, dilation_len), groups);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_mkldnn_convolution_out(gc_tensor out, gc_tensor self, gc_tensor weight, gc_tensor bias, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups) {
+  PROTECT(
+    torch::Tensor results__ = torch::mkldnn_convolution_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(weight), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(dilation_data, dilation_len), groups);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_mkldnn_linear(gc_tensor self, gc_tensor weight, gc_tensor bias) {
+  PROTECT(
+    torch::Tensor results__ = torch::mkldnn_linear(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(weight), (bias ? tensor_from_ocaml(bias) : torch::Tensor()));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_mkldnn_linear_backward_input(int64_t *input_size_data, int input_size_len, gc_tensor grad_output, gc_tensor weight) {
+  PROTECT(
+    torch::Tensor results__ = torch::mkldnn_linear_backward_input(torch::IntArrayRef(input_size_data, input_size_len), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(weight));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_mkldnn_linear_backward_input_out(gc_tensor out, int64_t *input_size_data, int input_size_len, gc_tensor grad_output, gc_tensor weight) {
+  PROTECT(
+    torch::Tensor results__ = torch::mkldnn_linear_backward_input_out(*tensor_ptr_from_ocaml(out), torch::IntArrayRef(input_size_data, input_size_len), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(weight));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_mkldnn_linear_backward_weights(raw_tensor *out__, gc_tensor grad_output, gc_tensor input, gc_tensor weight, int bias_defined) {
+  PROTECT(
+    auto results__ = torch::mkldnn_linear_backward_weights(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(weight), (bool)bias_defined);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_mkldnn_linear_backward_weights_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor grad_output, gc_tensor input, gc_tensor weight, int bias_defined) {
+  PROTECT(
+    auto results__ = torch::mkldnn_linear_backward_weights_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(weight), (bool)bias_defined);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_mkldnn_linear_out(gc_tensor out, gc_tensor self, gc_tensor weight, gc_tensor bias) {
+  PROTECT(
+    torch::Tensor results__ = torch::mkldnn_linear_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(weight), (bias ? tensor_from_ocaml(bias) : torch::Tensor()));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_mkldnn_max_pool2d(gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
+  PROTECT(
+    torch::Tensor results__ = torch::mkldnn_max_pool2d(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_mkldnn_max_pool2d_backward(gc_tensor grad_output, gc_tensor output, gc_tensor input, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
+  PROTECT(
+    torch::Tensor results__ = torch::mkldnn_max_pool2d_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(output), *tensor_ptr_from_ocaml(input), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_mkldnn_max_pool2d_backward_out(gc_tensor out, gc_tensor grad_output, gc_tensor output, gc_tensor input, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
+  PROTECT(
+    torch::Tensor results__ = torch::mkldnn_max_pool2d_backward_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(output), *tensor_ptr_from_ocaml(input), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_mkldnn_max_pool2d_out(gc_tensor out, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
+  PROTECT(
+    torch::Tensor results__ = torch::mkldnn_max_pool2d_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_mkldnn_max_pool3d(gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
+  PROTECT(
+    torch::Tensor results__ = torch::mkldnn_max_pool3d(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_mkldnn_max_pool3d_backward(gc_tensor grad_output, gc_tensor output, gc_tensor input, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
+  PROTECT(
+    torch::Tensor results__ = torch::mkldnn_max_pool3d_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(output), *tensor_ptr_from_ocaml(input), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_mkldnn_max_pool3d_backward_out(gc_tensor out, gc_tensor grad_output, gc_tensor output, gc_tensor input, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
+  PROTECT(
+    torch::Tensor results__ = torch::mkldnn_max_pool3d_backward_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(output), *tensor_ptr_from_ocaml(input), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_mkldnn_max_pool3d_out(gc_tensor out, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
+  PROTECT(
+    torch::Tensor results__ = torch::mkldnn_max_pool3d_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_mkldnn_reorder_conv2d_weight(gc_tensor self, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int64_t *input_size_data, int input_size_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::mkldnn_reorder_conv2d_weight(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(dilation_data, dilation_len), groups, input_size_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(input_size_data, input_size_len)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_mkldnn_reorder_conv2d_weight_out(gc_tensor out, gc_tensor self, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int64_t *input_size_data, int input_size_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::mkldnn_reorder_conv2d_weight_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(dilation_data, dilation_len), groups, input_size_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(input_size_data, input_size_len)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_mkldnn_reorder_conv3d_weight(gc_tensor self, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups) {
+  PROTECT(
+    torch::Tensor results__ = torch::mkldnn_reorder_conv3d_weight(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(dilation_data, dilation_len), groups);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_mkldnn_reorder_conv3d_weight_out(gc_tensor out, gc_tensor self, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups) {
+  PROTECT(
+    torch::Tensor results__ = torch::mkldnn_reorder_conv3d_weight_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(dilation_data, dilation_len), groups);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_mkldnn_rnn_layer(raw_tensor *out__, gc_tensor input, gc_tensor weight0, gc_tensor weight1, gc_tensor weight2, gc_tensor weight3, gc_tensor hx_, gc_tensor cx_, int reverse, int64_t *batch_sizes_data, int batch_sizes_len, int64_t mode, int64_t hidden_size, int64_t num_layers, int has_biases, int bidirectional, int batch_first, int train) {
+  PROTECT(
+    auto results__ = torch::mkldnn_rnn_layer(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(weight0), *tensor_ptr_from_ocaml(weight1), *tensor_ptr_from_ocaml(weight2), *tensor_ptr_from_ocaml(weight3), *tensor_ptr_from_ocaml(hx_), *tensor_ptr_from_ocaml(cx_), (bool)reverse, torch::IntArrayRef(batch_sizes_data, batch_sizes_len), mode, hidden_size, num_layers, (bool)has_biases, (bool)bidirectional, (bool)batch_first, (bool)train);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+    out__[3] = tensor_to_ocaml(std::get<3>(results__));
+  )
+}
+
+void atg_mkldnn_rnn_layer_backward(raw_tensor *out__, gc_tensor input, gc_tensor weight1, gc_tensor weight2, gc_tensor weight3, gc_tensor weight4, gc_tensor hx_, gc_tensor cx_tmp, gc_tensor output, gc_tensor hy_, gc_tensor cy_, gc_tensor grad_output, gc_tensor grad_hy, gc_tensor grad_cy, int reverse, int64_t mode, int64_t hidden_size, int64_t num_layers, int has_biases, int train, int bidirectional, int64_t *batch_sizes_data, int batch_sizes_len, int batch_first, gc_tensor workspace) {
+  PROTECT(
+    auto results__ = torch::mkldnn_rnn_layer_backward(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(weight1), *tensor_ptr_from_ocaml(weight2), *tensor_ptr_from_ocaml(weight3), *tensor_ptr_from_ocaml(weight4), *tensor_ptr_from_ocaml(hx_), *tensor_ptr_from_ocaml(cx_tmp), *tensor_ptr_from_ocaml(output), *tensor_ptr_from_ocaml(hy_), *tensor_ptr_from_ocaml(cy_), (grad_output ? tensor_from_ocaml(grad_output) : torch::Tensor()), (grad_hy ? tensor_from_ocaml(grad_hy) : torch::Tensor()), (grad_cy ? tensor_from_ocaml(grad_cy) : torch::Tensor()), (bool)reverse, mode, hidden_size, num_layers, (bool)has_biases, (bool)train, (bool)bidirectional, torch::IntArrayRef(batch_sizes_data, batch_sizes_len), (bool)batch_first, *tensor_ptr_from_ocaml(workspace));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+    out__[3] = tensor_to_ocaml(std::get<3>(results__));
+    out__[4] = tensor_to_ocaml(std::get<4>(results__));
+    out__[5] = tensor_to_ocaml(std::get<5>(results__));
+    out__[6] = tensor_to_ocaml(std::get<6>(results__));
+  )
+}
+
+void atg_mkldnn_rnn_layer_backward_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor out3, gc_tensor out4, gc_tensor out5, gc_tensor out6, gc_tensor input, gc_tensor weight1, gc_tensor weight2, gc_tensor weight3, gc_tensor weight4, gc_tensor hx_, gc_tensor cx_tmp, gc_tensor output, gc_tensor hy_, gc_tensor cy_, gc_tensor grad_output, gc_tensor grad_hy, gc_tensor grad_cy, int reverse, int64_t mode, int64_t hidden_size, int64_t num_layers, int has_biases, int train, int bidirectional, int64_t *batch_sizes_data, int batch_sizes_len, int batch_first, gc_tensor workspace) {
+  PROTECT(
+    auto results__ = torch::mkldnn_rnn_layer_backward_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(out2), *tensor_ptr_from_ocaml(out3), *tensor_ptr_from_ocaml(out4), *tensor_ptr_from_ocaml(out5), *tensor_ptr_from_ocaml(out6), *tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(weight1), *tensor_ptr_from_ocaml(weight2), *tensor_ptr_from_ocaml(weight3), *tensor_ptr_from_ocaml(weight4), *tensor_ptr_from_ocaml(hx_), *tensor_ptr_from_ocaml(cx_tmp), *tensor_ptr_from_ocaml(output), *tensor_ptr_from_ocaml(hy_), *tensor_ptr_from_ocaml(cy_), (grad_output ? tensor_from_ocaml(grad_output) : torch::Tensor()), (grad_hy ? tensor_from_ocaml(grad_hy) : torch::Tensor()), (grad_cy ? tensor_from_ocaml(grad_cy) : torch::Tensor()), (bool)reverse, mode, hidden_size, num_layers, (bool)has_biases, (bool)train, (bool)bidirectional, torch::IntArrayRef(batch_sizes_data, batch_sizes_len), (bool)batch_first, *tensor_ptr_from_ocaml(workspace));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+    out__[3] = tensor_to_ocaml(std::get<3>(results__));
+    out__[4] = tensor_to_ocaml(std::get<4>(results__));
+    out__[5] = tensor_to_ocaml(std::get<5>(results__));
+    out__[6] = tensor_to_ocaml(std::get<6>(results__));
+  )
+}
+
+void atg_mkldnn_rnn_layer_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor out3, gc_tensor input, gc_tensor weight0, gc_tensor weight1, gc_tensor weight2, gc_tensor weight3, gc_tensor hx_, gc_tensor cx_, int reverse, int64_t *batch_sizes_data, int batch_sizes_len, int64_t mode, int64_t hidden_size, int64_t num_layers, int has_biases, int bidirectional, int batch_first, int train) {
+  PROTECT(
+    auto results__ = torch::mkldnn_rnn_layer_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(out2), *tensor_ptr_from_ocaml(out3), *tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(weight0), *tensor_ptr_from_ocaml(weight1), *tensor_ptr_from_ocaml(weight2), *tensor_ptr_from_ocaml(weight3), *tensor_ptr_from_ocaml(hx_), *tensor_ptr_from_ocaml(cx_), (bool)reverse, torch::IntArrayRef(batch_sizes_data, batch_sizes_len), mode, hidden_size, num_layers, (bool)has_biases, (bool)bidirectional, (bool)batch_first, (bool)train);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+    out__[3] = tensor_to_ocaml(std::get<3>(results__));
+  )
+}
+
+raw_tensor atg_mm(gc_tensor self, gc_tensor mat2) {
+  PROTECT(
+    torch::Tensor results__ = torch::mm(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(mat2));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_mm_out(gc_tensor out, gc_tensor self, gc_tensor mat2) {
+  PROTECT(
+    torch::Tensor results__ = torch::mm_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(mat2));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_mode(raw_tensor *out__, gc_tensor self, int64_t dim, int keepdim) {
+  PROTECT(
+    auto results__ = torch::mode(*tensor_ptr_from_ocaml(self), dim, (bool)keepdim);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_mode_values(raw_tensor *out__, gc_tensor values, gc_tensor indices, gc_tensor self, int64_t dim, int keepdim) {
+  PROTECT(
+    auto results__ = torch::mode_out(*tensor_ptr_from_ocaml(values), *tensor_ptr_from_ocaml(indices), *tensor_ptr_from_ocaml(self), dim, (bool)keepdim);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_moveaxis(gc_tensor self, int64_t *source_data, int source_len, int64_t *destination_data, int destination_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::moveaxis(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(source_data, source_len), torch::IntArrayRef(destination_data, destination_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_moveaxis_int(gc_tensor self, int64_t source, int64_t destination) {
+  PROTECT(
+    torch::Tensor results__ = torch::moveaxis(*tensor_ptr_from_ocaml(self), source, destination);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_movedim(gc_tensor self, int64_t *source_data, int source_len, int64_t *destination_data, int destination_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::movedim(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(source_data, source_len), torch::IntArrayRef(destination_data, destination_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_movedim_int(gc_tensor self, int64_t source, int64_t destination) {
+  PROTECT(
+    torch::Tensor results__ = torch::movedim(*tensor_ptr_from_ocaml(self), source, destination);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_mse_loss(gc_tensor self, gc_tensor target, int64_t reduction) {
+  PROTECT(
+    torch::Tensor results__ = torch::mse_loss(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(target), reduction);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_mse_loss_backward(gc_tensor grad_output, gc_tensor self, gc_tensor target, int64_t reduction) {
+  PROTECT(
+    torch::Tensor results__ = torch::mse_loss_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(target), reduction);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_mse_loss_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, gc_tensor target, int64_t reduction) {
+  PROTECT(
+    torch::Tensor results__ = torch::mse_loss_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(target), reduction);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_mse_loss_out(gc_tensor out, gc_tensor self, gc_tensor target, int64_t reduction) {
+  PROTECT(
+    torch::Tensor results__ = torch::mse_loss_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(target), reduction);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_msort(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::msort(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_msort_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::msort_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_mt(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->mT();
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_mul(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::mul(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_mul_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->mul_(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_mul_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::mul_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_mul_scalar(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::mul(*tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_mul_scalar_(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->mul_(*other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_mul_scalar_out(gc_tensor out, gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::mul_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_multi_margin_loss_backward(gc_tensor grad_output, gc_tensor self, gc_tensor target, scalar p, scalar margin, gc_tensor weight, int64_t reduction) {
+  PROTECT(
+    torch::Tensor results__ = torch::multi_margin_loss_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(target), *p, *margin, (weight ? tensor_from_ocaml(weight) : torch::Tensor()), reduction);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_multi_margin_loss_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, gc_tensor target, scalar p, scalar margin, gc_tensor weight, int64_t reduction) {
+  PROTECT(
+    torch::Tensor results__ = torch::multi_margin_loss_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(target), *p, *margin, (weight ? tensor_from_ocaml(weight) : torch::Tensor()), reduction);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_multilabel_margin_loss(gc_tensor self, gc_tensor target, int64_t reduction) {
+  PROTECT(
+    torch::Tensor results__ = torch::multilabel_margin_loss(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(target), reduction);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_multilabel_margin_loss_backward(gc_tensor grad_output, gc_tensor self, gc_tensor target, int64_t reduction, gc_tensor is_target) {
+  PROTECT(
+    torch::Tensor results__ = torch::multilabel_margin_loss_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(target), reduction, *tensor_ptr_from_ocaml(is_target));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_multilabel_margin_loss_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, gc_tensor target, int64_t reduction, gc_tensor is_target) {
+  PROTECT(
+    torch::Tensor results__ = torch::multilabel_margin_loss_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(target), reduction, *tensor_ptr_from_ocaml(is_target));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_multilabel_margin_loss_out(gc_tensor out, gc_tensor self, gc_tensor target, int64_t reduction) {
+  PROTECT(
+    torch::Tensor results__ = torch::multilabel_margin_loss_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(target), reduction);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_multinomial(gc_tensor self, int64_t num_samples, int replacement) {
+  PROTECT(
+    torch::Tensor results__ = torch::multinomial(*tensor_ptr_from_ocaml(self), num_samples, (bool)replacement);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_multinomial_out(gc_tensor out, gc_tensor self, int64_t num_samples, int replacement) {
+  PROTECT(
+    torch::Tensor results__ = torch::multinomial_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), num_samples, (bool)replacement);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_multiply(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::multiply(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_multiply_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->multiply_(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_multiply_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::multiply_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_multiply_scalar(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::multiply(*tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_multiply_scalar_(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->multiply_(*other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_mv(gc_tensor self, gc_tensor vec) {
+  PROTECT(
+    torch::Tensor results__ = torch::mv(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(vec));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_mv_out(gc_tensor out, gc_tensor self, gc_tensor vec) {
+  PROTECT(
+    torch::Tensor results__ = torch::mv_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(vec));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_mvlgamma(gc_tensor self, int64_t p) {
+  PROTECT(
+    torch::Tensor results__ = torch::mvlgamma(*tensor_ptr_from_ocaml(self), p);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_mvlgamma_(gc_tensor self, int64_t p) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->mvlgamma_(p);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_mvlgamma_out(gc_tensor out, gc_tensor self, int64_t p) {
+  PROTECT(
+    torch::Tensor results__ = torch::mvlgamma_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), p);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_nan_to_num(gc_tensor self, double nan_v, int nan_null, double posinf_v, int posinf_null, double neginf_v, int neginf_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::nan_to_num(*tensor_ptr_from_ocaml(self), nan_null ? c10::nullopt : c10::optional<double>(nan_v), posinf_null ? c10::nullopt : c10::optional<double>(posinf_v), neginf_null ? c10::nullopt : c10::optional<double>(neginf_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_nan_to_num_(gc_tensor self, double nan_v, int nan_null, double posinf_v, int posinf_null, double neginf_v, int neginf_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::nan_to_num_(*tensor_ptr_from_ocaml(self), nan_null ? c10::nullopt : c10::optional<double>(nan_v), posinf_null ? c10::nullopt : c10::optional<double>(posinf_v), neginf_null ? c10::nullopt : c10::optional<double>(neginf_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_nan_to_num_out(gc_tensor out, gc_tensor self, double nan_v, int nan_null, double posinf_v, int posinf_null, double neginf_v, int neginf_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::nan_to_num_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), nan_null ? c10::nullopt : c10::optional<double>(nan_v), posinf_null ? c10::nullopt : c10::optional<double>(posinf_v), neginf_null ? c10::nullopt : c10::optional<double>(neginf_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_nanmean(gc_tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::nanmean(*tensor_ptr_from_ocaml(self), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), (bool)keepdim, torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_nanmean_out(gc_tensor out, gc_tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::nanmean_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), (bool)keepdim, torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_nanmedian(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::nanmedian(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_nanmedian_dim(raw_tensor *out__, gc_tensor self, int64_t dim, int keepdim) {
+  PROTECT(
+    auto results__ = torch::nanmedian(*tensor_ptr_from_ocaml(self), dim, (bool)keepdim);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_nanmedian_dim_values(raw_tensor *out__, gc_tensor values, gc_tensor indices, gc_tensor self, int64_t dim, int keepdim) {
+  PROTECT(
+    auto results__ = torch::nanmedian_out(*tensor_ptr_from_ocaml(values), *tensor_ptr_from_ocaml(indices), *tensor_ptr_from_ocaml(self), dim, (bool)keepdim);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_nanmedian_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::nanmedian_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_nanquantile(gc_tensor self, gc_tensor q, int64_t dim_v, int dim_null, int keepdim, char * interpolation) {
+  PROTECT(
+    torch::Tensor results__ = torch::nanquantile(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(q), dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v), (bool)keepdim, std::string(interpolation));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_nanquantile_out(gc_tensor out, gc_tensor self, gc_tensor q, int64_t dim_v, int dim_null, int keepdim, char * interpolation) {
+  PROTECT(
+    torch::Tensor results__ = torch::nanquantile_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(q), dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v), (bool)keepdim, std::string(interpolation));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_nanquantile_scalar(gc_tensor self, double q, int64_t dim_v, int dim_null, int keepdim, char * interpolation) {
+  PROTECT(
+    torch::Tensor results__ = torch::nanquantile(*tensor_ptr_from_ocaml(self), q, dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v), (bool)keepdim, std::string(interpolation));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_nanquantile_scalar_out(gc_tensor out, gc_tensor self, double q, int64_t dim_v, int dim_null, int keepdim, char * interpolation) {
+  PROTECT(
+    torch::Tensor results__ = torch::nanquantile_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), q, dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v), (bool)keepdim, std::string(interpolation));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_nansum(gc_tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::nansum(*tensor_ptr_from_ocaml(self), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), (bool)keepdim, torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_nansum_out(gc_tensor out, gc_tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::nansum_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), (bool)keepdim, torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_narrow(gc_tensor self, int64_t dim, int64_t start, int64_t length) {
+  PROTECT(
+    torch::Tensor results__ = torch::narrow(*tensor_ptr_from_ocaml(self), dim, start, length);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_narrow_copy(gc_tensor self, int64_t dim, int64_t start, int64_t length) {
+  PROTECT(
+    torch::Tensor results__ = torch::narrow_copy(*tensor_ptr_from_ocaml(self), dim, start, length);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_narrow_copy_out(gc_tensor out, gc_tensor self, int64_t dim, int64_t start, int64_t length) {
+  PROTECT(
+    torch::Tensor results__ = torch::narrow_copy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dim, start, length);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_narrow_tensor(gc_tensor self, int64_t dim, gc_tensor start, int64_t length) {
+  PROTECT(
+    torch::Tensor results__ = torch::narrow(*tensor_ptr_from_ocaml(self), dim, *tensor_ptr_from_ocaml(start), length);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_native_batch_norm(raw_tensor *out__, gc_tensor input, gc_tensor weight, gc_tensor bias, gc_tensor running_mean, gc_tensor running_var, int training, double momentum, double eps) {
+  PROTECT(
+    auto results__ = torch::native_batch_norm(*tensor_ptr_from_ocaml(input), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), (running_mean ? tensor_from_ocaml(running_mean) : torch::Tensor()), (running_var ? tensor_from_ocaml(running_var) : torch::Tensor()), (bool)training, momentum, eps);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+void atg_native_batch_norm_out(raw_tensor *out__, gc_tensor out, gc_tensor save_mean, gc_tensor save_invstd, gc_tensor input, gc_tensor weight, gc_tensor bias, gc_tensor running_mean, gc_tensor running_var, int training, double momentum, double eps) {
+  PROTECT(
+    auto results__ = torch::native_batch_norm_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(save_mean), *tensor_ptr_from_ocaml(save_invstd), *tensor_ptr_from_ocaml(input), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), (running_mean ? tensor_from_ocaml(running_mean) : torch::Tensor()), (running_var ? tensor_from_ocaml(running_var) : torch::Tensor()), (bool)training, momentum, eps);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+raw_tensor atg_native_channel_shuffle(gc_tensor self, int64_t groups) {
+  PROTECT(
+    torch::Tensor results__ = torch::native_channel_shuffle(*tensor_ptr_from_ocaml(self), groups);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_native_dropout(raw_tensor *out__, gc_tensor input, double p, int train) {
+  PROTECT(
+    auto results__ = torch::native_dropout(*tensor_ptr_from_ocaml(input), p, (bool)train);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_native_dropout_backward(gc_tensor grad_output, gc_tensor mask, double scale) {
+  PROTECT(
+    torch::Tensor results__ = torch::native_dropout_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(mask), scale);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_native_dropout_backward_out(gc_tensor out, gc_tensor grad_output, gc_tensor mask, double scale) {
+  PROTECT(
+    torch::Tensor results__ = torch::native_dropout_backward_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(mask), scale);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_native_dropout_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor input, double p, int train) {
+  PROTECT(
+    auto results__ = torch::native_dropout_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(input), p, (bool)train);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_native_group_norm(raw_tensor *out__, gc_tensor input, gc_tensor weight, gc_tensor bias, int64_t n, int64_t C, int64_t HxW, int64_t group, double eps) {
+  PROTECT(
+    auto results__ = torch::native_group_norm(*tensor_ptr_from_ocaml(input), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), n, C, HxW, group, eps);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+void atg_native_group_norm_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor input, gc_tensor weight, gc_tensor bias, int64_t n, int64_t C, int64_t HxW, int64_t group, double eps) {
+  PROTECT(
+    auto results__ = torch::native_group_norm_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(out2), *tensor_ptr_from_ocaml(input), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), n, C, HxW, group, eps);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+void atg_native_layer_norm(raw_tensor *out__, gc_tensor input, int64_t *normalized_shape_data, int normalized_shape_len, gc_tensor weight, gc_tensor bias, double eps) {
+  PROTECT(
+    auto results__ = torch::native_layer_norm(*tensor_ptr_from_ocaml(input), torch::IntArrayRef(normalized_shape_data, normalized_shape_len), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), eps);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+void atg_native_layer_norm_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor input, int64_t *normalized_shape_data, int normalized_shape_len, gc_tensor weight, gc_tensor bias, double eps) {
+  PROTECT(
+    auto results__ = torch::native_layer_norm_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(out2), *tensor_ptr_from_ocaml(input), torch::IntArrayRef(normalized_shape_data, normalized_shape_len), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), eps);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+raw_tensor atg_native_norm(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::native_norm(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_native_norm_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::native_norm_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_native_norm_scalaropt_dim_dtype(gc_tensor self, scalar p, int64_t *dim_data, int dim_len, int keepdim, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::native_norm(*tensor_ptr_from_ocaml(self), *p, torch::IntArrayRef(dim_data, dim_len), (bool)keepdim, torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_native_norm_scalaropt_dim_dtype_out(gc_tensor out, gc_tensor self, scalar p, int64_t *dim_data, int dim_len, int keepdim, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::native_norm_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *p, torch::IntArrayRef(dim_data, dim_len), (bool)keepdim, torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_ne(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::ne(*tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_ne_(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->ne_(*other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_ne_scalar_out(gc_tensor out, gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::ne_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_ne_tensor(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::ne(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_ne_tensor_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->ne_(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_ne_tensor_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::ne_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_neg(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::neg(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_neg_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::neg_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_neg_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::neg_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_negative(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::negative(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_negative_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::negative_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_negative_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::negative_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_nested_to_padded_tensor(gc_tensor self, double padding, int64_t *output_size_data, int output_size_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::nested_to_padded_tensor(*tensor_ptr_from_ocaml(self), padding, output_size_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(output_size_data, output_size_len)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_new_empty(gc_tensor self, int64_t *size_data, int size_len, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->new_empty(torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_new_empty_out(gc_tensor out, gc_tensor self, int64_t *size_data, int size_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::new_empty_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(size_data, size_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_new_empty_strided(gc_tensor self, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->new_empty_strided(torch::IntArrayRef(size_data, size_len), torch::IntArrayRef(stride_data, stride_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_new_empty_strided_out(gc_tensor out, gc_tensor self, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::new_empty_strided_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(size_data, size_len), torch::IntArrayRef(stride_data, stride_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_new_full(gc_tensor self, int64_t *size_data, int size_len, scalar fill_value, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->new_full(torch::IntArrayRef(size_data, size_len), *fill_value, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_new_full_out(gc_tensor out, gc_tensor self, int64_t *size_data, int size_len, scalar fill_value) {
+  PROTECT(
+    torch::Tensor results__ = torch::new_full_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(size_data, size_len), *fill_value);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_new_ones(gc_tensor self, int64_t *size_data, int size_len, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->new_ones(torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_new_ones_out(gc_tensor out, gc_tensor self, int64_t *size_data, int size_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::new_ones_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(size_data, size_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_new_zeros(gc_tensor self, int64_t *size_data, int size_len, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->new_zeros(torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_new_zeros_out(gc_tensor out, gc_tensor self, int64_t *size_data, int size_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::new_zeros_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(size_data, size_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_nextafter(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::nextafter(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_nextafter_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->nextafter_(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_nextafter_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::nextafter_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_nll_loss(gc_tensor self, gc_tensor target, gc_tensor weight, int64_t reduction, int64_t ignore_index) {
+  PROTECT(
+    torch::Tensor results__ = torch::nll_loss(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(target), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), reduction, ignore_index);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_nll_loss2d(gc_tensor self, gc_tensor target, gc_tensor weight, int64_t reduction, int64_t ignore_index) {
+  PROTECT(
+    torch::Tensor results__ = torch::nll_loss2d(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(target), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), reduction, ignore_index);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_nll_loss2d_backward(gc_tensor grad_output, gc_tensor self, gc_tensor target, gc_tensor weight, int64_t reduction, int64_t ignore_index, gc_tensor total_weight) {
+  PROTECT(
+    torch::Tensor results__ = torch::nll_loss2d_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(target), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), reduction, ignore_index, *tensor_ptr_from_ocaml(total_weight));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_nll_loss2d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, gc_tensor target, gc_tensor weight, int64_t reduction, int64_t ignore_index, gc_tensor total_weight) {
+  PROTECT(
+    torch::Tensor results__ = torch::nll_loss2d_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(target), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), reduction, ignore_index, *tensor_ptr_from_ocaml(total_weight));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_nll_loss2d_out(gc_tensor out, gc_tensor self, gc_tensor target, gc_tensor weight, int64_t reduction, int64_t ignore_index) {
+  PROTECT(
+    torch::Tensor results__ = torch::nll_loss2d_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(target), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), reduction, ignore_index);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_nll_loss_backward(gc_tensor grad_output, gc_tensor self, gc_tensor target, gc_tensor weight, int64_t reduction, int64_t ignore_index, gc_tensor total_weight) {
+  PROTECT(
+    torch::Tensor results__ = torch::nll_loss_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(target), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), reduction, ignore_index, *tensor_ptr_from_ocaml(total_weight));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_nll_loss_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, gc_tensor target, gc_tensor weight, int64_t reduction, int64_t ignore_index, gc_tensor total_weight) {
+  PROTECT(
+    torch::Tensor results__ = torch::nll_loss_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(target), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), reduction, ignore_index, *tensor_ptr_from_ocaml(total_weight));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_nll_loss_nd(gc_tensor self, gc_tensor target, gc_tensor weight, int64_t reduction, int64_t ignore_index) {
+  PROTECT(
+    torch::Tensor results__ = torch::nll_loss_nd(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(target), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), reduction, ignore_index);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_nll_loss_out(gc_tensor out, gc_tensor self, gc_tensor target, gc_tensor weight, int64_t reduction, int64_t ignore_index) {
+  PROTECT(
+    torch::Tensor results__ = torch::nll_loss_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(target), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), reduction, ignore_index);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_nonzero(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::nonzero(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor *atg_nonzero_numpy(gc_tensor self) {
+  PROTECT(
+    auto results__ = torch::nonzero_numpy(*tensor_ptr_from_ocaml(self));
+    int sz = results__.size();
+    raw_tensor *out__ = (raw_tensor*)malloc((sz + 1) * sizeof(raw_tensor));
+    for (int i = 0; i < sz; ++i)
+      out__[i] = tensor_to_ocaml(results__[i]);
+    out__[sz] = nullptr;
+    return out__;
+  )
+}
+
+raw_tensor atg_nonzero_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::nonzero_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_nonzero_static(gc_tensor self, int64_t size, int64_t fill_value) {
+  PROTECT(
+    torch::Tensor results__ = torch::nonzero_static(*tensor_ptr_from_ocaml(self), size, fill_value);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_nonzero_static_out(gc_tensor out, gc_tensor self, int64_t size, int64_t fill_value) {
+  PROTECT(
+    torch::Tensor results__ = torch::nonzero_static_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), size, fill_value);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_norm(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::norm(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_norm_dtype_out(gc_tensor out, gc_tensor self, scalar p, int64_t *dim_data, int dim_len, int keepdim, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::norm_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *p, torch::IntArrayRef(dim_data, dim_len), (bool)keepdim, torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_norm_except_dim(gc_tensor v, int64_t pow, int64_t dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::norm_except_dim(*tensor_ptr_from_ocaml(v), pow, dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_norm_out(gc_tensor out, gc_tensor self, scalar p, int64_t *dim_data, int dim_len, int keepdim) {
+  PROTECT(
+    torch::Tensor results__ = torch::norm_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *p, torch::IntArrayRef(dim_data, dim_len), (bool)keepdim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_norm_scalar_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::norm_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_norm_scalaropt_dim(gc_tensor self, scalar p, int64_t *dim_data, int dim_len, int keepdim) {
+  PROTECT(
+    torch::Tensor results__ = torch::norm(*tensor_ptr_from_ocaml(self), *p, torch::IntArrayRef(dim_data, dim_len), (bool)keepdim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_norm_scalaropt_dim_dtype(gc_tensor self, scalar p, int64_t *dim_data, int dim_len, int keepdim, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::norm(*tensor_ptr_from_ocaml(self), *p, torch::IntArrayRef(dim_data, dim_len), (bool)keepdim, torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_norm_scalaropt_dtype(gc_tensor self, scalar p, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::norm(*tensor_ptr_from_ocaml(self), *p, torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_norm_scalaropt_dtype_out(gc_tensor out, gc_tensor self, scalar p, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::norm_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *p, torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_normal_(gc_tensor self, double mean, double std) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->normal_(mean, std);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_normal_functional(gc_tensor self, double mean, double std) {
+  PROTECT(
+    torch::Tensor results__ = torch::normal_functional(*tensor_ptr_from_ocaml(self), mean, std);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_not_equal(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::not_equal(*tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_not_equal_(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->not_equal_(*other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_not_equal_scalar_out(gc_tensor out, gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::not_equal_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_not_equal_tensor(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::not_equal(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_not_equal_tensor_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->not_equal_(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_not_equal_tensor_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::not_equal_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_nuclear_norm(gc_tensor self, int keepdim) {
+  PROTECT(
+    torch::Tensor results__ = torch::nuclear_norm(*tensor_ptr_from_ocaml(self), (bool)keepdim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_nuclear_norm_dim(gc_tensor self, int64_t *dim_data, int dim_len, int keepdim) {
+  PROTECT(
+    torch::Tensor results__ = torch::nuclear_norm(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(dim_data, dim_len), (bool)keepdim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_nuclear_norm_dim_out(gc_tensor out, gc_tensor self, int64_t *dim_data, int dim_len, int keepdim) {
+  PROTECT(
+    torch::Tensor results__ = torch::nuclear_norm_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(dim_data, dim_len), (bool)keepdim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_nuclear_norm_out(gc_tensor out, gc_tensor self, int keepdim) {
+  PROTECT(
+    torch::Tensor results__ = torch::nuclear_norm_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), (bool)keepdim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_numpy_t(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->numpy_T();
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_one_hot(gc_tensor self, int64_t num_classes) {
+  PROTECT(
+    torch::Tensor results__ = torch::one_hot(*tensor_ptr_from_ocaml(self), num_classes);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_ones(int64_t *size_data, int size_len, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::ones(torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_ones_like(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::ones_like(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_ones_like_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::ones_like_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_ones_out(gc_tensor out, int64_t *size_data, int size_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::ones_out(*tensor_ptr_from_ocaml(out), torch::IntArrayRef(size_data, size_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_orgqr(gc_tensor self, gc_tensor input2) {
+  PROTECT(
+    torch::Tensor results__ = torch::orgqr(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(input2));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_orgqr_out(gc_tensor out, gc_tensor self, gc_tensor input2) {
+  PROTECT(
+    torch::Tensor results__ = torch::orgqr_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(input2));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_ormqr(gc_tensor self, gc_tensor input2, gc_tensor input3, int left, int transpose) {
+  PROTECT(
+    torch::Tensor results__ = torch::ormqr(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(input2), *tensor_ptr_from_ocaml(input3), (bool)left, (bool)transpose);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_ormqr_out(gc_tensor out, gc_tensor self, gc_tensor input2, gc_tensor input3, int left, int transpose) {
+  PROTECT(
+    torch::Tensor results__ = torch::ormqr_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(input2), *tensor_ptr_from_ocaml(input3), (bool)left, (bool)transpose);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_outer(gc_tensor self, gc_tensor vec2) {
+  PROTECT(
+    torch::Tensor results__ = torch::outer(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(vec2));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_outer_out(gc_tensor out, gc_tensor self, gc_tensor vec2) {
+  PROTECT(
+    torch::Tensor results__ = torch::outer_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(vec2));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+int64_t atg_output_nr(gc_tensor self) {
+  PROTECT(
+    return tensor_ptr_from_ocaml(self)->output_nr();
+  )
+  return 0;
+}
+
+raw_tensor atg_pad(gc_tensor self, int64_t *pad_data, int pad_len, char * mode, double value_v, int value_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::pad(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(pad_data, pad_len), std::string(mode), value_null ? c10::nullopt : c10::optional<double>(value_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_pad_sequence(gc_tensor *sequences_data, int sequences_len, int batch_first, double padding_value) {
+  PROTECT(
+    torch::Tensor results__ = torch::pad_sequence(of_carray_tensor(sequences_data, sequences_len), (bool)batch_first, padding_value);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_pairwise_distance(gc_tensor x1, gc_tensor x2, double p, double eps, int keepdim) {
+  PROTECT(
+    torch::Tensor results__ = torch::pairwise_distance(*tensor_ptr_from_ocaml(x1), *tensor_ptr_from_ocaml(x2), p, eps, (bool)keepdim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_pdist(gc_tensor self, double p) {
+  PROTECT(
+    torch::Tensor results__ = torch::pdist(*tensor_ptr_from_ocaml(self), p);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_permute(gc_tensor self, int64_t *dims_data, int dims_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::permute(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(dims_data, dims_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_permute_copy(gc_tensor self, int64_t *dims_data, int dims_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::permute_copy(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(dims_data, dims_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_permute_copy_out(gc_tensor out, gc_tensor self, int64_t *dims_data, int dims_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::permute_copy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(dims_data, dims_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_pin_memory(gc_tensor self, int device) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->pin_memory(device_of_int(device));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_pinverse(gc_tensor self, double rcond) {
+  PROTECT(
+    torch::Tensor results__ = torch::pinverse(*tensor_ptr_from_ocaml(self), rcond);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_pixel_shuffle(gc_tensor self, int64_t upscale_factor) {
+  PROTECT(
+    torch::Tensor results__ = torch::pixel_shuffle(*tensor_ptr_from_ocaml(self), upscale_factor);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_pixel_shuffle_out(gc_tensor out, gc_tensor self, int64_t upscale_factor) {
+  PROTECT(
+    torch::Tensor results__ = torch::pixel_shuffle_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), upscale_factor);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_pixel_unshuffle(gc_tensor self, int64_t downscale_factor) {
+  PROTECT(
+    torch::Tensor results__ = torch::pixel_unshuffle(*tensor_ptr_from_ocaml(self), downscale_factor);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_pixel_unshuffle_out(gc_tensor out, gc_tensor self, int64_t downscale_factor) {
+  PROTECT(
+    torch::Tensor results__ = torch::pixel_unshuffle_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), downscale_factor);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_poisson(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::poisson(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_poisson_nll_loss(gc_tensor input, gc_tensor target, int log_input, int full, double eps, int64_t reduction) {
+  PROTECT(
+    torch::Tensor results__ = torch::poisson_nll_loss(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(target), (bool)log_input, (bool)full, eps, reduction);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_poisson_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::poisson_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_polar(gc_tensor abs, gc_tensor angle) {
+  PROTECT(
+    torch::Tensor results__ = torch::polar(*tensor_ptr_from_ocaml(abs), *tensor_ptr_from_ocaml(angle));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_polar_out(gc_tensor out, gc_tensor abs, gc_tensor angle) {
+  PROTECT(
+    torch::Tensor results__ = torch::polar_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(abs), *tensor_ptr_from_ocaml(angle));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_polygamma(int64_t n, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::polygamma(n, *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_polygamma_(gc_tensor self, int64_t n) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->polygamma_(n);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_polygamma_out(gc_tensor out, int64_t n, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::polygamma_out(*tensor_ptr_from_ocaml(out), n, *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_positive(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::positive(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_pow(gc_tensor self, gc_tensor exponent) {
+  PROTECT(
+    torch::Tensor results__ = torch::pow(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(exponent));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_pow_(gc_tensor self, scalar exponent) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->pow_(*exponent);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_pow_scalar(scalar self, gc_tensor exponent) {
+  PROTECT(
+    torch::Tensor results__ = torch::pow(*self, *tensor_ptr_from_ocaml(exponent));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_pow_scalar_out(gc_tensor out, scalar self, gc_tensor exponent) {
+  PROTECT(
+    torch::Tensor results__ = torch::pow_out(*tensor_ptr_from_ocaml(out), *self, *tensor_ptr_from_ocaml(exponent));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_pow_tensor_(gc_tensor self, gc_tensor exponent) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->pow_(*tensor_ptr_from_ocaml(exponent));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_pow_tensor_scalar(gc_tensor self, scalar exponent) {
+  PROTECT(
+    torch::Tensor results__ = torch::pow(*tensor_ptr_from_ocaml(self), *exponent);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_pow_tensor_scalar_out(gc_tensor out, gc_tensor self, scalar exponent) {
+  PROTECT(
+    torch::Tensor results__ = torch::pow_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *exponent);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_pow_tensor_tensor_out(gc_tensor out, gc_tensor self, gc_tensor exponent) {
+  PROTECT(
+    torch::Tensor results__ = torch::pow_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(exponent));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_prelu(gc_tensor self, gc_tensor weight) {
+  PROTECT(
+    torch::Tensor results__ = torch::prelu(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(weight));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_prod(gc_tensor self, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::prod(*tensor_ptr_from_ocaml(self), torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_prod_dim_int(gc_tensor self, int64_t dim, int keepdim, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::prod(*tensor_ptr_from_ocaml(self), dim, (bool)keepdim, torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_prod_int_out(gc_tensor out, gc_tensor self, int64_t dim, int keepdim, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::prod_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dim, (bool)keepdim, torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_prod_out(gc_tensor out, gc_tensor self, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::prod_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_put(gc_tensor self, gc_tensor index, gc_tensor source, int accumulate) {
+  PROTECT(
+    torch::Tensor results__ = torch::put(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(index), *tensor_ptr_from_ocaml(source), (bool)accumulate);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_put_(gc_tensor self, gc_tensor index, gc_tensor source, int accumulate) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->put_(*tensor_ptr_from_ocaml(index), *tensor_ptr_from_ocaml(source), (bool)accumulate);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_put_out(gc_tensor out, gc_tensor self, gc_tensor index, gc_tensor source, int accumulate) {
+  PROTECT(
+    torch::Tensor results__ = torch::put_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(index), *tensor_ptr_from_ocaml(source), (bool)accumulate);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+int64_t atg_q_per_channel_axis(gc_tensor self) {
+  PROTECT(
+    return torch::q_per_channel_axis(*tensor_ptr_from_ocaml(self));
+  )
+  return 0;
+}
+
+raw_tensor atg_q_per_channel_scales(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::q_per_channel_scales(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_q_per_channel_scales_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::q_per_channel_scales_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_q_per_channel_zero_points(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::q_per_channel_zero_points(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_q_per_channel_zero_points_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::q_per_channel_zero_points_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+double atg_q_scale(gc_tensor self) {
+  PROTECT(
+    return torch::q_scale(*tensor_ptr_from_ocaml(self));
+  )
+  return 0;
+}
+
+int64_t atg_q_zero_point(gc_tensor self) {
+  PROTECT(
+    return torch::q_zero_point(*tensor_ptr_from_ocaml(self));
+  )
+  return 0;
+}
+
+void atg_qr(raw_tensor *out__, gc_tensor self, int some) {
+  PROTECT(
+    auto results__ = torch::qr(*tensor_ptr_from_ocaml(self), (bool)some);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_qr_q(raw_tensor *out__, gc_tensor Q, gc_tensor R, gc_tensor self, int some) {
+  PROTECT(
+    auto results__ = torch::qr_out(*tensor_ptr_from_ocaml(Q), *tensor_ptr_from_ocaml(R), *tensor_ptr_from_ocaml(self), (bool)some);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_quantile(gc_tensor self, gc_tensor q, int64_t dim_v, int dim_null, int keepdim, char * interpolation) {
+  PROTECT(
+    torch::Tensor results__ = torch::quantile(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(q), dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v), (bool)keepdim, std::string(interpolation));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_quantile_out(gc_tensor out, gc_tensor self, gc_tensor q, int64_t dim_v, int dim_null, int keepdim, char * interpolation) {
+  PROTECT(
+    torch::Tensor results__ = torch::quantile_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(q), dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v), (bool)keepdim, std::string(interpolation));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_quantile_scalar(gc_tensor self, double q, int64_t dim_v, int dim_null, int keepdim, char * interpolation) {
+  PROTECT(
+    torch::Tensor results__ = torch::quantile(*tensor_ptr_from_ocaml(self), q, dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v), (bool)keepdim, std::string(interpolation));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_quantile_scalar_out(gc_tensor out, gc_tensor self, double q, int64_t dim_v, int dim_null, int keepdim, char * interpolation) {
+  PROTECT(
+    torch::Tensor results__ = torch::quantile_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), q, dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v), (bool)keepdim, std::string(interpolation));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_quantize_per_channel(gc_tensor self, gc_tensor scales, gc_tensor zero_points, int64_t axis, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::quantize_per_channel(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(scales), *tensor_ptr_from_ocaml(zero_points), axis, torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_quantize_per_channel_out(gc_tensor out, gc_tensor self, gc_tensor scales, gc_tensor zero_points, int64_t axis, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::quantize_per_channel_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(scales), *tensor_ptr_from_ocaml(zero_points), axis, torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_quantize_per_tensor(gc_tensor self, double scale, int64_t zero_point, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::quantize_per_tensor(*tensor_ptr_from_ocaml(self), scale, zero_point, torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_quantize_per_tensor_dynamic(gc_tensor self, int dtype, int reduce_range) {
+  PROTECT(
+    torch::Tensor results__ = torch::quantize_per_tensor_dynamic(*tensor_ptr_from_ocaml(self), torch::ScalarType(dtype), (bool)reduce_range);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_quantize_per_tensor_dynamic_out(gc_tensor out, gc_tensor self, int dtype, int reduce_range) {
+  PROTECT(
+    torch::Tensor results__ = torch::quantize_per_tensor_dynamic_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::ScalarType(dtype), (bool)reduce_range);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_quantize_per_tensor_out(gc_tensor out, gc_tensor self, double scale, int64_t zero_point, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::quantize_per_tensor_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), scale, zero_point, torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_quantize_per_tensor_tensor_qparams(gc_tensor self, gc_tensor scale, gc_tensor zero_point, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::quantize_per_tensor(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(scale), *tensor_ptr_from_ocaml(zero_point), torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_quantize_per_tensor_tensor_qparams_out(gc_tensor out, gc_tensor self, gc_tensor scale, gc_tensor zero_point, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::quantize_per_tensor_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(scale), *tensor_ptr_from_ocaml(zero_point), torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor *atg_quantize_per_tensor_tensors(gc_tensor *tensors_data, int tensors_len, gc_tensor scales, gc_tensor zero_points, int dtype) {
+  PROTECT(
+    auto results__ = torch::quantize_per_tensor(of_carray_tensor(tensors_data, tensors_len), *tensor_ptr_from_ocaml(scales), *tensor_ptr_from_ocaml(zero_points), torch::ScalarType(dtype));
+    int sz = results__.size();
+    raw_tensor *out__ = (raw_tensor*)malloc((sz + 1) * sizeof(raw_tensor));
+    for (int i = 0; i < sz; ++i)
+      out__[i] = tensor_to_ocaml(results__[i]);
+    out__[sz] = nullptr;
+    return out__;
+  )
+}
+
+void atg_quantize_per_tensor_tensors_out(gc_tensor *out_data, int out_len, gc_tensor *tensors_data, int tensors_len, gc_tensor scales, gc_tensor zero_points, int dtype) {
+  PROTECT(
+    torch::quantize_per_tensor_out(of_carray_tensor(out_data, out_len), of_carray_tensor(tensors_data, tensors_len), *tensor_ptr_from_ocaml(scales), *tensor_ptr_from_ocaml(zero_points), torch::ScalarType(dtype));
+  )
+}
+
+raw_tensor atg_quantized_batch_norm(gc_tensor input, gc_tensor weight, gc_tensor bias, gc_tensor mean, gc_tensor var, double eps, double output_scale, int64_t output_zero_point) {
+  PROTECT(
+    torch::Tensor results__ = torch::quantized_batch_norm(*tensor_ptr_from_ocaml(input), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), *tensor_ptr_from_ocaml(mean), *tensor_ptr_from_ocaml(var), eps, output_scale, output_zero_point);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_quantized_batch_norm_out(gc_tensor out, gc_tensor input, gc_tensor weight, gc_tensor bias, gc_tensor mean, gc_tensor var, double eps, double output_scale, int64_t output_zero_point) {
+  PROTECT(
+    torch::Tensor results__ = torch::quantized_batch_norm_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(input), (weight ? tensor_from_ocaml(weight) : torch::Tensor()), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), *tensor_ptr_from_ocaml(mean), *tensor_ptr_from_ocaml(var), eps, output_scale, output_zero_point);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_quantized_gru_cell(gc_tensor input, gc_tensor hx, gc_tensor w_ih, gc_tensor w_hh, gc_tensor b_ih, gc_tensor b_hh, gc_tensor packed_ih, gc_tensor packed_hh, gc_tensor col_offsets_ih, gc_tensor col_offsets_hh, scalar scale_ih, scalar scale_hh, scalar zero_point_ih, scalar zero_point_hh) {
+  PROTECT(
+    torch::Tensor results__ = torch::quantized_gru_cell(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(hx), *tensor_ptr_from_ocaml(w_ih), *tensor_ptr_from_ocaml(w_hh), *tensor_ptr_from_ocaml(b_ih), *tensor_ptr_from_ocaml(b_hh), *tensor_ptr_from_ocaml(packed_ih), *tensor_ptr_from_ocaml(packed_hh), *tensor_ptr_from_ocaml(col_offsets_ih), *tensor_ptr_from_ocaml(col_offsets_hh), *scale_ih, *scale_hh, *zero_point_ih, *zero_point_hh);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_quantized_lstm_cell(raw_tensor *out__, gc_tensor input, gc_tensor *hx_data, int hx_len, gc_tensor w_ih, gc_tensor w_hh, gc_tensor b_ih, gc_tensor b_hh, gc_tensor packed_ih, gc_tensor packed_hh, gc_tensor col_offsets_ih, gc_tensor col_offsets_hh, scalar scale_ih, scalar scale_hh, scalar zero_point_ih, scalar zero_point_hh) {
+  PROTECT(
+    auto results__ = torch::quantized_lstm_cell(*tensor_ptr_from_ocaml(input), of_carray_tensor(hx_data, hx_len), *tensor_ptr_from_ocaml(w_ih), *tensor_ptr_from_ocaml(w_hh), *tensor_ptr_from_ocaml(b_ih), *tensor_ptr_from_ocaml(b_hh), *tensor_ptr_from_ocaml(packed_ih), *tensor_ptr_from_ocaml(packed_hh), *tensor_ptr_from_ocaml(col_offsets_ih), *tensor_ptr_from_ocaml(col_offsets_hh), *scale_ih, *scale_hh, *zero_point_ih, *zero_point_hh);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_quantized_max_pool1d(gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
+  PROTECT(
+    torch::Tensor results__ = torch::quantized_max_pool1d(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_quantized_max_pool1d_out(gc_tensor out, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
+  PROTECT(
+    torch::Tensor results__ = torch::quantized_max_pool1d_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_quantized_max_pool2d(gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
+  PROTECT(
+    torch::Tensor results__ = torch::quantized_max_pool2d(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_quantized_max_pool2d_out(gc_tensor out, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
+  PROTECT(
+    torch::Tensor results__ = torch::quantized_max_pool2d_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_quantized_max_pool3d(gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
+  PROTECT(
+    torch::Tensor results__ = torch::quantized_max_pool3d(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_quantized_max_pool3d_out(gc_tensor out, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
+  PROTECT(
+    torch::Tensor results__ = torch::quantized_max_pool3d_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_quantized_rnn_relu_cell(gc_tensor input, gc_tensor hx, gc_tensor w_ih, gc_tensor w_hh, gc_tensor b_ih, gc_tensor b_hh, gc_tensor packed_ih, gc_tensor packed_hh, gc_tensor col_offsets_ih, gc_tensor col_offsets_hh, scalar scale_ih, scalar scale_hh, scalar zero_point_ih, scalar zero_point_hh) {
+  PROTECT(
+    torch::Tensor results__ = torch::quantized_rnn_relu_cell(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(hx), *tensor_ptr_from_ocaml(w_ih), *tensor_ptr_from_ocaml(w_hh), *tensor_ptr_from_ocaml(b_ih), *tensor_ptr_from_ocaml(b_hh), *tensor_ptr_from_ocaml(packed_ih), *tensor_ptr_from_ocaml(packed_hh), *tensor_ptr_from_ocaml(col_offsets_ih), *tensor_ptr_from_ocaml(col_offsets_hh), *scale_ih, *scale_hh, *zero_point_ih, *zero_point_hh);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_quantized_rnn_tanh_cell(gc_tensor input, gc_tensor hx, gc_tensor w_ih, gc_tensor w_hh, gc_tensor b_ih, gc_tensor b_hh, gc_tensor packed_ih, gc_tensor packed_hh, gc_tensor col_offsets_ih, gc_tensor col_offsets_hh, scalar scale_ih, scalar scale_hh, scalar zero_point_ih, scalar zero_point_hh) {
+  PROTECT(
+    torch::Tensor results__ = torch::quantized_rnn_tanh_cell(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(hx), *tensor_ptr_from_ocaml(w_ih), *tensor_ptr_from_ocaml(w_hh), *tensor_ptr_from_ocaml(b_ih), *tensor_ptr_from_ocaml(b_hh), *tensor_ptr_from_ocaml(packed_ih), *tensor_ptr_from_ocaml(packed_hh), *tensor_ptr_from_ocaml(col_offsets_ih), *tensor_ptr_from_ocaml(col_offsets_hh), *scale_ih, *scale_hh, *zero_point_ih, *zero_point_hh);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_rad2deg(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::rad2deg(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_rad2deg_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::rad2deg_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_rad2deg_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::rad2deg_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_rand(int64_t *size_data, int size_len, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::rand(torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_rand_like(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::rand_like(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_rand_like_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::rand_like_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_rand_out(gc_tensor out, int64_t *size_data, int size_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::rand_out(*tensor_ptr_from_ocaml(out), torch::IntArrayRef(size_data, size_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_randint(int64_t high, int64_t *size_data, int size_len, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::randint(high, torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_randint_like(gc_tensor self, int64_t high) {
+  PROTECT(
+    torch::Tensor results__ = torch::randint_like(*tensor_ptr_from_ocaml(self), high);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_randint_like_low_dtype(gc_tensor self, int64_t low, int64_t high) {
+  PROTECT(
+    torch::Tensor results__ = torch::randint_like(*tensor_ptr_from_ocaml(self), low, high);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_randint_like_low_dtype_out(gc_tensor out, gc_tensor self, int64_t low, int64_t high) {
+  PROTECT(
+    torch::Tensor results__ = torch::randint_like_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), low, high);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_randint_like_out(gc_tensor out, gc_tensor self, int64_t high) {
+  PROTECT(
+    torch::Tensor results__ = torch::randint_like_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), high);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_randint_low(int64_t low, int64_t high, int64_t *size_data, int size_len, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::randint(low, high, torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_randint_low_out(gc_tensor out, int64_t low, int64_t high, int64_t *size_data, int size_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::randint_out(*tensor_ptr_from_ocaml(out), low, high, torch::IntArrayRef(size_data, size_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_randint_out(gc_tensor out, int64_t high, int64_t *size_data, int size_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::randint_out(*tensor_ptr_from_ocaml(out), high, torch::IntArrayRef(size_data, size_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_randn(int64_t *size_data, int size_len, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::randn(torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_randn_like(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::randn_like(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_randn_like_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::randn_like_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_randn_out(gc_tensor out, int64_t *size_data, int size_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::randn_out(*tensor_ptr_from_ocaml(out), torch::IntArrayRef(size_data, size_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_random(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::random(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_random_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->random_();
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_random_from(gc_tensor self, int64_t from, int64_t to_v, int to_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::random(*tensor_ptr_from_ocaml(self), from, to_null ? c10::nullopt : c10::optional<int64_t>(to_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_random_from_(gc_tensor self, int64_t from, int64_t to_v, int to_null) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->random_(from, to_null ? c10::nullopt : c10::optional<int64_t>(to_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_random_from_out(gc_tensor out, gc_tensor self, int64_t from, int64_t to_v, int to_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::random_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), from, to_null ? c10::nullopt : c10::optional<int64_t>(to_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_random_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::random_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_random_to(gc_tensor self, int64_t to) {
+  PROTECT(
+    torch::Tensor results__ = torch::random(*tensor_ptr_from_ocaml(self), to);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_random_to_(gc_tensor self, int64_t to) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->random_(to);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_random_to_out(gc_tensor out, gc_tensor self, int64_t to) {
+  PROTECT(
+    torch::Tensor results__ = torch::random_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), to);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_randperm(int64_t n, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::randperm(n, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_randperm_out(gc_tensor out, int64_t n) {
+  PROTECT(
+    torch::Tensor results__ = torch::randperm_out(*tensor_ptr_from_ocaml(out), n);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_range(scalar start, scalar end, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::range(*start, *end, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_range_out(gc_tensor out, scalar start, scalar end) {
+  PROTECT(
+    torch::Tensor results__ = torch::range_out(*tensor_ptr_from_ocaml(out), *start, *end);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_range_out_(gc_tensor out, scalar start, scalar end) {
+  PROTECT(
+    torch::Tensor results__ = torch::range_out(*tensor_ptr_from_ocaml(out), *start, *end);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_range_step(scalar start, scalar end, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::range(*start, *end, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_ravel(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::ravel(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_real(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::real(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_reciprocal(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::reciprocal(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_reciprocal_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::reciprocal_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_reciprocal_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::reciprocal_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_reflection_pad1d(gc_tensor self, int64_t *padding_data, int padding_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::reflection_pad1d(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(padding_data, padding_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_reflection_pad1d_backward(gc_tensor grad_output, gc_tensor self, int64_t *padding_data, int padding_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::reflection_pad1d_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(padding_data, padding_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_reflection_pad1d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, int64_t *padding_data, int padding_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::reflection_pad1d_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(padding_data, padding_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_reflection_pad1d_out(gc_tensor out, gc_tensor self, int64_t *padding_data, int padding_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::reflection_pad1d_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(padding_data, padding_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_reflection_pad2d(gc_tensor self, int64_t *padding_data, int padding_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::reflection_pad2d(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(padding_data, padding_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_reflection_pad2d_backward(gc_tensor grad_output, gc_tensor self, int64_t *padding_data, int padding_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::reflection_pad2d_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(padding_data, padding_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_reflection_pad2d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, int64_t *padding_data, int padding_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::reflection_pad2d_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(padding_data, padding_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_reflection_pad2d_out(gc_tensor out, gc_tensor self, int64_t *padding_data, int padding_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::reflection_pad2d_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(padding_data, padding_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_reflection_pad3d(gc_tensor self, int64_t *padding_data, int padding_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::reflection_pad3d(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(padding_data, padding_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_reflection_pad3d_backward(gc_tensor grad_output, gc_tensor self, int64_t *padding_data, int padding_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::reflection_pad3d_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(padding_data, padding_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_reflection_pad3d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, int64_t *padding_data, int padding_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::reflection_pad3d_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(padding_data, padding_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_reflection_pad3d_out(gc_tensor out, gc_tensor self, int64_t *padding_data, int padding_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::reflection_pad3d_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(padding_data, padding_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_relu(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::relu(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_relu6(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::relu6(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_relu6_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::relu6_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_relu_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::relu_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_relu_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::relu_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_remainder(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::remainder(*tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_remainder_(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->remainder_(*other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_remainder_scalar_out(gc_tensor out, gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::remainder_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_remainder_scalar_tensor(scalar self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::remainder(*self, *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_remainder_scalar_tensor_out(gc_tensor out, scalar self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::remainder_out(*tensor_ptr_from_ocaml(out), *self, *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_remainder_tensor(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::remainder(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_remainder_tensor_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->remainder_(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_remainder_tensor_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::remainder_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_renorm(gc_tensor self, scalar p, int64_t dim, scalar maxnorm) {
+  PROTECT(
+    torch::Tensor results__ = torch::renorm(*tensor_ptr_from_ocaml(self), *p, dim, *maxnorm);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_renorm_(gc_tensor self, scalar p, int64_t dim, scalar maxnorm) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->renorm_(*p, dim, *maxnorm);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_renorm_out(gc_tensor out, gc_tensor self, scalar p, int64_t dim, scalar maxnorm) {
+  PROTECT(
+    torch::Tensor results__ = torch::renorm_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *p, dim, *maxnorm);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_repeat(gc_tensor self, int64_t *repeats_data, int repeats_len) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->repeat(torch::IntArrayRef(repeats_data, repeats_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_repeat_interleave(gc_tensor repeats, int64_t output_size_v, int output_size_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::repeat_interleave(*tensor_ptr_from_ocaml(repeats), output_size_null ? c10::nullopt : c10::optional<int64_t>(output_size_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_repeat_interleave_self_int(gc_tensor self, int64_t repeats, int64_t dim_v, int dim_null, int64_t output_size_v, int output_size_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::repeat_interleave(*tensor_ptr_from_ocaml(self), repeats, dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v), output_size_null ? c10::nullopt : c10::optional<int64_t>(output_size_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_repeat_interleave_self_tensor(gc_tensor self, gc_tensor repeats, int64_t dim_v, int dim_null, int64_t output_size_v, int output_size_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::repeat_interleave(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(repeats), dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v), output_size_null ? c10::nullopt : c10::optional<int64_t>(output_size_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_repeat_interleave_tensor_out(gc_tensor out, gc_tensor repeats, int64_t output_size_v, int output_size_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::repeat_interleave_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(repeats), output_size_null ? c10::nullopt : c10::optional<int64_t>(output_size_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_repeat_out(gc_tensor out, gc_tensor self, int64_t *repeats_data, int repeats_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::repeat_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(repeats_data, repeats_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_replication_pad1d(gc_tensor self, int64_t *padding_data, int padding_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::replication_pad1d(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(padding_data, padding_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_replication_pad1d_backward(gc_tensor grad_output, gc_tensor self, int64_t *padding_data, int padding_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::replication_pad1d_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(padding_data, padding_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_replication_pad1d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, int64_t *padding_data, int padding_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::replication_pad1d_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(padding_data, padding_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_replication_pad1d_out(gc_tensor out, gc_tensor self, int64_t *padding_data, int padding_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::replication_pad1d_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(padding_data, padding_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_replication_pad2d(gc_tensor self, int64_t *padding_data, int padding_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::replication_pad2d(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(padding_data, padding_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_replication_pad2d_backward(gc_tensor grad_output, gc_tensor self, int64_t *padding_data, int padding_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::replication_pad2d_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(padding_data, padding_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_replication_pad2d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, int64_t *padding_data, int padding_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::replication_pad2d_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(padding_data, padding_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_replication_pad2d_out(gc_tensor out, gc_tensor self, int64_t *padding_data, int padding_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::replication_pad2d_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(padding_data, padding_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_replication_pad3d(gc_tensor self, int64_t *padding_data, int padding_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::replication_pad3d(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(padding_data, padding_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_replication_pad3d_backward(gc_tensor grad_output, gc_tensor self, int64_t *padding_data, int padding_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::replication_pad3d_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(padding_data, padding_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_replication_pad3d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, int64_t *padding_data, int padding_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::replication_pad3d_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(padding_data, padding_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_replication_pad3d_out(gc_tensor out, gc_tensor self, int64_t *padding_data, int padding_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::replication_pad3d_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(padding_data, padding_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_requires_grad_(gc_tensor self, int requires_grad) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->requires_grad_((bool)requires_grad);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_reshape(gc_tensor self, int64_t *shape_data, int shape_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::reshape(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(shape_data, shape_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_reshape_as(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->reshape_as(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_resize(gc_tensor self, int64_t *size_data, int size_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::resize(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(size_data, size_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_resize_(gc_tensor self, int64_t *size_data, int size_len) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->resize_(torch::IntArrayRef(size_data, size_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_resize_as(gc_tensor self, gc_tensor the_template) {
+  PROTECT(
+    torch::Tensor results__ = torch::resize_as(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(the_template));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_resize_as_(gc_tensor self, gc_tensor the_template) {
+  PROTECT(
+    torch::Tensor results__ = torch::resize_as_(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(the_template));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_resize_as_out(gc_tensor out, gc_tensor self, gc_tensor the_template) {
+  PROTECT(
+    torch::Tensor results__ = torch::resize_as_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(the_template));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_resize_as_sparse(gc_tensor self, gc_tensor the_template) {
+  PROTECT(
+    torch::Tensor results__ = torch::resize_as_sparse(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(the_template));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_resize_as_sparse_(gc_tensor self, gc_tensor the_template) {
+  PROTECT(
+    torch::Tensor results__ = torch::resize_as_sparse_(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(the_template));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_resize_as_sparse_out(gc_tensor out, gc_tensor self, gc_tensor the_template) {
+  PROTECT(
+    torch::Tensor results__ = torch::resize_as_sparse_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(the_template));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_resize_out(gc_tensor out, gc_tensor self, int64_t *size_data, int size_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::resize_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(size_data, size_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_resolve_conj(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::resolve_conj(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_resolve_neg(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::resolve_neg(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+int atg_retains_grad(gc_tensor self) {
+  PROTECT(
+    return tensor_ptr_from_ocaml(self)->retains_grad();
+  )
+  return 0;
+}
+
+void atg_rnn_relu(raw_tensor *out__, gc_tensor input, gc_tensor hx, gc_tensor *params_data, int params_len, int has_biases, int64_t num_layers, double dropout, int train, int bidirectional, int batch_first) {
+  PROTECT(
+    auto results__ = torch::rnn_relu(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(hx), of_carray_tensor(params_data, params_len), (bool)has_biases, num_layers, dropout, (bool)train, (bool)bidirectional, (bool)batch_first);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_rnn_relu_cell(gc_tensor input, gc_tensor hx, gc_tensor w_ih, gc_tensor w_hh, gc_tensor b_ih, gc_tensor b_hh) {
+  PROTECT(
+    torch::Tensor results__ = torch::rnn_relu_cell(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(hx), *tensor_ptr_from_ocaml(w_ih), *tensor_ptr_from_ocaml(w_hh), (b_ih ? tensor_from_ocaml(b_ih) : torch::Tensor()), (b_hh ? tensor_from_ocaml(b_hh) : torch::Tensor()));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_rnn_relu_data(raw_tensor *out__, gc_tensor data, gc_tensor batch_sizes, gc_tensor hx, gc_tensor *params_data, int params_len, int has_biases, int64_t num_layers, double dropout, int train, int bidirectional) {
+  PROTECT(
+    auto results__ = torch::rnn_relu(*tensor_ptr_from_ocaml(data), *tensor_ptr_from_ocaml(batch_sizes), *tensor_ptr_from_ocaml(hx), of_carray_tensor(params_data, params_len), (bool)has_biases, num_layers, dropout, (bool)train, (bool)bidirectional);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_rnn_tanh(raw_tensor *out__, gc_tensor input, gc_tensor hx, gc_tensor *params_data, int params_len, int has_biases, int64_t num_layers, double dropout, int train, int bidirectional, int batch_first) {
+  PROTECT(
+    auto results__ = torch::rnn_tanh(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(hx), of_carray_tensor(params_data, params_len), (bool)has_biases, num_layers, dropout, (bool)train, (bool)bidirectional, (bool)batch_first);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_rnn_tanh_cell(gc_tensor input, gc_tensor hx, gc_tensor w_ih, gc_tensor w_hh, gc_tensor b_ih, gc_tensor b_hh) {
+  PROTECT(
+    torch::Tensor results__ = torch::rnn_tanh_cell(*tensor_ptr_from_ocaml(input), *tensor_ptr_from_ocaml(hx), *tensor_ptr_from_ocaml(w_ih), *tensor_ptr_from_ocaml(w_hh), (b_ih ? tensor_from_ocaml(b_ih) : torch::Tensor()), (b_hh ? tensor_from_ocaml(b_hh) : torch::Tensor()));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_rnn_tanh_data(raw_tensor *out__, gc_tensor data, gc_tensor batch_sizes, gc_tensor hx, gc_tensor *params_data, int params_len, int has_biases, int64_t num_layers, double dropout, int train, int bidirectional) {
+  PROTECT(
+    auto results__ = torch::rnn_tanh(*tensor_ptr_from_ocaml(data), *tensor_ptr_from_ocaml(batch_sizes), *tensor_ptr_from_ocaml(hx), of_carray_tensor(params_data, params_len), (bool)has_biases, num_layers, dropout, (bool)train, (bool)bidirectional);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_roll(gc_tensor self, int64_t *shifts_data, int shifts_len, int64_t *dims_data, int dims_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::roll(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(shifts_data, shifts_len), torch::IntArrayRef(dims_data, dims_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_roll_out(gc_tensor out, gc_tensor self, int64_t *shifts_data, int shifts_len, int64_t *dims_data, int dims_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::roll_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(shifts_data, shifts_len), torch::IntArrayRef(dims_data, dims_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_rot90(gc_tensor self, int64_t k, int64_t *dims_data, int dims_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::rot90(*tensor_ptr_from_ocaml(self), k, torch::IntArrayRef(dims_data, dims_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_rot90_out(gc_tensor out, gc_tensor self, int64_t k, int64_t *dims_data, int dims_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::rot90_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), k, torch::IntArrayRef(dims_data, dims_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_round(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::round(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_round_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::round_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_round_decimals(gc_tensor self, int64_t decimals) {
+  PROTECT(
+    torch::Tensor results__ = torch::round(*tensor_ptr_from_ocaml(self), decimals);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_round_decimals_(gc_tensor self, int64_t decimals) {
+  PROTECT(
+    torch::Tensor results__ = torch::round_(*tensor_ptr_from_ocaml(self), decimals);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_round_decimals_out(gc_tensor out, gc_tensor self, int64_t decimals) {
+  PROTECT(
+    torch::Tensor results__ = torch::round_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), decimals);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_round_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::round_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_row_indices(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->row_indices();
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_row_indices_copy(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::row_indices_copy(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_row_indices_copy_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::row_indices_copy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_row_stack(gc_tensor *tensors_data, int tensors_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::row_stack(of_carray_tensor(tensors_data, tensors_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_row_stack_out(gc_tensor out, gc_tensor *tensors_data, int tensors_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::row_stack_out(*tensor_ptr_from_ocaml(out), of_carray_tensor(tensors_data, tensors_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_rrelu(gc_tensor self, int training) {
+  PROTECT(
+    torch::Tensor results__ = torch::rrelu(*tensor_ptr_from_ocaml(self), (bool)training);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_rrelu_(gc_tensor self, int training) {
+  PROTECT(
+    torch::Tensor results__ = torch::rrelu_(*tensor_ptr_from_ocaml(self), (bool)training);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_rrelu_with_noise(gc_tensor self, gc_tensor noise, int training) {
+  PROTECT(
+    torch::Tensor results__ = torch::rrelu_with_noise(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(noise), (bool)training);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_rrelu_with_noise_(gc_tensor self, gc_tensor noise, int training) {
+  PROTECT(
+    torch::Tensor results__ = torch::rrelu_with_noise_(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(noise), (bool)training);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_rrelu_with_noise_backward(gc_tensor grad_output, gc_tensor self, gc_tensor noise, scalar lower, scalar upper, int training, int self_is_result) {
+  PROTECT(
+    torch::Tensor results__ = torch::rrelu_with_noise_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(noise), *lower, *upper, (bool)training, (bool)self_is_result);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_rrelu_with_noise_backward_out(gc_tensor out, gc_tensor grad_output, gc_tensor self, gc_tensor noise, scalar lower, scalar upper, int training, int self_is_result) {
+  PROTECT(
+    torch::Tensor results__ = torch::rrelu_with_noise_backward_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(noise), *lower, *upper, (bool)training, (bool)self_is_result);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_rrelu_with_noise_out(gc_tensor out, gc_tensor self, gc_tensor noise, int training) {
+  PROTECT(
+    torch::Tensor results__ = torch::rrelu_with_noise_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(noise), (bool)training);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_rsqrt(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::rsqrt(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_rsqrt_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::rsqrt_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_rsqrt_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::rsqrt_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_rsub(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::rsub(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_rsub_scalar(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::rsub(*tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_rsub_scalar_out(gc_tensor out, gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::rsub_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_rsub_tensor_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::rsub_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_scalar_tensor(scalar s, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::scalar_tensor(*s, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_scalar_tensor_out(gc_tensor out, scalar s) {
+  PROTECT(
+    torch::Tensor results__ = torch::scalar_tensor_out(*tensor_ptr_from_ocaml(out), *s);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_scaled_dot_product_attention(gc_tensor query, gc_tensor key, gc_tensor value, gc_tensor attn_mask, double dropout_p, int is_causal, double scale_v, int scale_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::scaled_dot_product_attention(*tensor_ptr_from_ocaml(query), *tensor_ptr_from_ocaml(key), *tensor_ptr_from_ocaml(value), (attn_mask ? tensor_from_ocaml(attn_mask) : torch::Tensor()), dropout_p, (bool)is_causal, scale_null ? c10::nullopt : c10::optional<double>(scale_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_scatter(gc_tensor self, int64_t dim, gc_tensor index, gc_tensor src) {
+  PROTECT(
+    torch::Tensor results__ = torch::scatter(*tensor_ptr_from_ocaml(self), dim, *tensor_ptr_from_ocaml(index), *tensor_ptr_from_ocaml(src));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_scatter_(gc_tensor self, int64_t dim, gc_tensor index, gc_tensor src) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->scatter_(dim, *tensor_ptr_from_ocaml(index), *tensor_ptr_from_ocaml(src));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_scatter_add(gc_tensor self, int64_t dim, gc_tensor index, gc_tensor src) {
+  PROTECT(
+    torch::Tensor results__ = torch::scatter_add(*tensor_ptr_from_ocaml(self), dim, *tensor_ptr_from_ocaml(index), *tensor_ptr_from_ocaml(src));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_scatter_add_(gc_tensor self, int64_t dim, gc_tensor index, gc_tensor src) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->scatter_add_(dim, *tensor_ptr_from_ocaml(index), *tensor_ptr_from_ocaml(src));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_scatter_add_out(gc_tensor out, gc_tensor self, int64_t dim, gc_tensor index, gc_tensor src) {
+  PROTECT(
+    torch::Tensor results__ = torch::scatter_add_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dim, *tensor_ptr_from_ocaml(index), *tensor_ptr_from_ocaml(src));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_scatter_reduce(gc_tensor self, int64_t dim, gc_tensor index, gc_tensor src, char * reduce) {
+  PROTECT(
+    torch::Tensor results__ = torch::scatter(*tensor_ptr_from_ocaml(self), dim, *tensor_ptr_from_ocaml(index), *tensor_ptr_from_ocaml(src), std::string(reduce));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_scatter_reduce_(gc_tensor self, int64_t dim, gc_tensor index, gc_tensor src, char * reduce) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->scatter_(dim, *tensor_ptr_from_ocaml(index), *tensor_ptr_from_ocaml(src), std::string(reduce));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_scatter_reduce_out(gc_tensor out, gc_tensor self, int64_t dim, gc_tensor index, gc_tensor src, char * reduce) {
+  PROTECT(
+    torch::Tensor results__ = torch::scatter_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dim, *tensor_ptr_from_ocaml(index), *tensor_ptr_from_ocaml(src), std::string(reduce));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_scatter_src_out(gc_tensor out, gc_tensor self, int64_t dim, gc_tensor index, gc_tensor src) {
+  PROTECT(
+    torch::Tensor results__ = torch::scatter_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dim, *tensor_ptr_from_ocaml(index), *tensor_ptr_from_ocaml(src));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_scatter_value(gc_tensor self, int64_t dim, gc_tensor index, scalar value) {
+  PROTECT(
+    torch::Tensor results__ = torch::scatter(*tensor_ptr_from_ocaml(self), dim, *tensor_ptr_from_ocaml(index), *value);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_scatter_value_(gc_tensor self, int64_t dim, gc_tensor index, scalar value) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->scatter_(dim, *tensor_ptr_from_ocaml(index), *value);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_scatter_value_out(gc_tensor out, gc_tensor self, int64_t dim, gc_tensor index, scalar value) {
+  PROTECT(
+    torch::Tensor results__ = torch::scatter_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dim, *tensor_ptr_from_ocaml(index), *value);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_scatter_value_reduce(gc_tensor self, int64_t dim, gc_tensor index, scalar value, char * reduce) {
+  PROTECT(
+    torch::Tensor results__ = torch::scatter(*tensor_ptr_from_ocaml(self), dim, *tensor_ptr_from_ocaml(index), *value, std::string(reduce));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_scatter_value_reduce_(gc_tensor self, int64_t dim, gc_tensor index, scalar value, char * reduce) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->scatter_(dim, *tensor_ptr_from_ocaml(index), *value, std::string(reduce));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_scatter_value_reduce_out(gc_tensor out, gc_tensor self, int64_t dim, gc_tensor index, scalar value, char * reduce) {
+  PROTECT(
+    torch::Tensor results__ = torch::scatter_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dim, *tensor_ptr_from_ocaml(index), *value, std::string(reduce));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_searchsorted(gc_tensor sorted_sequence, gc_tensor self, int out_int32, int right, char * side, gc_tensor sorter) {
+  PROTECT(
+    torch::Tensor results__ = torch::searchsorted(*tensor_ptr_from_ocaml(sorted_sequence), *tensor_ptr_from_ocaml(self), (bool)out_int32, (bool)right, std::string(side), (sorter ? tensor_from_ocaml(sorter) : torch::Tensor()));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_searchsorted_scalar(gc_tensor sorted_sequence, scalar self, int out_int32, int right, char * side, gc_tensor sorter) {
+  PROTECT(
+    torch::Tensor results__ = torch::searchsorted(*tensor_ptr_from_ocaml(sorted_sequence), *self, (bool)out_int32, (bool)right, std::string(side), (sorter ? tensor_from_ocaml(sorter) : torch::Tensor()));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_searchsorted_scalar_out(gc_tensor out, gc_tensor sorted_sequence, scalar self, int out_int32, int right, char * side, gc_tensor sorter) {
+  PROTECT(
+    torch::Tensor results__ = torch::searchsorted_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(sorted_sequence), *self, (bool)out_int32, (bool)right, std::string(side), (sorter ? tensor_from_ocaml(sorter) : torch::Tensor()));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_searchsorted_tensor_out(gc_tensor out, gc_tensor sorted_sequence, gc_tensor self, int out_int32, int right, char * side, gc_tensor sorter) {
+  PROTECT(
+    torch::Tensor results__ = torch::searchsorted_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(sorted_sequence), *tensor_ptr_from_ocaml(self), (bool)out_int32, (bool)right, std::string(side), (sorter ? tensor_from_ocaml(sorter) : torch::Tensor()));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_segment_reduce(gc_tensor data, char * reduce, gc_tensor lengths, gc_tensor indices, gc_tensor offsets, int64_t axis, int unsafe, scalar initial) {
+  PROTECT(
+    torch::Tensor results__ = torch::segment_reduce(*tensor_ptr_from_ocaml(data), std::string(reduce), (lengths ? tensor_from_ocaml(lengths) : torch::Tensor()), (indices ? tensor_from_ocaml(indices) : torch::Tensor()), (offsets ? tensor_from_ocaml(offsets) : torch::Tensor()), axis, (bool)unsafe, *initial);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_segment_reduce_out(gc_tensor out, gc_tensor data, char * reduce, gc_tensor lengths, gc_tensor indices, gc_tensor offsets, int64_t axis, int unsafe, scalar initial) {
+  PROTECT(
+    torch::Tensor results__ = torch::segment_reduce_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(data), std::string(reduce), (lengths ? tensor_from_ocaml(lengths) : torch::Tensor()), (indices ? tensor_from_ocaml(indices) : torch::Tensor()), (offsets ? tensor_from_ocaml(offsets) : torch::Tensor()), axis, (bool)unsafe, *initial);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_select(gc_tensor self, int64_t dim, int64_t index) {
+  PROTECT(
+    torch::Tensor results__ = torch::select(*tensor_ptr_from_ocaml(self), dim, index);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_select_backward(gc_tensor grad_output, int64_t *input_sizes_data, int input_sizes_len, int64_t dim, int64_t index) {
+  PROTECT(
+    torch::Tensor results__ = torch::select_backward(*tensor_ptr_from_ocaml(grad_output), torch::IntArrayRef(input_sizes_data, input_sizes_len), dim, index);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_select_backward_out(gc_tensor out, gc_tensor grad_output, int64_t *input_sizes_data, int input_sizes_len, int64_t dim, int64_t index) {
+  PROTECT(
+    torch::Tensor results__ = torch::select_backward_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(grad_output), torch::IntArrayRef(input_sizes_data, input_sizes_len), dim, index);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_select_copy(gc_tensor self, int64_t dim, int64_t index) {
+  PROTECT(
+    torch::Tensor results__ = torch::select_copy(*tensor_ptr_from_ocaml(self), dim, index);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_select_copy_int_out(gc_tensor out, gc_tensor self, int64_t dim, int64_t index) {
+  PROTECT(
+    torch::Tensor results__ = torch::select_copy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dim, index);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_select_scatter(gc_tensor self, gc_tensor src, int64_t dim, int64_t index) {
+  PROTECT(
+    torch::Tensor results__ = torch::select_scatter(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(src), dim, index);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_select_scatter_out(gc_tensor out, gc_tensor self, gc_tensor src, int64_t dim, int64_t index) {
+  PROTECT(
+    torch::Tensor results__ = torch::select_scatter_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(src), dim, index);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_selu(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::selu(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_selu_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::selu_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_set(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::set(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_set_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->set_();
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_set_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::set_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_set_requires_grad(gc_tensor self, int r) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->set_requires_grad((bool)r);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_set_source_tensor(gc_tensor self, gc_tensor source) {
+  PROTECT(
+    torch::Tensor results__ = torch::set(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(source));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_set_source_tensor_(gc_tensor self, gc_tensor source) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->set_(*tensor_ptr_from_ocaml(source));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_set_source_tensor_out(gc_tensor out, gc_tensor self, gc_tensor source) {
+  PROTECT(
+    torch::Tensor results__ = torch::set_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(source));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_set_source_tensor_storage_offset_(gc_tensor self, gc_tensor source, int64_t storage_offset, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->set_(*tensor_ptr_from_ocaml(source), storage_offset, torch::IntArrayRef(size_data, size_len), torch::IntArrayRef(stride_data, stride_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sgn(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::sgn(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sgn_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->sgn_();
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sgn_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::sgn_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sigmoid(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::sigmoid(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sigmoid_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::sigmoid_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sigmoid_backward(gc_tensor grad_output, gc_tensor output) {
+  PROTECT(
+    torch::Tensor results__ = torch::sigmoid_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(output));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sigmoid_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor output) {
+  PROTECT(
+    torch::Tensor results__ = torch::sigmoid_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(output));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sigmoid_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::sigmoid_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sign(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::sign(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sign_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->sign_();
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sign_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::sign_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_signbit(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::signbit(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_signbit_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::signbit_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_silu(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::silu(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_silu_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::silu_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_silu_backward(gc_tensor grad_output, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::silu_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_silu_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::silu_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_silu_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::silu_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sin(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::sin(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sin_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::sin_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sin_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::sin_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sinc(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::sinc(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sinc_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::sinc_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sinc_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::sinc_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sinh(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::sinh(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sinh_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::sinh_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sinh_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::sinh_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+int64_t atg_size(gc_tensor self, int64_t dim) {
+  PROTECT(
+    return torch::size(*tensor_ptr_from_ocaml(self), dim);
+  )
+  return 0;
+}
+
+raw_tensor atg_slice(gc_tensor self, int64_t dim, int64_t start_v, int start_null, int64_t end_v, int end_null, int64_t step) {
+  PROTECT(
+    torch::Tensor results__ = torch::slice(*tensor_ptr_from_ocaml(self), dim, start_null ? c10::nullopt : c10::optional<int64_t>(start_v), end_null ? c10::nullopt : c10::optional<int64_t>(end_v), step);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_slice_backward(gc_tensor grad_output, int64_t *input_sizes_data, int input_sizes_len, int64_t dim, int64_t start, int64_t end, int64_t step) {
+  PROTECT(
+    torch::Tensor results__ = torch::slice_backward(*tensor_ptr_from_ocaml(grad_output), torch::IntArrayRef(input_sizes_data, input_sizes_len), dim, start, end, step);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_slice_backward_out(gc_tensor out, gc_tensor grad_output, int64_t *input_sizes_data, int input_sizes_len, int64_t dim, int64_t start, int64_t end, int64_t step) {
+  PROTECT(
+    torch::Tensor results__ = torch::slice_backward_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(grad_output), torch::IntArrayRef(input_sizes_data, input_sizes_len), dim, start, end, step);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_slice_copy(gc_tensor self, int64_t dim, int64_t start_v, int start_null, int64_t end_v, int end_null, int64_t step) {
+  PROTECT(
+    torch::Tensor results__ = torch::slice_copy(*tensor_ptr_from_ocaml(self), dim, start_null ? c10::nullopt : c10::optional<int64_t>(start_v), end_null ? c10::nullopt : c10::optional<int64_t>(end_v), step);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_slice_copy_tensor_out(gc_tensor out, gc_tensor self, int64_t dim, int64_t start_v, int start_null, int64_t end_v, int end_null, int64_t step) {
+  PROTECT(
+    torch::Tensor results__ = torch::slice_copy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dim, start_null ? c10::nullopt : c10::optional<int64_t>(start_v), end_null ? c10::nullopt : c10::optional<int64_t>(end_v), step);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_slice_scatter(gc_tensor self, gc_tensor src, int64_t dim, int64_t start_v, int start_null, int64_t end_v, int end_null, int64_t step) {
+  PROTECT(
+    torch::Tensor results__ = torch::slice_scatter(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(src), dim, start_null ? c10::nullopt : c10::optional<int64_t>(start_v), end_null ? c10::nullopt : c10::optional<int64_t>(end_v), step);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_slice_scatter_out(gc_tensor out, gc_tensor self, gc_tensor src, int64_t dim, int64_t start_v, int start_null, int64_t end_v, int end_null, int64_t step) {
+  PROTECT(
+    torch::Tensor results__ = torch::slice_scatter_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(src), dim, start_null ? c10::nullopt : c10::optional<int64_t>(start_v), end_null ? c10::nullopt : c10::optional<int64_t>(end_v), step);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_slogdet(raw_tensor *out__, gc_tensor self) {
+  PROTECT(
+    auto results__ = torch::slogdet(*tensor_ptr_from_ocaml(self));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_slogdet_out(raw_tensor *out__, gc_tensor sign, gc_tensor logabsdet, gc_tensor self) {
+  PROTECT(
+    auto results__ = torch::slogdet_out(*tensor_ptr_from_ocaml(sign), *tensor_ptr_from_ocaml(logabsdet), *tensor_ptr_from_ocaml(self));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_slow_conv3d(gc_tensor self, gc_tensor weight, int64_t *kernel_size_data, int kernel_size_len, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::slow_conv3d(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(weight), torch::IntArrayRef(kernel_size_data, kernel_size_len), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_slow_conv3d_out(gc_tensor out, gc_tensor self, gc_tensor weight, int64_t *kernel_size_data, int kernel_size_len, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::slow_conv3d_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(weight), torch::IntArrayRef(kernel_size_data, kernel_size_len), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_slow_conv_dilated2d(gc_tensor self, gc_tensor weight, int64_t *kernel_size_data, int kernel_size_len, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::slow_conv_dilated2d(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(weight), torch::IntArrayRef(kernel_size_data, kernel_size_len), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_slow_conv_dilated2d_out(gc_tensor out, gc_tensor self, gc_tensor weight, int64_t *kernel_size_data, int kernel_size_len, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::slow_conv_dilated2d_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(weight), torch::IntArrayRef(kernel_size_data, kernel_size_len), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_slow_conv_dilated3d(gc_tensor self, gc_tensor weight, int64_t *kernel_size_data, int kernel_size_len, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::slow_conv_dilated3d(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(weight), torch::IntArrayRef(kernel_size_data, kernel_size_len), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_slow_conv_dilated3d_out(gc_tensor out, gc_tensor self, gc_tensor weight, int64_t *kernel_size_data, int kernel_size_len, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::slow_conv_dilated3d_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(weight), torch::IntArrayRef(kernel_size_data, kernel_size_len), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_slow_conv_transpose2d(gc_tensor self, gc_tensor weight, int64_t *kernel_size_data, int kernel_size_len, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t *dilation_data, int dilation_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::slow_conv_transpose2d(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(weight), torch::IntArrayRef(kernel_size_data, kernel_size_len), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(output_padding_data, output_padding_len), torch::IntArrayRef(dilation_data, dilation_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_slow_conv_transpose2d_out(gc_tensor out, gc_tensor self, gc_tensor weight, int64_t *kernel_size_data, int kernel_size_len, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t *dilation_data, int dilation_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::slow_conv_transpose2d_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(weight), torch::IntArrayRef(kernel_size_data, kernel_size_len), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(output_padding_data, output_padding_len), torch::IntArrayRef(dilation_data, dilation_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_slow_conv_transpose3d(gc_tensor self, gc_tensor weight, int64_t *kernel_size_data, int kernel_size_len, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t *dilation_data, int dilation_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::slow_conv_transpose3d(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(weight), torch::IntArrayRef(kernel_size_data, kernel_size_len), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(output_padding_data, output_padding_len), torch::IntArrayRef(dilation_data, dilation_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_slow_conv_transpose3d_out(gc_tensor out, gc_tensor self, gc_tensor weight, int64_t *kernel_size_data, int kernel_size_len, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t *dilation_data, int dilation_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::slow_conv_transpose3d_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(weight), torch::IntArrayRef(kernel_size_data, kernel_size_len), (bias ? tensor_from_ocaml(bias) : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(output_padding_data, output_padding_len), torch::IntArrayRef(dilation_data, dilation_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_smm(gc_tensor self, gc_tensor mat2) {
+  PROTECT(
+    torch::Tensor results__ = torch::smm(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(mat2));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_smooth_l1_loss(gc_tensor self, gc_tensor target, int64_t reduction, double beta) {
+  PROTECT(
+    torch::Tensor results__ = torch::smooth_l1_loss(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(target), reduction, beta);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_smooth_l1_loss_backward(gc_tensor grad_output, gc_tensor self, gc_tensor target, int64_t reduction, double beta) {
+  PROTECT(
+    torch::Tensor results__ = torch::smooth_l1_loss_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(target), reduction, beta);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_smooth_l1_loss_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, gc_tensor target, int64_t reduction, double beta) {
+  PROTECT(
+    torch::Tensor results__ = torch::smooth_l1_loss_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(target), reduction, beta);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_smooth_l1_loss_out(gc_tensor out, gc_tensor self, gc_tensor target, int64_t reduction, double beta) {
+  PROTECT(
+    torch::Tensor results__ = torch::smooth_l1_loss_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(target), reduction, beta);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_soft_margin_loss(gc_tensor self, gc_tensor target, int64_t reduction) {
+  PROTECT(
+    torch::Tensor results__ = torch::soft_margin_loss(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(target), reduction);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_soft_margin_loss_backward(gc_tensor grad_output, gc_tensor self, gc_tensor target, int64_t reduction) {
+  PROTECT(
+    torch::Tensor results__ = torch::soft_margin_loss_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(target), reduction);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_soft_margin_loss_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, gc_tensor target, int64_t reduction) {
+  PROTECT(
+    torch::Tensor results__ = torch::soft_margin_loss_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(target), reduction);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_soft_margin_loss_out(gc_tensor out, gc_tensor self, gc_tensor target, int64_t reduction) {
+  PROTECT(
+    torch::Tensor results__ = torch::soft_margin_loss_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(target), reduction);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_softmax(gc_tensor self, int64_t dim, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::softmax(*tensor_ptr_from_ocaml(self), dim, torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_softmax_int_out(gc_tensor out, gc_tensor self, int64_t dim, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::softmax_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dim, torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_softplus(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::softplus(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_softplus_backward(gc_tensor grad_output, gc_tensor self, scalar beta, scalar threshold) {
+  PROTECT(
+    torch::Tensor results__ = torch::softplus_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), *beta, *threshold);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_softplus_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, scalar beta, scalar threshold) {
+  PROTECT(
+    torch::Tensor results__ = torch::softplus_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), *beta, *threshold);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_softplus_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::softplus_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_softshrink(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::softshrink(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_softshrink_backward(gc_tensor grad_output, gc_tensor self, scalar lambd) {
+  PROTECT(
+    torch::Tensor results__ = torch::softshrink_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), *lambd);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_softshrink_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, scalar lambd) {
+  PROTECT(
+    torch::Tensor results__ = torch::softshrink_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), *lambd);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_softshrink_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::softshrink_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_sort(raw_tensor *out__, gc_tensor self, int64_t dim, int descending) {
+  PROTECT(
+    auto results__ = torch::sort(*tensor_ptr_from_ocaml(self), dim, (bool)descending);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_sort_stable(raw_tensor *out__, gc_tensor self, int stable, int64_t dim, int descending) {
+  PROTECT(
+    auto results__ = torch::sort(*tensor_ptr_from_ocaml(self), (bool)stable, dim, (bool)descending);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_sort_values(raw_tensor *out__, gc_tensor values, gc_tensor indices, gc_tensor self, int64_t dim, int descending) {
+  PROTECT(
+    auto results__ = torch::sort_out(*tensor_ptr_from_ocaml(values), *tensor_ptr_from_ocaml(indices), *tensor_ptr_from_ocaml(self), dim, (bool)descending);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_sort_values_stable(raw_tensor *out__, gc_tensor values, gc_tensor indices, gc_tensor self, int stable, int64_t dim, int descending) {
+  PROTECT(
+    auto results__ = torch::sort_out(*tensor_ptr_from_ocaml(values), *tensor_ptr_from_ocaml(indices), *tensor_ptr_from_ocaml(self), (bool)stable, dim, (bool)descending);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_sparse_bsc_tensor(gc_tensor ccol_indices, gc_tensor row_indices, gc_tensor values, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::sparse_bsc_tensor(*tensor_ptr_from_ocaml(ccol_indices), *tensor_ptr_from_ocaml(row_indices), *tensor_ptr_from_ocaml(values), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sparse_bsc_tensor_ccol_row_value_size(gc_tensor ccol_indices, gc_tensor row_indices, gc_tensor values, int64_t *size_data, int size_len, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::sparse_bsc_tensor(*tensor_ptr_from_ocaml(ccol_indices), *tensor_ptr_from_ocaml(row_indices), *tensor_ptr_from_ocaml(values), torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sparse_bsr_tensor(gc_tensor crow_indices, gc_tensor col_indices, gc_tensor values, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::sparse_bsr_tensor(*tensor_ptr_from_ocaml(crow_indices), *tensor_ptr_from_ocaml(col_indices), *tensor_ptr_from_ocaml(values), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sparse_bsr_tensor_crow_col_value_size(gc_tensor crow_indices, gc_tensor col_indices, gc_tensor values, int64_t *size_data, int size_len, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::sparse_bsr_tensor(*tensor_ptr_from_ocaml(crow_indices), *tensor_ptr_from_ocaml(col_indices), *tensor_ptr_from_ocaml(values), torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sparse_compressed_tensor(gc_tensor compressed_indices, gc_tensor plain_indices, gc_tensor values, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::sparse_compressed_tensor(*tensor_ptr_from_ocaml(compressed_indices), *tensor_ptr_from_ocaml(plain_indices), *tensor_ptr_from_ocaml(values), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sparse_compressed_tensor_comp_plain_value_size(gc_tensor compressed_indices, gc_tensor plain_indices, gc_tensor values, int64_t *size_data, int size_len, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::sparse_compressed_tensor(*tensor_ptr_from_ocaml(compressed_indices), *tensor_ptr_from_ocaml(plain_indices), *tensor_ptr_from_ocaml(values), torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sparse_coo_tensor(int64_t *size_data, int size_len, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::sparse_coo_tensor(torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sparse_coo_tensor_indices(gc_tensor indices, gc_tensor values, int options_kind, int options_device, int is_coalesced) {
+  PROTECT(
+    torch::Tensor results__ = torch::sparse_coo_tensor(*tensor_ptr_from_ocaml(indices), *tensor_ptr_from_ocaml(values), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)), (bool)is_coalesced);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sparse_coo_tensor_indices_size(gc_tensor indices, gc_tensor values, int64_t *size_data, int size_len, int options_kind, int options_device, int is_coalesced) {
+  PROTECT(
+    torch::Tensor results__ = torch::sparse_coo_tensor(*tensor_ptr_from_ocaml(indices), *tensor_ptr_from_ocaml(values), torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)), (bool)is_coalesced);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sparse_coo_tensor_size_out(gc_tensor out, int64_t *size_data, int size_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::sparse_coo_tensor_out(*tensor_ptr_from_ocaml(out), torch::IntArrayRef(size_data, size_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sparse_csc_tensor(gc_tensor ccol_indices, gc_tensor row_indices, gc_tensor values, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::sparse_csc_tensor(*tensor_ptr_from_ocaml(ccol_indices), *tensor_ptr_from_ocaml(row_indices), *tensor_ptr_from_ocaml(values), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sparse_csc_tensor_ccol_row_value_size(gc_tensor ccol_indices, gc_tensor row_indices, gc_tensor values, int64_t *size_data, int size_len, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::sparse_csc_tensor(*tensor_ptr_from_ocaml(ccol_indices), *tensor_ptr_from_ocaml(row_indices), *tensor_ptr_from_ocaml(values), torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sparse_csr_tensor(gc_tensor crow_indices, gc_tensor col_indices, gc_tensor values, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::sparse_csr_tensor(*tensor_ptr_from_ocaml(crow_indices), *tensor_ptr_from_ocaml(col_indices), *tensor_ptr_from_ocaml(values), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sparse_csr_tensor_crow_col_value_size(gc_tensor crow_indices, gc_tensor col_indices, gc_tensor values, int64_t *size_data, int size_len, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::sparse_csr_tensor(*tensor_ptr_from_ocaml(crow_indices), *tensor_ptr_from_ocaml(col_indices), *tensor_ptr_from_ocaml(values), torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+int64_t atg_sparse_dim(gc_tensor self) {
+  PROTECT(
+    return tensor_ptr_from_ocaml(self)->sparse_dim();
+  )
+  return 0;
+}
+
+raw_tensor atg_sparse_mask(gc_tensor self, gc_tensor mask) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->sparse_mask(*tensor_ptr_from_ocaml(mask));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sparse_mask_out(gc_tensor out, gc_tensor self, gc_tensor mask) {
+  PROTECT(
+    torch::Tensor results__ = torch::sparse_mask_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(mask));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sparse_resize(gc_tensor self, int64_t *size_data, int size_len, int64_t sparse_dim, int64_t dense_dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::sparse_resize(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(size_data, size_len), sparse_dim, dense_dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sparse_resize_(gc_tensor self, int64_t *size_data, int size_len, int64_t sparse_dim, int64_t dense_dim) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->sparse_resize_(torch::IntArrayRef(size_data, size_len), sparse_dim, dense_dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sparse_resize_and_clear(gc_tensor self, int64_t *size_data, int size_len, int64_t sparse_dim, int64_t dense_dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::sparse_resize_and_clear(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(size_data, size_len), sparse_dim, dense_dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sparse_resize_and_clear_(gc_tensor self, int64_t *size_data, int size_len, int64_t sparse_dim, int64_t dense_dim) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->sparse_resize_and_clear_(torch::IntArrayRef(size_data, size_len), sparse_dim, dense_dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sparse_resize_and_clear_out(gc_tensor out, gc_tensor self, int64_t *size_data, int size_len, int64_t sparse_dim, int64_t dense_dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::sparse_resize_and_clear_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(size_data, size_len), sparse_dim, dense_dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sparse_resize_out(gc_tensor out, gc_tensor self, int64_t *size_data, int size_len, int64_t sparse_dim, int64_t dense_dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::sparse_resize_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(size_data, size_len), sparse_dim, dense_dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sparse_sampled_addmm(gc_tensor self, gc_tensor mat1, gc_tensor mat2) {
+  PROTECT(
+    torch::Tensor results__ = torch::sparse_sampled_addmm(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(mat1), *tensor_ptr_from_ocaml(mat2));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sparse_sampled_addmm_out(gc_tensor out, gc_tensor self, gc_tensor mat1, gc_tensor mat2) {
+  PROTECT(
+    torch::Tensor results__ = torch::sparse_sampled_addmm_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(mat1), *tensor_ptr_from_ocaml(mat2));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_airy_ai(gc_tensor x) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_airy_ai(*tensor_ptr_from_ocaml(x));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_airy_ai_out(gc_tensor out, gc_tensor x) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_airy_ai_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(x));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_bessel_j0(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_bessel_j0(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_bessel_j0_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_bessel_j0_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_bessel_j1(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_bessel_j1(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_bessel_j1_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_bessel_j1_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_bessel_y0(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_bessel_y0(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_bessel_y0_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_bessel_y0_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_bessel_y1(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_bessel_y1(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_bessel_y1_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_bessel_y1_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_chebyshev_polynomial_t(gc_tensor x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_chebyshev_polynomial_t(*tensor_ptr_from_ocaml(x), *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_chebyshev_polynomial_t_n_scalar(gc_tensor x, scalar n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_chebyshev_polynomial_t(*tensor_ptr_from_ocaml(x), *n);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_chebyshev_polynomial_t_n_scalar_out(gc_tensor out, gc_tensor x, scalar n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_chebyshev_polynomial_t_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(x), *n);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_chebyshev_polynomial_t_out(gc_tensor out, gc_tensor x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_chebyshev_polynomial_t_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(x), *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_chebyshev_polynomial_t_x_scalar(scalar x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_chebyshev_polynomial_t(*x, *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_chebyshev_polynomial_t_x_scalar_out(gc_tensor out, scalar x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_chebyshev_polynomial_t_out(*tensor_ptr_from_ocaml(out), *x, *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_chebyshev_polynomial_u(gc_tensor x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_chebyshev_polynomial_u(*tensor_ptr_from_ocaml(x), *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_chebyshev_polynomial_u_n_scalar(gc_tensor x, scalar n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_chebyshev_polynomial_u(*tensor_ptr_from_ocaml(x), *n);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_chebyshev_polynomial_u_n_scalar_out(gc_tensor out, gc_tensor x, scalar n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_chebyshev_polynomial_u_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(x), *n);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_chebyshev_polynomial_u_out(gc_tensor out, gc_tensor x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_chebyshev_polynomial_u_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(x), *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_chebyshev_polynomial_u_x_scalar(scalar x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_chebyshev_polynomial_u(*x, *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_chebyshev_polynomial_u_x_scalar_out(gc_tensor out, scalar x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_chebyshev_polynomial_u_out(*tensor_ptr_from_ocaml(out), *x, *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_chebyshev_polynomial_v(gc_tensor x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_chebyshev_polynomial_v(*tensor_ptr_from_ocaml(x), *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_chebyshev_polynomial_v_n_scalar(gc_tensor x, scalar n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_chebyshev_polynomial_v(*tensor_ptr_from_ocaml(x), *n);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_chebyshev_polynomial_v_n_scalar_out(gc_tensor out, gc_tensor x, scalar n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_chebyshev_polynomial_v_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(x), *n);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_chebyshev_polynomial_v_out(gc_tensor out, gc_tensor x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_chebyshev_polynomial_v_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(x), *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_chebyshev_polynomial_v_x_scalar(scalar x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_chebyshev_polynomial_v(*x, *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_chebyshev_polynomial_v_x_scalar_out(gc_tensor out, scalar x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_chebyshev_polynomial_v_out(*tensor_ptr_from_ocaml(out), *x, *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_chebyshev_polynomial_w(gc_tensor x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_chebyshev_polynomial_w(*tensor_ptr_from_ocaml(x), *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_chebyshev_polynomial_w_n_scalar(gc_tensor x, scalar n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_chebyshev_polynomial_w(*tensor_ptr_from_ocaml(x), *n);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_chebyshev_polynomial_w_n_scalar_out(gc_tensor out, gc_tensor x, scalar n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_chebyshev_polynomial_w_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(x), *n);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_chebyshev_polynomial_w_out(gc_tensor out, gc_tensor x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_chebyshev_polynomial_w_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(x), *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_chebyshev_polynomial_w_x_scalar(scalar x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_chebyshev_polynomial_w(*x, *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_chebyshev_polynomial_w_x_scalar_out(gc_tensor out, scalar x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_chebyshev_polynomial_w_out(*tensor_ptr_from_ocaml(out), *x, *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_digamma(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_digamma(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_digamma_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_digamma_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_entr(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_entr(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_entr_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_entr_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_erf(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_erf(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_erf_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_erf_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_erfc(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_erfc(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_erfc_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_erfc_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_erfcx(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_erfcx(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_erfcx_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_erfcx_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_erfinv(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_erfinv(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_erfinv_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_erfinv_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_exp2(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_exp2(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_exp2_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_exp2_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_expit(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_expit(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_expit_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_expit_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_expm1(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_expm1(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_expm1_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_expm1_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_gammainc(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_gammainc(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_gammainc_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_gammainc_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_gammaincc(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_gammaincc(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_gammaincc_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_gammaincc_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_gammaln(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_gammaln(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_gammaln_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_gammaln_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_hermite_polynomial_h(gc_tensor x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_hermite_polynomial_h(*tensor_ptr_from_ocaml(x), *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_hermite_polynomial_h_n_scalar(gc_tensor x, scalar n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_hermite_polynomial_h(*tensor_ptr_from_ocaml(x), *n);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_hermite_polynomial_h_n_scalar_out(gc_tensor out, gc_tensor x, scalar n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_hermite_polynomial_h_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(x), *n);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_hermite_polynomial_h_out(gc_tensor out, gc_tensor x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_hermite_polynomial_h_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(x), *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_hermite_polynomial_h_x_scalar(scalar x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_hermite_polynomial_h(*x, *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_hermite_polynomial_h_x_scalar_out(gc_tensor out, scalar x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_hermite_polynomial_h_out(*tensor_ptr_from_ocaml(out), *x, *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_hermite_polynomial_he(gc_tensor x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_hermite_polynomial_he(*tensor_ptr_from_ocaml(x), *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_hermite_polynomial_he_n_scalar(gc_tensor x, scalar n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_hermite_polynomial_he(*tensor_ptr_from_ocaml(x), *n);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_hermite_polynomial_he_n_scalar_out(gc_tensor out, gc_tensor x, scalar n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_hermite_polynomial_he_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(x), *n);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_hermite_polynomial_he_out(gc_tensor out, gc_tensor x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_hermite_polynomial_he_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(x), *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_hermite_polynomial_he_x_scalar(scalar x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_hermite_polynomial_he(*x, *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_hermite_polynomial_he_x_scalar_out(gc_tensor out, scalar x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_hermite_polynomial_he_out(*tensor_ptr_from_ocaml(out), *x, *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_i0(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_i0(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_i0_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_i0_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_i0e(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_i0e(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_i0e_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_i0e_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_i1(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_i1(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_i1_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_i1_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_i1e(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_i1e(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_i1e_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_i1e_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_laguerre_polynomial_l(gc_tensor x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_laguerre_polynomial_l(*tensor_ptr_from_ocaml(x), *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_laguerre_polynomial_l_n_scalar(gc_tensor x, scalar n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_laguerre_polynomial_l(*tensor_ptr_from_ocaml(x), *n);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_laguerre_polynomial_l_n_scalar_out(gc_tensor out, gc_tensor x, scalar n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_laguerre_polynomial_l_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(x), *n);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_laguerre_polynomial_l_out(gc_tensor out, gc_tensor x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_laguerre_polynomial_l_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(x), *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_laguerre_polynomial_l_x_scalar(scalar x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_laguerre_polynomial_l(*x, *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_laguerre_polynomial_l_x_scalar_out(gc_tensor out, scalar x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_laguerre_polynomial_l_out(*tensor_ptr_from_ocaml(out), *x, *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_legendre_polynomial_p(gc_tensor x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_legendre_polynomial_p(*tensor_ptr_from_ocaml(x), *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_legendre_polynomial_p_n_scalar(gc_tensor x, scalar n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_legendre_polynomial_p(*tensor_ptr_from_ocaml(x), *n);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_legendre_polynomial_p_n_scalar_out(gc_tensor out, gc_tensor x, scalar n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_legendre_polynomial_p_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(x), *n);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_legendre_polynomial_p_out(gc_tensor out, gc_tensor x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_legendre_polynomial_p_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(x), *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_legendre_polynomial_p_x_scalar(scalar x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_legendre_polynomial_p(*x, *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_legendre_polynomial_p_x_scalar_out(gc_tensor out, scalar x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_legendre_polynomial_p_out(*tensor_ptr_from_ocaml(out), *x, *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_log1p(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_log1p(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_log1p_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_log1p_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_log_ndtr(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_log_ndtr(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_log_ndtr_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_log_ndtr_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_log_softmax(gc_tensor self, int64_t dim, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_log_softmax(*tensor_ptr_from_ocaml(self), dim, torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_logit(gc_tensor self, double eps_v, int eps_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_logit(*tensor_ptr_from_ocaml(self), eps_null ? c10::nullopt : c10::optional<double>(eps_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_logit_out(gc_tensor out, gc_tensor self, double eps_v, int eps_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_logit_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), eps_null ? c10::nullopt : c10::optional<double>(eps_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_logsumexp(gc_tensor self, int64_t *dim_data, int dim_len, int keepdim) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_logsumexp(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(dim_data, dim_len), (bool)keepdim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_logsumexp_out(gc_tensor out, gc_tensor self, int64_t *dim_data, int dim_len, int keepdim) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_logsumexp_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(dim_data, dim_len), (bool)keepdim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_modified_bessel_i0(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_modified_bessel_i0(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_modified_bessel_i0_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_modified_bessel_i0_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_modified_bessel_i1(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_modified_bessel_i1(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_modified_bessel_i1_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_modified_bessel_i1_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_modified_bessel_k0(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_modified_bessel_k0(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_modified_bessel_k0_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_modified_bessel_k0_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_modified_bessel_k1(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_modified_bessel_k1(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_modified_bessel_k1_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_modified_bessel_k1_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_multigammaln(gc_tensor self, int64_t p) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_multigammaln(*tensor_ptr_from_ocaml(self), p);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_multigammaln_out(gc_tensor out, gc_tensor self, int64_t p) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_multigammaln_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), p);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_ndtr(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_ndtr(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_ndtr_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_ndtr_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_ndtri(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_ndtri(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_ndtri_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_ndtri_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_polygamma(int64_t n, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_polygamma(n, *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_polygamma_out(gc_tensor out, int64_t n, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_polygamma_out(*tensor_ptr_from_ocaml(out), n, *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_psi(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_psi(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_psi_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_psi_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_round(gc_tensor self, int64_t decimals) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_round(*tensor_ptr_from_ocaml(self), decimals);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_round_out(gc_tensor out, gc_tensor self, int64_t decimals) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_round_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), decimals);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_scaled_modified_bessel_k0(gc_tensor x) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_scaled_modified_bessel_k0(*tensor_ptr_from_ocaml(x));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_scaled_modified_bessel_k0_out(gc_tensor out, gc_tensor x) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_scaled_modified_bessel_k0_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(x));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_scaled_modified_bessel_k1(gc_tensor x) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_scaled_modified_bessel_k1(*tensor_ptr_from_ocaml(x));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_scaled_modified_bessel_k1_out(gc_tensor out, gc_tensor x) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_scaled_modified_bessel_k1_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(x));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_shifted_chebyshev_polynomial_t(gc_tensor x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_shifted_chebyshev_polynomial_t(*tensor_ptr_from_ocaml(x), *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_shifted_chebyshev_polynomial_t_n_scalar(gc_tensor x, scalar n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_shifted_chebyshev_polynomial_t(*tensor_ptr_from_ocaml(x), *n);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_shifted_chebyshev_polynomial_t_n_scalar_out(gc_tensor out, gc_tensor x, scalar n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_shifted_chebyshev_polynomial_t_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(x), *n);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_shifted_chebyshev_polynomial_t_out(gc_tensor out, gc_tensor x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_shifted_chebyshev_polynomial_t_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(x), *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_shifted_chebyshev_polynomial_t_x_scalar(scalar x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_shifted_chebyshev_polynomial_t(*x, *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_shifted_chebyshev_polynomial_t_x_scalar_out(gc_tensor out, scalar x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_shifted_chebyshev_polynomial_t_out(*tensor_ptr_from_ocaml(out), *x, *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_shifted_chebyshev_polynomial_u(gc_tensor x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_shifted_chebyshev_polynomial_u(*tensor_ptr_from_ocaml(x), *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_shifted_chebyshev_polynomial_u_n_scalar(gc_tensor x, scalar n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_shifted_chebyshev_polynomial_u(*tensor_ptr_from_ocaml(x), *n);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_shifted_chebyshev_polynomial_u_n_scalar_out(gc_tensor out, gc_tensor x, scalar n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_shifted_chebyshev_polynomial_u_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(x), *n);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_shifted_chebyshev_polynomial_u_out(gc_tensor out, gc_tensor x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_shifted_chebyshev_polynomial_u_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(x), *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_shifted_chebyshev_polynomial_u_x_scalar(scalar x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_shifted_chebyshev_polynomial_u(*x, *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_shifted_chebyshev_polynomial_u_x_scalar_out(gc_tensor out, scalar x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_shifted_chebyshev_polynomial_u_out(*tensor_ptr_from_ocaml(out), *x, *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_shifted_chebyshev_polynomial_v(gc_tensor x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_shifted_chebyshev_polynomial_v(*tensor_ptr_from_ocaml(x), *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_shifted_chebyshev_polynomial_v_n_scalar(gc_tensor x, scalar n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_shifted_chebyshev_polynomial_v(*tensor_ptr_from_ocaml(x), *n);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_shifted_chebyshev_polynomial_v_n_scalar_out(gc_tensor out, gc_tensor x, scalar n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_shifted_chebyshev_polynomial_v_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(x), *n);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_shifted_chebyshev_polynomial_v_out(gc_tensor out, gc_tensor x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_shifted_chebyshev_polynomial_v_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(x), *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_shifted_chebyshev_polynomial_v_x_scalar(scalar x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_shifted_chebyshev_polynomial_v(*x, *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_shifted_chebyshev_polynomial_v_x_scalar_out(gc_tensor out, scalar x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_shifted_chebyshev_polynomial_v_out(*tensor_ptr_from_ocaml(out), *x, *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_shifted_chebyshev_polynomial_w(gc_tensor x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_shifted_chebyshev_polynomial_w(*tensor_ptr_from_ocaml(x), *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_shifted_chebyshev_polynomial_w_n_scalar(gc_tensor x, scalar n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_shifted_chebyshev_polynomial_w(*tensor_ptr_from_ocaml(x), *n);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_shifted_chebyshev_polynomial_w_n_scalar_out(gc_tensor out, gc_tensor x, scalar n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_shifted_chebyshev_polynomial_w_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(x), *n);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_shifted_chebyshev_polynomial_w_out(gc_tensor out, gc_tensor x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_shifted_chebyshev_polynomial_w_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(x), *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_shifted_chebyshev_polynomial_w_x_scalar(scalar x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_shifted_chebyshev_polynomial_w(*x, *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_shifted_chebyshev_polynomial_w_x_scalar_out(gc_tensor out, scalar x, gc_tensor n) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_shifted_chebyshev_polynomial_w_out(*tensor_ptr_from_ocaml(out), *x, *tensor_ptr_from_ocaml(n));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_sinc(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_sinc(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_sinc_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_sinc_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_softmax(gc_tensor self, int64_t dim, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_softmax(*tensor_ptr_from_ocaml(self), dim, torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_spherical_bessel_j0(gc_tensor x) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_spherical_bessel_j0(*tensor_ptr_from_ocaml(x));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_spherical_bessel_j0_out(gc_tensor out, gc_tensor x) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_spherical_bessel_j0_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(x));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_xlog1py(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_xlog1py(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_xlog1py_other_scalar(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_xlog1py(*tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_xlog1py_other_scalar_out(gc_tensor out, gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_xlog1py_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_xlog1py_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_xlog1py_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_xlog1py_self_scalar(scalar self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_xlog1py(*self, *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_xlog1py_self_scalar_out(gc_tensor out, scalar self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_xlog1py_out(*tensor_ptr_from_ocaml(out), *self, *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_xlogy(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_xlogy(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_xlogy_other_scalar(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_xlogy(*tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_xlogy_other_scalar_out(gc_tensor out, gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_xlogy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_xlogy_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_xlogy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_xlogy_self_scalar(scalar self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_xlogy(*self, *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_xlogy_self_scalar_out(gc_tensor out, scalar self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_xlogy_out(*tensor_ptr_from_ocaml(out), *self, *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_zeta(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_zeta(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_zeta_other_scalar(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_zeta(*tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_zeta_other_scalar_out(gc_tensor out, gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_zeta_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_zeta_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_zeta_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_zeta_self_scalar(scalar self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_zeta(*self, *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_special_zeta_self_scalar_out(gc_tensor out, scalar self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::special_zeta_out(*tensor_ptr_from_ocaml(out), *self, *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor *atg_split(gc_tensor self, int64_t split_size, int64_t dim) {
+  PROTECT(
+    auto results__ = torch::split(*tensor_ptr_from_ocaml(self), split_size, dim);
+    int sz = results__.size();
+    raw_tensor *out__ = (raw_tensor*)malloc((sz + 1) * sizeof(raw_tensor));
+    for (int i = 0; i < sz; ++i)
+      out__[i] = tensor_to_ocaml(results__[i]);
+    out__[sz] = nullptr;
+    return out__;
+  )
+}
+
+raw_tensor *atg_split_copy(gc_tensor self, int64_t split_size, int64_t dim) {
+  PROTECT(
+    auto results__ = torch::split_copy(*tensor_ptr_from_ocaml(self), split_size, dim);
+    int sz = results__.size();
+    raw_tensor *out__ = (raw_tensor*)malloc((sz + 1) * sizeof(raw_tensor));
+    for (int i = 0; i < sz; ++i)
+      out__[i] = tensor_to_ocaml(results__[i]);
+    out__[sz] = nullptr;
+    return out__;
+  )
+}
+
+void atg_split_copy_tensor_out(gc_tensor *out_data, int out_len, gc_tensor self, int64_t split_size, int64_t dim) {
+  PROTECT(
+    torch::split_copy_out(of_carray_tensor(out_data, out_len), *tensor_ptr_from_ocaml(self), split_size, dim);
+  )
+}
+
+raw_tensor *atg_split_sizes(gc_tensor self, int64_t *split_size_data, int split_size_len, int64_t dim) {
+  PROTECT(
+    auto results__ = torch::split(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(split_size_data, split_size_len), dim);
+    int sz = results__.size();
+    raw_tensor *out__ = (raw_tensor*)malloc((sz + 1) * sizeof(raw_tensor));
+    for (int i = 0; i < sz; ++i)
+      out__[i] = tensor_to_ocaml(results__[i]);
+    out__[sz] = nullptr;
+    return out__;
+  )
+}
+
+raw_tensor *atg_split_with_sizes(gc_tensor self, int64_t *split_sizes_data, int split_sizes_len, int64_t dim) {
+  PROTECT(
+    auto results__ = torch::split_with_sizes(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(split_sizes_data, split_sizes_len), dim);
+    int sz = results__.size();
+    raw_tensor *out__ = (raw_tensor*)malloc((sz + 1) * sizeof(raw_tensor));
+    for (int i = 0; i < sz; ++i)
+      out__[i] = tensor_to_ocaml(results__[i]);
+    out__[sz] = nullptr;
+    return out__;
+  )
+}
+
+raw_tensor *atg_split_with_sizes_copy(gc_tensor self, int64_t *split_sizes_data, int split_sizes_len, int64_t dim) {
+  PROTECT(
+    auto results__ = torch::split_with_sizes_copy(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(split_sizes_data, split_sizes_len), dim);
+    int sz = results__.size();
+    raw_tensor *out__ = (raw_tensor*)malloc((sz + 1) * sizeof(raw_tensor));
+    for (int i = 0; i < sz; ++i)
+      out__[i] = tensor_to_ocaml(results__[i]);
+    out__[sz] = nullptr;
+    return out__;
+  )
+}
+
+void atg_split_with_sizes_copy_out(gc_tensor *out_data, int out_len, gc_tensor self, int64_t *split_sizes_data, int split_sizes_len, int64_t dim) {
+  PROTECT(
+    torch::split_with_sizes_copy_out(of_carray_tensor(out_data, out_len), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(split_sizes_data, split_sizes_len), dim);
+  )
+}
+
+raw_tensor atg_sqrt(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::sqrt(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sqrt_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::sqrt_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sqrt_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::sqrt_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_square(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::square(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_square_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::square_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_square_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::square_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_squeeze(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::squeeze(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_squeeze_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->squeeze_();
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_squeeze_copy(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::squeeze_copy(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_squeeze_copy_dim(gc_tensor self, int64_t dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::squeeze_copy(*tensor_ptr_from_ocaml(self), dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_squeeze_copy_dim_out(gc_tensor out, gc_tensor self, int64_t dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::squeeze_copy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_squeeze_copy_dims(gc_tensor self, int64_t *dim_data, int dim_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::squeeze_copy(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(dim_data, dim_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_squeeze_copy_dims_out(gc_tensor out, gc_tensor self, int64_t *dim_data, int dim_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::squeeze_copy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(dim_data, dim_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_squeeze_copy_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::squeeze_copy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_squeeze_dim(gc_tensor self, int64_t dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::squeeze(*tensor_ptr_from_ocaml(self), dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_squeeze_dim_(gc_tensor self, int64_t dim) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->squeeze_(dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_squeeze_dims(gc_tensor self, int64_t *dim_data, int dim_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::squeeze(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(dim_data, dim_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_squeeze_dims_(gc_tensor self, int64_t *dim_data, int dim_len) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->squeeze_(torch::IntArrayRef(dim_data, dim_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sspaddmm(gc_tensor self, gc_tensor mat1, gc_tensor mat2) {
+  PROTECT(
+    torch::Tensor results__ = torch::sspaddmm(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(mat1), *tensor_ptr_from_ocaml(mat2));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sspaddmm_out(gc_tensor out, gc_tensor self, gc_tensor mat1, gc_tensor mat2) {
+  PROTECT(
+    torch::Tensor results__ = torch::sspaddmm_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(mat1), *tensor_ptr_from_ocaml(mat2));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_stack(gc_tensor *tensors_data, int tensors_len, int64_t dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::stack(of_carray_tensor(tensors_data, tensors_len), dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_stack_out(gc_tensor out, gc_tensor *tensors_data, int tensors_len, int64_t dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::stack_out(*tensor_ptr_from_ocaml(out), of_carray_tensor(tensors_data, tensors_len), dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_std(gc_tensor self, int unbiased) {
+  PROTECT(
+    torch::Tensor results__ = torch::std(*tensor_ptr_from_ocaml(self), (bool)unbiased);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_std_correction(gc_tensor self, int64_t *dim_data, int dim_len, scalar correction, int keepdim) {
+  PROTECT(
+    torch::Tensor results__ = torch::std(*tensor_ptr_from_ocaml(self), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), *correction, (bool)keepdim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_std_correction_out(gc_tensor out, gc_tensor self, int64_t *dim_data, int dim_len, scalar correction, int keepdim) {
+  PROTECT(
+    torch::Tensor results__ = torch::std_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), *correction, (bool)keepdim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_std_dim(gc_tensor self, int64_t *dim_data, int dim_len, int unbiased, int keepdim) {
+  PROTECT(
+    torch::Tensor results__ = torch::std(*tensor_ptr_from_ocaml(self), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), (bool)unbiased, (bool)keepdim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_std_mean(raw_tensor *out__, gc_tensor self, int unbiased) {
+  PROTECT(
+    auto results__ = torch::std_mean(*tensor_ptr_from_ocaml(self), (bool)unbiased);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_std_mean_correction(raw_tensor *out__, gc_tensor self, int64_t *dim_data, int dim_len, scalar correction, int keepdim) {
+  PROTECT(
+    auto results__ = torch::std_mean(*tensor_ptr_from_ocaml(self), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), *correction, (bool)keepdim);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_std_mean_correction_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor self, int64_t *dim_data, int dim_len, scalar correction, int keepdim) {
+  PROTECT(
+    auto results__ = torch::std_mean_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(self), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), *correction, (bool)keepdim);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_std_mean_dim(raw_tensor *out__, gc_tensor self, int64_t *dim_data, int dim_len, int unbiased, int keepdim) {
+  PROTECT(
+    auto results__ = torch::std_mean(*tensor_ptr_from_ocaml(self), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), (bool)unbiased, (bool)keepdim);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_std_out(gc_tensor out, gc_tensor self, int64_t *dim_data, int dim_len, int unbiased, int keepdim) {
+  PROTECT(
+    torch::Tensor results__ = torch::std_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), (bool)unbiased, (bool)keepdim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_stft(gc_tensor self, int64_t n_fft, int64_t hop_length_v, int hop_length_null, int64_t win_length_v, int win_length_null, gc_tensor window, int normalized, int onesided, int return_complex) {
+  PROTECT(
+    torch::Tensor results__ = torch::stft(*tensor_ptr_from_ocaml(self), n_fft, hop_length_null ? c10::nullopt : c10::optional<int64_t>(hop_length_v), win_length_null ? c10::nullopt : c10::optional<int64_t>(win_length_v), (window ? tensor_from_ocaml(window) : torch::Tensor()), (bool)normalized, (bool)onesided, (bool)return_complex);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_stft_center(gc_tensor self, int64_t n_fft, int64_t hop_length_v, int hop_length_null, int64_t win_length_v, int win_length_null, gc_tensor window, int center, char * pad_mode, int normalized, int onesided, int return_complex) {
+  PROTECT(
+    torch::Tensor results__ = torch::stft(*tensor_ptr_from_ocaml(self), n_fft, hop_length_null ? c10::nullopt : c10::optional<int64_t>(hop_length_v), win_length_null ? c10::nullopt : c10::optional<int64_t>(win_length_v), (window ? tensor_from_ocaml(window) : torch::Tensor()), (bool)center, std::string(pad_mode), (bool)normalized, (bool)onesided, (bool)return_complex);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+int64_t atg_stride(gc_tensor self, int64_t dim) {
+  PROTECT(
+    return torch::stride(*tensor_ptr_from_ocaml(self), dim);
+  )
+  return 0;
+}
+
+raw_tensor atg_sub(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::sub(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sub_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->sub_(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sub_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::sub_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sub_scalar(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::sub(*tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sub_scalar_(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->sub_(*other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sub_scalar_out(gc_tensor out, gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::sub_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_subtract(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::subtract(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_subtract_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->subtract_(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_subtract_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::subtract_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_subtract_scalar(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::subtract(*tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_subtract_scalar_(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->subtract_(*other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sum(gc_tensor self, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::sum(*tensor_ptr_from_ocaml(self), torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sum_dim_intlist(gc_tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::sum(*tensor_ptr_from_ocaml(self), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), (bool)keepdim, torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sum_intlist_out(gc_tensor out, gc_tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::sum_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), (bool)keepdim, torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sum_out(gc_tensor out, gc_tensor self, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::sum_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_sum_to_size(gc_tensor self, int64_t *size_data, int size_len) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->sum_to_size(torch::IntArrayRef(size_data, size_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_svd(raw_tensor *out__, gc_tensor self, int some, int compute_uv) {
+  PROTECT(
+    auto results__ = torch::svd(*tensor_ptr_from_ocaml(self), (bool)some, (bool)compute_uv);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+void atg_svd_u(raw_tensor *out__, gc_tensor U, gc_tensor S, gc_tensor V, gc_tensor self, int some, int compute_uv) {
+  PROTECT(
+    auto results__ = torch::svd_out(*tensor_ptr_from_ocaml(U), *tensor_ptr_from_ocaml(S), *tensor_ptr_from_ocaml(V), *tensor_ptr_from_ocaml(self), (bool)some, (bool)compute_uv);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+raw_tensor atg_swapaxes(gc_tensor self, int64_t axis0, int64_t axis1) {
+  PROTECT(
+    torch::Tensor results__ = torch::swapaxes(*tensor_ptr_from_ocaml(self), axis0, axis1);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_swapaxes_(gc_tensor self, int64_t axis0, int64_t axis1) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->swapaxes_(axis0, axis1);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_swapdims(gc_tensor self, int64_t dim0, int64_t dim1) {
+  PROTECT(
+    torch::Tensor results__ = torch::swapdims(*tensor_ptr_from_ocaml(self), dim0, dim1);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_swapdims_(gc_tensor self, int64_t dim0, int64_t dim1) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->swapdims_(dim0, dim1);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_t(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::t(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_t_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->t_();
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_t_copy(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::t_copy(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_t_copy_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::t_copy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_take(gc_tensor self, gc_tensor index) {
+  PROTECT(
+    torch::Tensor results__ = torch::take(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(index));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_take_along_dim(gc_tensor self, gc_tensor indices, int64_t dim_v, int dim_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::take_along_dim(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(indices), dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_take_along_dim_out(gc_tensor out, gc_tensor self, gc_tensor indices, int64_t dim_v, int dim_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::take_along_dim_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(indices), dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_take_out(gc_tensor out, gc_tensor self, gc_tensor index) {
+  PROTECT(
+    torch::Tensor results__ = torch::take_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(index));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_tan(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::tan(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_tan_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::tan_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_tan_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::tan_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_tanh(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::tanh(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_tanh_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::tanh_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_tanh_backward(gc_tensor grad_output, gc_tensor output) {
+  PROTECT(
+    torch::Tensor results__ = torch::tanh_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(output));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_tanh_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor output) {
+  PROTECT(
+    torch::Tensor results__ = torch::tanh_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(output));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_tanh_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::tanh_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor *atg_tensor_split(gc_tensor self, int64_t sections, int64_t dim) {
+  PROTECT(
+    auto results__ = torch::tensor_split(*tensor_ptr_from_ocaml(self), sections, dim);
+    int sz = results__.size();
+    raw_tensor *out__ = (raw_tensor*)malloc((sz + 1) * sizeof(raw_tensor));
+    for (int i = 0; i < sz; ++i)
+      out__[i] = tensor_to_ocaml(results__[i]);
+    out__[sz] = nullptr;
+    return out__;
+  )
+}
+
+raw_tensor *atg_tensor_split_indices(gc_tensor self, int64_t *indices_data, int indices_len, int64_t dim) {
+  PROTECT(
+    auto results__ = torch::tensor_split(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(indices_data, indices_len), dim);
+    int sz = results__.size();
+    raw_tensor *out__ = (raw_tensor*)malloc((sz + 1) * sizeof(raw_tensor));
+    for (int i = 0; i < sz; ++i)
+      out__[i] = tensor_to_ocaml(results__[i]);
+    out__[sz] = nullptr;
+    return out__;
+  )
+}
+
+raw_tensor *atg_tensor_split_tensor_indices_or_sections(gc_tensor self, gc_tensor tensor_indices_or_sections, int64_t dim) {
+  PROTECT(
+    auto results__ = torch::tensor_split(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(tensor_indices_or_sections), dim);
+    int sz = results__.size();
+    raw_tensor *out__ = (raw_tensor*)malloc((sz + 1) * sizeof(raw_tensor));
+    for (int i = 0; i < sz; ++i)
+      out__[i] = tensor_to_ocaml(results__[i]);
+    out__[sz] = nullptr;
+    return out__;
+  )
+}
+
+raw_tensor atg_tensordot(gc_tensor self, gc_tensor other, int64_t *dims_self_data, int dims_self_len, int64_t *dims_other_data, int dims_other_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::tensordot(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other), torch::IntArrayRef(dims_self_data, dims_self_len), torch::IntArrayRef(dims_other_data, dims_other_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_tensordot_out(gc_tensor out, gc_tensor self, gc_tensor other, int64_t *dims_self_data, int dims_self_len, int64_t *dims_other_data, int dims_other_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::tensordot_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other), torch::IntArrayRef(dims_self_data, dims_self_len), torch::IntArrayRef(dims_other_data, dims_other_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_threshold(gc_tensor self, scalar threshold, scalar value) {
+  PROTECT(
+    torch::Tensor results__ = torch::threshold(*tensor_ptr_from_ocaml(self), *threshold, *value);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_threshold_(gc_tensor self, scalar threshold, scalar value) {
+  PROTECT(
+    torch::Tensor results__ = torch::threshold_(*tensor_ptr_from_ocaml(self), *threshold, *value);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_threshold_backward(gc_tensor grad_output, gc_tensor self, scalar threshold) {
+  PROTECT(
+    torch::Tensor results__ = torch::threshold_backward(*tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), *threshold);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_threshold_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, scalar threshold) {
+  PROTECT(
+    torch::Tensor results__ = torch::threshold_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), *tensor_ptr_from_ocaml(self), *threshold);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_threshold_out(gc_tensor out, gc_tensor self, scalar threshold, scalar value) {
+  PROTECT(
+    torch::Tensor results__ = torch::threshold_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *threshold, *value);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_tile(gc_tensor self, int64_t *dims_data, int dims_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::tile(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(dims_data, dims_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_to(gc_tensor self, int device) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->to(device_of_int(device));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_to_dense(gc_tensor self, int dtype, int masked_grad) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->to_dense(torch::ScalarType(dtype), (bool)masked_grad);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_to_dense_backward(gc_tensor grad, gc_tensor input, int masked_grad) {
+  PROTECT(
+    torch::Tensor results__ = torch::to_dense_backward(*tensor_ptr_from_ocaml(grad), *tensor_ptr_from_ocaml(input), (bool)masked_grad);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_to_device(gc_tensor self, int device, int dtype, int non_blocking, int copy) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->to(device_of_int(device), torch::ScalarType(dtype), (bool)non_blocking, (bool)copy);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_to_dtype(gc_tensor self, int dtype, int non_blocking, int copy) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->to(torch::ScalarType(dtype), (bool)non_blocking, (bool)copy);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_to_dtype_layout(gc_tensor self, int options_kind, int options_device, int non_blocking, int copy) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->to(at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)), (bool)non_blocking, (bool)copy);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_to_mkldnn(gc_tensor self, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->to_mkldnn(torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_to_mkldnn_backward(gc_tensor grad, gc_tensor input) {
+  PROTECT(
+    torch::Tensor results__ = torch::to_mkldnn_backward(*tensor_ptr_from_ocaml(grad), *tensor_ptr_from_ocaml(input));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_to_mkldnn_out(gc_tensor out, gc_tensor self, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::to_mkldnn_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_to_other(gc_tensor self, gc_tensor other, int non_blocking, int copy) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->to(*tensor_ptr_from_ocaml(other), (bool)non_blocking, (bool)copy);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_to_padded_tensor(gc_tensor self, double padding, int64_t *output_size_data, int output_size_len) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->to_padded_tensor(padding, output_size_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(output_size_data, output_size_len)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_to_padded_tensor_out(gc_tensor out, gc_tensor self, double padding, int64_t *output_size_data, int output_size_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::to_padded_tensor_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), padding, output_size_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(output_size_data, output_size_len)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_topk(raw_tensor *out__, gc_tensor self, int64_t k, int64_t dim, int largest, int sorted) {
+  PROTECT(
+    auto results__ = torch::topk(*tensor_ptr_from_ocaml(self), k, dim, (bool)largest, (bool)sorted);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_topk_values(raw_tensor *out__, gc_tensor values, gc_tensor indices, gc_tensor self, int64_t k, int64_t dim, int largest, int sorted) {
+  PROTECT(
+    auto results__ = torch::topk_out(*tensor_ptr_from_ocaml(values), *tensor_ptr_from_ocaml(indices), *tensor_ptr_from_ocaml(self), k, dim, (bool)largest, (bool)sorted);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_totype(gc_tensor self, int scalar_type) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->toType(torch::ScalarType(scalar_type));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_trace(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::trace(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_trace_backward(gc_tensor grad, int64_t *sizes_data, int sizes_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::trace_backward(*tensor_ptr_from_ocaml(grad), torch::IntArrayRef(sizes_data, sizes_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_trace_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::trace_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_transpose(gc_tensor self, int64_t dim0, int64_t dim1) {
+  PROTECT(
+    torch::Tensor results__ = torch::transpose(*tensor_ptr_from_ocaml(self), dim0, dim1);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_transpose_(gc_tensor self, int64_t dim0, int64_t dim1) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->transpose_(dim0, dim1);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_transpose_copy(gc_tensor self, int64_t dim0, int64_t dim1) {
+  PROTECT(
+    torch::Tensor results__ = torch::transpose_copy(*tensor_ptr_from_ocaml(self), dim0, dim1);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_transpose_copy_int_out(gc_tensor out, gc_tensor self, int64_t dim0, int64_t dim1) {
+  PROTECT(
+    torch::Tensor results__ = torch::transpose_copy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dim0, dim1);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_trapezoid(gc_tensor y, int64_t dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::trapezoid(*tensor_ptr_from_ocaml(y), dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_trapezoid_x(gc_tensor y, gc_tensor x, int64_t dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::trapezoid(*tensor_ptr_from_ocaml(y), *tensor_ptr_from_ocaml(x), dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_trapz(gc_tensor y, gc_tensor x, int64_t dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::trapz(*tensor_ptr_from_ocaml(y), *tensor_ptr_from_ocaml(x), dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_trapz_dx(gc_tensor y, double dx, int64_t dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::trapz(*tensor_ptr_from_ocaml(y), dx, dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_triangular_solve(raw_tensor *out__, gc_tensor self, gc_tensor A, int upper, int transpose, int unitriangular) {
+  PROTECT(
+    auto results__ = torch::triangular_solve(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(A), (bool)upper, (bool)transpose, (bool)unitriangular);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_triangular_solve_x(raw_tensor *out__, gc_tensor X, gc_tensor M, gc_tensor self, gc_tensor A, int upper, int transpose, int unitriangular) {
+  PROTECT(
+    auto results__ = torch::triangular_solve_out(*tensor_ptr_from_ocaml(X), *tensor_ptr_from_ocaml(M), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(A), (bool)upper, (bool)transpose, (bool)unitriangular);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_tril(gc_tensor self, int64_t diagonal) {
+  PROTECT(
+    torch::Tensor results__ = torch::tril(*tensor_ptr_from_ocaml(self), diagonal);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_tril_(gc_tensor self, int64_t diagonal) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->tril_(diagonal);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_tril_indices(int64_t row, int64_t col, int64_t offset, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::tril_indices(row, col, offset, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_tril_indices_out(gc_tensor out, int64_t row, int64_t col, int64_t offset) {
+  PROTECT(
+    torch::Tensor results__ = torch::tril_indices_out(*tensor_ptr_from_ocaml(out), row, col, offset);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_tril_out(gc_tensor out, gc_tensor self, int64_t diagonal) {
+  PROTECT(
+    torch::Tensor results__ = torch::tril_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), diagonal);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_triplet_margin_loss(gc_tensor anchor, gc_tensor positive, gc_tensor negative, double margin, double p, double eps, int swap, int64_t reduction) {
+  PROTECT(
+    torch::Tensor results__ = torch::triplet_margin_loss(*tensor_ptr_from_ocaml(anchor), *tensor_ptr_from_ocaml(positive), *tensor_ptr_from_ocaml(negative), margin, p, eps, (bool)swap, reduction);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_triu(gc_tensor self, int64_t diagonal) {
+  PROTECT(
+    torch::Tensor results__ = torch::triu(*tensor_ptr_from_ocaml(self), diagonal);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_triu_(gc_tensor self, int64_t diagonal) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->triu_(diagonal);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_triu_indices(int64_t row, int64_t col, int64_t offset, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::triu_indices(row, col, offset, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_triu_indices_out(gc_tensor out, int64_t row, int64_t col, int64_t offset) {
+  PROTECT(
+    torch::Tensor results__ = torch::triu_indices_out(*tensor_ptr_from_ocaml(out), row, col, offset);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_triu_out(gc_tensor out, gc_tensor self, int64_t diagonal) {
+  PROTECT(
+    torch::Tensor results__ = torch::triu_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), diagonal);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_true_divide(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::true_divide(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_true_divide_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->true_divide_(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_true_divide_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::true_divide_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_true_divide_scalar(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::true_divide(*tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_true_divide_scalar_(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->true_divide_(*other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_trunc(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::trunc(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_trunc_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::trunc_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_trunc_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::trunc_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_type_as(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->type_as(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor *atg_unbind(gc_tensor self, int64_t dim) {
+  PROTECT(
+    auto results__ = torch::unbind(*tensor_ptr_from_ocaml(self), dim);
+    int sz = results__.size();
+    raw_tensor *out__ = (raw_tensor*)malloc((sz + 1) * sizeof(raw_tensor));
+    for (int i = 0; i < sz; ++i)
+      out__[i] = tensor_to_ocaml(results__[i]);
+    out__[sz] = nullptr;
+    return out__;
+  )
+}
+
+raw_tensor *atg_unbind_copy(gc_tensor self, int64_t dim) {
+  PROTECT(
+    auto results__ = torch::unbind_copy(*tensor_ptr_from_ocaml(self), dim);
+    int sz = results__.size();
+    raw_tensor *out__ = (raw_tensor*)malloc((sz + 1) * sizeof(raw_tensor));
+    for (int i = 0; i < sz; ++i)
+      out__[i] = tensor_to_ocaml(results__[i]);
+    out__[sz] = nullptr;
+    return out__;
+  )
+}
+
+void atg_unbind_copy_int_out(gc_tensor *out_data, int out_len, gc_tensor self, int64_t dim) {
+  PROTECT(
+    torch::unbind_copy_out(of_carray_tensor(out_data, out_len), *tensor_ptr_from_ocaml(self), dim);
+  )
+}
+
+raw_tensor atg_unflatten(gc_tensor self, int64_t dim, int64_t *sizes_data, int sizes_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::unflatten(*tensor_ptr_from_ocaml(self), dim, torch::IntArrayRef(sizes_data, sizes_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor *atg_unflatten_dense_tensors(gc_tensor flat, gc_tensor *tensors_data, int tensors_len) {
+  PROTECT(
+    auto results__ = torch::unflatten_dense_tensors(*tensor_ptr_from_ocaml(flat), of_carray_tensor(tensors_data, tensors_len));
+    int sz = results__.size();
+    raw_tensor *out__ = (raw_tensor*)malloc((sz + 1) * sizeof(raw_tensor));
+    for (int i = 0; i < sz; ++i)
+      out__[i] = tensor_to_ocaml(results__[i]);
+    out__[sz] = nullptr;
+    return out__;
+  )
+}
+
+raw_tensor atg_unfold(gc_tensor self, int64_t dimension, int64_t size, int64_t step) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->unfold(dimension, size, step);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_unfold_backward(gc_tensor grad_in, int64_t *input_sizes_data, int input_sizes_len, int64_t dim, int64_t size, int64_t step) {
+  PROTECT(
+    torch::Tensor results__ = torch::unfold_backward(*tensor_ptr_from_ocaml(grad_in), torch::IntArrayRef(input_sizes_data, input_sizes_len), dim, size, step);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_unfold_backward_out(gc_tensor out, gc_tensor grad_in, int64_t *input_sizes_data, int input_sizes_len, int64_t dim, int64_t size, int64_t step) {
+  PROTECT(
+    torch::Tensor results__ = torch::unfold_backward_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(grad_in), torch::IntArrayRef(input_sizes_data, input_sizes_len), dim, size, step);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_unfold_copy(gc_tensor self, int64_t dimension, int64_t size, int64_t step) {
+  PROTECT(
+    torch::Tensor results__ = torch::unfold_copy(*tensor_ptr_from_ocaml(self), dimension, size, step);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_unfold_copy_out(gc_tensor out, gc_tensor self, int64_t dimension, int64_t size, int64_t step) {
+  PROTECT(
+    torch::Tensor results__ = torch::unfold_copy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dimension, size, step);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_uniform(gc_tensor self, double from, double to) {
+  PROTECT(
+    torch::Tensor results__ = torch::uniform(*tensor_ptr_from_ocaml(self), from, to);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_uniform_(gc_tensor self, double from, double to) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->uniform_(from, to);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_uniform_out(gc_tensor out, gc_tensor self, double from, double to) {
+  PROTECT(
+    torch::Tensor results__ = torch::uniform_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), from, to);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_unique_consecutive(raw_tensor *out__, gc_tensor self, int return_inverse, int return_counts, int64_t dim_v, int dim_null) {
+  PROTECT(
+    auto results__ = torch::unique_consecutive(*tensor_ptr_from_ocaml(self), (bool)return_inverse, (bool)return_counts, dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+void atg_unique_consecutive_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor self, int return_inverse, int return_counts, int64_t dim_v, int dim_null) {
+  PROTECT(
+    auto results__ = torch::unique_consecutive_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(out2), *tensor_ptr_from_ocaml(self), (bool)return_inverse, (bool)return_counts, dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v));
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+void atg_unique_dim(raw_tensor *out__, gc_tensor self, int64_t dim, int sorted, int return_inverse, int return_counts) {
+  PROTECT(
+    auto results__ = torch::unique_dim(*tensor_ptr_from_ocaml(self), dim, (bool)sorted, (bool)return_inverse, (bool)return_counts);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+void atg_unique_dim_consecutive(raw_tensor *out__, gc_tensor self, int64_t dim, int return_inverse, int return_counts) {
+  PROTECT(
+    auto results__ = torch::unique_dim_consecutive(*tensor_ptr_from_ocaml(self), dim, (bool)return_inverse, (bool)return_counts);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+void atg_unique_dim_consecutive_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor self, int64_t dim, int return_inverse, int return_counts) {
+  PROTECT(
+    auto results__ = torch::unique_dim_consecutive_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(out2), *tensor_ptr_from_ocaml(self), dim, (bool)return_inverse, (bool)return_counts);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+void atg_unique_dim_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor self, int64_t dim, int sorted, int return_inverse, int return_counts) {
+  PROTECT(
+    auto results__ = torch::unique_dim_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(out2), *tensor_ptr_from_ocaml(self), dim, (bool)sorted, (bool)return_inverse, (bool)return_counts);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+    out__[2] = tensor_to_ocaml(std::get<2>(results__));
+  )
+}
+
+raw_tensor *atg_unsafe_chunk(gc_tensor self, int64_t chunks, int64_t dim) {
+  PROTECT(
+    auto results__ = torch::unsafe_chunk(*tensor_ptr_from_ocaml(self), chunks, dim);
+    int sz = results__.size();
+    raw_tensor *out__ = (raw_tensor*)malloc((sz + 1) * sizeof(raw_tensor));
+    for (int i = 0; i < sz; ++i)
+      out__[i] = tensor_to_ocaml(results__[i]);
+    out__[sz] = nullptr;
+    return out__;
+  )
+}
+
+raw_tensor *atg_unsafe_split(gc_tensor self, int64_t split_size, int64_t dim) {
+  PROTECT(
+    auto results__ = torch::unsafe_split(*tensor_ptr_from_ocaml(self), split_size, dim);
+    int sz = results__.size();
+    raw_tensor *out__ = (raw_tensor*)malloc((sz + 1) * sizeof(raw_tensor));
+    for (int i = 0; i < sz; ++i)
+      out__[i] = tensor_to_ocaml(results__[i]);
+    out__[sz] = nullptr;
+    return out__;
+  )
+}
+
+void atg_unsafe_split_tensor_out(gc_tensor *out_data, int out_len, gc_tensor self, int64_t split_size, int64_t dim) {
+  PROTECT(
+    torch::unsafe_split_out(of_carray_tensor(out_data, out_len), *tensor_ptr_from_ocaml(self), split_size, dim);
+  )
+}
+
+raw_tensor *atg_unsafe_split_with_sizes(gc_tensor self, int64_t *split_sizes_data, int split_sizes_len, int64_t dim) {
+  PROTECT(
+    auto results__ = torch::unsafe_split_with_sizes(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(split_sizes_data, split_sizes_len), dim);
+    int sz = results__.size();
+    raw_tensor *out__ = (raw_tensor*)malloc((sz + 1) * sizeof(raw_tensor));
+    for (int i = 0; i < sz; ++i)
+      out__[i] = tensor_to_ocaml(results__[i]);
+    out__[sz] = nullptr;
+    return out__;
+  )
+}
+
+void atg_unsafe_split_with_sizes_out(gc_tensor *out_data, int out_len, gc_tensor self, int64_t *split_sizes_data, int split_sizes_len, int64_t dim) {
+  PROTECT(
+    torch::unsafe_split_with_sizes_out(of_carray_tensor(out_data, out_len), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(split_sizes_data, split_sizes_len), dim);
+  )
+}
+
+raw_tensor atg_unsqueeze(gc_tensor self, int64_t dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::unsqueeze(*tensor_ptr_from_ocaml(self), dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_unsqueeze_(gc_tensor self, int64_t dim) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->unsqueeze_(dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_unsqueeze_copy(gc_tensor self, int64_t dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::unsqueeze_copy(*tensor_ptr_from_ocaml(self), dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_unsqueeze_copy_out(gc_tensor out, gc_tensor self, int64_t dim) {
+  PROTECT(
+    torch::Tensor results__ = torch::unsqueeze_copy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_upsample_bicubic2d(gc_tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::upsample_bicubic2d(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(output_size_data, output_size_len), (bool)align_corners, scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_upsample_bicubic2d_backward(gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::upsample_bicubic2d_backward(*tensor_ptr_from_ocaml(grad_output), torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), (bool)align_corners, scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_upsample_bicubic2d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::upsample_bicubic2d_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), (bool)align_corners, scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_upsample_bicubic2d_out(gc_tensor out, gc_tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::upsample_bicubic2d_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(output_size_data, output_size_len), (bool)align_corners, scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_upsample_bicubic2d_vec(gc_tensor input, int64_t *output_size_data, int output_size_len, int align_corners, double *scale_factors_data, int scale_factors_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::upsample_bicubic2d(*tensor_ptr_from_ocaml(input), output_size_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(output_size_data, output_size_len)), (bool)align_corners, at::ArrayRef<double>(scale_factors_data, scale_factors_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_upsample_bilinear2d(gc_tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::upsample_bilinear2d(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(output_size_data, output_size_len), (bool)align_corners, scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_upsample_bilinear2d_backward(gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::upsample_bilinear2d_backward(*tensor_ptr_from_ocaml(grad_output), torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), (bool)align_corners, scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_upsample_bilinear2d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::upsample_bilinear2d_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), (bool)align_corners, scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_upsample_bilinear2d_out(gc_tensor out, gc_tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::upsample_bilinear2d_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(output_size_data, output_size_len), (bool)align_corners, scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_upsample_bilinear2d_vec(gc_tensor input, int64_t *output_size_data, int output_size_len, int align_corners, double *scale_factors_data, int scale_factors_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::upsample_bilinear2d(*tensor_ptr_from_ocaml(input), output_size_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(output_size_data, output_size_len)), (bool)align_corners, at::ArrayRef<double>(scale_factors_data, scale_factors_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_upsample_linear1d(gc_tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_v, int scales_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::upsample_linear1d(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(output_size_data, output_size_len), (bool)align_corners, scales_null ? c10::nullopt : c10::optional<double>(scales_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_upsample_linear1d_backward(gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_v, int scales_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::upsample_linear1d_backward(*tensor_ptr_from_ocaml(grad_output), torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), (bool)align_corners, scales_null ? c10::nullopt : c10::optional<double>(scales_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_upsample_linear1d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_v, int scales_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::upsample_linear1d_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), (bool)align_corners, scales_null ? c10::nullopt : c10::optional<double>(scales_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_upsample_linear1d_out(gc_tensor out, gc_tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_v, int scales_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::upsample_linear1d_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(output_size_data, output_size_len), (bool)align_corners, scales_null ? c10::nullopt : c10::optional<double>(scales_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_upsample_linear1d_vec(gc_tensor input, int64_t *output_size_data, int output_size_len, int align_corners, double *scale_factors_data, int scale_factors_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::upsample_linear1d(*tensor_ptr_from_ocaml(input), output_size_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(output_size_data, output_size_len)), (bool)align_corners, at::ArrayRef<double>(scale_factors_data, scale_factors_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_upsample_nearest1d(gc_tensor self, int64_t *output_size_data, int output_size_len, double scales_v, int scales_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::upsample_nearest1d(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(output_size_data, output_size_len), scales_null ? c10::nullopt : c10::optional<double>(scales_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_upsample_nearest1d_backward(gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_v, int scales_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::upsample_nearest1d_backward(*tensor_ptr_from_ocaml(grad_output), torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), scales_null ? c10::nullopt : c10::optional<double>(scales_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_upsample_nearest1d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_v, int scales_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::upsample_nearest1d_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), scales_null ? c10::nullopt : c10::optional<double>(scales_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_upsample_nearest1d_out(gc_tensor out, gc_tensor self, int64_t *output_size_data, int output_size_len, double scales_v, int scales_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::upsample_nearest1d_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(output_size_data, output_size_len), scales_null ? c10::nullopt : c10::optional<double>(scales_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_upsample_nearest1d_vec(gc_tensor input, int64_t *output_size_data, int output_size_len, double *scale_factors_data, int scale_factors_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::upsample_nearest1d(*tensor_ptr_from_ocaml(input), output_size_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(output_size_data, output_size_len)), at::ArrayRef<double>(scale_factors_data, scale_factors_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_upsample_nearest2d(gc_tensor self, int64_t *output_size_data, int output_size_len, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::upsample_nearest2d(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(output_size_data, output_size_len), scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_upsample_nearest2d_backward(gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::upsample_nearest2d_backward(*tensor_ptr_from_ocaml(grad_output), torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_upsample_nearest2d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::upsample_nearest2d_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_upsample_nearest2d_out(gc_tensor out, gc_tensor self, int64_t *output_size_data, int output_size_len, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::upsample_nearest2d_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(output_size_data, output_size_len), scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_upsample_nearest2d_vec(gc_tensor input, int64_t *output_size_data, int output_size_len, double *scale_factors_data, int scale_factors_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::upsample_nearest2d(*tensor_ptr_from_ocaml(input), output_size_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(output_size_data, output_size_len)), at::ArrayRef<double>(scale_factors_data, scale_factors_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_upsample_nearest3d(gc_tensor self, int64_t *output_size_data, int output_size_len, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::upsample_nearest3d(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(output_size_data, output_size_len), scales_d_null ? c10::nullopt : c10::optional<double>(scales_d_v), scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_upsample_nearest3d_backward(gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::upsample_nearest3d_backward(*tensor_ptr_from_ocaml(grad_output), torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), scales_d_null ? c10::nullopt : c10::optional<double>(scales_d_v), scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_upsample_nearest3d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::upsample_nearest3d_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), scales_d_null ? c10::nullopt : c10::optional<double>(scales_d_v), scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_upsample_nearest3d_out(gc_tensor out, gc_tensor self, int64_t *output_size_data, int output_size_len, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::upsample_nearest3d_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(output_size_data, output_size_len), scales_d_null ? c10::nullopt : c10::optional<double>(scales_d_v), scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_upsample_nearest3d_vec(gc_tensor input, int64_t *output_size_data, int output_size_len, double *scale_factors_data, int scale_factors_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::upsample_nearest3d(*tensor_ptr_from_ocaml(input), output_size_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(output_size_data, output_size_len)), at::ArrayRef<double>(scale_factors_data, scale_factors_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_upsample_trilinear3d(gc_tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::upsample_trilinear3d(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(output_size_data, output_size_len), (bool)align_corners, scales_d_null ? c10::nullopt : c10::optional<double>(scales_d_v), scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_upsample_trilinear3d_backward(gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::upsample_trilinear3d_backward(*tensor_ptr_from_ocaml(grad_output), torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), (bool)align_corners, scales_d_null ? c10::nullopt : c10::optional<double>(scales_d_v), scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_upsample_trilinear3d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::upsample_trilinear3d_backward_out(*tensor_ptr_from_ocaml(grad_input), *tensor_ptr_from_ocaml(grad_output), torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), (bool)align_corners, scales_d_null ? c10::nullopt : c10::optional<double>(scales_d_v), scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_upsample_trilinear3d_out(gc_tensor out, gc_tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
+  PROTECT(
+    torch::Tensor results__ = torch::upsample_trilinear3d_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(output_size_data, output_size_len), (bool)align_corners, scales_d_null ? c10::nullopt : c10::optional<double>(scales_d_v), scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_upsample_trilinear3d_vec(gc_tensor input, int64_t *output_size_data, int output_size_len, int align_corners, double *scale_factors_data, int scale_factors_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::upsample_trilinear3d(*tensor_ptr_from_ocaml(input), output_size_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(output_size_data, output_size_len)), (bool)align_corners, at::ArrayRef<double>(scale_factors_data, scale_factors_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_value_selecting_reduction_backward(gc_tensor grad, int64_t dim, gc_tensor indices, int64_t *sizes_data, int sizes_len, int keepdim) {
+  PROTECT(
+    torch::Tensor results__ = torch::value_selecting_reduction_backward(*tensor_ptr_from_ocaml(grad), dim, *tensor_ptr_from_ocaml(indices), torch::IntArrayRef(sizes_data, sizes_len), (bool)keepdim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_values(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->values();
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_values_copy(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::values_copy(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_values_copy_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::values_copy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_vander(gc_tensor x, int64_t n_v, int n_null, int increasing) {
+  PROTECT(
+    torch::Tensor results__ = torch::vander(*tensor_ptr_from_ocaml(x), n_null ? c10::nullopt : c10::optional<int64_t>(n_v), (bool)increasing);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_var(gc_tensor self, int unbiased) {
+  PROTECT(
+    torch::Tensor results__ = torch::var(*tensor_ptr_from_ocaml(self), (bool)unbiased);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_var_correction(gc_tensor self, int64_t *dim_data, int dim_len, scalar correction, int keepdim) {
+  PROTECT(
+    torch::Tensor results__ = torch::var(*tensor_ptr_from_ocaml(self), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), *correction, (bool)keepdim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_var_correction_out(gc_tensor out, gc_tensor self, int64_t *dim_data, int dim_len, scalar correction, int keepdim) {
+  PROTECT(
+    torch::Tensor results__ = torch::var_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), *correction, (bool)keepdim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_var_dim(gc_tensor self, int64_t *dim_data, int dim_len, int unbiased, int keepdim) {
+  PROTECT(
+    torch::Tensor results__ = torch::var(*tensor_ptr_from_ocaml(self), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), (bool)unbiased, (bool)keepdim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+void atg_var_mean(raw_tensor *out__, gc_tensor self, int unbiased) {
+  PROTECT(
+    auto results__ = torch::var_mean(*tensor_ptr_from_ocaml(self), (bool)unbiased);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_var_mean_correction(raw_tensor *out__, gc_tensor self, int64_t *dim_data, int dim_len, scalar correction, int keepdim) {
+  PROTECT(
+    auto results__ = torch::var_mean(*tensor_ptr_from_ocaml(self), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), *correction, (bool)keepdim);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_var_mean_correction_out(raw_tensor *out__, gc_tensor out0, gc_tensor out1, gc_tensor self, int64_t *dim_data, int dim_len, scalar correction, int keepdim) {
+  PROTECT(
+    auto results__ = torch::var_mean_out(*tensor_ptr_from_ocaml(out0), *tensor_ptr_from_ocaml(out1), *tensor_ptr_from_ocaml(self), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), *correction, (bool)keepdim);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+void atg_var_mean_dim(raw_tensor *out__, gc_tensor self, int64_t *dim_data, int dim_len, int unbiased, int keepdim) {
+  PROTECT(
+    auto results__ = torch::var_mean(*tensor_ptr_from_ocaml(self), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), (bool)unbiased, (bool)keepdim);
+    out__[0] = tensor_to_ocaml(std::get<0>(results__));
+    out__[1] = tensor_to_ocaml(std::get<1>(results__));
+  )
+}
+
+raw_tensor atg_var_out(gc_tensor out, gc_tensor self, int64_t *dim_data, int dim_len, int unbiased, int keepdim) {
+  PROTECT(
+    torch::Tensor results__ = torch::var_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), (bool)unbiased, (bool)keepdim);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_vdot(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::vdot(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_vdot_out(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::vdot_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_view(gc_tensor self, int64_t *size_data, int size_len) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->view(torch::IntArrayRef(size_data, size_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_view_as(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->view_as(*tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_view_as_complex(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::view_as_complex(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_view_as_complex_copy(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::view_as_complex_copy(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_view_as_complex_copy_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::view_as_complex_copy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_view_as_real(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::view_as_real(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_view_as_real_copy(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::view_as_real_copy(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_view_as_real_copy_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::view_as_real_copy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_view_copy(gc_tensor self, int64_t *size_data, int size_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::view_copy(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(size_data, size_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_view_copy_dtype(gc_tensor self, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::view_copy(*tensor_ptr_from_ocaml(self), torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_view_copy_dtype_out(gc_tensor out, gc_tensor self, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = torch::view_copy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_view_copy_out(gc_tensor out, gc_tensor self, int64_t *size_data, int size_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::view_copy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), torch::IntArrayRef(size_data, size_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_view_dtype(gc_tensor self, int dtype) {
+  PROTECT(
+    torch::Tensor results__ = tensor_ptr_from_ocaml(self)->view(torch::ScalarType(dtype));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor *atg_vsplit(gc_tensor self, int64_t sections) {
+  PROTECT(
+    auto results__ = torch::vsplit(*tensor_ptr_from_ocaml(self), sections);
+    int sz = results__.size();
+    raw_tensor *out__ = (raw_tensor*)malloc((sz + 1) * sizeof(raw_tensor));
+    for (int i = 0; i < sz; ++i)
+      out__[i] = tensor_to_ocaml(results__[i]);
+    out__[sz] = nullptr;
+    return out__;
+  )
+}
+
+raw_tensor *atg_vsplit_array(gc_tensor self, int64_t *indices_data, int indices_len) {
+  PROTECT(
+    auto results__ = torch::vsplit(*tensor_ptr_from_ocaml(self), torch::IntArrayRef(indices_data, indices_len));
+    int sz = results__.size();
+    raw_tensor *out__ = (raw_tensor*)malloc((sz + 1) * sizeof(raw_tensor));
+    for (int i = 0; i < sz; ++i)
+      out__[i] = tensor_to_ocaml(results__[i]);
+    out__[sz] = nullptr;
+    return out__;
+  )
+}
+
+raw_tensor atg_vstack(gc_tensor *tensors_data, int tensors_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::vstack(of_carray_tensor(tensors_data, tensors_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_vstack_out(gc_tensor out, gc_tensor *tensors_data, int tensors_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::vstack_out(*tensor_ptr_from_ocaml(out), of_carray_tensor(tensors_data, tensors_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor *atg_where(gc_tensor condition) {
+  PROTECT(
+    auto results__ = torch::where(*tensor_ptr_from_ocaml(condition));
+    int sz = results__.size();
+    raw_tensor *out__ = (raw_tensor*)malloc((sz + 1) * sizeof(raw_tensor));
+    for (int i = 0; i < sz; ++i)
+      out__[i] = tensor_to_ocaml(results__[i]);
+    out__[sz] = nullptr;
+    return out__;
+  )
+}
+
+raw_tensor atg_where_scalar(gc_tensor condition, scalar self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::where(*tensor_ptr_from_ocaml(condition), *self, *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_where_scalarother(gc_tensor condition, gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::where(*tensor_ptr_from_ocaml(condition), *tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_where_scalarself(gc_tensor condition, scalar self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::where(*tensor_ptr_from_ocaml(condition), *self, *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_where_self(gc_tensor condition, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::where(*tensor_ptr_from_ocaml(condition), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_where_self_out(gc_tensor out, gc_tensor condition, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::where_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(condition), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_xlogy(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::xlogy(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_xlogy_(gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::xlogy_(*tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_xlogy_outscalar_other(gc_tensor out, gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::xlogy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_xlogy_outscalar_self(gc_tensor out, scalar self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::xlogy_out(*tensor_ptr_from_ocaml(out), *self, *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_xlogy_outtensor(gc_tensor out, gc_tensor self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::xlogy_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self), *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_xlogy_scalar_other(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::xlogy(*tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_xlogy_scalar_other_(gc_tensor self, scalar other) {
+  PROTECT(
+    torch::Tensor results__ = torch::xlogy_(*tensor_ptr_from_ocaml(self), *other);
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_xlogy_scalar_self(scalar self, gc_tensor other) {
+  PROTECT(
+    torch::Tensor results__ = torch::xlogy(*self, *tensor_ptr_from_ocaml(other));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_zero(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::zero(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_zero_(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::zero_(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_zero_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::zero_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_zeros(int64_t *size_data, int size_len, int options_kind, int options_device) {
+  PROTECT(
+    torch::Tensor results__ = torch::zeros(torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_zeros_like(gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::zeros_like(*tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_zeros_like_out(gc_tensor out, gc_tensor self) {
+  PROTECT(
+    torch::Tensor results__ = torch::zeros_like_out(*tensor_ptr_from_ocaml(out), *tensor_ptr_from_ocaml(self));
+    return tensor_to_ocaml(results__);
+  )
+}
+
+raw_tensor atg_zeros_out(gc_tensor out, int64_t *size_data, int size_len) {
+  PROTECT(
+    torch::Tensor results__ = torch::zeros_out(*tensor_ptr_from_ocaml(out), torch::IntArrayRef(size_data, size_len));
+    return tensor_to_ocaml(results__);
+  )
+}
+
diff --git a/src/wrapper/torch_api_generated.cpp.h b/src/wrapper/torch_api_generated.cpp.h
deleted file mode 100644
index a223d65..0000000
--- a/src/wrapper/torch_api_generated.cpp.h
+++ /dev/null
@@ -1,18377 +0,0 @@
-// THIS FILE IS AUTOMATICALLY GENERATED, DO NOT EDIT BY HAND!
-
-void atg___and__(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::__and__(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg___and__tensor_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::__and__(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg___iand__(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = self->__iand__(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg___iand__tensor_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->__iand__(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg___ilshift__(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = self->__ilshift__(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg___ilshift__tensor_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->__ilshift__(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg___ior__(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = self->__ior__(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg___ior__tensor_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->__ior__(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg___irshift__(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = self->__irshift__(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg___irshift__tensor_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->__irshift__(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg___ixor__(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = self->__ixor__(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg___ixor__tensor_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->__ixor__(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg___lshift__(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::__lshift__(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg___lshift__scalar_out_(tensor *out__, tensor out, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::__lshift___out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg___lshift__tensor_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::__lshift__(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg___lshift__tensor_out_(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::__lshift___out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg___or__(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::__or__(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg___or__tensor_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::__or__(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg___rshift__(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::__rshift__(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg___rshift__scalar_out_(tensor *out__, tensor out, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::__rshift___out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg___rshift__tensor_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::__rshift__(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg___rshift__tensor_out_(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::__rshift___out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg___xor__(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::__xor__(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg___xor__tensor_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::__xor__(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__adaptive_avg_pool2d(tensor *out__, tensor self, int64_t *output_size_data, int output_size_len) {
-  PROTECT(
-    auto outputs__ = torch::_adaptive_avg_pool2d(*self, torch::IntArrayRef(output_size_data, output_size_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__adaptive_avg_pool2d_backward(tensor *out__, tensor grad_output, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::_adaptive_avg_pool2d_backward(*grad_output, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__adaptive_avg_pool2d_backward_out(tensor *out__, tensor out, tensor grad_output, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::_adaptive_avg_pool2d_backward_out(*out, *grad_output, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__adaptive_avg_pool2d_out(tensor *out__, tensor out, tensor self, int64_t *output_size_data, int output_size_len) {
-  PROTECT(
-    auto outputs__ = torch::_adaptive_avg_pool2d_out(*out, *self, torch::IntArrayRef(output_size_data, output_size_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__adaptive_avg_pool3d(tensor *out__, tensor self, int64_t *output_size_data, int output_size_len) {
-  PROTECT(
-    auto outputs__ = torch::_adaptive_avg_pool3d(*self, torch::IntArrayRef(output_size_data, output_size_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__adaptive_avg_pool3d_backward(tensor *out__, tensor grad_output, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::_adaptive_avg_pool3d_backward(*grad_output, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__adaptive_avg_pool3d_backward_out(tensor *out__, tensor out, tensor grad_output, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::_adaptive_avg_pool3d_backward_out(*out, *grad_output, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__adaptive_avg_pool3d_out(tensor *out__, tensor out, tensor self, int64_t *output_size_data, int output_size_len) {
-  PROTECT(
-    auto outputs__ = torch::_adaptive_avg_pool3d_out(*out, *self, torch::IntArrayRef(output_size_data, output_size_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__add_batch_dim(tensor *out__, tensor self, int64_t batch_dim, int64_t level) {
-  PROTECT(
-    auto outputs__ = torch::_add_batch_dim(*self, batch_dim, level);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__add_relu(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::_add_relu(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__add_relu_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::_add_relu_(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__add_relu_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::_add_relu_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__add_relu_scalar(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::_add_relu(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__add_relu_scalar_(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::_add_relu_(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__add_relu_scalar_out(tensor *out__, tensor out, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::_add_relu_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__addmm_activation(tensor *out__, tensor self, tensor mat1, tensor mat2, int use_gelu) {
-  PROTECT(
-    auto outputs__ = torch::_addmm_activation(*self, *mat1, *mat2, (bool)use_gelu);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__addmm_activation_out(tensor *out__, tensor out, tensor self, tensor mat1, tensor mat2, int use_gelu) {
-  PROTECT(
-    auto outputs__ = torch::_addmm_activation_out(*out, *self, *mat1, *mat2, (bool)use_gelu);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__aminmax(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::_aminmax(*self);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg__aminmax_dim(tensor *out__, tensor self, int64_t dim, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::_aminmax(*self, dim, (bool)keepdim);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg__aminmax_dim_out(tensor *out__, tensor out0, tensor out1, tensor self, int64_t dim, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::_aminmax_out(*out0, *out1, *self, dim, (bool)keepdim);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg__aminmax_out(tensor *out__, tensor out0, tensor out1, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::_aminmax_out(*out0, *out1, *self);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg__amp_update_scale(tensor *out__, tensor self, tensor growth_tracker, tensor found_inf, double scale_growth_factor, double scale_backoff_factor, int64_t growth_interval) {
-  PROTECT(
-    auto outputs__ = torch::_amp_update_scale(*self, *growth_tracker, *found_inf, scale_growth_factor, scale_backoff_factor, growth_interval);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg__amp_update_scale_(tensor *out__, tensor self, tensor growth_tracker, tensor found_inf, double scale_growth_factor, double scale_backoff_factor, int64_t growth_interval) {
-  PROTECT(
-    auto outputs__ = torch::_amp_update_scale_(*self, *growth_tracker, *found_inf, scale_growth_factor, scale_backoff_factor, growth_interval);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__amp_update_scale_out(tensor *out__, tensor out, tensor self, tensor growth_tracker, tensor found_inf, double scale_growth_factor, double scale_backoff_factor, int64_t growth_interval) {
-  PROTECT(
-    auto outputs__ = torch::_amp_update_scale_out(*out, *self, *growth_tracker, *found_inf, scale_growth_factor, scale_backoff_factor, growth_interval);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__assert_tensor_metadata(tensor a, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len, int dtype) {
-  PROTECT(
-    torch::_assert_tensor_metadata(*a, size_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(size_data, size_len)), stride_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(stride_data, stride_len)), torch::ScalarType(dtype));
-  )
-}
-
-void atg__autocast_to_full_precision(tensor *out__, tensor self, int cuda_enabled, int cpu_enabled) {
-  PROTECT(
-    auto outputs__ = self->_autocast_to_full_precision((bool)cuda_enabled, (bool)cpu_enabled);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__autocast_to_reduced_precision(tensor *out__, tensor self, int cuda_enabled, int cpu_enabled, int cuda_dtype, int cpu_dtype) {
-  PROTECT(
-    auto outputs__ = self->_autocast_to_reduced_precision((bool)cuda_enabled, (bool)cpu_enabled, torch::ScalarType(cuda_dtype), torch::ScalarType(cpu_dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__cast_byte(tensor *out__, tensor self, int non_blocking) {
-  PROTECT(
-    auto outputs__ = torch::_cast_Byte(*self, (bool)non_blocking);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__cast_char(tensor *out__, tensor self, int non_blocking) {
-  PROTECT(
-    auto outputs__ = torch::_cast_Char(*self, (bool)non_blocking);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__cast_double(tensor *out__, tensor self, int non_blocking) {
-  PROTECT(
-    auto outputs__ = torch::_cast_Double(*self, (bool)non_blocking);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__cast_float(tensor *out__, tensor self, int non_blocking) {
-  PROTECT(
-    auto outputs__ = torch::_cast_Float(*self, (bool)non_blocking);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__cast_half(tensor *out__, tensor self, int non_blocking) {
-  PROTECT(
-    auto outputs__ = torch::_cast_Half(*self, (bool)non_blocking);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__cast_int(tensor *out__, tensor self, int non_blocking) {
-  PROTECT(
-    auto outputs__ = torch::_cast_Int(*self, (bool)non_blocking);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__cast_long(tensor *out__, tensor self, int non_blocking) {
-  PROTECT(
-    auto outputs__ = torch::_cast_Long(*self, (bool)non_blocking);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__cast_short(tensor *out__, tensor self, int non_blocking) {
-  PROTECT(
-    auto outputs__ = torch::_cast_Short(*self, (bool)non_blocking);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__cdist_backward(tensor *out__, tensor grad, tensor x1, tensor x2, double p, tensor cdist) {
-  PROTECT(
-    auto outputs__ = torch::_cdist_backward(*grad, *x1, *x2, p, *cdist);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__cdist_backward_out(tensor *out__, tensor out, tensor grad, tensor x1, tensor x2, double p, tensor cdist) {
-  PROTECT(
-    auto outputs__ = torch::_cdist_backward_out(*out, *grad, *x1, *x2, p, *cdist);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__cholesky_solve_helper(tensor *out__, tensor self, tensor A, int upper) {
-  PROTECT(
-    auto outputs__ = torch::_cholesky_solve_helper(*self, *A, (bool)upper);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__cholesky_solve_helper_out(tensor *out__, tensor out, tensor self, tensor A, int upper) {
-  PROTECT(
-    auto outputs__ = torch::_cholesky_solve_helper_out(*out, *self, *A, (bool)upper);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__coalesce(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::_coalesce(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__coalesce_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::_coalesce_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__coalesced(tensor *out__, tensor self, int coalesced) {
-  PROTECT(
-    auto outputs__ = torch::_coalesced(*self, (bool)coalesced);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__coalesced_(tensor *out__, tensor self, int coalesced) {
-  PROTECT(
-    auto outputs__ = self->_coalesced_((bool)coalesced);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__coalesced_out(tensor *out__, tensor out, tensor self, int coalesced) {
-  PROTECT(
-    auto outputs__ = torch::_coalesced_out(*out, *self, (bool)coalesced);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__compute_linear_combination(tensor *out__, tensor input, tensor coefficients) {
-  PROTECT(
-    auto outputs__ = torch::_compute_linear_combination(*input, *coefficients);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__compute_linear_combination_out(tensor *out__, tensor out, tensor input, tensor coefficients) {
-  PROTECT(
-    auto outputs__ = torch::_compute_linear_combination_out(*out, *input, *coefficients);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__conj(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::_conj(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__conj_copy(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::_conj_copy(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__conj_copy_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::_conj_copy_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__conj_physical(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::_conj_physical(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__conj_physical_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::_conj_physical_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__conv_depthwise2d(tensor *out__, tensor self, tensor weight, int64_t *kernel_size_data, int kernel_size_len, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len) {
-  PROTECT(
-    auto outputs__ = torch::_conv_depthwise2d(*self, *weight, torch::IntArrayRef(kernel_size_data, kernel_size_len), (bias ? *bias : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__conv_depthwise2d_out(tensor *out__, tensor out, tensor self, tensor weight, int64_t *kernel_size_data, int kernel_size_len, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len) {
-  PROTECT(
-    auto outputs__ = torch::_conv_depthwise2d_out(*out, *self, *weight, torch::IntArrayRef(kernel_size_data, kernel_size_len), (bias ? *bias : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__convert_indices_from_coo_to_csr(tensor *out__, tensor self, int64_t size, int out_int32) {
-  PROTECT(
-    auto outputs__ = torch::_convert_indices_from_coo_to_csr(*self, size, (bool)out_int32);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__convert_indices_from_coo_to_csr_out(tensor *out__, tensor out, tensor self, int64_t size, int out_int32) {
-  PROTECT(
-    auto outputs__ = torch::_convert_indices_from_coo_to_csr_out(*out, *self, size, (bool)out_int32);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__convert_indices_from_csr_to_coo(tensor *out__, tensor crow_indices, tensor col_indices, int out_int32, int transpose) {
-  PROTECT(
-    auto outputs__ = torch::_convert_indices_from_csr_to_coo(*crow_indices, *col_indices, (bool)out_int32, (bool)transpose);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__convert_indices_from_csr_to_coo_out(tensor *out__, tensor out, tensor crow_indices, tensor col_indices, int out_int32, int transpose) {
-  PROTECT(
-    auto outputs__ = torch::_convert_indices_from_csr_to_coo_out(*out, *crow_indices, *col_indices, (bool)out_int32, (bool)transpose);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__convolution(tensor *out__, tensor input, tensor weight, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int transposed, int64_t *output_padding_data, int output_padding_len, int64_t groups, int benchmark, int deterministic, int cudnn_enabled, int allow_tf32) {
-  PROTECT(
-    auto outputs__ = torch::_convolution(*input, *weight, (bias ? *bias : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)transposed, torch::IntArrayRef(output_padding_data, output_padding_len), groups, (bool)benchmark, (bool)deterministic, (bool)cudnn_enabled, (bool)allow_tf32);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__convolution_deprecated(tensor *out__, tensor input, tensor weight, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int transposed, int64_t *output_padding_data, int output_padding_len, int64_t groups, int benchmark, int deterministic, int cudnn_enabled) {
-  PROTECT(
-    auto outputs__ = torch::_convolution(*input, *weight, (bias ? *bias : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)transposed, torch::IntArrayRef(output_padding_data, output_padding_len), groups, (bool)benchmark, (bool)deterministic, (bool)cudnn_enabled);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__convolution_mode(tensor *out__, tensor input, tensor weight, tensor bias, int64_t *stride_data, int stride_len, char * padding, int64_t *dilation_data, int dilation_len, int64_t groups) {
-  PROTECT(
-    auto outputs__ = torch::_convolution_mode(*input, *weight, (bias ? *bias : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), std::string(padding), torch::IntArrayRef(dilation_data, dilation_len), groups);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__convolution_out(tensor *out__, tensor out, tensor input, tensor weight, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int transposed, int64_t *output_padding_data, int output_padding_len, int64_t groups, int benchmark, int deterministic, int cudnn_enabled, int allow_tf32) {
-  PROTECT(
-    auto outputs__ = torch::_convolution_out(*out, *input, *weight, (bias ? *bias : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)transposed, torch::IntArrayRef(output_padding_data, output_padding_len), groups, (bool)benchmark, (bool)deterministic, (bool)cudnn_enabled, (bool)allow_tf32);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__copy_from(tensor *out__, tensor self, tensor dst, int non_blocking) {
-  PROTECT(
-    auto outputs__ = torch::_copy_from(*self, *dst, (bool)non_blocking);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__copy_from_and_resize(tensor *out__, tensor self, tensor dst) {
-  PROTECT(
-    auto outputs__ = torch::_copy_from_and_resize(*self, *dst);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__copy_from_and_resize_out(tensor *out__, tensor out, tensor self, tensor dst) {
-  PROTECT(
-    auto outputs__ = torch::_copy_from_and_resize_out(*out, *self, *dst);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__copy_from_out(tensor *out__, tensor out, tensor self, tensor dst, int non_blocking) {
-  PROTECT(
-    auto outputs__ = torch::_copy_from_out(*out, *self, *dst, (bool)non_blocking);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__cslt_compress(tensor *out__, tensor input) {
-  PROTECT(
-    auto outputs__ = torch::_cslt_compress(*input);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__cslt_sparse_mm(tensor *out__, tensor compressed_A, tensor dense_B, tensor bias, int transpose_result) {
-  PROTECT(
-    auto outputs__ = torch::_cslt_sparse_mm(*compressed_A, *dense_B, (bias ? *bias : torch::Tensor()), (bool)transpose_result);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__ctc_loss(tensor *out__, tensor log_probs, tensor targets, int64_t *input_lengths_data, int input_lengths_len, int64_t *target_lengths_data, int target_lengths_len, int64_t blank, int zero_infinity) {
-  PROTECT(
-    auto outputs__ = torch::_ctc_loss(*log_probs, *targets, torch::IntArrayRef(input_lengths_data, input_lengths_len), torch::IntArrayRef(target_lengths_data, target_lengths_len), blank, (bool)zero_infinity);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg__ctc_loss_backward(tensor *out__, tensor grad, tensor log_probs, tensor targets, int64_t *input_lengths_data, int input_lengths_len, int64_t *target_lengths_data, int target_lengths_len, tensor neg_log_likelihood, tensor log_alpha, int64_t blank, int zero_infinity) {
-  PROTECT(
-    auto outputs__ = torch::_ctc_loss_backward(*grad, *log_probs, *targets, torch::IntArrayRef(input_lengths_data, input_lengths_len), torch::IntArrayRef(target_lengths_data, target_lengths_len), *neg_log_likelihood, *log_alpha, blank, (bool)zero_infinity);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__ctc_loss_backward_out(tensor *out__, tensor out, tensor grad, tensor log_probs, tensor targets, int64_t *input_lengths_data, int input_lengths_len, int64_t *target_lengths_data, int target_lengths_len, tensor neg_log_likelihood, tensor log_alpha, int64_t blank, int zero_infinity) {
-  PROTECT(
-    auto outputs__ = torch::_ctc_loss_backward_out(*out, *grad, *log_probs, *targets, torch::IntArrayRef(input_lengths_data, input_lengths_len), torch::IntArrayRef(target_lengths_data, target_lengths_len), *neg_log_likelihood, *log_alpha, blank, (bool)zero_infinity);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__ctc_loss_backward_tensor(tensor *out__, tensor grad, tensor log_probs, tensor targets, tensor input_lengths, tensor target_lengths, tensor neg_log_likelihood, tensor log_alpha, int64_t blank, int zero_infinity) {
-  PROTECT(
-    auto outputs__ = torch::_ctc_loss_backward(*grad, *log_probs, *targets, *input_lengths, *target_lengths, *neg_log_likelihood, *log_alpha, blank, (bool)zero_infinity);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__ctc_loss_out(tensor *out__, tensor out0, tensor out1, tensor log_probs, tensor targets, int64_t *input_lengths_data, int input_lengths_len, int64_t *target_lengths_data, int target_lengths_len, int64_t blank, int zero_infinity) {
-  PROTECT(
-    auto outputs__ = torch::_ctc_loss_out(*out0, *out1, *log_probs, *targets, torch::IntArrayRef(input_lengths_data, input_lengths_len), torch::IntArrayRef(target_lengths_data, target_lengths_len), blank, (bool)zero_infinity);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg__ctc_loss_tensor(tensor *out__, tensor log_probs, tensor targets, tensor input_lengths, tensor target_lengths, int64_t blank, int zero_infinity) {
-  PROTECT(
-    auto outputs__ = torch::_ctc_loss(*log_probs, *targets, *input_lengths, *target_lengths, blank, (bool)zero_infinity);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg__ctc_loss_tensor_out(tensor *out__, tensor out0, tensor out1, tensor log_probs, tensor targets, tensor input_lengths, tensor target_lengths, int64_t blank, int zero_infinity) {
-  PROTECT(
-    auto outputs__ = torch::_ctc_loss_out(*out0, *out1, *log_probs, *targets, *input_lengths, *target_lengths, blank, (bool)zero_infinity);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg__cudnn_ctc_loss(tensor *out__, tensor log_probs, tensor targets, int64_t *input_lengths_data, int input_lengths_len, int64_t *target_lengths_data, int target_lengths_len, int64_t blank, int deterministic, int zero_infinity) {
-  PROTECT(
-    auto outputs__ = torch::_cudnn_ctc_loss(*log_probs, *targets, torch::IntArrayRef(input_lengths_data, input_lengths_len), torch::IntArrayRef(target_lengths_data, target_lengths_len), blank, (bool)deterministic, (bool)zero_infinity);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg__cudnn_ctc_loss_out(tensor *out__, tensor out0, tensor out1, tensor log_probs, tensor targets, int64_t *input_lengths_data, int input_lengths_len, int64_t *target_lengths_data, int target_lengths_len, int64_t blank, int deterministic, int zero_infinity) {
-  PROTECT(
-    auto outputs__ = torch::_cudnn_ctc_loss_out(*out0, *out1, *log_probs, *targets, torch::IntArrayRef(input_lengths_data, input_lengths_len), torch::IntArrayRef(target_lengths_data, target_lengths_len), blank, (bool)deterministic, (bool)zero_infinity);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg__cudnn_ctc_loss_tensor(tensor *out__, tensor log_probs, tensor targets, tensor input_lengths, tensor target_lengths, int64_t blank, int deterministic, int zero_infinity) {
-  PROTECT(
-    auto outputs__ = torch::_cudnn_ctc_loss(*log_probs, *targets, *input_lengths, *target_lengths, blank, (bool)deterministic, (bool)zero_infinity);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg__cudnn_init_dropout_state(tensor *out__, double dropout, int train, int64_t dropout_seed, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::_cudnn_init_dropout_state(dropout, (bool)train, dropout_seed, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__cudnn_init_dropout_state_out(tensor *out__, tensor out, double dropout, int train, int64_t dropout_seed) {
-  PROTECT(
-    auto outputs__ = torch::_cudnn_init_dropout_state_out(*out, dropout, (bool)train, dropout_seed);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__cudnn_rnn(tensor *out__, tensor input, tensor *weight_data, int weight_len, int64_t weight_stride0, tensor weight_buf, tensor hx, tensor cx, int64_t mode, int64_t hidden_size, int64_t proj_size, int64_t num_layers, int batch_first, double dropout, int train, int bidirectional, int64_t *batch_sizes_data, int batch_sizes_len, tensor dropout_state) {
-  PROTECT(
-    auto outputs__ = torch::_cudnn_rnn(*input, of_carray_tensor(weight_data, weight_len), weight_stride0, (weight_buf ? *weight_buf : torch::Tensor()), *hx, (cx ? *cx : torch::Tensor()), mode, hidden_size, proj_size, num_layers, (bool)batch_first, dropout, (bool)train, (bool)bidirectional, torch::IntArrayRef(batch_sizes_data, batch_sizes_len), (dropout_state ? *dropout_state : torch::Tensor()));
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-    out__[3] = new torch::Tensor(std::get<3>(outputs__));
-    out__[4] = new torch::Tensor(std::get<4>(outputs__));
-  )
-}
-
-void atg__cudnn_rnn_flatten_weight(tensor *out__, tensor *weight_arr_data, int weight_arr_len, int64_t weight_stride0, int64_t input_size, int64_t mode, int64_t hidden_size, int64_t proj_size, int64_t num_layers, int batch_first, int bidirectional) {
-  PROTECT(
-    auto outputs__ = torch::_cudnn_rnn_flatten_weight(of_carray_tensor(weight_arr_data, weight_arr_len), weight_stride0, input_size, mode, hidden_size, proj_size, num_layers, (bool)batch_first, (bool)bidirectional);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__cudnn_rnn_flatten_weight_out(tensor *out__, tensor out, tensor *weight_arr_data, int weight_arr_len, int64_t weight_stride0, int64_t input_size, int64_t mode, int64_t hidden_size, int64_t proj_size, int64_t num_layers, int batch_first, int bidirectional) {
-  PROTECT(
-    auto outputs__ = torch::_cudnn_rnn_flatten_weight_out(*out, of_carray_tensor(weight_arr_data, weight_arr_len), weight_stride0, input_size, mode, hidden_size, proj_size, num_layers, (bool)batch_first, (bool)bidirectional);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__cudnn_rnn_out(tensor *out__, tensor out0, tensor out1, tensor out2, tensor out3, tensor out4, tensor input, tensor *weight_data, int weight_len, int64_t weight_stride0, tensor weight_buf, tensor hx, tensor cx, int64_t mode, int64_t hidden_size, int64_t proj_size, int64_t num_layers, int batch_first, double dropout, int train, int bidirectional, int64_t *batch_sizes_data, int batch_sizes_len, tensor dropout_state) {
-  PROTECT(
-    auto outputs__ = torch::_cudnn_rnn_out(*out0, *out1, *out2, *out3, *out4, *input, of_carray_tensor(weight_data, weight_len), weight_stride0, (weight_buf ? *weight_buf : torch::Tensor()), *hx, (cx ? *cx : torch::Tensor()), mode, hidden_size, proj_size, num_layers, (bool)batch_first, dropout, (bool)train, (bool)bidirectional, torch::IntArrayRef(batch_sizes_data, batch_sizes_len), (dropout_state ? *dropout_state : torch::Tensor()));
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-    out__[3] = new torch::Tensor(std::get<3>(outputs__));
-    out__[4] = new torch::Tensor(std::get<4>(outputs__));
-  )
-}
-
-int64_t atg__debug_has_internal_overlap(tensor self) {
-  PROTECT(
-    return torch::_debug_has_internal_overlap(*self);
-  )
-  return 0;
-}
-
-void atg__dim_arange(tensor *out__, tensor like, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::_dim_arange(*like, dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-int64_t atg__dimi(tensor self) {
-  PROTECT(
-    return self->_dimI();
-  )
-  return 0;
-}
-
-int64_t atg__dimv(tensor self) {
-  PROTECT(
-    return self->_dimV();
-  )
-  return 0;
-}
-
-void atg__dirichlet_grad(tensor *out__, tensor x, tensor alpha, tensor total) {
-  PROTECT(
-    auto outputs__ = torch::_dirichlet_grad(*x, *alpha, *total);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__dirichlet_grad_out(tensor *out__, tensor out, tensor x, tensor alpha, tensor total) {
-  PROTECT(
-    auto outputs__ = torch::_dirichlet_grad_out(*out, *x, *alpha, *total);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__efficient_attention_backward(tensor *out__, tensor grad_out_, tensor query, tensor key, tensor value, tensor bias, tensor out, tensor cu_seqlens_q, tensor cu_seqlens_k, int64_t max_seqlen_k, int64_t max_seqlen_q, tensor logsumexp, double dropout_p, tensor philox_seed, tensor philox_offset, int64_t custom_mask_type, int bias_requires_grad, double scale_v, int scale_null, int64_t num_splits_key_v, int num_splits_key_null) {
-  PROTECT(
-    auto outputs__ = torch::_efficient_attention_backward(*grad_out_, *query, *key, *value, (bias ? *bias : torch::Tensor()), *out, (cu_seqlens_q ? *cu_seqlens_q : torch::Tensor()), (cu_seqlens_k ? *cu_seqlens_k : torch::Tensor()), max_seqlen_k, max_seqlen_q, *logsumexp, dropout_p, *philox_seed, *philox_offset, custom_mask_type, (bool)bias_requires_grad, scale_null ? c10::nullopt : c10::optional<double>(scale_v), num_splits_key_null ? c10::nullopt : c10::optional<int64_t>(num_splits_key_v));
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-    out__[3] = new torch::Tensor(std::get<3>(outputs__));
-  )
-}
-
-void atg__efficientzerotensor(tensor *out__, int64_t *size_data, int size_len, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::_efficientzerotensor(torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__efficientzerotensor_out(tensor *out__, tensor out, int64_t *size_data, int size_len) {
-  PROTECT(
-    auto outputs__ = torch::_efficientzerotensor_out(*out, torch::IntArrayRef(size_data, size_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__embedding_bag(tensor *out__, tensor weight, tensor indices, tensor offsets, int scale_grad_by_freq, int64_t mode, int sparse, tensor per_sample_weights, int include_last_offset, int64_t padding_idx) {
-  PROTECT(
-    auto outputs__ = torch::_embedding_bag(*weight, *indices, *offsets, (bool)scale_grad_by_freq, mode, (bool)sparse, (per_sample_weights ? *per_sample_weights : torch::Tensor()), (bool)include_last_offset, padding_idx);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-    out__[3] = new torch::Tensor(std::get<3>(outputs__));
-  )
-}
-
-void atg__embedding_bag_backward(tensor *out__, tensor grad, tensor indices, tensor offsets, tensor offset2bag, tensor bag_size, tensor maximum_indices, int64_t num_weights, int scale_grad_by_freq, int64_t mode, int sparse, tensor per_sample_weights, int64_t padding_idx) {
-  PROTECT(
-    auto outputs__ = torch::_embedding_bag_backward(*grad, *indices, *offsets, *offset2bag, *bag_size, *maximum_indices, num_weights, (bool)scale_grad_by_freq, mode, (bool)sparse, (per_sample_weights ? *per_sample_weights : torch::Tensor()), padding_idx);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__embedding_bag_dense_backward(tensor *out__, tensor grad, tensor indices, tensor offset2bag, tensor bag_size, tensor maximum_indices, int64_t num_weights, int scale_grad_by_freq, int64_t mode, tensor per_sample_weights, int64_t padding_idx) {
-  PROTECT(
-    auto outputs__ = torch::_embedding_bag_dense_backward(*grad, *indices, *offset2bag, *bag_size, *maximum_indices, num_weights, (bool)scale_grad_by_freq, mode, (per_sample_weights ? *per_sample_weights : torch::Tensor()), padding_idx);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__embedding_bag_dense_backward_out(tensor *out__, tensor out, tensor grad, tensor indices, tensor offset2bag, tensor bag_size, tensor maximum_indices, int64_t num_weights, int scale_grad_by_freq, int64_t mode, tensor per_sample_weights, int64_t padding_idx) {
-  PROTECT(
-    auto outputs__ = torch::_embedding_bag_dense_backward_out(*out, *grad, *indices, *offset2bag, *bag_size, *maximum_indices, num_weights, (bool)scale_grad_by_freq, mode, (per_sample_weights ? *per_sample_weights : torch::Tensor()), padding_idx);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__embedding_bag_forward_only(tensor *out__, tensor weight, tensor indices, tensor offsets, int scale_grad_by_freq, int64_t mode, int sparse, tensor per_sample_weights, int include_last_offset, int64_t padding_idx) {
-  PROTECT(
-    auto outputs__ = torch::_embedding_bag_forward_only(*weight, *indices, *offsets, (bool)scale_grad_by_freq, mode, (bool)sparse, (per_sample_weights ? *per_sample_weights : torch::Tensor()), (bool)include_last_offset, padding_idx);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-    out__[3] = new torch::Tensor(std::get<3>(outputs__));
-  )
-}
-
-void atg__embedding_bag_forward_only_out(tensor *out__, tensor out0, tensor out1, tensor out2, tensor out3, tensor weight, tensor indices, tensor offsets, int scale_grad_by_freq, int64_t mode, int sparse, tensor per_sample_weights, int include_last_offset, int64_t padding_idx) {
-  PROTECT(
-    auto outputs__ = torch::_embedding_bag_forward_only_out(*out0, *out1, *out2, *out3, *weight, *indices, *offsets, (bool)scale_grad_by_freq, mode, (bool)sparse, (per_sample_weights ? *per_sample_weights : torch::Tensor()), (bool)include_last_offset, padding_idx);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-    out__[3] = new torch::Tensor(std::get<3>(outputs__));
-  )
-}
-
-void atg__embedding_bag_out(tensor *out__, tensor out0, tensor out1, tensor out2, tensor out3, tensor weight, tensor indices, tensor offsets, int scale_grad_by_freq, int64_t mode, int sparse, tensor per_sample_weights, int include_last_offset, int64_t padding_idx) {
-  PROTECT(
-    auto outputs__ = torch::_embedding_bag_out(*out0, *out1, *out2, *out3, *weight, *indices, *offsets, (bool)scale_grad_by_freq, mode, (bool)sparse, (per_sample_weights ? *per_sample_weights : torch::Tensor()), (bool)include_last_offset, padding_idx);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-    out__[3] = new torch::Tensor(std::get<3>(outputs__));
-  )
-}
-
-void atg__embedding_bag_per_sample_weights_backward(tensor *out__, tensor grad, tensor weight, tensor indices, tensor offsets, tensor offset2bag, int64_t mode, int64_t padding_idx) {
-  PROTECT(
-    auto outputs__ = torch::_embedding_bag_per_sample_weights_backward(*grad, *weight, *indices, *offsets, *offset2bag, mode, padding_idx);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__embedding_bag_per_sample_weights_backward_out(tensor *out__, tensor out, tensor grad, tensor weight, tensor indices, tensor offsets, tensor offset2bag, int64_t mode, int64_t padding_idx) {
-  PROTECT(
-    auto outputs__ = torch::_embedding_bag_per_sample_weights_backward_out(*out, *grad, *weight, *indices, *offsets, *offset2bag, mode, padding_idx);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__embedding_bag_sparse_backward(tensor *out__, tensor grad, tensor indices, tensor offsets, tensor offset2bag, tensor bag_size, int64_t num_weights, int scale_grad_by_freq, int64_t mode, tensor per_sample_weights, int64_t padding_idx) {
-  PROTECT(
-    auto outputs__ = torch::_embedding_bag_sparse_backward(*grad, *indices, *offsets, *offset2bag, *bag_size, num_weights, (bool)scale_grad_by_freq, mode, (per_sample_weights ? *per_sample_weights : torch::Tensor()), padding_idx);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__empty_affine_quantized(tensor *out__, int64_t *size_data, int size_len, int options_kind, int options_device, double scale, int64_t zero_point) {
-  PROTECT(
-    auto outputs__ = torch::_empty_affine_quantized(torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)), scale, zero_point);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__empty_affine_quantized_out(tensor *out__, tensor out, int64_t *size_data, int size_len, double scale, int64_t zero_point) {
-  PROTECT(
-    auto outputs__ = torch::_empty_affine_quantized_out(*out, torch::IntArrayRef(size_data, size_len), scale, zero_point);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__empty_per_channel_affine_quantized(tensor *out__, int64_t *size_data, int size_len, tensor scales, tensor zero_points, int64_t axis, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::_empty_per_channel_affine_quantized(torch::IntArrayRef(size_data, size_len), *scales, *zero_points, axis, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__empty_per_channel_affine_quantized_out(tensor *out__, tensor out, int64_t *size_data, int size_len, tensor scales, tensor zero_points, int64_t axis) {
-  PROTECT(
-    auto outputs__ = torch::_empty_per_channel_affine_quantized_out(*out, torch::IntArrayRef(size_data, size_len), *scales, *zero_points, axis);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__euclidean_dist(tensor *out__, tensor x1, tensor x2) {
-  PROTECT(
-    auto outputs__ = torch::_euclidean_dist(*x1, *x2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__euclidean_dist_out(tensor *out__, tensor out, tensor x1, tensor x2) {
-  PROTECT(
-    auto outputs__ = torch::_euclidean_dist_out(*out, *x1, *x2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__fake_quantize_learnable_per_channel_affine(tensor *out__, tensor self, tensor scale, tensor zero_point, int64_t axis, int64_t quant_min, int64_t quant_max, double grad_factor) {
-  PROTECT(
-    auto outputs__ = torch::_fake_quantize_learnable_per_channel_affine(*self, *scale, *zero_point, axis, quant_min, quant_max, grad_factor);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__fake_quantize_learnable_per_channel_affine_backward(tensor *out__, tensor grad, tensor self, tensor scale, tensor zero_point, int64_t axis, int64_t quant_min, int64_t quant_max, double grad_factor) {
-  PROTECT(
-    auto outputs__ = torch::_fake_quantize_learnable_per_channel_affine_backward(*grad, *self, *scale, *zero_point, axis, quant_min, quant_max, grad_factor);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg__fake_quantize_learnable_per_channel_affine_out(tensor *out__, tensor out, tensor self, tensor scale, tensor zero_point, int64_t axis, int64_t quant_min, int64_t quant_max, double grad_factor) {
-  PROTECT(
-    auto outputs__ = torch::_fake_quantize_learnable_per_channel_affine_out(*out, *self, *scale, *zero_point, axis, quant_min, quant_max, grad_factor);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__fake_quantize_learnable_per_tensor_affine(tensor *out__, tensor self, tensor scale, tensor zero_point, int64_t quant_min, int64_t quant_max, double grad_factor) {
-  PROTECT(
-    auto outputs__ = torch::_fake_quantize_learnable_per_tensor_affine(*self, *scale, *zero_point, quant_min, quant_max, grad_factor);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__fake_quantize_learnable_per_tensor_affine_backward(tensor *out__, tensor grad, tensor self, tensor scale, tensor zero_point, int64_t quant_min, int64_t quant_max, double grad_factor) {
-  PROTECT(
-    auto outputs__ = torch::_fake_quantize_learnable_per_tensor_affine_backward(*grad, *self, *scale, *zero_point, quant_min, quant_max, grad_factor);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg__fake_quantize_learnable_per_tensor_affine_out(tensor *out__, tensor out, tensor self, tensor scale, tensor zero_point, int64_t quant_min, int64_t quant_max, double grad_factor) {
-  PROTECT(
-    auto outputs__ = torch::_fake_quantize_learnable_per_tensor_affine_out(*out, *self, *scale, *zero_point, quant_min, quant_max, grad_factor);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__fake_quantize_per_tensor_affine_cachemask_tensor_qparams(tensor *out__, tensor self, tensor scale, tensor zero_point, tensor fake_quant_enabled, int64_t quant_min, int64_t quant_max) {
-  PROTECT(
-    auto outputs__ = torch::_fake_quantize_per_tensor_affine_cachemask_tensor_qparams(*self, *scale, *zero_point, *fake_quant_enabled, quant_min, quant_max);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg__fake_quantize_per_tensor_affine_cachemask_tensor_qparams_out(tensor *out__, tensor out0, tensor out1, tensor self, tensor scale, tensor zero_point, tensor fake_quant_enabled, int64_t quant_min, int64_t quant_max) {
-  PROTECT(
-    auto outputs__ = torch::_fake_quantize_per_tensor_affine_cachemask_tensor_qparams_out(*out0, *out1, *self, *scale, *zero_point, *fake_quant_enabled, quant_min, quant_max);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg__fft_c2c(tensor *out__, tensor self, int64_t *dim_data, int dim_len, int64_t normalization, int forward) {
-  PROTECT(
-    auto outputs__ = torch::_fft_c2c(*self, torch::IntArrayRef(dim_data, dim_len), normalization, (bool)forward);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__fft_c2c_out(tensor *out__, tensor out, tensor self, int64_t *dim_data, int dim_len, int64_t normalization, int forward) {
-  PROTECT(
-    auto outputs__ = torch::_fft_c2c_out(*out, *self, torch::IntArrayRef(dim_data, dim_len), normalization, (bool)forward);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__fft_c2r(tensor *out__, tensor self, int64_t *dim_data, int dim_len, int64_t normalization, int64_t last_dim_size) {
-  PROTECT(
-    auto outputs__ = torch::_fft_c2r(*self, torch::IntArrayRef(dim_data, dim_len), normalization, last_dim_size);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__fft_c2r_out(tensor *out__, tensor out, tensor self, int64_t *dim_data, int dim_len, int64_t normalization, int64_t last_dim_size) {
-  PROTECT(
-    auto outputs__ = torch::_fft_c2r_out(*out, *self, torch::IntArrayRef(dim_data, dim_len), normalization, last_dim_size);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__fft_r2c(tensor *out__, tensor self, int64_t *dim_data, int dim_len, int64_t normalization, int onesided) {
-  PROTECT(
-    auto outputs__ = torch::_fft_r2c(*self, torch::IntArrayRef(dim_data, dim_len), normalization, (bool)onesided);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__fft_r2c_out(tensor *out__, tensor out, tensor self, int64_t *dim_data, int dim_len, int64_t normalization, int onesided) {
-  PROTECT(
-    auto outputs__ = torch::_fft_r2c_out(*out, *self, torch::IntArrayRef(dim_data, dim_len), normalization, (bool)onesided);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__fill_mem_eff_dropout_mask_(tensor *out__, tensor self, double dropout_p, int64_t seed, int64_t offset) {
-  PROTECT(
-    auto outputs__ = torch::_fill_mem_eff_dropout_mask_(*self, dropout_p, seed, offset);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__flash_attention_backward(tensor *out__, tensor grad_out, tensor query, tensor key, tensor value, tensor out, tensor logsumexp, tensor cum_seq_q, tensor cum_seq_k, int64_t max_q, int64_t max_k, double dropout_p, int is_causal, tensor philox_seed, tensor philox_offset, double scale_v, int scale_null) {
-  PROTECT(
-    auto outputs__ = torch::_flash_attention_backward(*grad_out, *query, *key, *value, *out, *logsumexp, *cum_seq_q, *cum_seq_k, max_q, max_k, dropout_p, (bool)is_causal, *philox_seed, *philox_offset, scale_null ? c10::nullopt : c10::optional<double>(scale_v));
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg__foobar(tensor *out__, tensor self, int arg1, int arg2, int arg3) {
-  PROTECT(
-    auto outputs__ = torch::_foobar(*self, (bool)arg1, (bool)arg2, (bool)arg3);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__foobar_out(tensor *out__, tensor out, tensor self, int arg1, int arg2, int arg3) {
-  PROTECT(
-    auto outputs__ = torch::_foobar_out(*out, *self, (bool)arg1, (bool)arg2, (bool)arg3);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__functional_assert_async(tensor *out__, tensor self, char * assert_msg, tensor dep_token) {
-  PROTECT(
-    auto outputs__ = torch::_functional_assert_async(*self, std::string(assert_msg), *dep_token);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__functional_sym_constrain_range(tensor *out__, scalar size, int64_t min_v, int min_null, int64_t max_v, int max_null, tensor dep_token) {
-  PROTECT(
-    auto outputs__ = torch::_functional_sym_constrain_range(*size, min_null ? c10::nullopt : c10::optional<int64_t>(min_v), max_null ? c10::nullopt : c10::optional<int64_t>(max_v), *dep_token);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__functional_sym_constrain_range_for_size(tensor *out__, scalar size, int64_t min_v, int min_null, int64_t max_v, int max_null, tensor dep_token) {
-  PROTECT(
-    auto outputs__ = torch::_functional_sym_constrain_range_for_size(*size, min_null ? c10::nullopt : c10::optional<int64_t>(min_v), max_null ? c10::nullopt : c10::optional<int64_t>(max_v), *dep_token);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__fused_adam(tensor *out_data, int out_len, tensor *self_data, int self_len, tensor *grads_data, int grads_len, tensor *exp_avgs_data, int exp_avgs_len, tensor *exp_avg_sqs_data, int exp_avg_sqs_len, tensor *max_exp_avg_sqs_data, int max_exp_avg_sqs_len, tensor *state_steps_data, int state_steps_len, double lr, double beta1, double beta2, double weight_decay, double eps, int amsgrad, int maximize, tensor grad_scale, tensor found_inf) {
-  PROTECT(
-    torch::_fused_adam_out(of_carray_tensor(out_data, out_len), of_carray_tensor(self_data, self_len), of_carray_tensor(grads_data, grads_len), of_carray_tensor(exp_avgs_data, exp_avgs_len), of_carray_tensor(exp_avg_sqs_data, exp_avg_sqs_len), of_carray_tensor(max_exp_avg_sqs_data, max_exp_avg_sqs_len), of_carray_tensor(state_steps_data, state_steps_len), lr, beta1, beta2, weight_decay, eps, (bool)amsgrad, (bool)maximize, (grad_scale ? *grad_scale : torch::Tensor()), (found_inf ? *found_inf : torch::Tensor()));
-  )
-}
-
-void atg__fused_adam_(tensor *self_data, int self_len, tensor *grads_data, int grads_len, tensor *exp_avgs_data, int exp_avgs_len, tensor *exp_avg_sqs_data, int exp_avg_sqs_len, tensor *max_exp_avg_sqs_data, int max_exp_avg_sqs_len, tensor *state_steps_data, int state_steps_len, double lr, double beta1, double beta2, double weight_decay, double eps, int amsgrad, int maximize, tensor grad_scale, tensor found_inf) {
-  PROTECT(
-    torch::_fused_adam_(of_carray_tensor(self_data, self_len), of_carray_tensor(grads_data, grads_len), of_carray_tensor(exp_avgs_data, exp_avgs_len), of_carray_tensor(exp_avg_sqs_data, exp_avg_sqs_len), of_carray_tensor(max_exp_avg_sqs_data, max_exp_avg_sqs_len), of_carray_tensor(state_steps_data, state_steps_len), lr, beta1, beta2, weight_decay, eps, (bool)amsgrad, (bool)maximize, (grad_scale ? *grad_scale : torch::Tensor()), (found_inf ? *found_inf : torch::Tensor()));
-  )
-}
-
-void atg__fused_adam_tensor_lr_(tensor *self_data, int self_len, tensor *grads_data, int grads_len, tensor *exp_avgs_data, int exp_avgs_len, tensor *exp_avg_sqs_data, int exp_avg_sqs_len, tensor *max_exp_avg_sqs_data, int max_exp_avg_sqs_len, tensor *state_steps_data, int state_steps_len, tensor lr, double beta1, double beta2, double weight_decay, double eps, int amsgrad, int maximize, tensor grad_scale, tensor found_inf) {
-  PROTECT(
-    torch::_fused_adam_(of_carray_tensor(self_data, self_len), of_carray_tensor(grads_data, grads_len), of_carray_tensor(exp_avgs_data, exp_avgs_len), of_carray_tensor(exp_avg_sqs_data, exp_avg_sqs_len), of_carray_tensor(max_exp_avg_sqs_data, max_exp_avg_sqs_len), of_carray_tensor(state_steps_data, state_steps_len), *lr, beta1, beta2, weight_decay, eps, (bool)amsgrad, (bool)maximize, (grad_scale ? *grad_scale : torch::Tensor()), (found_inf ? *found_inf : torch::Tensor()));
-  )
-}
-
-void atg__fused_adam_tensor_lr_out(tensor *out_data, int out_len, tensor *self_data, int self_len, tensor *grads_data, int grads_len, tensor *exp_avgs_data, int exp_avgs_len, tensor *exp_avg_sqs_data, int exp_avg_sqs_len, tensor *max_exp_avg_sqs_data, int max_exp_avg_sqs_len, tensor *state_steps_data, int state_steps_len, tensor lr, double beta1, double beta2, double weight_decay, double eps, int amsgrad, int maximize, tensor grad_scale, tensor found_inf) {
-  PROTECT(
-    torch::_fused_adam_out(of_carray_tensor(out_data, out_len), of_carray_tensor(self_data, self_len), of_carray_tensor(grads_data, grads_len), of_carray_tensor(exp_avgs_data, exp_avgs_len), of_carray_tensor(exp_avg_sqs_data, exp_avg_sqs_len), of_carray_tensor(max_exp_avg_sqs_data, max_exp_avg_sqs_len), of_carray_tensor(state_steps_data, state_steps_len), *lr, beta1, beta2, weight_decay, eps, (bool)amsgrad, (bool)maximize, (grad_scale ? *grad_scale : torch::Tensor()), (found_inf ? *found_inf : torch::Tensor()));
-  )
-}
-
-void atg__fused_adamw(tensor *out_data, int out_len, tensor *self_data, int self_len, tensor *grads_data, int grads_len, tensor *exp_avgs_data, int exp_avgs_len, tensor *exp_avg_sqs_data, int exp_avg_sqs_len, tensor *max_exp_avg_sqs_data, int max_exp_avg_sqs_len, tensor *state_steps_data, int state_steps_len, double lr, double beta1, double beta2, double weight_decay, double eps, int amsgrad, int maximize, tensor grad_scale, tensor found_inf) {
-  PROTECT(
-    torch::_fused_adamw_out(of_carray_tensor(out_data, out_len), of_carray_tensor(self_data, self_len), of_carray_tensor(grads_data, grads_len), of_carray_tensor(exp_avgs_data, exp_avgs_len), of_carray_tensor(exp_avg_sqs_data, exp_avg_sqs_len), of_carray_tensor(max_exp_avg_sqs_data, max_exp_avg_sqs_len), of_carray_tensor(state_steps_data, state_steps_len), lr, beta1, beta2, weight_decay, eps, (bool)amsgrad, (bool)maximize, (grad_scale ? *grad_scale : torch::Tensor()), (found_inf ? *found_inf : torch::Tensor()));
-  )
-}
-
-void atg__fused_adamw_(tensor *self_data, int self_len, tensor *grads_data, int grads_len, tensor *exp_avgs_data, int exp_avgs_len, tensor *exp_avg_sqs_data, int exp_avg_sqs_len, tensor *max_exp_avg_sqs_data, int max_exp_avg_sqs_len, tensor *state_steps_data, int state_steps_len, double lr, double beta1, double beta2, double weight_decay, double eps, int amsgrad, int maximize, tensor grad_scale, tensor found_inf) {
-  PROTECT(
-    torch::_fused_adamw_(of_carray_tensor(self_data, self_len), of_carray_tensor(grads_data, grads_len), of_carray_tensor(exp_avgs_data, exp_avgs_len), of_carray_tensor(exp_avg_sqs_data, exp_avg_sqs_len), of_carray_tensor(max_exp_avg_sqs_data, max_exp_avg_sqs_len), of_carray_tensor(state_steps_data, state_steps_len), lr, beta1, beta2, weight_decay, eps, (bool)amsgrad, (bool)maximize, (grad_scale ? *grad_scale : torch::Tensor()), (found_inf ? *found_inf : torch::Tensor()));
-  )
-}
-
-void atg__fused_adamw_tensor_lr_(tensor *self_data, int self_len, tensor *grads_data, int grads_len, tensor *exp_avgs_data, int exp_avgs_len, tensor *exp_avg_sqs_data, int exp_avg_sqs_len, tensor *max_exp_avg_sqs_data, int max_exp_avg_sqs_len, tensor *state_steps_data, int state_steps_len, tensor lr, double beta1, double beta2, double weight_decay, double eps, int amsgrad, int maximize, tensor grad_scale, tensor found_inf) {
-  PROTECT(
-    torch::_fused_adamw_(of_carray_tensor(self_data, self_len), of_carray_tensor(grads_data, grads_len), of_carray_tensor(exp_avgs_data, exp_avgs_len), of_carray_tensor(exp_avg_sqs_data, exp_avg_sqs_len), of_carray_tensor(max_exp_avg_sqs_data, max_exp_avg_sqs_len), of_carray_tensor(state_steps_data, state_steps_len), *lr, beta1, beta2, weight_decay, eps, (bool)amsgrad, (bool)maximize, (grad_scale ? *grad_scale : torch::Tensor()), (found_inf ? *found_inf : torch::Tensor()));
-  )
-}
-
-void atg__fused_adamw_tensor_lr_out(tensor *out_data, int out_len, tensor *self_data, int self_len, tensor *grads_data, int grads_len, tensor *exp_avgs_data, int exp_avgs_len, tensor *exp_avg_sqs_data, int exp_avg_sqs_len, tensor *max_exp_avg_sqs_data, int max_exp_avg_sqs_len, tensor *state_steps_data, int state_steps_len, tensor lr, double beta1, double beta2, double weight_decay, double eps, int amsgrad, int maximize, tensor grad_scale, tensor found_inf) {
-  PROTECT(
-    torch::_fused_adamw_out(of_carray_tensor(out_data, out_len), of_carray_tensor(self_data, self_len), of_carray_tensor(grads_data, grads_len), of_carray_tensor(exp_avgs_data, exp_avgs_len), of_carray_tensor(exp_avg_sqs_data, exp_avg_sqs_len), of_carray_tensor(max_exp_avg_sqs_data, max_exp_avg_sqs_len), of_carray_tensor(state_steps_data, state_steps_len), *lr, beta1, beta2, weight_decay, eps, (bool)amsgrad, (bool)maximize, (grad_scale ? *grad_scale : torch::Tensor()), (found_inf ? *found_inf : torch::Tensor()));
-  )
-}
-
-void atg__fused_dropout(tensor *out__, tensor self, double p) {
-  PROTECT(
-    auto outputs__ = torch::_fused_dropout(*self, p);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg__fused_dropout_out(tensor *out__, tensor out0, tensor out1, tensor self, double p) {
-  PROTECT(
-    auto outputs__ = torch::_fused_dropout_out(*out0, *out1, *self, p);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg__fused_moving_avg_obs_fq_helper(tensor *out__, tensor self, tensor observer_on, tensor fake_quant_on, tensor running_min, tensor running_max, tensor scale, tensor zero_point, double averaging_const, int64_t quant_min, int64_t quant_max, int64_t ch_axis, int per_row_fake_quant, int symmetric_quant) {
-  PROTECT(
-    auto outputs__ = torch::_fused_moving_avg_obs_fq_helper(*self, *observer_on, *fake_quant_on, *running_min, *running_max, *scale, *zero_point, averaging_const, quant_min, quant_max, ch_axis, (bool)per_row_fake_quant, (bool)symmetric_quant);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg__fused_moving_avg_obs_fq_helper_functional(tensor *out__, tensor self, tensor observer_on, tensor fake_quant_on, tensor running_min, tensor running_max, tensor scale, tensor zero_point, double averaging_const, int64_t quant_min, int64_t quant_max, int64_t ch_axis, int per_row_fake_quant, int symmetric_quant) {
-  PROTECT(
-    auto outputs__ = torch::_fused_moving_avg_obs_fq_helper_functional(*self, *observer_on, *fake_quant_on, *running_min, *running_max, *scale, *zero_point, averaging_const, quant_min, quant_max, ch_axis, (bool)per_row_fake_quant, (bool)symmetric_quant);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-    out__[3] = new torch::Tensor(std::get<3>(outputs__));
-    out__[4] = new torch::Tensor(std::get<4>(outputs__));
-    out__[5] = new torch::Tensor(std::get<5>(outputs__));
-  )
-}
-
-void atg__fused_moving_avg_obs_fq_helper_out(tensor *out__, tensor out0, tensor out1, tensor self, tensor observer_on, tensor fake_quant_on, tensor running_min, tensor running_max, tensor scale, tensor zero_point, double averaging_const, int64_t quant_min, int64_t quant_max, int64_t ch_axis, int per_row_fake_quant, int symmetric_quant) {
-  PROTECT(
-    auto outputs__ = torch::_fused_moving_avg_obs_fq_helper_out(*out0, *out1, *self, *observer_on, *fake_quant_on, *running_min, *running_max, *scale, *zero_point, averaging_const, quant_min, quant_max, ch_axis, (bool)per_row_fake_quant, (bool)symmetric_quant);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-int64_t atg__fused_sdp_choice(tensor query, tensor key, tensor value, tensor attn_mask, double dropout_p, int is_causal, double scale_v, int scale_null) {
-  PROTECT(
-    return torch::_fused_sdp_choice(*query, *key, *value, (attn_mask ? *attn_mask : torch::Tensor()), dropout_p, (bool)is_causal, scale_null ? c10::nullopt : c10::optional<double>(scale_v));
-  )
-  return 0;
-}
-
-void atg__fw_primal(tensor *out__, tensor self, int64_t level) {
-  PROTECT(
-    auto outputs__ = self->_fw_primal(level);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__fw_primal_copy(tensor *out__, tensor self, int64_t level) {
-  PROTECT(
-    auto outputs__ = torch::_fw_primal_copy(*self, level);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__fw_primal_copy_out(tensor *out__, tensor out, tensor self, int64_t level) {
-  PROTECT(
-    auto outputs__ = torch::_fw_primal_copy_out(*out, *self, level);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__gather_sparse_backward(tensor *out__, tensor self, int64_t dim, tensor index, tensor grad) {
-  PROTECT(
-    auto outputs__ = torch::_gather_sparse_backward(*self, dim, *index, *grad);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__grid_sampler_2d_cpu_fallback(tensor *out__, tensor input, tensor grid, int64_t interpolation_mode, int64_t padding_mode, int align_corners) {
-  PROTECT(
-    auto outputs__ = torch::_grid_sampler_2d_cpu_fallback(*input, *grid, interpolation_mode, padding_mode, (bool)align_corners);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__grid_sampler_2d_cpu_fallback_backward(tensor *out__, tensor grad_output, tensor input, tensor grid, int64_t interpolation_mode, int64_t padding_mode, int align_corners) {
-  PROTECT(
-    auto outputs__ = torch::_grid_sampler_2d_cpu_fallback_backward(*grad_output, *input, *grid, interpolation_mode, padding_mode, (bool)align_corners);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg__grid_sampler_2d_cpu_fallback_out(tensor *out__, tensor out, tensor input, tensor grid, int64_t interpolation_mode, int64_t padding_mode, int align_corners) {
-  PROTECT(
-    auto outputs__ = torch::_grid_sampler_2d_cpu_fallback_out(*out, *input, *grid, interpolation_mode, padding_mode, (bool)align_corners);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-int atg__has_compatible_shallow_copy_type(tensor self, tensor from) {
-  PROTECT(
-    return torch::_has_compatible_shallow_copy_type(*self, *from);
-  )
-  return 0;
-}
-
-int atg__has_same_storage_numel(tensor self, tensor other) {
-  PROTECT(
-    return torch::_has_same_storage_numel(*self, *other);
-  )
-  return 0;
-}
-
-tensor *atg__histogramdd_bin_edges(tensor self, int64_t *bins_data, int bins_len, double *range_data, int range_len, tensor weight, int density) {
-  PROTECT(
-    auto outputs__ = torch::_histogramdd_bin_edges(*self, torch::IntArrayRef(bins_data, bins_len), at::ArrayRef<double>(range_data, range_len), (weight ? *weight : torch::Tensor()), (bool)density);
-    int sz = outputs__.size();
-    torch::Tensor **out__ = (torch::Tensor**)malloc((sz + 1) * sizeof(torch::Tensor*));
-    for (int i = 0; i < sz; ++i)
-      out__[i] = new torch::Tensor(outputs__[i]);
-    out__[sz] = nullptr;
-    return out__;
-  )
-}
-
-void atg__histogramdd_bin_edges_out(tensor *out_data, int out_len, tensor self, int64_t *bins_data, int bins_len, double *range_data, int range_len, tensor weight, int density) {
-  PROTECT(
-    torch::_histogramdd_bin_edges_out(of_carray_tensor(out_data, out_len), *self, torch::IntArrayRef(bins_data, bins_len), at::ArrayRef<double>(range_data, range_len), (weight ? *weight : torch::Tensor()), (bool)density);
-  )
-}
-
-void atg__histogramdd_from_bin_cts(tensor *out__, tensor self, int64_t *bins_data, int bins_len, double *range_data, int range_len, tensor weight, int density) {
-  PROTECT(
-    auto outputs__ = torch::_histogramdd_from_bin_cts(*self, torch::IntArrayRef(bins_data, bins_len), at::ArrayRef<double>(range_data, range_len), (weight ? *weight : torch::Tensor()), (bool)density);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__histogramdd_from_bin_cts_out(tensor *out__, tensor out, tensor self, int64_t *bins_data, int bins_len, double *range_data, int range_len, tensor weight, int density) {
-  PROTECT(
-    auto outputs__ = torch::_histogramdd_from_bin_cts_out(*out, *self, torch::IntArrayRef(bins_data, bins_len), at::ArrayRef<double>(range_data, range_len), (weight ? *weight : torch::Tensor()), (bool)density);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__histogramdd_from_bin_tensors(tensor *out__, tensor self, tensor *bins_data, int bins_len, tensor weight, int density) {
-  PROTECT(
-    auto outputs__ = torch::_histogramdd_from_bin_tensors(*self, of_carray_tensor(bins_data, bins_len), (weight ? *weight : torch::Tensor()), (bool)density);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__histogramdd_from_bin_tensors_out(tensor *out__, tensor out, tensor self, tensor *bins_data, int bins_len, tensor weight, int density) {
-  PROTECT(
-    auto outputs__ = torch::_histogramdd_from_bin_tensors_out(*out, *self, of_carray_tensor(bins_data, bins_len), (weight ? *weight : torch::Tensor()), (bool)density);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__index_put_impl(tensor *out__, tensor self, tensor *indices_data, int indices_len, tensor values, int accumulate, int unsafe) {
-  PROTECT(
-    auto outputs__ = torch::_index_put_impl(*self, of_carray_tensor_opt(indices_data, indices_len), *values, (bool)accumulate, (bool)unsafe);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__index_put_impl_(tensor *out__, tensor self, tensor *indices_data, int indices_len, tensor values, int accumulate, int unsafe) {
-  PROTECT(
-    auto outputs__ = torch::_index_put_impl_(*self, of_carray_tensor_opt(indices_data, indices_len), *values, (bool)accumulate, (bool)unsafe);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__index_put_impl_out(tensor *out__, tensor out, tensor self, tensor *indices_data, int indices_len, tensor values, int accumulate, int unsafe) {
-  PROTECT(
-    auto outputs__ = torch::_index_put_impl_out(*out, *self, of_carray_tensor_opt(indices_data, indices_len), *values, (bool)accumulate, (bool)unsafe);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__indices(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = self->_indices();
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__indices_copy(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::_indices_copy(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__indices_copy_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::_indices_copy_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__int_mm(tensor *out__, tensor self, tensor mat2) {
-  PROTECT(
-    auto outputs__ = torch::_int_mm(*self, *mat2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__int_mm_out(tensor *out__, tensor out, tensor self, tensor mat2) {
-  PROTECT(
-    auto outputs__ = torch::_int_mm_out(*out, *self, *mat2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__is_all_true(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::_is_all_true(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__is_any_true(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::_is_any_true(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-int atg__is_zerotensor(tensor self) {
-  PROTECT(
-    return torch::_is_zerotensor(*self);
-  )
-  return 0;
-}
-
-void atg__linalg_check_errors(tensor info, char * api_name, int is_matrix) {
-  PROTECT(
-    torch::_linalg_check_errors(*info, std::string(api_name), (bool)is_matrix);
-  )
-}
-
-void atg__linalg_det(tensor *out__, tensor A) {
-  PROTECT(
-    auto outputs__ = torch::_linalg_det(*A);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg__linalg_det_result(tensor *out__, tensor result, tensor LU, tensor pivots, tensor A) {
-  PROTECT(
-    auto outputs__ = torch::_linalg_det_out(*result, *LU, *pivots, *A);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg__linalg_eigh(tensor *out__, tensor A, char * UPLO, int compute_v) {
-  PROTECT(
-    auto outputs__ = torch::_linalg_eigh(*A, std::string(UPLO), (bool)compute_v);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg__linalg_eigh_eigenvalues(tensor *out__, tensor eigenvalues, tensor eigenvectors, tensor A, char * UPLO, int compute_v) {
-  PROTECT(
-    auto outputs__ = torch::_linalg_eigh_out(*eigenvalues, *eigenvectors, *A, std::string(UPLO), (bool)compute_v);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg__linalg_slogdet(tensor *out__, tensor A) {
-  PROTECT(
-    auto outputs__ = torch::_linalg_slogdet(*A);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-    out__[3] = new torch::Tensor(std::get<3>(outputs__));
-  )
-}
-
-void atg__linalg_slogdet_sign(tensor *out__, tensor sign, tensor logabsdet, tensor LU, tensor pivots, tensor A) {
-  PROTECT(
-    auto outputs__ = torch::_linalg_slogdet_out(*sign, *logabsdet, *LU, *pivots, *A);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-    out__[3] = new torch::Tensor(std::get<3>(outputs__));
-  )
-}
-
-void atg__linalg_solve_ex(tensor *out__, tensor A, tensor B, int left, int check_errors) {
-  PROTECT(
-    auto outputs__ = torch::_linalg_solve_ex(*A, *B, (bool)left, (bool)check_errors);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-    out__[3] = new torch::Tensor(std::get<3>(outputs__));
-  )
-}
-
-void atg__linalg_solve_ex_result(tensor *out__, tensor result, tensor LU, tensor pivots, tensor info, tensor A, tensor B, int left, int check_errors) {
-  PROTECT(
-    auto outputs__ = torch::_linalg_solve_ex_out(*result, *LU, *pivots, *info, *A, *B, (bool)left, (bool)check_errors);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-    out__[3] = new torch::Tensor(std::get<3>(outputs__));
-  )
-}
-
-void atg__linalg_svd(tensor *out__, tensor A, int full_matrices, int compute_uv, char * driver) {
-  PROTECT(
-    auto outputs__ = torch::_linalg_svd(*A, (bool)full_matrices, (bool)compute_uv, std::string(driver));
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg__linalg_svd_u(tensor *out__, tensor U, tensor S, tensor Vh, tensor A, int full_matrices, int compute_uv, char * driver) {
-  PROTECT(
-    auto outputs__ = torch::_linalg_svd_out(*U, *S, *Vh, *A, (bool)full_matrices, (bool)compute_uv, std::string(driver));
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg__log_softmax(tensor *out__, tensor self, int64_t dim, int half_to_float) {
-  PROTECT(
-    auto outputs__ = torch::_log_softmax(*self, dim, (bool)half_to_float);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__log_softmax_backward_data(tensor *out__, tensor grad_output, tensor output, int64_t dim, int input_dtype) {
-  PROTECT(
-    auto outputs__ = torch::_log_softmax_backward_data(*grad_output, *output, dim, torch::ScalarType(input_dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__log_softmax_backward_data_out(tensor *out__, tensor out, tensor grad_output, tensor output, int64_t dim, int input_dtype) {
-  PROTECT(
-    auto outputs__ = torch::_log_softmax_backward_data_out(*out, *grad_output, *output, dim, torch::ScalarType(input_dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__log_softmax_out(tensor *out__, tensor out, tensor self, int64_t dim, int half_to_float) {
-  PROTECT(
-    auto outputs__ = torch::_log_softmax_out(*out, *self, dim, (bool)half_to_float);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__logcumsumexp(tensor *out__, tensor self, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::_logcumsumexp(*self, dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__logcumsumexp_out(tensor *out__, tensor out, tensor self, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::_logcumsumexp_out(*out, *self, dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__lstm_mps(tensor *out__, tensor input, tensor *hx_data, int hx_len, tensor *params_data, int params_len, int has_biases, int64_t num_layers, double dropout, int train, int bidirectional, int batch_first) {
-  PROTECT(
-    auto outputs__ = torch::_lstm_mps(*input, of_carray_tensor(hx_data, hx_len), of_carray_tensor(params_data, params_len), (bool)has_biases, num_layers, dropout, (bool)train, (bool)bidirectional, (bool)batch_first);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-    out__[3] = new torch::Tensor(std::get<3>(outputs__));
-    out__[4] = new torch::Tensor(std::get<4>(outputs__));
-    out__[5] = new torch::Tensor(std::get<5>(outputs__));
-  )
-}
-
-void atg__lstm_mps_out(tensor *out__, tensor out0, tensor out1, tensor out2, tensor out3, tensor out4, tensor out5, tensor input, tensor *hx_data, int hx_len, tensor *params_data, int params_len, int has_biases, int64_t num_layers, double dropout, int train, int bidirectional, int batch_first) {
-  PROTECT(
-    auto outputs__ = torch::_lstm_mps_out(*out0, *out1, *out2, *out3, *out4, *out5, *input, of_carray_tensor(hx_data, hx_len), of_carray_tensor(params_data, params_len), (bool)has_biases, num_layers, dropout, (bool)train, (bool)bidirectional, (bool)batch_first);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-    out__[3] = new torch::Tensor(std::get<3>(outputs__));
-    out__[4] = new torch::Tensor(std::get<4>(outputs__));
-    out__[5] = new torch::Tensor(std::get<5>(outputs__));
-  )
-}
-
-void atg__lu_with_info(tensor *out__, tensor self, int pivot, int check_errors) {
-  PROTECT(
-    auto outputs__ = torch::_lu_with_info(*self, (bool)pivot, (bool)check_errors);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg__make_dep_token(tensor *out__, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::_make_dep_token(at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__make_dual(tensor *out__, tensor primal, tensor tangent, int64_t level) {
-  PROTECT(
-    auto outputs__ = torch::_make_dual(*primal, *tangent, level);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__make_dual_copy(tensor *out__, tensor primal, tensor tangent, int64_t level) {
-  PROTECT(
-    auto outputs__ = torch::_make_dual_copy(*primal, *tangent, level);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__make_dual_copy_out(tensor *out__, tensor out, tensor primal, tensor tangent, int64_t level) {
-  PROTECT(
-    auto outputs__ = torch::_make_dual_copy_out(*out, *primal, *tangent, level);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__make_per_channel_quantized_tensor(tensor *out__, tensor self, tensor scale, tensor zero_point, int64_t axis) {
-  PROTECT(
-    auto outputs__ = torch::_make_per_channel_quantized_tensor(*self, *scale, *zero_point, axis);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__make_per_channel_quantized_tensor_out(tensor *out__, tensor out, tensor self, tensor scale, tensor zero_point, int64_t axis) {
-  PROTECT(
-    auto outputs__ = torch::_make_per_channel_quantized_tensor_out(*out, *self, *scale, *zero_point, axis);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__make_per_tensor_quantized_tensor(tensor *out__, tensor self, double scale, int64_t zero_point) {
-  PROTECT(
-    auto outputs__ = torch::_make_per_tensor_quantized_tensor(*self, scale, zero_point);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__make_per_tensor_quantized_tensor_out(tensor *out__, tensor out, tensor self, double scale, int64_t zero_point) {
-  PROTECT(
-    auto outputs__ = torch::_make_per_tensor_quantized_tensor_out(*out, *self, scale, zero_point);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__masked_scale(tensor *out__, tensor self, tensor mask, double scale) {
-  PROTECT(
-    auto outputs__ = torch::_masked_scale(*self, *mask, scale);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__masked_scale_out(tensor *out__, tensor out, tensor self, tensor mask, double scale) {
-  PROTECT(
-    auto outputs__ = torch::_masked_scale_out(*out, *self, *mask, scale);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__masked_softmax(tensor *out__, tensor self, tensor mask, int64_t dim_v, int dim_null, int64_t mask_type_v, int mask_type_null) {
-  PROTECT(
-    auto outputs__ = torch::_masked_softmax(*self, *mask, dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v), mask_type_null ? c10::nullopt : c10::optional<int64_t>(mask_type_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__masked_softmax_backward(tensor *out__, tensor grad_output, tensor output, tensor mask, int64_t dim_v, int dim_null) {
-  PROTECT(
-    auto outputs__ = torch::_masked_softmax_backward(*grad_output, *output, *mask, dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__masked_softmax_backward_out(tensor *out__, tensor out, tensor grad_output, tensor output, tensor mask, int64_t dim_v, int dim_null) {
-  PROTECT(
-    auto outputs__ = torch::_masked_softmax_backward_out(*out, *grad_output, *output, *mask, dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__masked_softmax_out(tensor *out__, tensor out, tensor self, tensor mask, int64_t dim_v, int dim_null, int64_t mask_type_v, int mask_type_null) {
-  PROTECT(
-    auto outputs__ = torch::_masked_softmax_out(*out, *self, *mask, dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v), mask_type_null ? c10::nullopt : c10::optional<int64_t>(mask_type_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__mkldnn_reshape(tensor *out__, tensor self, int64_t *shape_data, int shape_len) {
-  PROTECT(
-    auto outputs__ = torch::_mkldnn_reshape(*self, torch::IntArrayRef(shape_data, shape_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__mkldnn_reshape_out(tensor *out__, tensor out, tensor self, int64_t *shape_data, int shape_len) {
-  PROTECT(
-    auto outputs__ = torch::_mkldnn_reshape_out(*out, *self, torch::IntArrayRef(shape_data, shape_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__mkldnn_transpose(tensor *out__, tensor self, int64_t dim0, int64_t dim1) {
-  PROTECT(
-    auto outputs__ = torch::_mkldnn_transpose(*self, dim0, dim1);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__mkldnn_transpose_(tensor *out__, tensor self, int64_t dim0, int64_t dim1) {
-  PROTECT(
-    auto outputs__ = torch::_mkldnn_transpose_(*self, dim0, dim1);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__mkldnn_transpose_out(tensor *out__, tensor out, tensor self, int64_t dim0, int64_t dim1) {
-  PROTECT(
-    auto outputs__ = torch::_mkldnn_transpose_out(*out, *self, dim0, dim1);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__mps_convolution(tensor *out__, tensor self, tensor weight, tensor bias, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups) {
-  PROTECT(
-    auto outputs__ = torch::_mps_convolution(*self, *weight, (bias ? *bias : torch::Tensor()), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(dilation_data, dilation_len), groups);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__mps_convolution_out(tensor *out__, tensor out, tensor self, tensor weight, tensor bias, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups) {
-  PROTECT(
-    auto outputs__ = torch::_mps_convolution_out(*out, *self, *weight, (bias ? *bias : torch::Tensor()), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(dilation_data, dilation_len), groups);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__mps_convolution_transpose(tensor *out__, tensor self, tensor weight, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups) {
-  PROTECT(
-    auto outputs__ = torch::_mps_convolution_transpose(*self, *weight, torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(output_padding_data, output_padding_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(dilation_data, dilation_len), groups);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__mps_convolution_transpose_out(tensor *out__, tensor out, tensor self, tensor weight, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups) {
-  PROTECT(
-    auto outputs__ = torch::_mps_convolution_transpose_out(*out, *self, *weight, torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(output_padding_data, output_padding_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(dilation_data, dilation_len), groups);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__native_batch_norm_legit(tensor *out__, tensor input, tensor weight, tensor bias, tensor running_mean, tensor running_var, int training, double momentum, double eps) {
-  PROTECT(
-    auto outputs__ = torch::_native_batch_norm_legit(*input, (weight ? *weight : torch::Tensor()), (bias ? *bias : torch::Tensor()), *running_mean, *running_var, (bool)training, momentum, eps);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg__native_batch_norm_legit_functional(tensor *out__, tensor input, tensor weight, tensor bias, tensor running_mean, tensor running_var, int training, double momentum, double eps) {
-  PROTECT(
-    auto outputs__ = torch::_native_batch_norm_legit_functional(*input, (weight ? *weight : torch::Tensor()), (bias ? *bias : torch::Tensor()), *running_mean, *running_var, (bool)training, momentum, eps);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-    out__[3] = new torch::Tensor(std::get<3>(outputs__));
-    out__[4] = new torch::Tensor(std::get<4>(outputs__));
-  )
-}
-
-void atg__native_batch_norm_legit_no_stats(tensor *out__, tensor input, tensor weight, tensor bias, int training, double momentum, double eps) {
-  PROTECT(
-    auto outputs__ = torch::_native_batch_norm_legit(*input, (weight ? *weight : torch::Tensor()), (bias ? *bias : torch::Tensor()), (bool)training, momentum, eps);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg__native_batch_norm_legit_no_stats_out(tensor *out__, tensor out, tensor save_mean, tensor save_invstd, tensor input, tensor weight, tensor bias, int training, double momentum, double eps) {
-  PROTECT(
-    auto outputs__ = torch::_native_batch_norm_legit_out(*out, *save_mean, *save_invstd, *input, (weight ? *weight : torch::Tensor()), (bias ? *bias : torch::Tensor()), (bool)training, momentum, eps);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg__native_batch_norm_legit_no_training(tensor *out__, tensor input, tensor weight, tensor bias, tensor running_mean, tensor running_var, double momentum, double eps) {
-  PROTECT(
-    auto outputs__ = torch::_native_batch_norm_legit_no_training(*input, (weight ? *weight : torch::Tensor()), (bias ? *bias : torch::Tensor()), *running_mean, *running_var, momentum, eps);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg__native_batch_norm_legit_no_training_out(tensor *out__, tensor out0, tensor out1, tensor out2, tensor input, tensor weight, tensor bias, tensor running_mean, tensor running_var, double momentum, double eps) {
-  PROTECT(
-    auto outputs__ = torch::_native_batch_norm_legit_no_training_out(*out0, *out1, *out2, *input, (weight ? *weight : torch::Tensor()), (bias ? *bias : torch::Tensor()), *running_mean, *running_var, momentum, eps);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg__native_batch_norm_legit_out(tensor *out__, tensor out, tensor save_mean, tensor save_invstd, tensor input, tensor weight, tensor bias, tensor running_mean, tensor running_var, int training, double momentum, double eps) {
-  PROTECT(
-    auto outputs__ = torch::_native_batch_norm_legit_out(*out, *save_mean, *save_invstd, *input, (weight ? *weight : torch::Tensor()), (bias ? *bias : torch::Tensor()), *running_mean, *running_var, (bool)training, momentum, eps);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg__native_multi_head_attention(tensor *out__, tensor query, tensor key, tensor value, int64_t embed_dim, int64_t num_head, tensor qkv_weight, tensor qkv_bias, tensor proj_weight, tensor proj_bias, tensor mask, int need_weights, int average_attn_weights, int64_t mask_type_v, int mask_type_null) {
-  PROTECT(
-    auto outputs__ = torch::_native_multi_head_attention(*query, *key, *value, embed_dim, num_head, *qkv_weight, *qkv_bias, *proj_weight, *proj_bias, (mask ? *mask : torch::Tensor()), (bool)need_weights, (bool)average_attn_weights, mask_type_null ? c10::nullopt : c10::optional<int64_t>(mask_type_v));
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg__native_multi_head_attention_out(tensor *out__, tensor out0, tensor out1, tensor query, tensor key, tensor value, int64_t embed_dim, int64_t num_head, tensor qkv_weight, tensor qkv_bias, tensor proj_weight, tensor proj_bias, tensor mask, int need_weights, int average_attn_weights, int64_t mask_type_v, int mask_type_null) {
-  PROTECT(
-    auto outputs__ = torch::_native_multi_head_attention_out(*out0, *out1, *query, *key, *value, embed_dim, num_head, *qkv_weight, *qkv_bias, *proj_weight, *proj_bias, (mask ? *mask : torch::Tensor()), (bool)need_weights, (bool)average_attn_weights, mask_type_null ? c10::nullopt : c10::optional<int64_t>(mask_type_v));
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg__neg_view(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::_neg_view(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__neg_view_copy(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::_neg_view_copy(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__neg_view_copy_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::_neg_view_copy_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__nested_from_padded(tensor *out__, tensor padded, tensor cpu_nested_shape_example, int fuse_transform_0213) {
-  PROTECT(
-    auto outputs__ = torch::_nested_from_padded(*padded, *cpu_nested_shape_example, (bool)fuse_transform_0213);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__nested_from_padded_and_nested_example(tensor *out__, tensor padded, tensor nt_example) {
-  PROTECT(
-    auto outputs__ = torch::_nested_from_padded_and_nested_example(*padded, *nt_example);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__nested_from_padded_and_nested_example_out(tensor *out__, tensor out, tensor padded, tensor nt_example) {
-  PROTECT(
-    auto outputs__ = torch::_nested_from_padded_and_nested_example_out(*out, *padded, *nt_example);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__nested_from_padded_out(tensor *out__, tensor out, tensor padded, tensor cpu_nested_shape_example, int fuse_transform_0213) {
-  PROTECT(
-    auto outputs__ = torch::_nested_from_padded_out(*out, *padded, *cpu_nested_shape_example, (bool)fuse_transform_0213);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__nested_select_backward(tensor *out__, tensor grad_output, tensor self, int64_t dim, int64_t index) {
-  PROTECT(
-    auto outputs__ = torch::_nested_select_backward(*grad_output, *self, dim, index);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__nested_sum_backward(tensor *out__, tensor grad, tensor self, int64_t *dim_data, int dim_len, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::_nested_sum_backward(*grad, *self, dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), (bool)keepdim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__nested_view_from_buffer(tensor *out__, tensor self, tensor nested_size, tensor nested_strides, tensor offsets) {
-  PROTECT(
-    auto outputs__ = torch::_nested_view_from_buffer(*self, *nested_size, *nested_strides, *offsets);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__nested_view_from_buffer_copy(tensor *out__, tensor self, tensor nested_size, tensor nested_strides, tensor offsets) {
-  PROTECT(
-    auto outputs__ = torch::_nested_view_from_buffer_copy(*self, *nested_size, *nested_strides, *offsets);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__nested_view_from_buffer_copy_out(tensor *out__, tensor out, tensor self, tensor nested_size, tensor nested_strides, tensor offsets) {
-  PROTECT(
-    auto outputs__ = torch::_nested_view_from_buffer_copy_out(*out, *self, *nested_size, *nested_strides, *offsets);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__new_zeros_with_same_feature_meta(tensor *out__, tensor self, tensor other, int64_t self_num_batch_dims) {
-  PROTECT(
-    auto outputs__ = torch::_new_zeros_with_same_feature_meta(*self, *other, self_num_batch_dims);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__new_zeros_with_same_feature_meta_out(tensor *out__, tensor out, tensor self, tensor other, int64_t self_num_batch_dims) {
-  PROTECT(
-    auto outputs__ = torch::_new_zeros_with_same_feature_meta_out(*out, *self, *other, self_num_batch_dims);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-int atg__nnpack_available() {
-  PROTECT(
-    return torch::_nnpack_available();
-  )
-  return 0;
-}
-
-void atg__nnpack_spatial_convolution(tensor *out__, tensor input, tensor weight, tensor bias, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len) {
-  PROTECT(
-    auto outputs__ = torch::_nnpack_spatial_convolution(*input, *weight, (bias ? *bias : torch::Tensor()), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(stride_data, stride_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__nnpack_spatial_convolution_out(tensor *out__, tensor out, tensor input, tensor weight, tensor bias, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len) {
-  PROTECT(
-    auto outputs__ = torch::_nnpack_spatial_convolution_out(*out, *input, *weight, (bias ? *bias : torch::Tensor()), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(stride_data, stride_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-int64_t atg__nnz(tensor self) {
-  PROTECT(
-    return self->_nnz();
-  )
-  return 0;
-}
-
-void atg__pack_padded_sequence(tensor *out__, tensor input, tensor lengths, int batch_first) {
-  PROTECT(
-    auto outputs__ = torch::_pack_padded_sequence(*input, *lengths, (bool)batch_first);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg__pack_padded_sequence_backward(tensor *out__, tensor grad, int64_t *input_size_data, int input_size_len, tensor batch_sizes, int batch_first) {
-  PROTECT(
-    auto outputs__ = torch::_pack_padded_sequence_backward(*grad, torch::IntArrayRef(input_size_data, input_size_len), *batch_sizes, (bool)batch_first);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__pack_padded_sequence_out(tensor *out__, tensor out0, tensor out1, tensor input, tensor lengths, int batch_first) {
-  PROTECT(
-    auto outputs__ = torch::_pack_padded_sequence_out(*out0, *out1, *input, *lengths, (bool)batch_first);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg__pad_circular(tensor *out__, tensor self, int64_t *pad_data, int pad_len) {
-  PROTECT(
-    auto outputs__ = torch::_pad_circular(*self, torch::IntArrayRef(pad_data, pad_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__pad_enum(tensor *out__, tensor self, int64_t *pad_data, int pad_len, int64_t mode, double value_v, int value_null) {
-  PROTECT(
-    auto outputs__ = torch::_pad_enum(*self, torch::IntArrayRef(pad_data, pad_len), mode, value_null ? c10::nullopt : c10::optional<double>(value_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__pad_packed_sequence(tensor *out__, tensor data, tensor batch_sizes, int batch_first, scalar padding_value, int64_t total_length) {
-  PROTECT(
-    auto outputs__ = torch::_pad_packed_sequence(*data, *batch_sizes, (bool)batch_first, *padding_value, total_length);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg__pdist_backward(tensor *out__, tensor grad, tensor self, double p, tensor pdist) {
-  PROTECT(
-    auto outputs__ = torch::_pdist_backward(*grad, *self, p, *pdist);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__pdist_backward_out(tensor *out__, tensor out, tensor grad, tensor self, double p, tensor pdist) {
-  PROTECT(
-    auto outputs__ = torch::_pdist_backward_out(*out, *grad, *self, p, *pdist);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__pin_memory(tensor *out__, tensor self, int device) {
-  PROTECT(
-    auto outputs__ = torch::_pin_memory(*self, device_of_int(device));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__pin_memory_out(tensor *out__, tensor out, tensor self, int device) {
-  PROTECT(
-    auto outputs__ = torch::_pin_memory_out(*out, *self, device_of_int(device));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__prelu_kernel(tensor *out__, tensor self, tensor weight) {
-  PROTECT(
-    auto outputs__ = torch::_prelu_kernel(*self, *weight);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__prelu_kernel_backward(tensor *out__, tensor grad_output, tensor self, tensor weight) {
-  PROTECT(
-    auto outputs__ = torch::_prelu_kernel_backward(*grad_output, *self, *weight);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg__propagate_xla_data(tensor input, tensor output) {
-  PROTECT(
-    torch::_propagate_xla_data(*input, *output);
-  )
-}
-
-void atg__remove_batch_dim(tensor *out__, tensor self, int64_t level, int64_t batch_size, int64_t out_dim) {
-  PROTECT(
-    auto outputs__ = torch::_remove_batch_dim(*self, level, batch_size, out_dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__reshape_alias(tensor *out__, tensor self, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len) {
-  PROTECT(
-    auto outputs__ = torch::_reshape_alias(*self, torch::IntArrayRef(size_data, size_len), torch::IntArrayRef(stride_data, stride_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__reshape_alias_copy(tensor *out__, tensor self, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len) {
-  PROTECT(
-    auto outputs__ = torch::_reshape_alias_copy(*self, torch::IntArrayRef(size_data, size_len), torch::IntArrayRef(stride_data, stride_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__reshape_alias_copy_out(tensor *out__, tensor out, tensor self, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len) {
-  PROTECT(
-    auto outputs__ = torch::_reshape_alias_copy_out(*out, *self, torch::IntArrayRef(size_data, size_len), torch::IntArrayRef(stride_data, stride_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__reshape_copy(tensor *out__, tensor self, int64_t *size_data, int size_len) {
-  PROTECT(
-    auto outputs__ = torch::_reshape_copy(*self, torch::IntArrayRef(size_data, size_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__reshape_from_tensor(tensor *out__, tensor self, tensor shape) {
-  PROTECT(
-    auto outputs__ = torch::_reshape_from_tensor(*self, *shape);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__resize_output(tensor *out__, tensor self, int64_t *size_data, int size_len, int device) {
-  PROTECT(
-    auto outputs__ = torch::_resize_output(*self, torch::IntArrayRef(size_data, size_len), device_of_int(device));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__resize_output_(tensor *out__, tensor self, int64_t *size_data, int size_len, int device) {
-  PROTECT(
-    auto outputs__ = torch::_resize_output_(*self, torch::IntArrayRef(size_data, size_len), device_of_int(device));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__resize_output_out(tensor *out__, tensor out, tensor self, int64_t *size_data, int size_len, int device) {
-  PROTECT(
-    auto outputs__ = torch::_resize_output_out(*out, *self, torch::IntArrayRef(size_data, size_len), device_of_int(device));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__rowwise_prune(tensor *out__, tensor weight, tensor mask, int compressed_indices_dtype) {
-  PROTECT(
-    auto outputs__ = torch::_rowwise_prune(*weight, *mask, torch::ScalarType(compressed_indices_dtype));
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg__sample_dirichlet(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::_sample_dirichlet(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sample_dirichlet_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::_sample_dirichlet_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__saturate_weight_to_fp16(tensor *out__, tensor weight) {
-  PROTECT(
-    auto outputs__ = torch::_saturate_weight_to_fp16(*weight);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__scaled_dot_product_attention_math(tensor *out__, tensor query, tensor key, tensor value, tensor attn_mask, double dropout_p, int is_causal, tensor dropout_mask, double scale_v, int scale_null) {
-  PROTECT(
-    auto outputs__ = torch::_scaled_dot_product_attention_math(*query, *key, *value, (attn_mask ? *attn_mask : torch::Tensor()), dropout_p, (bool)is_causal, (dropout_mask ? *dropout_mask : torch::Tensor()), scale_null ? c10::nullopt : c10::optional<double>(scale_v));
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg__scaled_dot_product_efficient_attention(tensor *out__, tensor query, tensor key, tensor value, tensor attn_bias, int compute_log_sumexp, double dropout_p, int is_causal, double scale_v, int scale_null) {
-  PROTECT(
-    auto outputs__ = torch::_scaled_dot_product_efficient_attention(*query, *key, *value, (attn_bias ? *attn_bias : torch::Tensor()), (bool)compute_log_sumexp, dropout_p, (bool)is_causal, scale_null ? c10::nullopt : c10::optional<double>(scale_v));
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-    out__[3] = new torch::Tensor(std::get<3>(outputs__));
-  )
-}
-
-void atg__scaled_dot_product_flash_attention_backward(tensor *out__, tensor grad_out, tensor query, tensor key, tensor value, tensor out, tensor logsumexp, tensor cum_seq_q, tensor cum_seq_k, int64_t max_q, int64_t max_k, double dropout_p, int is_causal, tensor philox_seed, tensor philox_offset, double scale_v, int scale_null) {
-  PROTECT(
-    auto outputs__ = torch::_scaled_dot_product_flash_attention_backward(*grad_out, *query, *key, *value, *out, *logsumexp, *cum_seq_q, *cum_seq_k, max_q, max_k, dropout_p, (bool)is_causal, *philox_seed, *philox_offset, scale_null ? c10::nullopt : c10::optional<double>(scale_v));
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg__scaled_mm(tensor *out__, tensor self, tensor mat2, tensor bias, int out_dtype, tensor scale_a, tensor scale_b, tensor scale_result) {
-  PROTECT(
-    auto outputs__ = torch::_scaled_mm(*self, *mat2, (bias ? *bias : torch::Tensor()), torch::ScalarType(out_dtype), (scale_a ? *scale_a : torch::Tensor()), (scale_b ? *scale_b : torch::Tensor()), (scale_result ? *scale_result : torch::Tensor()));
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg__scaled_mm_out(tensor *out__, tensor out, tensor out_amax, tensor self, tensor mat2, tensor bias, int out_dtype, tensor scale_a, tensor scale_b, tensor scale_result) {
-  PROTECT(
-    auto outputs__ = torch::_scaled_mm_out(*out, *out_amax, *self, *mat2, (bias ? *bias : torch::Tensor()), torch::ScalarType(out_dtype), (scale_a ? *scale_a : torch::Tensor()), (scale_b ? *scale_b : torch::Tensor()), (scale_result ? *scale_result : torch::Tensor()));
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg__scatter_reduce(tensor *out__, tensor self, int64_t dim, tensor index, tensor src, char * reduce, int include_self) {
-  PROTECT(
-    auto outputs__ = torch::scatter_reduce(*self, dim, *index, *src, std::string(reduce), (bool)include_self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__scatter_reduce_(tensor *out__, tensor self, int64_t dim, tensor index, tensor src, char * reduce, int include_self) {
-  PROTECT(
-    auto outputs__ = self->scatter_reduce_(dim, *index, *src, std::string(reduce), (bool)include_self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__scatter_reduce_two_out(tensor *out__, tensor out, tensor self, int64_t dim, tensor index, tensor src, char * reduce, int include_self) {
-  PROTECT(
-    auto outputs__ = torch::scatter_reduce_out(*out, *self, dim, *index, *src, std::string(reduce), (bool)include_self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__segment_reduce_backward(tensor *out__, tensor grad, tensor output, tensor data, char * reduce, tensor lengths, tensor offsets, int64_t axis, scalar initial) {
-  PROTECT(
-    auto outputs__ = torch::_segment_reduce_backward(*grad, *output, *data, std::string(reduce), (lengths ? *lengths : torch::Tensor()), (offsets ? *offsets : torch::Tensor()), axis, *initial);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__segment_reduce_backward_out(tensor *out__, tensor out, tensor grad, tensor output, tensor data, char * reduce, tensor lengths, tensor offsets, int64_t axis, scalar initial) {
-  PROTECT(
-    auto outputs__ = torch::_segment_reduce_backward_out(*out, *grad, *output, *data, std::string(reduce), (lengths ? *lengths : torch::Tensor()), (offsets ? *offsets : torch::Tensor()), axis, *initial);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__shape_as_tensor(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::_shape_as_tensor(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__slow_conv2d_backward(tensor *out__, tensor grad_input, tensor grad_weight, tensor grad_bias, tensor grad_output, tensor self, tensor weight, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len) {
-  PROTECT(
-    auto outputs__ = torch::_slow_conv2d_backward_out(*grad_input, *grad_weight, *grad_bias, *grad_output, *self, *weight, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len));
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg__sobol_engine_draw(tensor *out__, tensor quasi, int64_t n, tensor sobolstate, int64_t dimension, int64_t num_generated, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::_sobol_engine_draw(*quasi, n, *sobolstate, dimension, num_generated, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg__sobol_engine_ff_(tensor *out__, tensor self, int64_t n, tensor sobolstate, int64_t dimension, int64_t num_generated) {
-  PROTECT(
-    auto outputs__ = torch::_sobol_engine_ff_(*self, n, *sobolstate, dimension, num_generated);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sobol_engine_initialize_state_(tensor *out__, tensor self, int64_t dimension) {
-  PROTECT(
-    auto outputs__ = torch::_sobol_engine_initialize_state_(*self, dimension);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sobol_engine_scramble_(tensor *out__, tensor self, tensor ltm, int64_t dimension) {
-  PROTECT(
-    auto outputs__ = torch::_sobol_engine_scramble_(*self, *ltm, dimension);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__softmax(tensor *out__, tensor self, int64_t dim, int half_to_float) {
-  PROTECT(
-    auto outputs__ = torch::_softmax(*self, dim, (bool)half_to_float);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__softmax_backward_data(tensor *out__, tensor grad_output, tensor output, int64_t dim, int input_dtype) {
-  PROTECT(
-    auto outputs__ = torch::_softmax_backward_data(*grad_output, *output, dim, torch::ScalarType(input_dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__softmax_backward_data_out(tensor *out__, tensor grad_input, tensor grad_output, tensor output, int64_t dim, int input_dtype) {
-  PROTECT(
-    auto outputs__ = torch::_softmax_backward_data_out(*grad_input, *grad_output, *output, dim, torch::ScalarType(input_dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__softmax_out(tensor *out__, tensor out, tensor self, int64_t dim, int half_to_float) {
-  PROTECT(
-    auto outputs__ = torch::_softmax_out(*out, *self, dim, (bool)half_to_float);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sparse_addmm(tensor *out__, tensor self, tensor mat1, tensor mat2) {
-  PROTECT(
-    auto outputs__ = torch::_sparse_addmm(*self, *mat1, *mat2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sparse_addmm_out(tensor *out__, tensor out, tensor self, tensor mat1, tensor mat2) {
-  PROTECT(
-    auto outputs__ = torch::_sparse_addmm_out(*out, *self, *mat1, *mat2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sparse_broadcast_to(tensor *out__, tensor self, int64_t *size_data, int size_len) {
-  PROTECT(
-    auto outputs__ = torch::_sparse_broadcast_to(*self, torch::IntArrayRef(size_data, size_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sparse_broadcast_to_copy(tensor *out__, tensor self, int64_t *size_data, int size_len) {
-  PROTECT(
-    auto outputs__ = torch::_sparse_broadcast_to_copy(*self, torch::IntArrayRef(size_data, size_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sparse_broadcast_to_copy_out(tensor *out__, tensor out, tensor self, int64_t *size_data, int size_len) {
-  PROTECT(
-    auto outputs__ = torch::_sparse_broadcast_to_copy_out(*out, *self, torch::IntArrayRef(size_data, size_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sparse_bsc_tensor_unsafe(tensor *out__, tensor ccol_indices, tensor row_indices, tensor values, int64_t *size_data, int size_len, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::_sparse_bsc_tensor_unsafe(*ccol_indices, *row_indices, *values, torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sparse_bsr_tensor_unsafe(tensor *out__, tensor crow_indices, tensor col_indices, tensor values, int64_t *size_data, int size_len, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::_sparse_bsr_tensor_unsafe(*crow_indices, *col_indices, *values, torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sparse_compressed_tensor_unsafe(tensor *out__, tensor compressed_indices, tensor plain_indices, tensor values, int64_t *size_data, int size_len, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::_sparse_compressed_tensor_unsafe(*compressed_indices, *plain_indices, *values, torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sparse_coo_tensor_unsafe(tensor *out__, tensor indices, tensor values, int64_t *size_data, int size_len, int options_kind, int options_device, int is_coalesced) {
-  PROTECT(
-    auto outputs__ = torch::_sparse_coo_tensor_unsafe(*indices, *values, torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)), (bool)is_coalesced);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sparse_coo_tensor_with_dims(tensor *out__, int64_t sparse_dim, int64_t dense_dim, int64_t *size_data, int size_len, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::_sparse_coo_tensor_with_dims(sparse_dim, dense_dim, torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sparse_coo_tensor_with_dims_and_tensors(tensor *out__, int64_t sparse_dim, int64_t dense_dim, int64_t *size_data, int size_len, tensor indices, tensor values, int options_kind, int options_device, int is_coalesced) {
-  PROTECT(
-    auto outputs__ = torch::_sparse_coo_tensor_with_dims_and_tensors(sparse_dim, dense_dim, torch::IntArrayRef(size_data, size_len), *indices, *values, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)), (bool)is_coalesced);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sparse_coo_tensor_with_dims_and_tensors_out(tensor *out__, tensor out, int64_t sparse_dim, int64_t dense_dim, int64_t *size_data, int size_len, tensor indices, tensor values, int is_coalesced) {
-  PROTECT(
-    auto outputs__ = torch::_sparse_coo_tensor_with_dims_and_tensors_out(*out, sparse_dim, dense_dim, torch::IntArrayRef(size_data, size_len), *indices, *values, (bool)is_coalesced);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sparse_coo_tensor_with_dims_out(tensor *out__, tensor out, int64_t sparse_dim, int64_t dense_dim, int64_t *size_data, int size_len) {
-  PROTECT(
-    auto outputs__ = torch::_sparse_coo_tensor_with_dims_out(*out, sparse_dim, dense_dim, torch::IntArrayRef(size_data, size_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sparse_csc_tensor_unsafe(tensor *out__, tensor ccol_indices, tensor row_indices, tensor values, int64_t *size_data, int size_len, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::_sparse_csc_tensor_unsafe(*ccol_indices, *row_indices, *values, torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sparse_csr_prod(tensor *out__, tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::_sparse_csr_prod(*self, torch::IntArrayRef(dim_data, dim_len), (bool)keepdim, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sparse_csr_prod_dim_dtype_out(tensor *out__, tensor out, tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::_sparse_csr_prod_out(*out, *self, torch::IntArrayRef(dim_data, dim_len), (bool)keepdim, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sparse_csr_sum(tensor *out__, tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::_sparse_csr_sum(*self, torch::IntArrayRef(dim_data, dim_len), (bool)keepdim, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sparse_csr_sum_dim_dtype_out(tensor *out__, tensor out, tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::_sparse_csr_sum_out(*out, *self, torch::IntArrayRef(dim_data, dim_len), (bool)keepdim, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sparse_csr_tensor_unsafe(tensor *out__, tensor crow_indices, tensor col_indices, tensor values, int64_t *size_data, int size_len, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::_sparse_csr_tensor_unsafe(*crow_indices, *col_indices, *values, torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sparse_log_softmax(tensor *out__, tensor self, int64_t dim, int half_to_float) {
-  PROTECT(
-    auto outputs__ = torch::_sparse_log_softmax(*self, dim, (bool)half_to_float);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sparse_log_softmax_backward_data(tensor *out__, tensor grad_output, tensor output, int64_t dim, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::_sparse_log_softmax_backward_data(*grad_output, *output, dim, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sparse_log_softmax_backward_data_out(tensor *out__, tensor out, tensor grad_output, tensor output, int64_t dim, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::_sparse_log_softmax_backward_data_out(*out, *grad_output, *output, dim, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sparse_log_softmax_int(tensor *out__, tensor self, int64_t dim, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::_sparse_log_softmax(*self, dim, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sparse_log_softmax_out(tensor *out__, tensor out, tensor self, int64_t dim, int half_to_float) {
-  PROTECT(
-    auto outputs__ = torch::_sparse_log_softmax_out(*out, *self, dim, (bool)half_to_float);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sparse_mask_projection(tensor *out__, tensor self, tensor mask, int accumulate_matches) {
-  PROTECT(
-    auto outputs__ = self->_sparse_mask_projection(*mask, (bool)accumulate_matches);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sparse_mask_projection_out(tensor *out__, tensor out, tensor self, tensor mask, int accumulate_matches) {
-  PROTECT(
-    auto outputs__ = torch::_sparse_mask_projection_out(*out, *self, *mask, (bool)accumulate_matches);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sparse_mm(tensor *out__, tensor sparse, tensor dense) {
-  PROTECT(
-    auto outputs__ = torch::_sparse_mm(*sparse, *dense);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sparse_mm_reduce(tensor *out__, tensor sparse, tensor dense, char * reduce) {
-  PROTECT(
-    auto outputs__ = torch::_sparse_mm(*sparse, *dense, std::string(reduce));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sparse_mm_reduce_impl(tensor *out__, tensor self, tensor other, char * reduce) {
-  PROTECT(
-    auto outputs__ = torch::_sparse_mm_reduce_impl(*self, *other, std::string(reduce));
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg__sparse_semi_structured_linear(tensor *out__, tensor input, tensor weight, tensor meta, tensor bias, char * activation) {
-  PROTECT(
-    auto outputs__ = torch::_sparse_semi_structured_linear(*input, *weight, *meta, (bias ? *bias : torch::Tensor()), std::string(activation));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sparse_softmax(tensor *out__, tensor self, int64_t dim, int half_to_float) {
-  PROTECT(
-    auto outputs__ = torch::_sparse_softmax(*self, dim, (bool)half_to_float);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sparse_softmax_backward_data(tensor *out__, tensor grad_output, tensor output, int64_t dim, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::_sparse_softmax_backward_data(*grad_output, *output, dim, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sparse_softmax_backward_data_out(tensor *out__, tensor out, tensor grad_output, tensor output, int64_t dim, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::_sparse_softmax_backward_data_out(*out, *grad_output, *output, dim, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sparse_softmax_int(tensor *out__, tensor self, int64_t dim, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::_sparse_softmax(*self, dim, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sparse_softmax_out(tensor *out__, tensor out, tensor self, int64_t dim, int half_to_float) {
-  PROTECT(
-    auto outputs__ = torch::_sparse_softmax_out(*out, *self, dim, (bool)half_to_float);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sparse_sparse_matmul(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::_sparse_sparse_matmul(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sparse_sparse_matmul_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::_sparse_sparse_matmul_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sparse_sum(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::_sparse_sum(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sparse_sum_backward(tensor *out__, tensor grad, tensor self, int64_t *dim_data, int dim_len) {
-  PROTECT(
-    auto outputs__ = torch::_sparse_sum_backward(*grad, *self, torch::IntArrayRef(dim_data, dim_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sparse_sum_backward_out(tensor *out__, tensor out, tensor grad, tensor self, int64_t *dim_data, int dim_len) {
-  PROTECT(
-    auto outputs__ = torch::_sparse_sum_backward_out(*out, *grad, *self, torch::IntArrayRef(dim_data, dim_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sparse_sum_dim(tensor *out__, tensor self, int64_t *dim_data, int dim_len) {
-  PROTECT(
-    auto outputs__ = torch::_sparse_sum(*self, torch::IntArrayRef(dim_data, dim_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sparse_sum_dim_dtype(tensor *out__, tensor self, int64_t *dim_data, int dim_len, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::_sparse_sum(*self, torch::IntArrayRef(dim_data, dim_len), torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sparse_sum_dim_out(tensor *out__, tensor out, tensor self, int64_t *dim_data, int dim_len) {
-  PROTECT(
-    auto outputs__ = torch::_sparse_sum_out(*out, *self, torch::IntArrayRef(dim_data, dim_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__sparse_sum_dtype(tensor *out__, tensor self, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::_sparse_sum(*self, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__spdiags(tensor *out__, tensor diagonals, tensor offsets, int64_t *shape_data, int shape_len) {
-  PROTECT(
-    auto outputs__ = torch::_spdiags(*diagonals, *offsets, torch::IntArrayRef(shape_data, shape_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__spdiags_out(tensor *out__, tensor out, tensor diagonals, tensor offsets, int64_t *shape_data, int shape_len) {
-  PROTECT(
-    auto outputs__ = torch::_spdiags_out(*out, *diagonals, *offsets, torch::IntArrayRef(shape_data, shape_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__stack(tensor *out__, tensor *tensors_data, int tensors_len, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::_stack(of_carray_tensor(tensors_data, tensors_len), dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__stack_out(tensor *out__, tensor out, tensor *tensors_data, int tensors_len, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::_stack_out(*out, of_carray_tensor(tensors_data, tensors_len), dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__standard_gamma(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::_standard_gamma(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__standard_gamma_grad(tensor *out__, tensor self, tensor output) {
-  PROTECT(
-    auto outputs__ = torch::_standard_gamma_grad(*self, *output);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__standard_gamma_grad_out(tensor *out__, tensor out, tensor self, tensor output) {
-  PROTECT(
-    auto outputs__ = torch::_standard_gamma_grad_out(*out, *self, *output);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__standard_gamma_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::_standard_gamma_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__test_ambiguous_defaults(tensor *out__, tensor dummy, int64_t a, int64_t b) {
-  PROTECT(
-    auto outputs__ = torch::_test_ambiguous_defaults(*dummy, a, b);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__test_ambiguous_defaults_b(tensor *out__, tensor dummy, int64_t a, char * b) {
-  PROTECT(
-    auto outputs__ = torch::_test_ambiguous_defaults(*dummy, a, std::string(b));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__test_autograd_multiple_dispatch(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::_test_autograd_multiple_dispatch(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__test_autograd_multiple_dispatch_fullcoverage_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::_test_autograd_multiple_dispatch_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__test_autograd_multiple_dispatch_ntonly(tensor *out__, tensor self, int b) {
-  PROTECT(
-    auto outputs__ = torch::_test_autograd_multiple_dispatch(*self, (bool)b);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__test_autograd_multiple_dispatch_view(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::_test_autograd_multiple_dispatch_view(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__test_autograd_multiple_dispatch_view_copy(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::_test_autograd_multiple_dispatch_view_copy(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__test_autograd_multiple_dispatch_view_copy_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::_test_autograd_multiple_dispatch_view_copy_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__test_check_tensor(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::_test_check_tensor(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__test_functorch_fallback(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::_test_functorch_fallback(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__test_functorch_fallback_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::_test_functorch_fallback_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__test_optional_filled_intlist(tensor *out__, tensor values, int64_t *addends_data, int addends_len) {
-  PROTECT(
-    auto outputs__ = torch::_test_optional_filled_intlist(*values, addends_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(addends_data, addends_len)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__test_optional_filled_intlist_out(tensor *out__, tensor out, tensor values, int64_t *addends_data, int addends_len) {
-  PROTECT(
-    auto outputs__ = torch::_test_optional_filled_intlist_out(*out, *values, addends_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(addends_data, addends_len)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__test_optional_floatlist(tensor *out__, tensor values, double *addends_data, int addends_len) {
-  PROTECT(
-    auto outputs__ = torch::_test_optional_floatlist(*values, at::ArrayRef<double>(addends_data, addends_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__test_optional_floatlist_out(tensor *out__, tensor out, tensor values, double *addends_data, int addends_len) {
-  PROTECT(
-    auto outputs__ = torch::_test_optional_floatlist_out(*out, *values, at::ArrayRef<double>(addends_data, addends_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__test_optional_intlist(tensor *out__, tensor values, int64_t *addends_data, int addends_len) {
-  PROTECT(
-    auto outputs__ = torch::_test_optional_intlist(*values, addends_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(addends_data, addends_len)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__test_optional_intlist_out(tensor *out__, tensor out, tensor values, int64_t *addends_data, int addends_len) {
-  PROTECT(
-    auto outputs__ = torch::_test_optional_intlist_out(*out, *values, addends_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(addends_data, addends_len)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__test_serialization_subcmul(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::_test_serialization_subcmul(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__test_string_default(tensor *out__, tensor dummy, char * a, char * b) {
-  PROTECT(
-    auto outputs__ = torch::_test_string_default(*dummy, std::string(a), std::string(b));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__test_warn_in_autograd(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::_test_warn_in_autograd(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__test_warn_in_autograd_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::_test_warn_in_autograd_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__thnn_differentiable_gru_cell_backward(tensor *out__, tensor grad_hy, tensor input_gates, tensor hidden_gates, tensor hx, tensor input_bias, tensor hidden_bias) {
-  PROTECT(
-    auto outputs__ = torch::_thnn_differentiable_gru_cell_backward(*grad_hy, *input_gates, *hidden_gates, *hx, (input_bias ? *input_bias : torch::Tensor()), (hidden_bias ? *hidden_bias : torch::Tensor()));
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-    out__[3] = new torch::Tensor(std::get<3>(outputs__));
-    out__[4] = new torch::Tensor(std::get<4>(outputs__));
-  )
-}
-
-void atg__thnn_differentiable_lstm_cell_backward(tensor *out__, tensor grad_hy, tensor grad_cy, tensor input_gates, tensor hidden_gates, tensor input_bias, tensor hidden_bias, tensor cx, tensor cy) {
-  PROTECT(
-    auto outputs__ = torch::_thnn_differentiable_lstm_cell_backward((grad_hy ? *grad_hy : torch::Tensor()), (grad_cy ? *grad_cy : torch::Tensor()), *input_gates, *hidden_gates, (input_bias ? *input_bias : torch::Tensor()), (hidden_bias ? *hidden_bias : torch::Tensor()), *cx, *cy);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-    out__[3] = new torch::Tensor(std::get<3>(outputs__));
-    out__[4] = new torch::Tensor(std::get<4>(outputs__));
-  )
-}
-
-void atg__thnn_fused_gru_cell(tensor *out__, tensor input_gates, tensor hidden_gates, tensor hx, tensor input_bias, tensor hidden_bias) {
-  PROTECT(
-    auto outputs__ = torch::_thnn_fused_gru_cell(*input_gates, *hidden_gates, *hx, (input_bias ? *input_bias : torch::Tensor()), (hidden_bias ? *hidden_bias : torch::Tensor()));
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg__thnn_fused_gru_cell_backward(tensor *out__, tensor grad_hy, tensor workspace, int has_bias) {
-  PROTECT(
-    auto outputs__ = torch::_thnn_fused_gru_cell_backward(*grad_hy, *workspace, (bool)has_bias);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-    out__[3] = new torch::Tensor(std::get<3>(outputs__));
-    out__[4] = new torch::Tensor(std::get<4>(outputs__));
-  )
-}
-
-void atg__thnn_fused_gru_cell_backward_out(tensor *out__, tensor out0, tensor out1, tensor out2, tensor out3, tensor out4, tensor grad_hy, tensor workspace, int has_bias) {
-  PROTECT(
-    auto outputs__ = torch::_thnn_fused_gru_cell_backward_out(*out0, *out1, *out2, *out3, *out4, *grad_hy, *workspace, (bool)has_bias);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-    out__[3] = new torch::Tensor(std::get<3>(outputs__));
-    out__[4] = new torch::Tensor(std::get<4>(outputs__));
-  )
-}
-
-void atg__thnn_fused_gru_cell_out(tensor *out__, tensor out0, tensor out1, tensor input_gates, tensor hidden_gates, tensor hx, tensor input_bias, tensor hidden_bias) {
-  PROTECT(
-    auto outputs__ = torch::_thnn_fused_gru_cell_out(*out0, *out1, *input_gates, *hidden_gates, *hx, (input_bias ? *input_bias : torch::Tensor()), (hidden_bias ? *hidden_bias : torch::Tensor()));
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg__thnn_fused_lstm_cell(tensor *out__, tensor input_gates, tensor hidden_gates, tensor cx, tensor input_bias, tensor hidden_bias) {
-  PROTECT(
-    auto outputs__ = torch::_thnn_fused_lstm_cell(*input_gates, *hidden_gates, *cx, (input_bias ? *input_bias : torch::Tensor()), (hidden_bias ? *hidden_bias : torch::Tensor()));
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg__thnn_fused_lstm_cell_backward(tensor *out__, tensor grad_hy, tensor grad_cy, tensor cx, tensor cy, tensor workspace, int has_bias) {
-  PROTECT(
-    auto outputs__ = torch::_thnn_fused_lstm_cell_backward((grad_hy ? *grad_hy : torch::Tensor()), (grad_cy ? *grad_cy : torch::Tensor()), *cx, *cy, *workspace, (bool)has_bias);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-    out__[3] = new torch::Tensor(std::get<3>(outputs__));
-    out__[4] = new torch::Tensor(std::get<4>(outputs__));
-  )
-}
-
-void atg__thnn_fused_lstm_cell_backward_impl(tensor *out__, tensor grad_hy, tensor grad_cy, tensor cx, tensor cy, tensor workspace, int has_bias) {
-  PROTECT(
-    auto outputs__ = torch::_thnn_fused_lstm_cell_backward_impl((grad_hy ? *grad_hy : torch::Tensor()), (grad_cy ? *grad_cy : torch::Tensor()), *cx, *cy, *workspace, (bool)has_bias);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg__thnn_fused_lstm_cell_backward_impl_out(tensor *out__, tensor out0, tensor out1, tensor out2, tensor grad_hy, tensor grad_cy, tensor cx, tensor cy, tensor workspace, int has_bias) {
-  PROTECT(
-    auto outputs__ = torch::_thnn_fused_lstm_cell_backward_impl_out(*out0, *out1, *out2, (grad_hy ? *grad_hy : torch::Tensor()), (grad_cy ? *grad_cy : torch::Tensor()), *cx, *cy, *workspace, (bool)has_bias);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg__thnn_fused_lstm_cell_out(tensor *out__, tensor out0, tensor out1, tensor out2, tensor input_gates, tensor hidden_gates, tensor cx, tensor input_bias, tensor hidden_bias) {
-  PROTECT(
-    auto outputs__ = torch::_thnn_fused_lstm_cell_out(*out0, *out1, *out2, *input_gates, *hidden_gates, *cx, (input_bias ? *input_bias : torch::Tensor()), (hidden_bias ? *hidden_bias : torch::Tensor()));
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg__to_copy(tensor *out__, tensor self, int options_kind, int options_device, int non_blocking) {
-  PROTECT(
-    auto outputs__ = torch::_to_copy(*self, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)), (bool)non_blocking);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__to_copy_out(tensor *out__, tensor out, tensor self, int non_blocking) {
-  PROTECT(
-    auto outputs__ = torch::_to_copy_out(*out, *self, (bool)non_blocking);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-tensor *atg__to_cpu(tensor *tensors_data, int tensors_len) {
-  PROTECT(
-    auto outputs__ = torch::_to_cpu(of_carray_tensor(tensors_data, tensors_len));
-    int sz = outputs__.size();
-    torch::Tensor **out__ = (torch::Tensor**)malloc((sz + 1) * sizeof(torch::Tensor*));
-    for (int i = 0; i < sz; ++i)
-      out__[i] = new torch::Tensor(outputs__[i]);
-    out__[sz] = nullptr;
-    return out__;
-  )
-}
-
-void atg__to_dense(tensor *out__, tensor self, int dtype, int masked_grad) {
-  PROTECT(
-    auto outputs__ = self->_to_dense(torch::ScalarType(dtype), (bool)masked_grad);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__to_dense_out(tensor *out__, tensor out, tensor self, int dtype, int masked_grad) {
-  PROTECT(
-    auto outputs__ = torch::_to_dense_out(*out, *self, torch::ScalarType(dtype), (bool)masked_grad);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__to_sparse_bsc(tensor *out__, tensor self, int64_t *blocksize_data, int blocksize_len, int64_t dense_dim_v, int dense_dim_null) {
-  PROTECT(
-    auto outputs__ = self->_to_sparse_bsc(torch::IntArrayRef(blocksize_data, blocksize_len), dense_dim_null ? c10::nullopt : c10::optional<int64_t>(dense_dim_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__to_sparse_bsc_out(tensor *out__, tensor out, tensor self, int64_t *blocksize_data, int blocksize_len, int64_t dense_dim_v, int dense_dim_null) {
-  PROTECT(
-    auto outputs__ = torch::_to_sparse_bsc_out(*out, *self, torch::IntArrayRef(blocksize_data, blocksize_len), dense_dim_null ? c10::nullopt : c10::optional<int64_t>(dense_dim_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__to_sparse_bsr(tensor *out__, tensor self, int64_t *blocksize_data, int blocksize_len, int64_t dense_dim_v, int dense_dim_null) {
-  PROTECT(
-    auto outputs__ = self->_to_sparse_bsr(torch::IntArrayRef(blocksize_data, blocksize_len), dense_dim_null ? c10::nullopt : c10::optional<int64_t>(dense_dim_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__to_sparse_bsr_out(tensor *out__, tensor out, tensor self, int64_t *blocksize_data, int blocksize_len, int64_t dense_dim_v, int dense_dim_null) {
-  PROTECT(
-    auto outputs__ = torch::_to_sparse_bsr_out(*out, *self, torch::IntArrayRef(blocksize_data, blocksize_len), dense_dim_null ? c10::nullopt : c10::optional<int64_t>(dense_dim_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__to_sparse_csc(tensor *out__, tensor self, int64_t dense_dim_v, int dense_dim_null) {
-  PROTECT(
-    auto outputs__ = self->_to_sparse_csc(dense_dim_null ? c10::nullopt : c10::optional<int64_t>(dense_dim_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__to_sparse_csc_out(tensor *out__, tensor out, tensor self, int64_t dense_dim_v, int dense_dim_null) {
-  PROTECT(
-    auto outputs__ = torch::_to_sparse_csc_out(*out, *self, dense_dim_null ? c10::nullopt : c10::optional<int64_t>(dense_dim_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__to_sparse_csr(tensor *out__, tensor self, int64_t dense_dim_v, int dense_dim_null) {
-  PROTECT(
-    auto outputs__ = self->_to_sparse_csr(dense_dim_null ? c10::nullopt : c10::optional<int64_t>(dense_dim_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__to_sparse_csr_out(tensor *out__, tensor out, tensor self, int64_t dense_dim_v, int dense_dim_null) {
-  PROTECT(
-    auto outputs__ = torch::_to_sparse_csr_out(*out, *self, dense_dim_null ? c10::nullopt : c10::optional<int64_t>(dense_dim_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__to_sparse_semi_structured(tensor *out__, tensor dense) {
-  PROTECT(
-    auto outputs__ = torch::_to_sparse_semi_structured(*dense);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg__transform_bias_rescale_qkv(tensor *out__, tensor qkv, tensor qkv_bias, int64_t num_heads) {
-  PROTECT(
-    auto outputs__ = torch::_transform_bias_rescale_qkv(*qkv, *qkv_bias, num_heads);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg__transform_bias_rescale_qkv_out(tensor *out__, tensor out0, tensor out1, tensor out2, tensor qkv, tensor qkv_bias, int64_t num_heads) {
-  PROTECT(
-    auto outputs__ = torch::_transform_bias_rescale_qkv_out(*out0, *out1, *out2, *qkv, *qkv_bias, num_heads);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg__transformer_encoder_layer_fwd(tensor *out__, tensor src, int64_t embed_dim, int64_t num_heads, tensor qkv_weight, tensor qkv_bias, tensor proj_weight, tensor proj_bias, int use_gelu, int norm_first, double eps, tensor norm_weight_1, tensor norm_bias_1, tensor norm_weight_2, tensor norm_bias_2, tensor ffn_weight_1, tensor ffn_bias_1, tensor ffn_weight_2, tensor ffn_bias_2, tensor mask, int64_t mask_type_v, int mask_type_null) {
-  PROTECT(
-    auto outputs__ = torch::_transformer_encoder_layer_fwd(*src, embed_dim, num_heads, *qkv_weight, *qkv_bias, *proj_weight, *proj_bias, (bool)use_gelu, (bool)norm_first, eps, *norm_weight_1, *norm_bias_1, *norm_weight_2, *norm_bias_2, *ffn_weight_1, *ffn_bias_1, *ffn_weight_2, *ffn_bias_2, (mask ? *mask : torch::Tensor()), mask_type_null ? c10::nullopt : c10::optional<int64_t>(mask_type_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__transformer_encoder_layer_fwd_out(tensor *out__, tensor out, tensor src, int64_t embed_dim, int64_t num_heads, tensor qkv_weight, tensor qkv_bias, tensor proj_weight, tensor proj_bias, int use_gelu, int norm_first, double eps, tensor norm_weight_1, tensor norm_bias_1, tensor norm_weight_2, tensor norm_bias_2, tensor ffn_weight_1, tensor ffn_bias_1, tensor ffn_weight_2, tensor ffn_bias_2, tensor mask, int64_t mask_type_v, int mask_type_null) {
-  PROTECT(
-    auto outputs__ = torch::_transformer_encoder_layer_fwd_out(*out, *src, embed_dim, num_heads, *qkv_weight, *qkv_bias, *proj_weight, *proj_bias, (bool)use_gelu, (bool)norm_first, eps, *norm_weight_1, *norm_bias_1, *norm_weight_2, *norm_bias_2, *ffn_weight_1, *ffn_bias_1, *ffn_weight_2, *ffn_bias_2, (mask ? *mask : torch::Tensor()), mask_type_null ? c10::nullopt : c10::optional<int64_t>(mask_type_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__trilinear(tensor *out__, tensor i1, tensor i2, tensor i3, int64_t *expand1_data, int expand1_len, int64_t *expand2_data, int expand2_len, int64_t *expand3_data, int expand3_len, int64_t *sumdim_data, int sumdim_len, int64_t unroll_dim) {
-  PROTECT(
-    auto outputs__ = torch::_trilinear(*i1, *i2, *i3, torch::IntArrayRef(expand1_data, expand1_len), torch::IntArrayRef(expand2_data, expand2_len), torch::IntArrayRef(expand3_data, expand3_len), torch::IntArrayRef(sumdim_data, sumdim_len), unroll_dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__trilinear_out(tensor *out__, tensor out, tensor i1, tensor i2, tensor i3, int64_t *expand1_data, int expand1_len, int64_t *expand2_data, int expand2_len, int64_t *expand3_data, int expand3_len, int64_t *sumdim_data, int sumdim_len, int64_t unroll_dim) {
-  PROTECT(
-    auto outputs__ = torch::_trilinear_out(*out, *i1, *i2, *i3, torch::IntArrayRef(expand1_data, expand1_len), torch::IntArrayRef(expand2_data, expand2_len), torch::IntArrayRef(expand3_data, expand3_len), torch::IntArrayRef(sumdim_data, sumdim_len), unroll_dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__triton_multi_head_attention(tensor *out__, tensor query, tensor key, tensor value, int64_t embed_dim, int64_t num_head, tensor qkv_weight, tensor qkv_bias, tensor proj_weight, tensor proj_bias, tensor mask) {
-  PROTECT(
-    auto outputs__ = torch::_triton_multi_head_attention(*query, *key, *value, embed_dim, num_head, *qkv_weight, *qkv_bias, *proj_weight, *proj_bias, (mask ? *mask : torch::Tensor()));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__triton_multi_head_attention_out(tensor *out__, tensor out, tensor query, tensor key, tensor value, int64_t embed_dim, int64_t num_head, tensor qkv_weight, tensor qkv_bias, tensor proj_weight, tensor proj_bias, tensor mask) {
-  PROTECT(
-    auto outputs__ = torch::_triton_multi_head_attention_out(*out, *query, *key, *value, embed_dim, num_head, *qkv_weight, *qkv_bias, *proj_weight, *proj_bias, (mask ? *mask : torch::Tensor()));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__triton_scaled_dot_attention(tensor *out__, tensor q, tensor k, tensor v, double dropout_p) {
-  PROTECT(
-    auto outputs__ = torch::_triton_scaled_dot_attention(*q, *k, *v, dropout_p);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__triton_scaled_dot_attention_out(tensor *out__, tensor out, tensor q, tensor k, tensor v, double dropout_p) {
-  PROTECT(
-    auto outputs__ = torch::_triton_scaled_dot_attention_out(*out, *q, *k, *v, dropout_p);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__unique(tensor *out__, tensor self, int sorted, int return_inverse) {
-  PROTECT(
-    auto outputs__ = torch::_unique(*self, (bool)sorted, (bool)return_inverse);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg__unique2(tensor *out__, tensor self, int sorted, int return_inverse, int return_counts) {
-  PROTECT(
-    auto outputs__ = torch::_unique2(*self, (bool)sorted, (bool)return_inverse, (bool)return_counts);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg__unique2_out(tensor *out__, tensor out0, tensor out1, tensor out2, tensor self, int sorted, int return_inverse, int return_counts) {
-  PROTECT(
-    auto outputs__ = torch::_unique2_out(*out0, *out1, *out2, *self, (bool)sorted, (bool)return_inverse, (bool)return_counts);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg__unique_out(tensor *out__, tensor out0, tensor out1, tensor self, int sorted, int return_inverse) {
-  PROTECT(
-    auto outputs__ = torch::_unique_out(*out0, *out1, *self, (bool)sorted, (bool)return_inverse);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg__unpack_dual(tensor *out__, tensor dual, int64_t level) {
-  PROTECT(
-    auto outputs__ = torch::_unpack_dual(*dual, level);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg__unsafe_index(tensor *out__, tensor self, tensor *indices_data, int indices_len) {
-  PROTECT(
-    auto outputs__ = torch::_unsafe_index(*self, of_carray_tensor_opt(indices_data, indices_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__unsafe_index_put(tensor *out__, tensor self, tensor *indices_data, int indices_len, tensor values, int accumulate) {
-  PROTECT(
-    auto outputs__ = torch::_unsafe_index_put(*self, of_carray_tensor_opt(indices_data, indices_len), *values, (bool)accumulate);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__unsafe_view(tensor *out__, tensor self, int64_t *size_data, int size_len) {
-  PROTECT(
-    auto outputs__ = torch::_unsafe_view(*self, torch::IntArrayRef(size_data, size_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__unsafe_view_out(tensor *out__, tensor out, tensor self, int64_t *size_data, int size_len) {
-  PROTECT(
-    auto outputs__ = torch::_unsafe_view_out(*out, *self, torch::IntArrayRef(size_data, size_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__upsample_bicubic2d_aa(tensor *out__, tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
-  PROTECT(
-    auto outputs__ = torch::_upsample_bicubic2d_aa(*self, torch::IntArrayRef(output_size_data, output_size_len), (bool)align_corners, scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__upsample_bicubic2d_aa_backward(tensor *out__, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
-  PROTECT(
-    auto outputs__ = torch::_upsample_bicubic2d_aa_backward(*grad_output, torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), (bool)align_corners, scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__upsample_bicubic2d_aa_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
-  PROTECT(
-    auto outputs__ = torch::_upsample_bicubic2d_aa_backward_out(*grad_input, *grad_output, torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), (bool)align_corners, scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__upsample_bicubic2d_aa_out(tensor *out__, tensor out, tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
-  PROTECT(
-    auto outputs__ = torch::_upsample_bicubic2d_aa_out(*out, *self, torch::IntArrayRef(output_size_data, output_size_len), (bool)align_corners, scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__upsample_bicubic2d_aa_vec(tensor *out__, tensor input, int64_t *output_size_data, int output_size_len, int align_corners, double *scale_factors_data, int scale_factors_len) {
-  PROTECT(
-    auto outputs__ = torch::_upsample_bicubic2d_aa(*input, output_size_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(output_size_data, output_size_len)), (bool)align_corners, at::ArrayRef<double>(scale_factors_data, scale_factors_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__upsample_bilinear2d_aa(tensor *out__, tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
-  PROTECT(
-    auto outputs__ = torch::_upsample_bilinear2d_aa(*self, torch::IntArrayRef(output_size_data, output_size_len), (bool)align_corners, scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__upsample_bilinear2d_aa_backward(tensor *out__, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
-  PROTECT(
-    auto outputs__ = torch::_upsample_bilinear2d_aa_backward(*grad_output, torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), (bool)align_corners, scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__upsample_bilinear2d_aa_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
-  PROTECT(
-    auto outputs__ = torch::_upsample_bilinear2d_aa_backward_out(*grad_input, *grad_output, torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), (bool)align_corners, scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__upsample_bilinear2d_aa_out(tensor *out__, tensor out, tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
-  PROTECT(
-    auto outputs__ = torch::_upsample_bilinear2d_aa_out(*out, *self, torch::IntArrayRef(output_size_data, output_size_len), (bool)align_corners, scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__upsample_bilinear2d_aa_vec(tensor *out__, tensor input, int64_t *output_size_data, int output_size_len, int align_corners, double *scale_factors_data, int scale_factors_len) {
-  PROTECT(
-    auto outputs__ = torch::_upsample_bilinear2d_aa(*input, output_size_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(output_size_data, output_size_len)), (bool)align_corners, at::ArrayRef<double>(scale_factors_data, scale_factors_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__upsample_nearest_exact1d(tensor *out__, tensor self, int64_t *output_size_data, int output_size_len, double scales_v, int scales_null) {
-  PROTECT(
-    auto outputs__ = torch::_upsample_nearest_exact1d(*self, torch::IntArrayRef(output_size_data, output_size_len), scales_null ? c10::nullopt : c10::optional<double>(scales_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__upsample_nearest_exact1d_backward(tensor *out__, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_v, int scales_null) {
-  PROTECT(
-    auto outputs__ = torch::_upsample_nearest_exact1d_backward(*grad_output, torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), scales_null ? c10::nullopt : c10::optional<double>(scales_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__upsample_nearest_exact1d_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_v, int scales_null) {
-  PROTECT(
-    auto outputs__ = torch::_upsample_nearest_exact1d_backward_out(*grad_input, *grad_output, torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), scales_null ? c10::nullopt : c10::optional<double>(scales_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__upsample_nearest_exact1d_out(tensor *out__, tensor out, tensor self, int64_t *output_size_data, int output_size_len, double scales_v, int scales_null) {
-  PROTECT(
-    auto outputs__ = torch::_upsample_nearest_exact1d_out(*out, *self, torch::IntArrayRef(output_size_data, output_size_len), scales_null ? c10::nullopt : c10::optional<double>(scales_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__upsample_nearest_exact1d_vec(tensor *out__, tensor input, int64_t *output_size_data, int output_size_len, double *scale_factors_data, int scale_factors_len) {
-  PROTECT(
-    auto outputs__ = torch::_upsample_nearest_exact1d(*input, output_size_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(output_size_data, output_size_len)), at::ArrayRef<double>(scale_factors_data, scale_factors_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__upsample_nearest_exact2d(tensor *out__, tensor self, int64_t *output_size_data, int output_size_len, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
-  PROTECT(
-    auto outputs__ = torch::_upsample_nearest_exact2d(*self, torch::IntArrayRef(output_size_data, output_size_len), scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__upsample_nearest_exact2d_backward(tensor *out__, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
-  PROTECT(
-    auto outputs__ = torch::_upsample_nearest_exact2d_backward(*grad_output, torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__upsample_nearest_exact2d_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
-  PROTECT(
-    auto outputs__ = torch::_upsample_nearest_exact2d_backward_out(*grad_input, *grad_output, torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__upsample_nearest_exact2d_out(tensor *out__, tensor out, tensor self, int64_t *output_size_data, int output_size_len, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
-  PROTECT(
-    auto outputs__ = torch::_upsample_nearest_exact2d_out(*out, *self, torch::IntArrayRef(output_size_data, output_size_len), scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__upsample_nearest_exact2d_vec(tensor *out__, tensor input, int64_t *output_size_data, int output_size_len, double *scale_factors_data, int scale_factors_len) {
-  PROTECT(
-    auto outputs__ = torch::_upsample_nearest_exact2d(*input, output_size_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(output_size_data, output_size_len)), at::ArrayRef<double>(scale_factors_data, scale_factors_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__upsample_nearest_exact3d(tensor *out__, tensor self, int64_t *output_size_data, int output_size_len, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
-  PROTECT(
-    auto outputs__ = torch::_upsample_nearest_exact3d(*self, torch::IntArrayRef(output_size_data, output_size_len), scales_d_null ? c10::nullopt : c10::optional<double>(scales_d_v), scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__upsample_nearest_exact3d_backward(tensor *out__, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
-  PROTECT(
-    auto outputs__ = torch::_upsample_nearest_exact3d_backward(*grad_output, torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), scales_d_null ? c10::nullopt : c10::optional<double>(scales_d_v), scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__upsample_nearest_exact3d_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
-  PROTECT(
-    auto outputs__ = torch::_upsample_nearest_exact3d_backward_out(*grad_input, *grad_output, torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), scales_d_null ? c10::nullopt : c10::optional<double>(scales_d_v), scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__upsample_nearest_exact3d_out(tensor *out__, tensor out, tensor self, int64_t *output_size_data, int output_size_len, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
-  PROTECT(
-    auto outputs__ = torch::_upsample_nearest_exact3d_out(*out, *self, torch::IntArrayRef(output_size_data, output_size_len), scales_d_null ? c10::nullopt : c10::optional<double>(scales_d_v), scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__upsample_nearest_exact3d_vec(tensor *out__, tensor input, int64_t *output_size_data, int output_size_len, double *scale_factors_data, int scale_factors_len) {
-  PROTECT(
-    auto outputs__ = torch::_upsample_nearest_exact3d(*input, output_size_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(output_size_data, output_size_len)), at::ArrayRef<double>(scale_factors_data, scale_factors_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-int atg__use_cudnn_ctc_loss(tensor log_probs, tensor targets, int64_t *input_lengths_data, int input_lengths_len, int64_t *target_lengths_data, int target_lengths_len, int64_t blank) {
-  PROTECT(
-    return torch::_use_cudnn_ctc_loss(*log_probs, *targets, torch::IntArrayRef(input_lengths_data, input_lengths_len), torch::IntArrayRef(target_lengths_data, target_lengths_len), blank);
-  )
-  return 0;
-}
-
-int atg__use_cudnn_ctc_loss_tensor(tensor log_probs, tensor targets, tensor input_lengths, tensor target_lengths, int64_t blank) {
-  PROTECT(
-    return torch::_use_cudnn_ctc_loss(*log_probs, *targets, *input_lengths, *target_lengths, blank);
-  )
-  return 0;
-}
-
-int atg__use_cudnn_rnn_flatten_weight() {
-  PROTECT(
-    return torch::_use_cudnn_rnn_flatten_weight();
-  )
-  return 0;
-}
-
-void atg__validate_compressed_sparse_indices(int is_crow, tensor compressed_idx, tensor plain_idx, int64_t cdim, int64_t dim, int64_t nnz) {
-  PROTECT(
-    torch::_validate_compressed_sparse_indices((bool)is_crow, *compressed_idx, *plain_idx, cdim, dim, nnz);
-  )
-}
-
-void atg__validate_sparse_bsc_tensor_args(tensor ccol_indices, tensor row_indices, tensor values, int64_t *size_data, int size_len) {
-  PROTECT(
-    torch::_validate_sparse_bsc_tensor_args(*ccol_indices, *row_indices, *values, torch::IntArrayRef(size_data, size_len));
-  )
-}
-
-void atg__validate_sparse_bsr_tensor_args(tensor crow_indices, tensor col_indices, tensor values, int64_t *size_data, int size_len) {
-  PROTECT(
-    torch::_validate_sparse_bsr_tensor_args(*crow_indices, *col_indices, *values, torch::IntArrayRef(size_data, size_len));
-  )
-}
-
-void atg__validate_sparse_csc_tensor_args(tensor ccol_indices, tensor row_indices, tensor values, int64_t *size_data, int size_len) {
-  PROTECT(
-    torch::_validate_sparse_csc_tensor_args(*ccol_indices, *row_indices, *values, torch::IntArrayRef(size_data, size_len));
-  )
-}
-
-void atg__values(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = self->_values();
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__values_copy(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::_values_copy(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__values_copy_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::_values_copy_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-int64_t atg__version(tensor self) {
-  PROTECT(
-    return self->_version();
-  )
-  return 0;
-}
-
-void atg__weight_norm(tensor *out__, tensor v, tensor g, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::_weight_norm(*v, *g, dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg__weight_norm_differentiable_backward(tensor *out__, tensor grad_w, tensor saved_v, tensor saved_g, tensor saved_norms, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::_weight_norm_differentiable_backward(*grad_w, *saved_v, *saved_g, *saved_norms, dim);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg__weight_norm_interface(tensor *out__, tensor v, tensor g, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::_weight_norm_interface(*v, *g, dim);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg__weight_norm_interface_backward(tensor *out__, tensor grad_w, tensor saved_v, tensor saved_g, tensor saved_norms, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::_weight_norm_interface_backward(*grad_w, *saved_v, *saved_g, *saved_norms, dim);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg__weight_norm_interface_backward_out(tensor *out__, tensor out0, tensor out1, tensor grad_w, tensor saved_v, tensor saved_g, tensor saved_norms, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::_weight_norm_interface_backward_out(*out0, *out1, *grad_w, *saved_v, *saved_g, *saved_norms, dim);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg__weight_norm_interface_out(tensor *out__, tensor out0, tensor out1, tensor v, tensor g, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::_weight_norm_interface_out(*out0, *out1, *v, *g, dim);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_abs(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::abs(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_abs_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::abs_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_abs_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::abs_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_absolute(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::absolute(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_absolute_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = self->absolute_();
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_absolute_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::absolute_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_acos(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::acos(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_acos_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::acos_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_acos_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::acos_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_acosh(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::acosh(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_acosh_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::acosh_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_acosh_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::acosh_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_adaptive_avg_pool1d(tensor *out__, tensor self, int64_t *output_size_data, int output_size_len) {
-  PROTECT(
-    auto outputs__ = torch::adaptive_avg_pool1d(*self, torch::IntArrayRef(output_size_data, output_size_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_adaptive_avg_pool2d(tensor *out__, tensor self, int64_t *output_size_data, int output_size_len) {
-  PROTECT(
-    auto outputs__ = torch::adaptive_avg_pool2d(*self, torch::IntArrayRef(output_size_data, output_size_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_adaptive_avg_pool2d_out(tensor *out__, tensor out, tensor self, int64_t *output_size_data, int output_size_len) {
-  PROTECT(
-    auto outputs__ = torch::adaptive_avg_pool2d_out(*out, *self, torch::IntArrayRef(output_size_data, output_size_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_adaptive_avg_pool3d(tensor *out__, tensor self, int64_t *output_size_data, int output_size_len) {
-  PROTECT(
-    auto outputs__ = torch::adaptive_avg_pool3d(*self, torch::IntArrayRef(output_size_data, output_size_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_adaptive_avg_pool3d_backward(tensor *out__, tensor grad_input, tensor grad_output, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::adaptive_avg_pool3d_backward_out(*grad_input, *grad_output, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_adaptive_avg_pool3d_out(tensor *out__, tensor out, tensor self, int64_t *output_size_data, int output_size_len) {
-  PROTECT(
-    auto outputs__ = torch::adaptive_avg_pool3d_out(*out, *self, torch::IntArrayRef(output_size_data, output_size_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_adaptive_max_pool1d(tensor *out__, tensor self, int64_t *output_size_data, int output_size_len) {
-  PROTECT(
-    auto outputs__ = torch::adaptive_max_pool1d(*self, torch::IntArrayRef(output_size_data, output_size_len));
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_adaptive_max_pool2d(tensor *out__, tensor self, int64_t *output_size_data, int output_size_len) {
-  PROTECT(
-    auto outputs__ = torch::adaptive_max_pool2d(*self, torch::IntArrayRef(output_size_data, output_size_len));
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_adaptive_max_pool2d_backward(tensor *out__, tensor grad_output, tensor self, tensor indices) {
-  PROTECT(
-    auto outputs__ = torch::adaptive_max_pool2d_backward(*grad_output, *self, *indices);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_adaptive_max_pool2d_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, tensor self, tensor indices) {
-  PROTECT(
-    auto outputs__ = torch::adaptive_max_pool2d_backward_out(*grad_input, *grad_output, *self, *indices);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_adaptive_max_pool2d_out(tensor *out__, tensor out, tensor indices, tensor self, int64_t *output_size_data, int output_size_len) {
-  PROTECT(
-    auto outputs__ = torch::adaptive_max_pool2d_out(*out, *indices, *self, torch::IntArrayRef(output_size_data, output_size_len));
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_adaptive_max_pool3d(tensor *out__, tensor self, int64_t *output_size_data, int output_size_len) {
-  PROTECT(
-    auto outputs__ = torch::adaptive_max_pool3d(*self, torch::IntArrayRef(output_size_data, output_size_len));
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_adaptive_max_pool3d_backward(tensor *out__, tensor grad_output, tensor self, tensor indices) {
-  PROTECT(
-    auto outputs__ = torch::adaptive_max_pool3d_backward(*grad_output, *self, *indices);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_adaptive_max_pool3d_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, tensor self, tensor indices) {
-  PROTECT(
-    auto outputs__ = torch::adaptive_max_pool3d_backward_out(*grad_input, *grad_output, *self, *indices);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_adaptive_max_pool3d_out(tensor *out__, tensor out, tensor indices, tensor self, int64_t *output_size_data, int output_size_len) {
-  PROTECT(
-    auto outputs__ = torch::adaptive_max_pool3d_out(*out, *indices, *self, torch::IntArrayRef(output_size_data, output_size_len));
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_add(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::add(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_add_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->add_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_add_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::add_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_add_scalar(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::add(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_add_scalar_(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = self->add_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_add_scalar_out(tensor *out__, tensor out, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::add_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_addbmm(tensor *out__, tensor self, tensor batch1, tensor batch2) {
-  PROTECT(
-    auto outputs__ = torch::addbmm(*self, *batch1, *batch2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_addbmm_(tensor *out__, tensor self, tensor batch1, tensor batch2) {
-  PROTECT(
-    auto outputs__ = self->addbmm_(*batch1, *batch2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_addbmm_out(tensor *out__, tensor out, tensor self, tensor batch1, tensor batch2) {
-  PROTECT(
-    auto outputs__ = torch::addbmm_out(*out, *self, *batch1, *batch2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_addcdiv(tensor *out__, tensor self, tensor tensor1, tensor tensor2) {
-  PROTECT(
-    auto outputs__ = torch::addcdiv(*self, *tensor1, *tensor2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_addcdiv_(tensor *out__, tensor self, tensor tensor1, tensor tensor2) {
-  PROTECT(
-    auto outputs__ = self->addcdiv_(*tensor1, *tensor2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_addcdiv_out(tensor *out__, tensor out, tensor self, tensor tensor1, tensor tensor2) {
-  PROTECT(
-    auto outputs__ = torch::addcdiv_out(*out, *self, *tensor1, *tensor2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_addcmul(tensor *out__, tensor self, tensor tensor1, tensor tensor2) {
-  PROTECT(
-    auto outputs__ = torch::addcmul(*self, *tensor1, *tensor2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_addcmul_(tensor *out__, tensor self, tensor tensor1, tensor tensor2) {
-  PROTECT(
-    auto outputs__ = self->addcmul_(*tensor1, *tensor2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_addcmul_out(tensor *out__, tensor out, tensor self, tensor tensor1, tensor tensor2) {
-  PROTECT(
-    auto outputs__ = torch::addcmul_out(*out, *self, *tensor1, *tensor2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_addmm(tensor *out__, tensor self, tensor mat1, tensor mat2) {
-  PROTECT(
-    auto outputs__ = torch::addmm(*self, *mat1, *mat2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_addmm_(tensor *out__, tensor self, tensor mat1, tensor mat2) {
-  PROTECT(
-    auto outputs__ = self->addmm_(*mat1, *mat2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_addmm_out(tensor *out__, tensor out, tensor self, tensor mat1, tensor mat2) {
-  PROTECT(
-    auto outputs__ = torch::addmm_out(*out, *self, *mat1, *mat2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_addmv(tensor *out__, tensor self, tensor mat, tensor vec) {
-  PROTECT(
-    auto outputs__ = torch::addmv(*self, *mat, *vec);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_addmv_(tensor *out__, tensor self, tensor mat, tensor vec) {
-  PROTECT(
-    auto outputs__ = torch::addmv_(*self, *mat, *vec);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_addmv_out(tensor *out__, tensor out, tensor self, tensor mat, tensor vec) {
-  PROTECT(
-    auto outputs__ = torch::addmv_out(*out, *self, *mat, *vec);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_addr(tensor *out__, tensor self, tensor vec1, tensor vec2) {
-  PROTECT(
-    auto outputs__ = torch::addr(*self, *vec1, *vec2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_addr_(tensor *out__, tensor self, tensor vec1, tensor vec2) {
-  PROTECT(
-    auto outputs__ = self->addr_(*vec1, *vec2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_addr_out(tensor *out__, tensor out, tensor self, tensor vec1, tensor vec2) {
-  PROTECT(
-    auto outputs__ = torch::addr_out(*out, *self, *vec1, *vec2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_adjoint(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::adjoint(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_affine_grid_generator(tensor *out__, tensor theta, int64_t *size_data, int size_len, int align_corners) {
-  PROTECT(
-    auto outputs__ = torch::affine_grid_generator(*theta, torch::IntArrayRef(size_data, size_len), (bool)align_corners);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_affine_grid_generator_backward(tensor *out__, tensor grad, int64_t *size_data, int size_len, int align_corners) {
-  PROTECT(
-    auto outputs__ = torch::affine_grid_generator_backward(*grad, torch::IntArrayRef(size_data, size_len), (bool)align_corners);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_affine_grid_generator_out(tensor *out__, tensor out, tensor theta, int64_t *size_data, int size_len, int align_corners) {
-  PROTECT(
-    auto outputs__ = torch::affine_grid_generator_out(*out, *theta, torch::IntArrayRef(size_data, size_len), (bool)align_corners);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_alias(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::alias(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_alias_copy(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::alias_copy(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_alias_copy_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::alias_copy_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_align_as(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->align_as(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-tensor *atg_align_tensors(tensor *tensors_data, int tensors_len) {
-  PROTECT(
-    auto outputs__ = torch::align_tensors(of_carray_tensor(tensors_data, tensors_len));
-    int sz = outputs__.size();
-    torch::Tensor **out__ = (torch::Tensor**)malloc((sz + 1) * sizeof(torch::Tensor*));
-    for (int i = 0; i < sz; ++i)
-      out__[i] = new torch::Tensor(outputs__[i]);
-    out__[sz] = nullptr;
-    return out__;
-  )
-}
-
-void atg_all(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::all(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_all_all_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::all_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_all_dim(tensor *out__, tensor self, int64_t dim, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::all(*self, dim, (bool)keepdim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_all_out(tensor *out__, tensor out, tensor self, int64_t dim, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::all_out(*out, *self, dim, (bool)keepdim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-int atg_allclose(tensor self, tensor other, double rtol, double atol, int equal_nan) {
-  PROTECT(
-    return torch::allclose(*self, *other, rtol, atol, (bool)equal_nan);
-  )
-  return 0;
-}
-
-void atg_alpha_dropout(tensor *out__, tensor input, double p, int train) {
-  PROTECT(
-    auto outputs__ = torch::alpha_dropout(*input, p, (bool)train);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_alpha_dropout_(tensor *out__, tensor self, double p, int train) {
-  PROTECT(
-    auto outputs__ = torch::alpha_dropout_(*self, p, (bool)train);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_amax(tensor *out__, tensor self, int64_t *dim_data, int dim_len, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::amax(*self, torch::IntArrayRef(dim_data, dim_len), (bool)keepdim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_amax_out(tensor *out__, tensor out, tensor self, int64_t *dim_data, int dim_len, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::amax_out(*out, *self, torch::IntArrayRef(dim_data, dim_len), (bool)keepdim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_amin(tensor *out__, tensor self, int64_t *dim_data, int dim_len, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::amin(*self, torch::IntArrayRef(dim_data, dim_len), (bool)keepdim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_amin_out(tensor *out__, tensor out, tensor self, int64_t *dim_data, int dim_len, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::amin_out(*out, *self, torch::IntArrayRef(dim_data, dim_len), (bool)keepdim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_aminmax(tensor *out__, tensor self, int64_t dim_v, int dim_null, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::aminmax(*self, dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v), (bool)keepdim);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_aminmax_out(tensor *out__, tensor min, tensor max, tensor self, int64_t dim_v, int dim_null, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::aminmax_out(*min, *max, *self, dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v), (bool)keepdim);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_angle(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::angle(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_angle_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::angle_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_any(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::any(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_any_all_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::any_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_any_dim(tensor *out__, tensor self, int64_t dim, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::any(*self, dim, (bool)keepdim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_any_out(tensor *out__, tensor out, tensor self, int64_t dim, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::any_out(*out, *self, dim, (bool)keepdim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_arange(tensor *out__, scalar end, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::arange(*end, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_arange_start(tensor *out__, scalar start, scalar end, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::arange(*start, *end, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_arange_start_step(tensor *out__, scalar start, scalar end, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::arange(*start, *end, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_arccos(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::arccos(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_arccos_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::arccos_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_arccos_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::arccos_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_arccosh(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::arccosh(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_arccosh_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::arccosh_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_arccosh_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::arccosh_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_arcsin(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::arcsin(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_arcsin_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::arcsin_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_arcsin_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::arcsin_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_arcsinh(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::arcsinh(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_arcsinh_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::arcsinh_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_arcsinh_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::arcsinh_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_arctan(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::arctan(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_arctan2(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::arctan2(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_arctan2_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->arctan2_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_arctan2_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::arctan2_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_arctan_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::arctan_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_arctan_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::arctan_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_arctanh(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::arctanh(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_arctanh_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::arctanh_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_arctanh_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::arctanh_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_argmax(tensor *out__, tensor self, int64_t dim_v, int dim_null, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::argmax(*self, dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v), (bool)keepdim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_argmax_out(tensor *out__, tensor out, tensor self, int64_t dim_v, int dim_null, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::argmax_out(*out, *self, dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v), (bool)keepdim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_argmin(tensor *out__, tensor self, int64_t dim_v, int dim_null, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::argmin(*self, dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v), (bool)keepdim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_argmin_out(tensor *out__, tensor out, tensor self, int64_t dim_v, int dim_null, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::argmin_out(*out, *self, dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v), (bool)keepdim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_argsort(tensor *out__, tensor self, int64_t dim, int descending) {
-  PROTECT(
-    auto outputs__ = torch::argsort(*self, dim, (bool)descending);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_argsort_stable(tensor *out__, tensor self, int stable, int64_t dim, int descending) {
-  PROTECT(
-    auto outputs__ = torch::argsort(*self, (bool)stable, dim, (bool)descending);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_argsort_stable_out(tensor *out__, tensor out, tensor self, int stable, int64_t dim, int descending) {
-  PROTECT(
-    auto outputs__ = torch::argsort_out(*out, *self, (bool)stable, dim, (bool)descending);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_argwhere(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::argwhere(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_as_strided(tensor *out__, tensor self, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len, int64_t storage_offset_v, int storage_offset_null) {
-  PROTECT(
-    auto outputs__ = torch::as_strided(*self, torch::IntArrayRef(size_data, size_len), torch::IntArrayRef(stride_data, stride_len), storage_offset_null ? c10::nullopt : c10::optional<int64_t>(storage_offset_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_as_strided_(tensor *out__, tensor self, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len, int64_t storage_offset_v, int storage_offset_null) {
-  PROTECT(
-    auto outputs__ = torch::as_strided_(*self, torch::IntArrayRef(size_data, size_len), torch::IntArrayRef(stride_data, stride_len), storage_offset_null ? c10::nullopt : c10::optional<int64_t>(storage_offset_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_as_strided_copy(tensor *out__, tensor self, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len, int64_t storage_offset_v, int storage_offset_null) {
-  PROTECT(
-    auto outputs__ = torch::as_strided_copy(*self, torch::IntArrayRef(size_data, size_len), torch::IntArrayRef(stride_data, stride_len), storage_offset_null ? c10::nullopt : c10::optional<int64_t>(storage_offset_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_as_strided_copy_out(tensor *out__, tensor out, tensor self, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len, int64_t storage_offset_v, int storage_offset_null) {
-  PROTECT(
-    auto outputs__ = torch::as_strided_copy_out(*out, *self, torch::IntArrayRef(size_data, size_len), torch::IntArrayRef(stride_data, stride_len), storage_offset_null ? c10::nullopt : c10::optional<int64_t>(storage_offset_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_as_strided_scatter(tensor *out__, tensor self, tensor src, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len, int64_t storage_offset_v, int storage_offset_null) {
-  PROTECT(
-    auto outputs__ = torch::as_strided_scatter(*self, *src, torch::IntArrayRef(size_data, size_len), torch::IntArrayRef(stride_data, stride_len), storage_offset_null ? c10::nullopt : c10::optional<int64_t>(storage_offset_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_as_strided_scatter_out(tensor *out__, tensor out, tensor self, tensor src, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len, int64_t storage_offset_v, int storage_offset_null) {
-  PROTECT(
-    auto outputs__ = torch::as_strided_scatter_out(*out, *self, *src, torch::IntArrayRef(size_data, size_len), torch::IntArrayRef(stride_data, stride_len), storage_offset_null ? c10::nullopt : c10::optional<int64_t>(storage_offset_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_asin(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::asin(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_asin_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::asin_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_asin_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::asin_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_asinh(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::asinh(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_asinh_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::asinh_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_asinh_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::asinh_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_atan(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::atan(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_atan2(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::atan2(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_atan2_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->atan2_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_atan2_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::atan2_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_atan_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::atan_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_atan_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::atan_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_atanh(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::atanh(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_atanh_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::atanh_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_atanh_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::atanh_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_atleast_1d(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::atleast_1d(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-tensor *atg_atleast_1d_sequence(tensor *tensors_data, int tensors_len) {
-  PROTECT(
-    auto outputs__ = torch::atleast_1d(of_carray_tensor(tensors_data, tensors_len));
-    int sz = outputs__.size();
-    torch::Tensor **out__ = (torch::Tensor**)malloc((sz + 1) * sizeof(torch::Tensor*));
-    for (int i = 0; i < sz; ++i)
-      out__[i] = new torch::Tensor(outputs__[i]);
-    out__[sz] = nullptr;
-    return out__;
-  )
-}
-
-void atg_atleast_2d(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::atleast_2d(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-tensor *atg_atleast_2d_sequence(tensor *tensors_data, int tensors_len) {
-  PROTECT(
-    auto outputs__ = torch::atleast_2d(of_carray_tensor(tensors_data, tensors_len));
-    int sz = outputs__.size();
-    torch::Tensor **out__ = (torch::Tensor**)malloc((sz + 1) * sizeof(torch::Tensor*));
-    for (int i = 0; i < sz; ++i)
-      out__[i] = new torch::Tensor(outputs__[i]);
-    out__[sz] = nullptr;
-    return out__;
-  )
-}
-
-void atg_atleast_3d(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::atleast_3d(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-tensor *atg_atleast_3d_sequence(tensor *tensors_data, int tensors_len) {
-  PROTECT(
-    auto outputs__ = torch::atleast_3d(of_carray_tensor(tensors_data, tensors_len));
-    int sz = outputs__.size();
-    torch::Tensor **out__ = (torch::Tensor**)malloc((sz + 1) * sizeof(torch::Tensor*));
-    for (int i = 0; i < sz; ++i)
-      out__[i] = new torch::Tensor(outputs__[i]);
-    out__[sz] = nullptr;
-    return out__;
-  )
-}
-
-void atg_avg_pool1d(tensor *out__, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int ceil_mode, int count_include_pad) {
-  PROTECT(
-    auto outputs__ = torch::avg_pool1d(*self, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), (bool)ceil_mode, (bool)count_include_pad);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_avg_pool2d(tensor *out__, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int ceil_mode, int count_include_pad, int64_t divisor_override_v, int divisor_override_null) {
-  PROTECT(
-    auto outputs__ = torch::avg_pool2d(*self, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), (bool)ceil_mode, (bool)count_include_pad, divisor_override_null ? c10::nullopt : c10::optional<int64_t>(divisor_override_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_avg_pool2d_backward(tensor *out__, tensor grad_output, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int ceil_mode, int count_include_pad, int64_t divisor_override_v, int divisor_override_null) {
-  PROTECT(
-    auto outputs__ = torch::avg_pool2d_backward(*grad_output, *self, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), (bool)ceil_mode, (bool)count_include_pad, divisor_override_null ? c10::nullopt : c10::optional<int64_t>(divisor_override_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_avg_pool2d_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int ceil_mode, int count_include_pad, int64_t divisor_override_v, int divisor_override_null) {
-  PROTECT(
-    auto outputs__ = torch::avg_pool2d_backward_out(*grad_input, *grad_output, *self, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), (bool)ceil_mode, (bool)count_include_pad, divisor_override_null ? c10::nullopt : c10::optional<int64_t>(divisor_override_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_avg_pool2d_out(tensor *out__, tensor out, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int ceil_mode, int count_include_pad, int64_t divisor_override_v, int divisor_override_null) {
-  PROTECT(
-    auto outputs__ = torch::avg_pool2d_out(*out, *self, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), (bool)ceil_mode, (bool)count_include_pad, divisor_override_null ? c10::nullopt : c10::optional<int64_t>(divisor_override_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_avg_pool3d(tensor *out__, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int ceil_mode, int count_include_pad, int64_t divisor_override_v, int divisor_override_null) {
-  PROTECT(
-    auto outputs__ = torch::avg_pool3d(*self, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), (bool)ceil_mode, (bool)count_include_pad, divisor_override_null ? c10::nullopt : c10::optional<int64_t>(divisor_override_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_avg_pool3d_backward(tensor *out__, tensor grad_output, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int ceil_mode, int count_include_pad, int64_t divisor_override_v, int divisor_override_null) {
-  PROTECT(
-    auto outputs__ = torch::avg_pool3d_backward(*grad_output, *self, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), (bool)ceil_mode, (bool)count_include_pad, divisor_override_null ? c10::nullopt : c10::optional<int64_t>(divisor_override_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_avg_pool3d_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int ceil_mode, int count_include_pad, int64_t divisor_override_v, int divisor_override_null) {
-  PROTECT(
-    auto outputs__ = torch::avg_pool3d_backward_out(*grad_input, *grad_output, *self, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), (bool)ceil_mode, (bool)count_include_pad, divisor_override_null ? c10::nullopt : c10::optional<int64_t>(divisor_override_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_avg_pool3d_out(tensor *out__, tensor out, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int ceil_mode, int count_include_pad, int64_t divisor_override_v, int divisor_override_null) {
-  PROTECT(
-    auto outputs__ = torch::avg_pool3d_out(*out, *self, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), (bool)ceil_mode, (bool)count_include_pad, divisor_override_null ? c10::nullopt : c10::optional<int64_t>(divisor_override_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_baddbmm(tensor *out__, tensor self, tensor batch1, tensor batch2) {
-  PROTECT(
-    auto outputs__ = torch::baddbmm(*self, *batch1, *batch2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_baddbmm_(tensor *out__, tensor self, tensor batch1, tensor batch2) {
-  PROTECT(
-    auto outputs__ = self->baddbmm_(*batch1, *batch2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_baddbmm_out(tensor *out__, tensor out, tensor self, tensor batch1, tensor batch2) {
-  PROTECT(
-    auto outputs__ = torch::baddbmm_out(*out, *self, *batch1, *batch2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bartlett_window(tensor *out__, int64_t window_length, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::bartlett_window(window_length, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bartlett_window_out(tensor *out__, tensor out, int64_t window_length) {
-  PROTECT(
-    auto outputs__ = torch::bartlett_window_out(*out, window_length);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bartlett_window_periodic(tensor *out__, int64_t window_length, int periodic, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::bartlett_window(window_length, (bool)periodic, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bartlett_window_periodic_out(tensor *out__, tensor out, int64_t window_length, int periodic) {
-  PROTECT(
-    auto outputs__ = torch::bartlett_window_out(*out, window_length, (bool)periodic);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_batch_norm(tensor *out__, tensor input, tensor weight, tensor bias, tensor running_mean, tensor running_var, int training, double momentum, double eps, int cudnn_enabled) {
-  PROTECT(
-    auto outputs__ = torch::batch_norm(*input, (weight ? *weight : torch::Tensor()), (bias ? *bias : torch::Tensor()), (running_mean ? *running_mean : torch::Tensor()), (running_var ? *running_var : torch::Tensor()), (bool)training, momentum, eps, (bool)cudnn_enabled);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_batch_norm_backward_elemt(tensor *out__, tensor grad_out, tensor input, tensor mean, tensor invstd, tensor weight, tensor sum_dy, tensor sum_dy_xmu, tensor count) {
-  PROTECT(
-    auto outputs__ = torch::batch_norm_backward_elemt(*grad_out, *input, *mean, *invstd, (weight ? *weight : torch::Tensor()), *sum_dy, *sum_dy_xmu, *count);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_batch_norm_backward_elemt_out(tensor *out__, tensor out, tensor grad_out, tensor input, tensor mean, tensor invstd, tensor weight, tensor sum_dy, tensor sum_dy_xmu, tensor count) {
-  PROTECT(
-    auto outputs__ = torch::batch_norm_backward_elemt_out(*out, *grad_out, *input, *mean, *invstd, (weight ? *weight : torch::Tensor()), *sum_dy, *sum_dy_xmu, *count);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_batch_norm_backward_reduce(tensor *out__, tensor grad_out, tensor input, tensor mean, tensor invstd, tensor weight, int input_g, int weight_g, int bias_g) {
-  PROTECT(
-    auto outputs__ = torch::batch_norm_backward_reduce(*grad_out, *input, *mean, *invstd, (weight ? *weight : torch::Tensor()), (bool)input_g, (bool)weight_g, (bool)bias_g);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-    out__[3] = new torch::Tensor(std::get<3>(outputs__));
-  )
-}
-
-void atg_batch_norm_backward_reduce_out(tensor *out__, tensor out0, tensor out1, tensor out2, tensor out3, tensor grad_out, tensor input, tensor mean, tensor invstd, tensor weight, int input_g, int weight_g, int bias_g) {
-  PROTECT(
-    auto outputs__ = torch::batch_norm_backward_reduce_out(*out0, *out1, *out2, *out3, *grad_out, *input, *mean, *invstd, (weight ? *weight : torch::Tensor()), (bool)input_g, (bool)weight_g, (bool)bias_g);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-    out__[3] = new torch::Tensor(std::get<3>(outputs__));
-  )
-}
-
-void atg_batch_norm_elemt(tensor *out__, tensor input, tensor weight, tensor bias, tensor mean, tensor invstd, double eps) {
-  PROTECT(
-    auto outputs__ = torch::batch_norm_elemt(*input, (weight ? *weight : torch::Tensor()), (bias ? *bias : torch::Tensor()), *mean, *invstd, eps);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_batch_norm_elemt_out(tensor *out__, tensor out, tensor input, tensor weight, tensor bias, tensor mean, tensor invstd, double eps) {
-  PROTECT(
-    auto outputs__ = torch::batch_norm_elemt_out(*out, *input, (weight ? *weight : torch::Tensor()), (bias ? *bias : torch::Tensor()), *mean, *invstd, eps);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_batch_norm_gather_stats(tensor *out__, tensor input, tensor mean, tensor invstd, tensor running_mean, tensor running_var, double momentum, double eps, int64_t count) {
-  PROTECT(
-    auto outputs__ = torch::batch_norm_gather_stats(*input, *mean, *invstd, (running_mean ? *running_mean : torch::Tensor()), (running_var ? *running_var : torch::Tensor()), momentum, eps, count);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_batch_norm_gather_stats_out(tensor *out__, tensor out0, tensor out1, tensor input, tensor mean, tensor invstd, tensor running_mean, tensor running_var, double momentum, double eps, int64_t count) {
-  PROTECT(
-    auto outputs__ = torch::batch_norm_gather_stats_out(*out0, *out1, *input, *mean, *invstd, (running_mean ? *running_mean : torch::Tensor()), (running_var ? *running_var : torch::Tensor()), momentum, eps, count);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_batch_norm_gather_stats_with_counts(tensor *out__, tensor input, tensor mean, tensor invstd, tensor running_mean, tensor running_var, double momentum, double eps, tensor counts) {
-  PROTECT(
-    auto outputs__ = torch::batch_norm_gather_stats_with_counts(*input, *mean, *invstd, (running_mean ? *running_mean : torch::Tensor()), (running_var ? *running_var : torch::Tensor()), momentum, eps, *counts);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_batch_norm_gather_stats_with_counts_out(tensor *out__, tensor out0, tensor out1, tensor input, tensor mean, tensor invstd, tensor running_mean, tensor running_var, double momentum, double eps, tensor counts) {
-  PROTECT(
-    auto outputs__ = torch::batch_norm_gather_stats_with_counts_out(*out0, *out1, *input, *mean, *invstd, (running_mean ? *running_mean : torch::Tensor()), (running_var ? *running_var : torch::Tensor()), momentum, eps, *counts);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_batch_norm_stats(tensor *out__, tensor input, double eps) {
-  PROTECT(
-    auto outputs__ = torch::batch_norm_stats(*input, eps);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_batch_norm_stats_out(tensor *out__, tensor out0, tensor out1, tensor input, double eps) {
-  PROTECT(
-    auto outputs__ = torch::batch_norm_stats_out(*out0, *out1, *input, eps);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_batch_norm_update_stats(tensor *out__, tensor input, tensor running_mean, tensor running_var, double momentum) {
-  PROTECT(
-    auto outputs__ = torch::batch_norm_update_stats(*input, (running_mean ? *running_mean : torch::Tensor()), (running_var ? *running_var : torch::Tensor()), momentum);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_batch_norm_update_stats_out(tensor *out__, tensor out0, tensor out1, tensor input, tensor running_mean, tensor running_var, double momentum) {
-  PROTECT(
-    auto outputs__ = torch::batch_norm_update_stats_out(*out0, *out1, *input, (running_mean ? *running_mean : torch::Tensor()), (running_var ? *running_var : torch::Tensor()), momentum);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_bernoulli(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::bernoulli(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bernoulli_(tensor *out__, tensor self, tensor p) {
-  PROTECT(
-    auto outputs__ = self->bernoulli_(*p);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bernoulli_float_(tensor *out__, tensor self, double p) {
-  PROTECT(
-    auto outputs__ = self->bernoulli_(p);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bernoulli_p(tensor *out__, tensor self, double p) {
-  PROTECT(
-    auto outputs__ = torch::bernoulli(*self, p);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bernoulli_tensor(tensor *out__, tensor self, tensor p) {
-  PROTECT(
-    auto outputs__ = torch::bernoulli(*self, *p);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bilinear(tensor *out__, tensor input1, tensor input2, tensor weight, tensor bias) {
-  PROTECT(
-    auto outputs__ = torch::bilinear(*input1, *input2, *weight, (bias ? *bias : torch::Tensor()));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_binary_cross_entropy(tensor *out__, tensor self, tensor target, tensor weight, int64_t reduction) {
-  PROTECT(
-    auto outputs__ = torch::binary_cross_entropy(*self, *target, (weight ? *weight : torch::Tensor()), reduction);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_binary_cross_entropy_backward(tensor *out__, tensor grad_output, tensor self, tensor target, tensor weight, int64_t reduction) {
-  PROTECT(
-    auto outputs__ = torch::binary_cross_entropy_backward(*grad_output, *self, *target, (weight ? *weight : torch::Tensor()), reduction);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_binary_cross_entropy_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, tensor self, tensor target, tensor weight, int64_t reduction) {
-  PROTECT(
-    auto outputs__ = torch::binary_cross_entropy_backward_out(*grad_input, *grad_output, *self, *target, (weight ? *weight : torch::Tensor()), reduction);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_binary_cross_entropy_out(tensor *out__, tensor out, tensor self, tensor target, tensor weight, int64_t reduction) {
-  PROTECT(
-    auto outputs__ = torch::binary_cross_entropy_out(*out, *self, *target, (weight ? *weight : torch::Tensor()), reduction);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_binary_cross_entropy_with_logits(tensor *out__, tensor self, tensor target, tensor weight, tensor pos_weight, int64_t reduction) {
-  PROTECT(
-    auto outputs__ = torch::binary_cross_entropy_with_logits(*self, *target, (weight ? *weight : torch::Tensor()), (pos_weight ? *pos_weight : torch::Tensor()), reduction);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_binary_cross_entropy_with_logits_out(tensor *out__, tensor out, tensor self, tensor target, tensor weight, tensor pos_weight, int64_t reduction) {
-  PROTECT(
-    auto outputs__ = torch::binary_cross_entropy_with_logits_out(*out, *self, *target, (weight ? *weight : torch::Tensor()), (pos_weight ? *pos_weight : torch::Tensor()), reduction);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bincount(tensor *out__, tensor self, tensor weights, int64_t minlength) {
-  PROTECT(
-    auto outputs__ = torch::bincount(*self, (weights ? *weights : torch::Tensor()), minlength);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bincount_out(tensor *out__, tensor out, tensor self, tensor weights, int64_t minlength) {
-  PROTECT(
-    auto outputs__ = torch::bincount_out(*out, *self, (weights ? *weights : torch::Tensor()), minlength);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_binomial(tensor *out__, tensor count, tensor prob) {
-  PROTECT(
-    auto outputs__ = torch::binomial(*count, *prob);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_binomial_out(tensor *out__, tensor out, tensor count, tensor prob) {
-  PROTECT(
-    auto outputs__ = torch::binomial_out(*out, *count, *prob);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bitwise_and(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::bitwise_and(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bitwise_and_(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = self->bitwise_and_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bitwise_and_scalar_out(tensor *out__, tensor out, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::bitwise_and_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bitwise_and_scalar_tensor(tensor *out__, scalar self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::bitwise_and(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bitwise_and_scalar_tensor_out(tensor *out__, tensor out, scalar self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::bitwise_and_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bitwise_and_tensor(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::bitwise_and(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bitwise_and_tensor_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->bitwise_and_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bitwise_and_tensor_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::bitwise_and_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bitwise_left_shift(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::bitwise_left_shift(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bitwise_left_shift_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->bitwise_left_shift_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bitwise_left_shift_scalar_tensor(tensor *out__, scalar self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::bitwise_left_shift(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bitwise_left_shift_scalar_tensor_out(tensor *out__, tensor out, scalar self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::bitwise_left_shift_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bitwise_left_shift_tensor_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::bitwise_left_shift_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bitwise_left_shift_tensor_scalar(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::bitwise_left_shift(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bitwise_left_shift_tensor_scalar_(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = self->bitwise_left_shift_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bitwise_left_shift_tensor_scalar_out(tensor *out__, tensor out, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::bitwise_left_shift_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bitwise_not(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::bitwise_not(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bitwise_not_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = self->bitwise_not_();
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bitwise_not_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::bitwise_not_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bitwise_or(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::bitwise_or(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bitwise_or_(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = self->bitwise_or_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bitwise_or_scalar_out(tensor *out__, tensor out, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::bitwise_or_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bitwise_or_scalar_tensor(tensor *out__, scalar self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::bitwise_or(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bitwise_or_scalar_tensor_out(tensor *out__, tensor out, scalar self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::bitwise_or_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bitwise_or_tensor(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::bitwise_or(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bitwise_or_tensor_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->bitwise_or_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bitwise_or_tensor_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::bitwise_or_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bitwise_right_shift(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::bitwise_right_shift(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bitwise_right_shift_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->bitwise_right_shift_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bitwise_right_shift_scalar_tensor(tensor *out__, scalar self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::bitwise_right_shift(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bitwise_right_shift_scalar_tensor_out(tensor *out__, tensor out, scalar self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::bitwise_right_shift_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bitwise_right_shift_tensor_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::bitwise_right_shift_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bitwise_right_shift_tensor_scalar(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::bitwise_right_shift(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bitwise_right_shift_tensor_scalar_(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = self->bitwise_right_shift_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bitwise_right_shift_tensor_scalar_out(tensor *out__, tensor out, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::bitwise_right_shift_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bitwise_xor(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::bitwise_xor(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bitwise_xor_(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = self->bitwise_xor_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bitwise_xor_scalar_out(tensor *out__, tensor out, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::bitwise_xor_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bitwise_xor_scalar_tensor(tensor *out__, scalar self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::bitwise_xor(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bitwise_xor_scalar_tensor_out(tensor *out__, tensor out, scalar self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::bitwise_xor_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bitwise_xor_tensor(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::bitwise_xor(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bitwise_xor_tensor_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->bitwise_xor_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bitwise_xor_tensor_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::bitwise_xor_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_blackman_window(tensor *out__, int64_t window_length, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::blackman_window(window_length, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_blackman_window_out(tensor *out__, tensor out, int64_t window_length) {
-  PROTECT(
-    auto outputs__ = torch::blackman_window_out(*out, window_length);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_blackman_window_periodic(tensor *out__, int64_t window_length, int periodic, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::blackman_window(window_length, (bool)periodic, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_blackman_window_periodic_out(tensor *out__, tensor out, int64_t window_length, int periodic) {
-  PROTECT(
-    auto outputs__ = torch::blackman_window_out(*out, window_length, (bool)periodic);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_block_diag(tensor *out__, tensor *tensors_data, int tensors_len) {
-  PROTECT(
-    auto outputs__ = torch::block_diag(of_carray_tensor(tensors_data, tensors_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_block_diag_out(tensor *out__, tensor out, tensor *tensors_data, int tensors_len) {
-  PROTECT(
-    auto outputs__ = torch::block_diag_out(*out, of_carray_tensor(tensors_data, tensors_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bmm(tensor *out__, tensor self, tensor mat2) {
-  PROTECT(
-    auto outputs__ = torch::bmm(*self, *mat2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bmm_out(tensor *out__, tensor out, tensor self, tensor mat2) {
-  PROTECT(
-    auto outputs__ = torch::bmm_out(*out, *self, *mat2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-tensor *atg_broadcast_tensors(tensor *tensors_data, int tensors_len) {
-  PROTECT(
-    auto outputs__ = torch::broadcast_tensors(of_carray_tensor(tensors_data, tensors_len));
-    int sz = outputs__.size();
-    torch::Tensor **out__ = (torch::Tensor**)malloc((sz + 1) * sizeof(torch::Tensor*));
-    for (int i = 0; i < sz; ++i)
-      out__[i] = new torch::Tensor(outputs__[i]);
-    out__[sz] = nullptr;
-    return out__;
-  )
-}
-
-void atg_broadcast_to(tensor *out__, tensor self, int64_t *size_data, int size_len) {
-  PROTECT(
-    auto outputs__ = torch::broadcast_to(*self, torch::IntArrayRef(size_data, size_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bucketize(tensor *out__, tensor self, tensor boundaries, int out_int32, int right) {
-  PROTECT(
-    auto outputs__ = torch::bucketize(*self, *boundaries, (bool)out_int32, (bool)right);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bucketize_scalar(tensor *out__, scalar self, tensor boundaries, int out_int32, int right) {
-  PROTECT(
-    auto outputs__ = torch::bucketize(*self, *boundaries, (bool)out_int32, (bool)right);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bucketize_scalar_out(tensor *out__, tensor out, scalar self, tensor boundaries, int out_int32, int right) {
-  PROTECT(
-    auto outputs__ = torch::bucketize_out(*out, *self, *boundaries, (bool)out_int32, (bool)right);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_bucketize_tensor_out(tensor *out__, tensor out, tensor self, tensor boundaries, int out_int32, int right) {
-  PROTECT(
-    auto outputs__ = torch::bucketize_out(*out, *self, *boundaries, (bool)out_int32, (bool)right);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-int atg_can_cast(int from, int to) {
-  PROTECT(
-    return torch::can_cast(torch::ScalarType(from), torch::ScalarType(to));
-  )
-  return 0;
-}
-
-void atg_cartesian_prod(tensor *out__, tensor *tensors_data, int tensors_len) {
-  PROTECT(
-    auto outputs__ = torch::cartesian_prod(of_carray_tensor(tensors_data, tensors_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cat(tensor *out__, tensor *tensors_data, int tensors_len, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::cat(of_carray_tensor(tensors_data, tensors_len), dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cat_out(tensor *out__, tensor out, tensor *tensors_data, int tensors_len, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::cat_out(*out, of_carray_tensor(tensors_data, tensors_len), dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cauchy(tensor *out__, tensor self, double median, double sigma) {
-  PROTECT(
-    auto outputs__ = torch::cauchy(*self, median, sigma);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cauchy_(tensor *out__, tensor self, double median, double sigma) {
-  PROTECT(
-    auto outputs__ = self->cauchy_(median, sigma);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cauchy_out(tensor *out__, tensor out, tensor self, double median, double sigma) {
-  PROTECT(
-    auto outputs__ = torch::cauchy_out(*out, *self, median, sigma);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_ccol_indices(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = self->ccol_indices();
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_ccol_indices_copy(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::ccol_indices_copy(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_ccol_indices_copy_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::ccol_indices_copy_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cdist(tensor *out__, tensor x1, tensor x2, double p, int64_t compute_mode_v, int compute_mode_null) {
-  PROTECT(
-    auto outputs__ = torch::cdist(*x1, *x2, p, compute_mode_null ? c10::nullopt : c10::optional<int64_t>(compute_mode_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_ceil(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::ceil(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_ceil_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::ceil_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_ceil_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::ceil_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_celu(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::celu(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_celu_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::celu_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_celu_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::celu_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_chain_matmul(tensor *out__, tensor *matrices_data, int matrices_len) {
-  PROTECT(
-    auto outputs__ = torch::chain_matmul(of_carray_tensor(matrices_data, matrices_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_chain_matmul_out(tensor *out__, tensor out, tensor *matrices_data, int matrices_len) {
-  PROTECT(
-    auto outputs__ = torch::chain_matmul_out(*out, of_carray_tensor(matrices_data, matrices_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_chalf(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = self->chalf();
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_channel_shuffle(tensor *out__, tensor self, int64_t groups) {
-  PROTECT(
-    auto outputs__ = torch::channel_shuffle(*self, groups);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_channel_shuffle_out(tensor *out__, tensor out, tensor self, int64_t groups) {
-  PROTECT(
-    auto outputs__ = torch::channel_shuffle_out(*out, *self, groups);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cholesky(tensor *out__, tensor self, int upper) {
-  PROTECT(
-    auto outputs__ = torch::cholesky(*self, (bool)upper);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cholesky_inverse(tensor *out__, tensor self, int upper) {
-  PROTECT(
-    auto outputs__ = torch::cholesky_inverse(*self, (bool)upper);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cholesky_inverse_out(tensor *out__, tensor out, tensor self, int upper) {
-  PROTECT(
-    auto outputs__ = torch::cholesky_inverse_out(*out, *self, (bool)upper);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cholesky_out(tensor *out__, tensor out, tensor self, int upper) {
-  PROTECT(
-    auto outputs__ = torch::cholesky_out(*out, *self, (bool)upper);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cholesky_solve(tensor *out__, tensor self, tensor input2, int upper) {
-  PROTECT(
-    auto outputs__ = torch::cholesky_solve(*self, *input2, (bool)upper);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cholesky_solve_out(tensor *out__, tensor out, tensor self, tensor input2, int upper) {
-  PROTECT(
-    auto outputs__ = torch::cholesky_solve_out(*out, *self, *input2, (bool)upper);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_choose_qparams_optimized(tensor *out__, tensor input, int64_t numel, int64_t n_bins, double ratio, int64_t bit_width) {
-  PROTECT(
-    auto outputs__ = torch::choose_qparams_optimized(*input, numel, n_bins, ratio, bit_width);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-tensor *atg_chunk(tensor self, int64_t chunks, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::chunk(*self, chunks, dim);
-    int sz = outputs__.size();
-    torch::Tensor **out__ = (torch::Tensor**)malloc((sz + 1) * sizeof(torch::Tensor*));
-    for (int i = 0; i < sz; ++i)
-      out__[i] = new torch::Tensor(outputs__[i]);
-    out__[sz] = nullptr;
-    return out__;
-  )
-}
-
-void atg_clamp(tensor *out__, tensor self, scalar min, scalar max) {
-  PROTECT(
-    auto outputs__ = torch::clamp(*self, *min, *max);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_clamp_(tensor *out__, tensor self, scalar min, scalar max) {
-  PROTECT(
-    auto outputs__ = torch::clamp_(*self, *min, *max);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_clamp_max(tensor *out__, tensor self, scalar max) {
-  PROTECT(
-    auto outputs__ = torch::clamp_max(*self, *max);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_clamp_max_(tensor *out__, tensor self, scalar max) {
-  PROTECT(
-    auto outputs__ = torch::clamp_max_(*self, *max);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_clamp_max_out(tensor *out__, tensor out, tensor self, scalar max) {
-  PROTECT(
-    auto outputs__ = torch::clamp_max_out(*out, *self, *max);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_clamp_max_tensor(tensor *out__, tensor self, tensor max) {
-  PROTECT(
-    auto outputs__ = torch::clamp_max(*self, *max);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_clamp_max_tensor_(tensor *out__, tensor self, tensor max) {
-  PROTECT(
-    auto outputs__ = torch::clamp_max_(*self, *max);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_clamp_max_tensor_out(tensor *out__, tensor out, tensor self, tensor max) {
-  PROTECT(
-    auto outputs__ = torch::clamp_max_out(*out, *self, *max);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_clamp_min(tensor *out__, tensor self, scalar min) {
-  PROTECT(
-    auto outputs__ = torch::clamp_min(*self, *min);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_clamp_min_(tensor *out__, tensor self, scalar min) {
-  PROTECT(
-    auto outputs__ = torch::clamp_min_(*self, *min);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_clamp_min_out(tensor *out__, tensor out, tensor self, scalar min) {
-  PROTECT(
-    auto outputs__ = torch::clamp_min_out(*out, *self, *min);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_clamp_min_tensor(tensor *out__, tensor self, tensor min) {
-  PROTECT(
-    auto outputs__ = torch::clamp_min(*self, *min);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_clamp_min_tensor_(tensor *out__, tensor self, tensor min) {
-  PROTECT(
-    auto outputs__ = torch::clamp_min_(*self, *min);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_clamp_min_tensor_out(tensor *out__, tensor out, tensor self, tensor min) {
-  PROTECT(
-    auto outputs__ = torch::clamp_min_out(*out, *self, *min);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_clamp_out(tensor *out__, tensor out, tensor self, scalar min, scalar max) {
-  PROTECT(
-    auto outputs__ = torch::clamp_out(*out, *self, *min, *max);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_clamp_tensor(tensor *out__, tensor self, tensor min, tensor max) {
-  PROTECT(
-    auto outputs__ = torch::clamp(*self, (min ? *min : torch::Tensor()), (max ? *max : torch::Tensor()));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_clamp_tensor_(tensor *out__, tensor self, tensor min, tensor max) {
-  PROTECT(
-    auto outputs__ = torch::clamp_(*self, (min ? *min : torch::Tensor()), (max ? *max : torch::Tensor()));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_clamp_tensor_out(tensor *out__, tensor out, tensor self, tensor min, tensor max) {
-  PROTECT(
-    auto outputs__ = torch::clamp_out(*out, *self, (min ? *min : torch::Tensor()), (max ? *max : torch::Tensor()));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_clip(tensor *out__, tensor self, scalar min, scalar max) {
-  PROTECT(
-    auto outputs__ = torch::clip(*self, *min, *max);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_clip_(tensor *out__, tensor self, scalar min, scalar max) {
-  PROTECT(
-    auto outputs__ = torch::clip_(*self, *min, *max);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_clip_out(tensor *out__, tensor out, tensor self, scalar min, scalar max) {
-  PROTECT(
-    auto outputs__ = torch::clip_out(*out, *self, *min, *max);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_clip_tensor(tensor *out__, tensor self, tensor min, tensor max) {
-  PROTECT(
-    auto outputs__ = torch::clip(*self, (min ? *min : torch::Tensor()), (max ? *max : torch::Tensor()));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_clip_tensor_(tensor *out__, tensor self, tensor min, tensor max) {
-  PROTECT(
-    auto outputs__ = torch::clip_(*self, (min ? *min : torch::Tensor()), (max ? *max : torch::Tensor()));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_clip_tensor_out(tensor *out__, tensor out, tensor self, tensor min, tensor max) {
-  PROTECT(
-    auto outputs__ = torch::clip_out(*out, *self, (min ? *min : torch::Tensor()), (max ? *max : torch::Tensor()));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_clone(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::clone(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_clone_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::clone_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_coalesce(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = self->coalesce();
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_col2im(tensor *out__, tensor self, int64_t *output_size_data, int output_size_len, int64_t *kernel_size_data, int kernel_size_len, int64_t *dilation_data, int dilation_len, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len) {
-  PROTECT(
-    auto outputs__ = torch::col2im(*self, torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(dilation_data, dilation_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(stride_data, stride_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_col2im_out(tensor *out__, tensor out, tensor self, int64_t *output_size_data, int output_size_len, int64_t *kernel_size_data, int kernel_size_len, int64_t *dilation_data, int dilation_len, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len) {
-  PROTECT(
-    auto outputs__ = torch::col2im_out(*out, *self, torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(dilation_data, dilation_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(stride_data, stride_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_col_indices(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = self->col_indices();
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_col_indices_copy(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::col_indices_copy(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_col_indices_copy_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::col_indices_copy_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_column_stack(tensor *out__, tensor *tensors_data, int tensors_len) {
-  PROTECT(
-    auto outputs__ = torch::column_stack(of_carray_tensor(tensors_data, tensors_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_column_stack_out(tensor *out__, tensor out, tensor *tensors_data, int tensors_len) {
-  PROTECT(
-    auto outputs__ = torch::column_stack_out(*out, of_carray_tensor(tensors_data, tensors_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_combinations(tensor *out__, tensor self, int64_t r, int with_replacement) {
-  PROTECT(
-    auto outputs__ = torch::combinations(*self, r, (bool)with_replacement);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_complex(tensor *out__, tensor real, tensor imag) {
-  PROTECT(
-    auto outputs__ = torch::complex(*real, *imag);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_complex_out(tensor *out__, tensor out, tensor real, tensor imag) {
-  PROTECT(
-    auto outputs__ = torch::complex_out(*out, *real, *imag);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_concat(tensor *out__, tensor *tensors_data, int tensors_len, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::concat(of_carray_tensor(tensors_data, tensors_len), dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_concat_out(tensor *out__, tensor out, tensor *tensors_data, int tensors_len, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::concat_out(*out, of_carray_tensor(tensors_data, tensors_len), dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_concatenate(tensor *out__, tensor *tensors_data, int tensors_len, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::concatenate(of_carray_tensor(tensors_data, tensors_len), dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_concatenate_out(tensor *out__, tensor out, tensor *tensors_data, int tensors_len, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::concatenate_out(*out, of_carray_tensor(tensors_data, tensors_len), dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_conj(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::conj(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_conj_physical(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::conj_physical(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_conj_physical_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::conj_physical_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_conj_physical_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::conj_physical_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_constant_pad_nd(tensor *out__, tensor self, int64_t *pad_data, int pad_len) {
-  PROTECT(
-    auto outputs__ = torch::constant_pad_nd(*self, torch::IntArrayRef(pad_data, pad_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_constant_pad_nd_out(tensor *out__, tensor out, tensor self, int64_t *pad_data, int pad_len) {
-  PROTECT(
-    auto outputs__ = torch::constant_pad_nd_out(*out, *self, torch::IntArrayRef(pad_data, pad_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_contiguous(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = self->contiguous();
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_conv1d(tensor *out__, tensor input, tensor weight, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int64_t groups) {
-  PROTECT(
-    auto outputs__ = torch::conv1d(*input, *weight, (bias ? *bias : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), groups);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_conv1d_padding(tensor *out__, tensor input, tensor weight, tensor bias, int64_t *stride_data, int stride_len, char * padding, int64_t *dilation_data, int dilation_len, int64_t groups) {
-  PROTECT(
-    auto outputs__ = torch::conv1d(*input, *weight, (bias ? *bias : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), std::string(padding), torch::IntArrayRef(dilation_data, dilation_len), groups);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_conv2d(tensor *out__, tensor input, tensor weight, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int64_t groups) {
-  PROTECT(
-    auto outputs__ = torch::conv2d(*input, *weight, (bias ? *bias : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), groups);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_conv2d_padding(tensor *out__, tensor input, tensor weight, tensor bias, int64_t *stride_data, int stride_len, char * padding, int64_t *dilation_data, int dilation_len, int64_t groups) {
-  PROTECT(
-    auto outputs__ = torch::conv2d(*input, *weight, (bias ? *bias : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), std::string(padding), torch::IntArrayRef(dilation_data, dilation_len), groups);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_conv3d(tensor *out__, tensor input, tensor weight, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int64_t groups) {
-  PROTECT(
-    auto outputs__ = torch::conv3d(*input, *weight, (bias ? *bias : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), groups);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_conv3d_padding(tensor *out__, tensor input, tensor weight, tensor bias, int64_t *stride_data, int stride_len, char * padding, int64_t *dilation_data, int dilation_len, int64_t groups) {
-  PROTECT(
-    auto outputs__ = torch::conv3d(*input, *weight, (bias ? *bias : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), std::string(padding), torch::IntArrayRef(dilation_data, dilation_len), groups);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_conv_depthwise3d(tensor *out__, tensor self, tensor weight, int64_t *kernel_size_data, int kernel_size_len, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len) {
-  PROTECT(
-    auto outputs__ = torch::conv_depthwise3d(*self, *weight, torch::IntArrayRef(kernel_size_data, kernel_size_len), (bias ? *bias : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_conv_depthwise3d_out(tensor *out__, tensor out, tensor self, tensor weight, int64_t *kernel_size_data, int kernel_size_len, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len) {
-  PROTECT(
-    auto outputs__ = torch::conv_depthwise3d_out(*out, *self, *weight, torch::IntArrayRef(kernel_size_data, kernel_size_len), (bias ? *bias : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_conv_tbc(tensor *out__, tensor self, tensor weight, tensor bias, int64_t pad) {
-  PROTECT(
-    auto outputs__ = torch::conv_tbc(*self, *weight, *bias, pad);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_conv_tbc_backward(tensor *out__, tensor self, tensor input, tensor weight, tensor bias, int64_t pad) {
-  PROTECT(
-    auto outputs__ = torch::conv_tbc_backward(*self, *input, *weight, *bias, pad);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg_conv_tbc_out(tensor *out__, tensor out, tensor self, tensor weight, tensor bias, int64_t pad) {
-  PROTECT(
-    auto outputs__ = torch::conv_tbc_out(*out, *self, *weight, *bias, pad);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_conv_transpose1d(tensor *out__, tensor input, tensor weight, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t groups, int64_t *dilation_data, int dilation_len) {
-  PROTECT(
-    auto outputs__ = torch::conv_transpose1d(*input, *weight, (bias ? *bias : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(output_padding_data, output_padding_len), groups, torch::IntArrayRef(dilation_data, dilation_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_conv_transpose2d(tensor *out__, tensor input, tensor weight, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t groups, int64_t *dilation_data, int dilation_len) {
-  PROTECT(
-    auto outputs__ = torch::conv_transpose2d(*input, *weight, (bias ? *bias : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(output_padding_data, output_padding_len), groups, torch::IntArrayRef(dilation_data, dilation_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_conv_transpose3d(tensor *out__, tensor input, tensor weight, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t groups, int64_t *dilation_data, int dilation_len) {
-  PROTECT(
-    auto outputs__ = torch::conv_transpose3d(*input, *weight, (bias ? *bias : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(output_padding_data, output_padding_len), groups, torch::IntArrayRef(dilation_data, dilation_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_convolution(tensor *out__, tensor input, tensor weight, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int transposed, int64_t *output_padding_data, int output_padding_len, int64_t groups) {
-  PROTECT(
-    auto outputs__ = torch::convolution(*input, *weight, (bias ? *bias : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)transposed, torch::IntArrayRef(output_padding_data, output_padding_len), groups);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_convolution_out(tensor *out__, tensor out, tensor input, tensor weight, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int transposed, int64_t *output_padding_data, int output_padding_len, int64_t groups) {
-  PROTECT(
-    auto outputs__ = torch::convolution_out(*out, *input, *weight, (bias ? *bias : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)transposed, torch::IntArrayRef(output_padding_data, output_padding_len), groups);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_convolution_overrideable(tensor *out__, tensor input, tensor weight, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int transposed, int64_t *output_padding_data, int output_padding_len, int64_t groups) {
-  PROTECT(
-    auto outputs__ = torch::convolution_overrideable(*input, *weight, (bias ? *bias : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)transposed, torch::IntArrayRef(output_padding_data, output_padding_len), groups);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_convolution_overrideable_out(tensor *out__, tensor out, tensor input, tensor weight, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int transposed, int64_t *output_padding_data, int output_padding_len, int64_t groups) {
-  PROTECT(
-    auto outputs__ = torch::convolution_overrideable_out(*out, *input, *weight, (bias ? *bias : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)transposed, torch::IntArrayRef(output_padding_data, output_padding_len), groups);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_copy(tensor *out__, tensor self, tensor src, int non_blocking) {
-  PROTECT(
-    auto outputs__ = torch::copy(*self, *src, (bool)non_blocking);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_copy_out(tensor *out__, tensor out, tensor self, tensor src, int non_blocking) {
-  PROTECT(
-    auto outputs__ = torch::copy_out(*out, *self, *src, (bool)non_blocking);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_copy_sparse_to_sparse(tensor *out__, tensor self, tensor src, int non_blocking) {
-  PROTECT(
-    auto outputs__ = torch::copy_sparse_to_sparse(*self, *src, (bool)non_blocking);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_copy_sparse_to_sparse_(tensor *out__, tensor self, tensor src, int non_blocking) {
-  PROTECT(
-    auto outputs__ = torch::copy_sparse_to_sparse_(*self, *src, (bool)non_blocking);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_copy_sparse_to_sparse_out(tensor *out__, tensor out, tensor self, tensor src, int non_blocking) {
-  PROTECT(
-    auto outputs__ = torch::copy_sparse_to_sparse_out(*out, *self, *src, (bool)non_blocking);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_copysign(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::copysign(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_copysign_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->copysign_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_copysign_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::copysign_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_copysign_scalar(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::copysign(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_copysign_scalar_(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = self->copysign_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_copysign_scalar_out(tensor *out__, tensor out, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::copysign_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_corrcoef(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::corrcoef(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cos(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::cos(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cos_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::cos_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cos_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::cos_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cosh(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::cosh(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cosh_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::cosh_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cosh_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::cosh_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cosine_embedding_loss(tensor *out__, tensor input1, tensor input2, tensor target, double margin, int64_t reduction) {
-  PROTECT(
-    auto outputs__ = torch::cosine_embedding_loss(*input1, *input2, *target, margin, reduction);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cosine_similarity(tensor *out__, tensor x1, tensor x2, int64_t dim, double eps) {
-  PROTECT(
-    auto outputs__ = torch::cosine_similarity(*x1, *x2, dim, eps);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_count_nonzero(tensor *out__, tensor out, tensor self, int64_t *dim_data, int dim_len) {
-  PROTECT(
-    auto outputs__ = torch::count_nonzero_out(*out, *self, torch::IntArrayRef(dim_data, dim_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_count_nonzero_out(tensor *out__, tensor out, tensor self, int64_t dim_v, int dim_null) {
-  PROTECT(
-    auto outputs__ = torch::count_nonzero_out(*out, *self, dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cov(tensor *out__, tensor self, int64_t correction, tensor fweights, tensor aweights) {
-  PROTECT(
-    auto outputs__ = torch::cov(*self, correction, (fweights ? *fweights : torch::Tensor()), (aweights ? *aweights : torch::Tensor()));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cross(tensor *out__, tensor self, tensor other, int64_t dim_v, int dim_null) {
-  PROTECT(
-    auto outputs__ = torch::cross(*self, *other, dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cross_entropy_loss(tensor *out__, tensor self, tensor target, tensor weight, int64_t reduction, int64_t ignore_index, double label_smoothing) {
-  PROTECT(
-    auto outputs__ = torch::cross_entropy_loss(*self, *target, (weight ? *weight : torch::Tensor()), reduction, ignore_index, label_smoothing);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cross_out(tensor *out__, tensor out, tensor self, tensor other, int64_t dim_v, int dim_null) {
-  PROTECT(
-    auto outputs__ = torch::cross_out(*out, *self, *other, dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_crow_indices(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = self->crow_indices();
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_crow_indices_copy(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::crow_indices_copy(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_crow_indices_copy_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::crow_indices_copy_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_ctc_loss(tensor *out__, tensor log_probs, tensor targets, int64_t *input_lengths_data, int input_lengths_len, int64_t *target_lengths_data, int target_lengths_len, int64_t blank, int64_t reduction, int zero_infinity) {
-  PROTECT(
-    auto outputs__ = torch::ctc_loss(*log_probs, *targets, torch::IntArrayRef(input_lengths_data, input_lengths_len), torch::IntArrayRef(target_lengths_data, target_lengths_len), blank, reduction, (bool)zero_infinity);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_ctc_loss_tensor(tensor *out__, tensor log_probs, tensor targets, tensor input_lengths, tensor target_lengths, int64_t blank, int64_t reduction, int zero_infinity) {
-  PROTECT(
-    auto outputs__ = torch::ctc_loss(*log_probs, *targets, *input_lengths, *target_lengths, blank, reduction, (bool)zero_infinity);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cudnn_affine_grid_generator(tensor *out__, tensor theta, int64_t n, int64_t C, int64_t H, int64_t W) {
-  PROTECT(
-    auto outputs__ = torch::cudnn_affine_grid_generator(*theta, n, C, H, W);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cudnn_affine_grid_generator_backward(tensor *out__, tensor grad, int64_t n, int64_t C, int64_t H, int64_t W) {
-  PROTECT(
-    auto outputs__ = torch::cudnn_affine_grid_generator_backward(*grad, n, C, H, W);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cudnn_affine_grid_generator_backward_out(tensor *out__, tensor out, tensor grad, int64_t n, int64_t C, int64_t H, int64_t W) {
-  PROTECT(
-    auto outputs__ = torch::cudnn_affine_grid_generator_backward_out(*out, *grad, n, C, H, W);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cudnn_affine_grid_generator_out(tensor *out__, tensor out, tensor theta, int64_t n, int64_t C, int64_t H, int64_t W) {
-  PROTECT(
-    auto outputs__ = torch::cudnn_affine_grid_generator_out(*out, *theta, n, C, H, W);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cudnn_batch_norm(tensor *out__, tensor input, tensor weight, tensor bias, tensor running_mean, tensor running_var, int training, double exponential_average_factor, double epsilon) {
-  PROTECT(
-    auto outputs__ = torch::cudnn_batch_norm(*input, *weight, (bias ? *bias : torch::Tensor()), (running_mean ? *running_mean : torch::Tensor()), (running_var ? *running_var : torch::Tensor()), (bool)training, exponential_average_factor, epsilon);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-    out__[3] = new torch::Tensor(std::get<3>(outputs__));
-  )
-}
-
-void atg_cudnn_batch_norm_backward(tensor *out__, tensor input, tensor grad_output, tensor weight, tensor running_mean, tensor running_var, tensor save_mean, tensor save_var, double epsilon, tensor reserveSpace) {
-  PROTECT(
-    auto outputs__ = torch::cudnn_batch_norm_backward(*input, *grad_output, *weight, (running_mean ? *running_mean : torch::Tensor()), (running_var ? *running_var : torch::Tensor()), (save_mean ? *save_mean : torch::Tensor()), (save_var ? *save_var : torch::Tensor()), epsilon, *reserveSpace);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg_cudnn_batch_norm_backward_out(tensor *out__, tensor out0, tensor out1, tensor out2, tensor input, tensor grad_output, tensor weight, tensor running_mean, tensor running_var, tensor save_mean, tensor save_var, double epsilon, tensor reserveSpace) {
-  PROTECT(
-    auto outputs__ = torch::cudnn_batch_norm_backward_out(*out0, *out1, *out2, *input, *grad_output, *weight, (running_mean ? *running_mean : torch::Tensor()), (running_var ? *running_var : torch::Tensor()), (save_mean ? *save_mean : torch::Tensor()), (save_var ? *save_var : torch::Tensor()), epsilon, *reserveSpace);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg_cudnn_batch_norm_out(tensor *out__, tensor out0, tensor out1, tensor out2, tensor out3, tensor input, tensor weight, tensor bias, tensor running_mean, tensor running_var, int training, double exponential_average_factor, double epsilon) {
-  PROTECT(
-    auto outputs__ = torch::cudnn_batch_norm_out(*out0, *out1, *out2, *out3, *input, *weight, (bias ? *bias : torch::Tensor()), (running_mean ? *running_mean : torch::Tensor()), (running_var ? *running_var : torch::Tensor()), (bool)training, exponential_average_factor, epsilon);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-    out__[3] = new torch::Tensor(std::get<3>(outputs__));
-  )
-}
-
-void atg_cudnn_convolution(tensor *out__, tensor self, tensor weight, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int benchmark, int deterministic, int allow_tf32) {
-  PROTECT(
-    auto outputs__ = torch::cudnn_convolution(*self, *weight, torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(dilation_data, dilation_len), groups, (bool)benchmark, (bool)deterministic, (bool)allow_tf32);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cudnn_convolution_add_relu(tensor *out__, tensor self, tensor weight, tensor z, scalar alpha, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int64_t groups) {
-  PROTECT(
-    auto outputs__ = torch::cudnn_convolution_add_relu(*self, *weight, *z, *alpha, (bias ? *bias : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), groups);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cudnn_convolution_add_relu_out(tensor *out__, tensor out, tensor self, tensor weight, tensor z, scalar alpha, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int64_t groups) {
-  PROTECT(
-    auto outputs__ = torch::cudnn_convolution_add_relu_out(*out, *self, *weight, *z, *alpha, (bias ? *bias : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), groups);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cudnn_convolution_out(tensor *out__, tensor out, tensor self, tensor weight, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int benchmark, int deterministic, int allow_tf32) {
-  PROTECT(
-    auto outputs__ = torch::cudnn_convolution_out(*out, *self, *weight, torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(dilation_data, dilation_len), groups, (bool)benchmark, (bool)deterministic, (bool)allow_tf32);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cudnn_convolution_relu(tensor *out__, tensor self, tensor weight, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int64_t groups) {
-  PROTECT(
-    auto outputs__ = torch::cudnn_convolution_relu(*self, *weight, (bias ? *bias : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), groups);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cudnn_convolution_relu_out(tensor *out__, tensor out, tensor self, tensor weight, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int64_t groups) {
-  PROTECT(
-    auto outputs__ = torch::cudnn_convolution_relu_out(*out, *self, *weight, (bias ? *bias : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), groups);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cudnn_convolution_transpose(tensor *out__, tensor self, tensor weight, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int benchmark, int deterministic, int allow_tf32) {
-  PROTECT(
-    auto outputs__ = torch::cudnn_convolution_transpose(*self, *weight, torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(output_padding_data, output_padding_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(dilation_data, dilation_len), groups, (bool)benchmark, (bool)deterministic, (bool)allow_tf32);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cudnn_convolution_transpose_out(tensor *out__, tensor out, tensor self, tensor weight, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int benchmark, int deterministic, int allow_tf32) {
-  PROTECT(
-    auto outputs__ = torch::cudnn_convolution_transpose_out(*out, *self, *weight, torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(output_padding_data, output_padding_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(dilation_data, dilation_len), groups, (bool)benchmark, (bool)deterministic, (bool)allow_tf32);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cudnn_grid_sampler(tensor *out__, tensor self, tensor grid) {
-  PROTECT(
-    auto outputs__ = torch::cudnn_grid_sampler(*self, *grid);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cudnn_grid_sampler_backward(tensor *out__, tensor self, tensor grid, tensor grad_output) {
-  PROTECT(
-    auto outputs__ = torch::cudnn_grid_sampler_backward(*self, *grid, *grad_output);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_cudnn_grid_sampler_backward_out(tensor *out__, tensor out0, tensor out1, tensor self, tensor grid, tensor grad_output) {
-  PROTECT(
-    auto outputs__ = torch::cudnn_grid_sampler_backward_out(*out0, *out1, *self, *grid, *grad_output);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_cudnn_grid_sampler_out(tensor *out__, tensor out, tensor self, tensor grid) {
-  PROTECT(
-    auto outputs__ = torch::cudnn_grid_sampler_out(*out, *self, *grid);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-int atg_cudnn_is_acceptable(tensor self) {
-  PROTECT(
-    return torch::cudnn_is_acceptable(*self);
-  )
-  return 0;
-}
-
-void atg_cummax(tensor *out__, tensor self, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::cummax(*self, dim);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_cummax_out(tensor *out__, tensor values, tensor indices, tensor self, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::cummax_out(*values, *indices, *self, dim);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_cummaxmin_backward(tensor *out__, tensor grad, tensor input, tensor indices, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::cummaxmin_backward(*grad, *input, *indices, dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cummin(tensor *out__, tensor self, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::cummin(*self, dim);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_cummin_out(tensor *out__, tensor values, tensor indices, tensor self, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::cummin_out(*values, *indices, *self, dim);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_cumprod(tensor *out__, tensor self, int64_t dim, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::cumprod(*self, dim, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cumprod_(tensor *out__, tensor self, int64_t dim, int dtype) {
-  PROTECT(
-    auto outputs__ = self->cumprod_(dim, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cumprod_backward(tensor *out__, tensor grad, tensor input, int64_t dim, tensor output) {
-  PROTECT(
-    auto outputs__ = torch::cumprod_backward(*grad, *input, dim, *output);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cumprod_out(tensor *out__, tensor out, tensor self, int64_t dim, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::cumprod_out(*out, *self, dim, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cumsum(tensor *out__, tensor self, int64_t dim, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::cumsum(*self, dim, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cumsum_(tensor *out__, tensor self, int64_t dim, int dtype) {
-  PROTECT(
-    auto outputs__ = self->cumsum_(dim, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cumsum_out(tensor *out__, tensor out, tensor self, int64_t dim, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::cumsum_out(*out, *self, dim, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cumulative_trapezoid(tensor *out__, tensor y, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::cumulative_trapezoid(*y, dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_cumulative_trapezoid_x(tensor *out__, tensor y, tensor x, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::cumulative_trapezoid(*y, *x, dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_data(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = self->data();
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_deg2rad(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::deg2rad(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_deg2rad_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::deg2rad_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_deg2rad_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::deg2rad_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-int64_t atg_dense_dim(tensor self) {
-  PROTECT(
-    return self->dense_dim();
-  )
-  return 0;
-}
-
-void atg_dequantize(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::dequantize(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_dequantize_self_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::dequantize_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-tensor *atg_dequantize_tensors(tensor *tensors_data, int tensors_len) {
-  PROTECT(
-    auto outputs__ = torch::dequantize(of_carray_tensor(tensors_data, tensors_len));
-    int sz = outputs__.size();
-    torch::Tensor **out__ = (torch::Tensor**)malloc((sz + 1) * sizeof(torch::Tensor*));
-    for (int i = 0; i < sz; ++i)
-      out__[i] = new torch::Tensor(outputs__[i]);
-    out__[sz] = nullptr;
-    return out__;
-  )
-}
-
-void atg_dequantize_tensors_out(tensor *out_data, int out_len, tensor *tensors_data, int tensors_len) {
-  PROTECT(
-    torch::dequantize_out(of_carray_tensor(out_data, out_len), of_carray_tensor(tensors_data, tensors_len));
-  )
-}
-
-void atg_det(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::det(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_detach(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::detach(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_detach_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::detach_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_detach_copy(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::detach_copy(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_detach_copy_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::detach_copy_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_diag(tensor *out__, tensor self, int64_t diagonal) {
-  PROTECT(
-    auto outputs__ = torch::diag(*self, diagonal);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_diag_embed(tensor *out__, tensor self, int64_t offset, int64_t dim1, int64_t dim2) {
-  PROTECT(
-    auto outputs__ = torch::diag_embed(*self, offset, dim1, dim2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_diag_embed_out(tensor *out__, tensor out, tensor self, int64_t offset, int64_t dim1, int64_t dim2) {
-  PROTECT(
-    auto outputs__ = torch::diag_embed_out(*out, *self, offset, dim1, dim2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_diag_out(tensor *out__, tensor out, tensor self, int64_t diagonal) {
-  PROTECT(
-    auto outputs__ = torch::diag_out(*out, *self, diagonal);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_diagflat(tensor *out__, tensor self, int64_t offset) {
-  PROTECT(
-    auto outputs__ = torch::diagflat(*self, offset);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_diagonal(tensor *out__, tensor self, int64_t offset, int64_t dim1, int64_t dim2) {
-  PROTECT(
-    auto outputs__ = torch::diagonal(*self, offset, dim1, dim2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_diagonal_backward(tensor *out__, tensor grad_output, int64_t *input_sizes_data, int input_sizes_len, int64_t offset, int64_t dim1, int64_t dim2) {
-  PROTECT(
-    auto outputs__ = torch::diagonal_backward(*grad_output, torch::IntArrayRef(input_sizes_data, input_sizes_len), offset, dim1, dim2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_diagonal_backward_out(tensor *out__, tensor out, tensor grad_output, int64_t *input_sizes_data, int input_sizes_len, int64_t offset, int64_t dim1, int64_t dim2) {
-  PROTECT(
-    auto outputs__ = torch::diagonal_backward_out(*out, *grad_output, torch::IntArrayRef(input_sizes_data, input_sizes_len), offset, dim1, dim2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_diagonal_copy(tensor *out__, tensor self, int64_t offset, int64_t dim1, int64_t dim2) {
-  PROTECT(
-    auto outputs__ = torch::diagonal_copy(*self, offset, dim1, dim2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_diagonal_copy_out(tensor *out__, tensor out, tensor self, int64_t offset, int64_t dim1, int64_t dim2) {
-  PROTECT(
-    auto outputs__ = torch::diagonal_copy_out(*out, *self, offset, dim1, dim2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_diagonal_scatter(tensor *out__, tensor self, tensor src, int64_t offset, int64_t dim1, int64_t dim2) {
-  PROTECT(
-    auto outputs__ = torch::diagonal_scatter(*self, *src, offset, dim1, dim2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_diagonal_scatter_out(tensor *out__, tensor out, tensor self, tensor src, int64_t offset, int64_t dim1, int64_t dim2) {
-  PROTECT(
-    auto outputs__ = torch::diagonal_scatter_out(*out, *self, *src, offset, dim1, dim2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_diff(tensor *out__, tensor self, int64_t n, int64_t dim, tensor prepend, tensor append) {
-  PROTECT(
-    auto outputs__ = torch::diff(*self, n, dim, (prepend ? *prepend : torch::Tensor()), (append ? *append : torch::Tensor()));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_diff_out(tensor *out__, tensor out, tensor self, int64_t n, int64_t dim, tensor prepend, tensor append) {
-  PROTECT(
-    auto outputs__ = torch::diff_out(*out, *self, n, dim, (prepend ? *prepend : torch::Tensor()), (append ? *append : torch::Tensor()));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_digamma(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::digamma(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_digamma_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = self->digamma_();
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_digamma_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::digamma_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_dist(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::dist(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_dist_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::dist_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_div(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::div(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_div_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->div_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_div_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::div_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_div_out_mode(tensor *out__, tensor out, tensor self, tensor other, char * rounding_mode) {
-  PROTECT(
-    auto outputs__ = torch::div_out(*out, *self, *other, std::string(rounding_mode));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_div_scalar(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::div(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_div_scalar_(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = self->div_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_div_scalar_mode(tensor *out__, tensor self, scalar other, char * rounding_mode) {
-  PROTECT(
-    auto outputs__ = torch::div(*self, *other, std::string(rounding_mode));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_div_scalar_mode_(tensor *out__, tensor self, scalar other, char * rounding_mode) {
-  PROTECT(
-    auto outputs__ = self->div_(*other, std::string(rounding_mode));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_div_scalar_mode_out(tensor *out__, tensor out, tensor self, scalar other, char * rounding_mode) {
-  PROTECT(
-    auto outputs__ = torch::div_out(*out, *self, *other, std::string(rounding_mode));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_div_scalar_out(tensor *out__, tensor out, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::div_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_div_tensor_mode(tensor *out__, tensor self, tensor other, char * rounding_mode) {
-  PROTECT(
-    auto outputs__ = torch::div(*self, *other, std::string(rounding_mode));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_div_tensor_mode_(tensor *out__, tensor self, tensor other, char * rounding_mode) {
-  PROTECT(
-    auto outputs__ = self->div_(*other, std::string(rounding_mode));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_divide(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::divide(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_divide_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->divide_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_divide_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::divide_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_divide_out_mode(tensor *out__, tensor out, tensor self, tensor other, char * rounding_mode) {
-  PROTECT(
-    auto outputs__ = torch::divide_out(*out, *self, *other, std::string(rounding_mode));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_divide_scalar(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::divide(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_divide_scalar_(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = self->divide_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_divide_scalar_mode(tensor *out__, tensor self, scalar other, char * rounding_mode) {
-  PROTECT(
-    auto outputs__ = torch::divide(*self, *other, std::string(rounding_mode));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_divide_scalar_mode_(tensor *out__, tensor self, scalar other, char * rounding_mode) {
-  PROTECT(
-    auto outputs__ = self->divide_(*other, std::string(rounding_mode));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_divide_tensor_mode(tensor *out__, tensor self, tensor other, char * rounding_mode) {
-  PROTECT(
-    auto outputs__ = torch::divide(*self, *other, std::string(rounding_mode));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_divide_tensor_mode_(tensor *out__, tensor self, tensor other, char * rounding_mode) {
-  PROTECT(
-    auto outputs__ = self->divide_(*other, std::string(rounding_mode));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_dot(tensor *out__, tensor self, tensor tensor) {
-  PROTECT(
-    auto outputs__ = torch::dot(*self, *tensor);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_dot_out(tensor *out__, tensor out, tensor self, tensor tensor) {
-  PROTECT(
-    auto outputs__ = torch::dot_out(*out, *self, *tensor);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_dropout(tensor *out__, tensor input, double p, int train) {
-  PROTECT(
-    auto outputs__ = torch::dropout(*input, p, (bool)train);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_dropout_(tensor *out__, tensor self, double p, int train) {
-  PROTECT(
-    auto outputs__ = torch::dropout_(*self, p, (bool)train);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-tensor *atg_dsplit(tensor self, int64_t sections) {
-  PROTECT(
-    auto outputs__ = torch::dsplit(*self, sections);
-    int sz = outputs__.size();
-    torch::Tensor **out__ = (torch::Tensor**)malloc((sz + 1) * sizeof(torch::Tensor*));
-    for (int i = 0; i < sz; ++i)
-      out__[i] = new torch::Tensor(outputs__[i]);
-    out__[sz] = nullptr;
-    return out__;
-  )
-}
-
-tensor *atg_dsplit_array(tensor self, int64_t *indices_data, int indices_len) {
-  PROTECT(
-    auto outputs__ = torch::dsplit(*self, torch::IntArrayRef(indices_data, indices_len));
-    int sz = outputs__.size();
-    torch::Tensor **out__ = (torch::Tensor**)malloc((sz + 1) * sizeof(torch::Tensor*));
-    for (int i = 0; i < sz; ++i)
-      out__[i] = new torch::Tensor(outputs__[i]);
-    out__[sz] = nullptr;
-    return out__;
-  )
-}
-
-void atg_dstack(tensor *out__, tensor *tensors_data, int tensors_len) {
-  PROTECT(
-    auto outputs__ = torch::dstack(of_carray_tensor(tensors_data, tensors_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_dstack_out(tensor *out__, tensor out, tensor *tensors_data, int tensors_len) {
-  PROTECT(
-    auto outputs__ = torch::dstack_out(*out, of_carray_tensor(tensors_data, tensors_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_einsum(tensor *out__, char * equation, tensor *tensors_data, int tensors_len, int64_t *path_data, int path_len) {
-  PROTECT(
-    auto outputs__ = torch::einsum(std::string(equation), of_carray_tensor(tensors_data, tensors_len), path_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(path_data, path_len)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_elu(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::elu(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_elu_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::elu_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_elu_backward(tensor *out__, tensor grad_output, scalar alpha, scalar scale, scalar input_scale, int is_result, tensor self_or_result) {
-  PROTECT(
-    auto outputs__ = torch::elu_backward(*grad_output, *alpha, *scale, *input_scale, (bool)is_result, *self_or_result);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_elu_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, scalar alpha, scalar scale, scalar input_scale, int is_result, tensor self_or_result) {
-  PROTECT(
-    auto outputs__ = torch::elu_backward_out(*grad_input, *grad_output, *alpha, *scale, *input_scale, (bool)is_result, *self_or_result);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_elu_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::elu_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_embedding(tensor *out__, tensor weight, tensor indices, int64_t padding_idx, int scale_grad_by_freq, int sparse) {
-  PROTECT(
-    auto outputs__ = torch::embedding(*weight, *indices, padding_idx, (bool)scale_grad_by_freq, (bool)sparse);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_embedding_backward(tensor *out__, tensor grad, tensor indices, int64_t num_weights, int64_t padding_idx, int scale_grad_by_freq, int sparse) {
-  PROTECT(
-    auto outputs__ = torch::embedding_backward(*grad, *indices, num_weights, padding_idx, (bool)scale_grad_by_freq, (bool)sparse);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_embedding_bag(tensor *out__, tensor weight, tensor indices, tensor offsets, int scale_grad_by_freq, int64_t mode, int sparse, tensor per_sample_weights, int include_last_offset) {
-  PROTECT(
-    auto outputs__ = torch::embedding_bag(*weight, *indices, *offsets, (bool)scale_grad_by_freq, mode, (bool)sparse, (per_sample_weights ? *per_sample_weights : torch::Tensor()), (bool)include_last_offset);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-    out__[3] = new torch::Tensor(std::get<3>(outputs__));
-  )
-}
-
-void atg_embedding_bag_padding_idx(tensor *out__, tensor weight, tensor indices, tensor offsets, int scale_grad_by_freq, int64_t mode, int sparse, tensor per_sample_weights, int include_last_offset, int64_t padding_idx_v, int padding_idx_null) {
-  PROTECT(
-    auto outputs__ = torch::embedding_bag(*weight, *indices, *offsets, (bool)scale_grad_by_freq, mode, (bool)sparse, (per_sample_weights ? *per_sample_weights : torch::Tensor()), (bool)include_last_offset, padding_idx_null ? c10::nullopt : c10::optional<int64_t>(padding_idx_v));
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-    out__[3] = new torch::Tensor(std::get<3>(outputs__));
-  )
-}
-
-void atg_embedding_dense_backward(tensor *out__, tensor grad_output, tensor indices, int64_t num_weights, int64_t padding_idx, int scale_grad_by_freq) {
-  PROTECT(
-    auto outputs__ = torch::embedding_dense_backward(*grad_output, *indices, num_weights, padding_idx, (bool)scale_grad_by_freq);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_embedding_dense_backward_out(tensor *out__, tensor out, tensor grad_output, tensor indices, int64_t num_weights, int64_t padding_idx, int scale_grad_by_freq) {
-  PROTECT(
-    auto outputs__ = torch::embedding_dense_backward_out(*out, *grad_output, *indices, num_weights, padding_idx, (bool)scale_grad_by_freq);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_embedding_out(tensor *out__, tensor out, tensor weight, tensor indices, int64_t padding_idx, int scale_grad_by_freq, int sparse) {
-  PROTECT(
-    auto outputs__ = torch::embedding_out(*out, *weight, *indices, padding_idx, (bool)scale_grad_by_freq, (bool)sparse);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_embedding_renorm(tensor *out__, tensor self, tensor indices, double max_norm, double norm_type) {
-  PROTECT(
-    auto outputs__ = torch::embedding_renorm(*self, *indices, max_norm, norm_type);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_embedding_renorm_(tensor *out__, tensor self, tensor indices, double max_norm, double norm_type) {
-  PROTECT(
-    auto outputs__ = torch::embedding_renorm_(*self, *indices, max_norm, norm_type);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_embedding_renorm_out(tensor *out__, tensor out, tensor self, tensor indices, double max_norm, double norm_type) {
-  PROTECT(
-    auto outputs__ = torch::embedding_renorm_out(*out, *self, *indices, max_norm, norm_type);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_embedding_sparse_backward(tensor *out__, tensor grad, tensor indices, int64_t num_weights, int64_t padding_idx, int scale_grad_by_freq) {
-  PROTECT(
-    auto outputs__ = torch::embedding_sparse_backward(*grad, *indices, num_weights, padding_idx, (bool)scale_grad_by_freq);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_empty(tensor *out__, int64_t *size_data, int size_len, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::empty(torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_empty_like(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::empty_like(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_empty_like_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::empty_like_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_empty_out(tensor *out__, tensor out, int64_t *size_data, int size_len) {
-  PROTECT(
-    auto outputs__ = torch::empty_out(*out, torch::IntArrayRef(size_data, size_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_empty_permuted(tensor *out__, int64_t *size_data, int size_len, int64_t *physical_layout_data, int physical_layout_len, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::empty_permuted(torch::IntArrayRef(size_data, size_len), torch::IntArrayRef(physical_layout_data, physical_layout_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_empty_permuted_out(tensor *out__, tensor out, int64_t *size_data, int size_len, int64_t *physical_layout_data, int physical_layout_len) {
-  PROTECT(
-    auto outputs__ = torch::empty_permuted_out(*out, torch::IntArrayRef(size_data, size_len), torch::IntArrayRef(physical_layout_data, physical_layout_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_empty_quantized(tensor *out__, int64_t *size_data, int size_len, tensor qtensor, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::empty_quantized(torch::IntArrayRef(size_data, size_len), *qtensor, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_empty_quantized_out(tensor *out__, tensor out, int64_t *size_data, int size_len, tensor qtensor) {
-  PROTECT(
-    auto outputs__ = torch::empty_quantized_out(*out, torch::IntArrayRef(size_data, size_len), *qtensor);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_empty_strided(tensor *out__, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::empty_strided(torch::IntArrayRef(size_data, size_len), torch::IntArrayRef(stride_data, stride_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_empty_strided_out(tensor *out__, tensor out, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len) {
-  PROTECT(
-    auto outputs__ = torch::empty_strided_out(*out, torch::IntArrayRef(size_data, size_len), torch::IntArrayRef(stride_data, stride_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_eq(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::eq(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_eq_(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = self->eq_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_eq_scalar_out(tensor *out__, tensor out, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::eq_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_eq_tensor(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::eq(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_eq_tensor_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->eq_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_eq_tensor_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::eq_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-int atg_equal(tensor self, tensor other) {
-  PROTECT(
-    return torch::equal(*self, *other);
-  )
-  return 0;
-}
-
-void atg_erf(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::erf(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_erf_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::erf_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_erf_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::erf_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_erfc(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::erfc(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_erfc_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::erfc_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_erfc_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::erfc_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_erfinv(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::erfinv(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_erfinv_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = self->erfinv_();
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_erfinv_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::erfinv_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_exp(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::exp(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_exp2(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::exp2(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_exp2_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::exp2_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_exp2_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::exp2_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_exp_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::exp_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_exp_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::exp_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_expand(tensor *out__, tensor self, int64_t *size_data, int size_len, int implicit) {
-  PROTECT(
-    auto outputs__ = self->expand(torch::IntArrayRef(size_data, size_len), (bool)implicit);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_expand_as(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->expand_as(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_expand_copy(tensor *out__, tensor self, int64_t *size_data, int size_len, int implicit) {
-  PROTECT(
-    auto outputs__ = torch::expand_copy(*self, torch::IntArrayRef(size_data, size_len), (bool)implicit);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_expand_copy_out(tensor *out__, tensor out, tensor self, int64_t *size_data, int size_len, int implicit) {
-  PROTECT(
-    auto outputs__ = torch::expand_copy_out(*out, *self, torch::IntArrayRef(size_data, size_len), (bool)implicit);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_expm1(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::expm1(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_expm1_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::expm1_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_expm1_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::expm1_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_exponential(tensor *out__, tensor self, double lambd) {
-  PROTECT(
-    auto outputs__ = torch::exponential(*self, lambd);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_exponential_(tensor *out__, tensor self, double lambd) {
-  PROTECT(
-    auto outputs__ = self->exponential_(lambd);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_exponential_out(tensor *out__, tensor out, tensor self, double lambd) {
-  PROTECT(
-    auto outputs__ = torch::exponential_out(*out, *self, lambd);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_eye(tensor *out__, int64_t n, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::eye(n, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_eye_m(tensor *out__, int64_t n, int64_t m, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::eye(n, m, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_eye_m_out(tensor *out__, tensor out, int64_t n, int64_t m) {
-  PROTECT(
-    auto outputs__ = torch::eye_out(*out, n, m);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_eye_out(tensor *out__, tensor out, int64_t n) {
-  PROTECT(
-    auto outputs__ = torch::eye_out(*out, n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fake_quantize_per_channel_affine(tensor *out__, tensor self, tensor scale, tensor zero_point, int64_t axis, int64_t quant_min, int64_t quant_max) {
-  PROTECT(
-    auto outputs__ = torch::fake_quantize_per_channel_affine(*self, *scale, *zero_point, axis, quant_min, quant_max);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fake_quantize_per_channel_affine_cachemask(tensor *out__, tensor self, tensor scale, tensor zero_point, int64_t axis, int64_t quant_min, int64_t quant_max) {
-  PROTECT(
-    auto outputs__ = torch::fake_quantize_per_channel_affine_cachemask(*self, *scale, *zero_point, axis, quant_min, quant_max);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_fake_quantize_per_channel_affine_cachemask_backward(tensor *out__, tensor grad, tensor mask) {
-  PROTECT(
-    auto outputs__ = torch::fake_quantize_per_channel_affine_cachemask_backward(*grad, *mask);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fake_quantize_per_channel_affine_cachemask_out(tensor *out__, tensor out0, tensor out1, tensor self, tensor scale, tensor zero_point, int64_t axis, int64_t quant_min, int64_t quant_max) {
-  PROTECT(
-    auto outputs__ = torch::fake_quantize_per_channel_affine_cachemask_out(*out0, *out1, *self, *scale, *zero_point, axis, quant_min, quant_max);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_fake_quantize_per_tensor_affine(tensor *out__, tensor self, double scale, int64_t zero_point, int64_t quant_min, int64_t quant_max) {
-  PROTECT(
-    auto outputs__ = torch::fake_quantize_per_tensor_affine(*self, scale, zero_point, quant_min, quant_max);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fake_quantize_per_tensor_affine_cachemask(tensor *out__, tensor self, double scale, int64_t zero_point, int64_t quant_min, int64_t quant_max) {
-  PROTECT(
-    auto outputs__ = torch::fake_quantize_per_tensor_affine_cachemask(*self, scale, zero_point, quant_min, quant_max);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_fake_quantize_per_tensor_affine_cachemask_backward(tensor *out__, tensor grad, tensor mask) {
-  PROTECT(
-    auto outputs__ = torch::fake_quantize_per_tensor_affine_cachemask_backward(*grad, *mask);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fake_quantize_per_tensor_affine_cachemask_out(tensor *out__, tensor out0, tensor out1, tensor self, double scale, int64_t zero_point, int64_t quant_min, int64_t quant_max) {
-  PROTECT(
-    auto outputs__ = torch::fake_quantize_per_tensor_affine_cachemask_out(*out0, *out1, *self, scale, zero_point, quant_min, quant_max);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_fake_quantize_per_tensor_affine_tensor_qparams(tensor *out__, tensor self, tensor scale, tensor zero_point, int64_t quant_min, int64_t quant_max) {
-  PROTECT(
-    auto outputs__ = torch::fake_quantize_per_tensor_affine(*self, *scale, *zero_point, quant_min, quant_max);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fbgemm_linear_fp16_weight(tensor *out__, tensor input, tensor packed_weight, tensor bias) {
-  PROTECT(
-    auto outputs__ = torch::fbgemm_linear_fp16_weight(*input, *packed_weight, *bias);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fbgemm_linear_fp16_weight_fp32_activation(tensor *out__, tensor input, tensor packed_weight, tensor bias) {
-  PROTECT(
-    auto outputs__ = torch::fbgemm_linear_fp16_weight_fp32_activation(*input, *packed_weight, *bias);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fbgemm_linear_int8_weight(tensor *out__, tensor input, tensor weight, tensor packed, tensor col_offsets, scalar weight_scale, scalar weight_zero_point, tensor bias) {
-  PROTECT(
-    auto outputs__ = torch::fbgemm_linear_int8_weight(*input, *weight, *packed, *col_offsets, *weight_scale, *weight_zero_point, *bias);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fbgemm_linear_int8_weight_fp32_activation(tensor *out__, tensor input, tensor weight, tensor packed, tensor col_offsets, scalar weight_scale, scalar weight_zero_point, tensor bias) {
-  PROTECT(
-    auto outputs__ = torch::fbgemm_linear_int8_weight_fp32_activation(*input, *weight, *packed, *col_offsets, *weight_scale, *weight_zero_point, *bias);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fbgemm_pack_gemm_matrix_fp16(tensor *out__, tensor input) {
-  PROTECT(
-    auto outputs__ = torch::fbgemm_pack_gemm_matrix_fp16(*input);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fbgemm_pack_quantized_matrix(tensor *out__, tensor input) {
-  PROTECT(
-    auto outputs__ = torch::fbgemm_pack_quantized_matrix(*input);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fbgemm_pack_quantized_matrix_kn(tensor *out__, tensor input, int64_t K, int64_t n) {
-  PROTECT(
-    auto outputs__ = torch::fbgemm_pack_quantized_matrix(*input, K, n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_feature_alpha_dropout(tensor *out__, tensor input, double p, int train) {
-  PROTECT(
-    auto outputs__ = torch::feature_alpha_dropout(*input, p, (bool)train);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_feature_alpha_dropout_(tensor *out__, tensor self, double p, int train) {
-  PROTECT(
-    auto outputs__ = torch::feature_alpha_dropout_(*self, p, (bool)train);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_feature_dropout(tensor *out__, tensor input, double p, int train) {
-  PROTECT(
-    auto outputs__ = torch::feature_dropout(*input, p, (bool)train);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_feature_dropout_(tensor *out__, tensor self, double p, int train) {
-  PROTECT(
-    auto outputs__ = torch::feature_dropout_(*self, p, (bool)train);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fft_fft(tensor *out__, tensor self, int64_t n_v, int n_null, int64_t dim, char * norm) {
-  PROTECT(
-    auto outputs__ = torch::fft_fft(*self, n_null ? c10::nullopt : c10::optional<int64_t>(n_v), dim, std::string(norm));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fft_fft2(tensor *out__, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
-  PROTECT(
-    auto outputs__ = torch::fft_fft2(*self, s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), torch::IntArrayRef(dim_data, dim_len), std::string(norm));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fft_fft2_out(tensor *out__, tensor out, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
-  PROTECT(
-    auto outputs__ = torch::fft_fft2_out(*out, *self, s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), torch::IntArrayRef(dim_data, dim_len), std::string(norm));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fft_fft_out(tensor *out__, tensor out, tensor self, int64_t n_v, int n_null, int64_t dim, char * norm) {
-  PROTECT(
-    auto outputs__ = torch::fft_fft_out(*out, *self, n_null ? c10::nullopt : c10::optional<int64_t>(n_v), dim, std::string(norm));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fft_fftfreq(tensor *out__, int64_t n, double d, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::fft_fftfreq(n, d, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fft_fftfreq_out(tensor *out__, tensor out, int64_t n, double d) {
-  PROTECT(
-    auto outputs__ = torch::fft_fftfreq_out(*out, n, d);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fft_fftn(tensor *out__, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
-  PROTECT(
-    auto outputs__ = torch::fft_fftn(*self, s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), std::string(norm));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fft_fftn_out(tensor *out__, tensor out, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
-  PROTECT(
-    auto outputs__ = torch::fft_fftn_out(*out, *self, s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), std::string(norm));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fft_fftshift(tensor *out__, tensor self, int64_t *dim_data, int dim_len) {
-  PROTECT(
-    auto outputs__ = torch::fft_fftshift(*self, dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fft_hfft(tensor *out__, tensor self, int64_t n_v, int n_null, int64_t dim, char * norm) {
-  PROTECT(
-    auto outputs__ = torch::fft_hfft(*self, n_null ? c10::nullopt : c10::optional<int64_t>(n_v), dim, std::string(norm));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fft_hfft2(tensor *out__, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
-  PROTECT(
-    auto outputs__ = torch::fft_hfft2(*self, s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), torch::IntArrayRef(dim_data, dim_len), std::string(norm));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fft_hfft2_out(tensor *out__, tensor out, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
-  PROTECT(
-    auto outputs__ = torch::fft_hfft2_out(*out, *self, s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), torch::IntArrayRef(dim_data, dim_len), std::string(norm));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fft_hfft_out(tensor *out__, tensor out, tensor self, int64_t n_v, int n_null, int64_t dim, char * norm) {
-  PROTECT(
-    auto outputs__ = torch::fft_hfft_out(*out, *self, n_null ? c10::nullopt : c10::optional<int64_t>(n_v), dim, std::string(norm));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fft_hfftn(tensor *out__, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
-  PROTECT(
-    auto outputs__ = torch::fft_hfftn(*self, s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), std::string(norm));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fft_hfftn_out(tensor *out__, tensor out, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
-  PROTECT(
-    auto outputs__ = torch::fft_hfftn_out(*out, *self, s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), std::string(norm));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fft_ifft(tensor *out__, tensor self, int64_t n_v, int n_null, int64_t dim, char * norm) {
-  PROTECT(
-    auto outputs__ = torch::fft_ifft(*self, n_null ? c10::nullopt : c10::optional<int64_t>(n_v), dim, std::string(norm));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fft_ifft2(tensor *out__, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
-  PROTECT(
-    auto outputs__ = torch::fft_ifft2(*self, s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), torch::IntArrayRef(dim_data, dim_len), std::string(norm));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fft_ifft2_out(tensor *out__, tensor out, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
-  PROTECT(
-    auto outputs__ = torch::fft_ifft2_out(*out, *self, s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), torch::IntArrayRef(dim_data, dim_len), std::string(norm));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fft_ifft_out(tensor *out__, tensor out, tensor self, int64_t n_v, int n_null, int64_t dim, char * norm) {
-  PROTECT(
-    auto outputs__ = torch::fft_ifft_out(*out, *self, n_null ? c10::nullopt : c10::optional<int64_t>(n_v), dim, std::string(norm));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fft_ifftn(tensor *out__, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
-  PROTECT(
-    auto outputs__ = torch::fft_ifftn(*self, s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), std::string(norm));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fft_ifftn_out(tensor *out__, tensor out, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
-  PROTECT(
-    auto outputs__ = torch::fft_ifftn_out(*out, *self, s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), std::string(norm));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fft_ifftshift(tensor *out__, tensor self, int64_t *dim_data, int dim_len) {
-  PROTECT(
-    auto outputs__ = torch::fft_ifftshift(*self, dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fft_ihfft(tensor *out__, tensor self, int64_t n_v, int n_null, int64_t dim, char * norm) {
-  PROTECT(
-    auto outputs__ = torch::fft_ihfft(*self, n_null ? c10::nullopt : c10::optional<int64_t>(n_v), dim, std::string(norm));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fft_ihfft2(tensor *out__, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
-  PROTECT(
-    auto outputs__ = torch::fft_ihfft2(*self, s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), torch::IntArrayRef(dim_data, dim_len), std::string(norm));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fft_ihfft2_out(tensor *out__, tensor out, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
-  PROTECT(
-    auto outputs__ = torch::fft_ihfft2_out(*out, *self, s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), torch::IntArrayRef(dim_data, dim_len), std::string(norm));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fft_ihfft_out(tensor *out__, tensor out, tensor self, int64_t n_v, int n_null, int64_t dim, char * norm) {
-  PROTECT(
-    auto outputs__ = torch::fft_ihfft_out(*out, *self, n_null ? c10::nullopt : c10::optional<int64_t>(n_v), dim, std::string(norm));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fft_ihfftn(tensor *out__, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
-  PROTECT(
-    auto outputs__ = torch::fft_ihfftn(*self, s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), std::string(norm));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fft_ihfftn_out(tensor *out__, tensor out, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
-  PROTECT(
-    auto outputs__ = torch::fft_ihfftn_out(*out, *self, s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), std::string(norm));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fft_irfft(tensor *out__, tensor self, int64_t n_v, int n_null, int64_t dim, char * norm) {
-  PROTECT(
-    auto outputs__ = torch::fft_irfft(*self, n_null ? c10::nullopt : c10::optional<int64_t>(n_v), dim, std::string(norm));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fft_irfft2(tensor *out__, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
-  PROTECT(
-    auto outputs__ = torch::fft_irfft2(*self, s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), torch::IntArrayRef(dim_data, dim_len), std::string(norm));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fft_irfft2_out(tensor *out__, tensor out, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
-  PROTECT(
-    auto outputs__ = torch::fft_irfft2_out(*out, *self, s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), torch::IntArrayRef(dim_data, dim_len), std::string(norm));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fft_irfft_out(tensor *out__, tensor out, tensor self, int64_t n_v, int n_null, int64_t dim, char * norm) {
-  PROTECT(
-    auto outputs__ = torch::fft_irfft_out(*out, *self, n_null ? c10::nullopt : c10::optional<int64_t>(n_v), dim, std::string(norm));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fft_irfftn(tensor *out__, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
-  PROTECT(
-    auto outputs__ = torch::fft_irfftn(*self, s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), std::string(norm));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fft_irfftn_out(tensor *out__, tensor out, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
-  PROTECT(
-    auto outputs__ = torch::fft_irfftn_out(*out, *self, s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), std::string(norm));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fft_rfft(tensor *out__, tensor self, int64_t n_v, int n_null, int64_t dim, char * norm) {
-  PROTECT(
-    auto outputs__ = torch::fft_rfft(*self, n_null ? c10::nullopt : c10::optional<int64_t>(n_v), dim, std::string(norm));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fft_rfft2(tensor *out__, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
-  PROTECT(
-    auto outputs__ = torch::fft_rfft2(*self, s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), torch::IntArrayRef(dim_data, dim_len), std::string(norm));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fft_rfft2_out(tensor *out__, tensor out, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
-  PROTECT(
-    auto outputs__ = torch::fft_rfft2_out(*out, *self, s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), torch::IntArrayRef(dim_data, dim_len), std::string(norm));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fft_rfft_out(tensor *out__, tensor out, tensor self, int64_t n_v, int n_null, int64_t dim, char * norm) {
-  PROTECT(
-    auto outputs__ = torch::fft_rfft_out(*out, *self, n_null ? c10::nullopt : c10::optional<int64_t>(n_v), dim, std::string(norm));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fft_rfftfreq(tensor *out__, int64_t n, double d, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::fft_rfftfreq(n, d, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fft_rfftfreq_out(tensor *out__, tensor out, int64_t n, double d) {
-  PROTECT(
-    auto outputs__ = torch::fft_rfftfreq_out(*out, n, d);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fft_rfftn(tensor *out__, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
-  PROTECT(
-    auto outputs__ = torch::fft_rfftn(*self, s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), std::string(norm));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fft_rfftn_out(tensor *out__, tensor out, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm) {
-  PROTECT(
-    auto outputs__ = torch::fft_rfftn_out(*out, *self, s_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(s_data, s_len)), dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), std::string(norm));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fill(tensor *out__, tensor self, scalar value) {
-  PROTECT(
-    auto outputs__ = torch::fill(*self, *value);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fill_(tensor *out__, tensor self, scalar value) {
-  PROTECT(
-    auto outputs__ = torch::fill_(*self, *value);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fill_diagonal_(tensor *out__, tensor self, scalar fill_value, int wrap) {
-  PROTECT(
-    auto outputs__ = self->fill_diagonal_(*fill_value, (bool)wrap);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fill_scalar_out(tensor *out__, tensor out, tensor self, scalar value) {
-  PROTECT(
-    auto outputs__ = torch::fill_out(*out, *self, *value);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fill_tensor(tensor *out__, tensor self, tensor value) {
-  PROTECT(
-    auto outputs__ = torch::fill(*self, *value);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fill_tensor_(tensor *out__, tensor self, tensor value) {
-  PROTECT(
-    auto outputs__ = torch::fill_(*self, *value);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fill_tensor_out(tensor *out__, tensor out, tensor self, tensor value) {
-  PROTECT(
-    auto outputs__ = torch::fill_out(*out, *self, *value);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fix(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::fix(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fix_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::fix_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fix_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::fix_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_flatten(tensor *out__, tensor self, int64_t start_dim, int64_t end_dim) {
-  PROTECT(
-    auto outputs__ = torch::flatten(*self, start_dim, end_dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_flatten_dense_tensors(tensor *out__, tensor *tensors_data, int tensors_len) {
-  PROTECT(
-    auto outputs__ = torch::flatten_dense_tensors(of_carray_tensor(tensors_data, tensors_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_flip(tensor *out__, tensor self, int64_t *dims_data, int dims_len) {
-  PROTECT(
-    auto outputs__ = torch::flip(*self, torch::IntArrayRef(dims_data, dims_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_flip_out(tensor *out__, tensor out, tensor self, int64_t *dims_data, int dims_len) {
-  PROTECT(
-    auto outputs__ = torch::flip_out(*out, *self, torch::IntArrayRef(dims_data, dims_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fliplr(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::fliplr(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_flipud(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::flipud(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_float_power(tensor *out__, tensor self, tensor exponent) {
-  PROTECT(
-    auto outputs__ = torch::float_power(*self, *exponent);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_float_power_(tensor *out__, tensor self, scalar exponent) {
-  PROTECT(
-    auto outputs__ = self->float_power_(*exponent);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_float_power_scalar(tensor *out__, scalar self, tensor exponent) {
-  PROTECT(
-    auto outputs__ = torch::float_power(*self, *exponent);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_float_power_scalar_out(tensor *out__, tensor out, scalar self, tensor exponent) {
-  PROTECT(
-    auto outputs__ = torch::float_power_out(*out, *self, *exponent);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_float_power_tensor_(tensor *out__, tensor self, tensor exponent) {
-  PROTECT(
-    auto outputs__ = self->float_power_(*exponent);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_float_power_tensor_scalar(tensor *out__, tensor self, scalar exponent) {
-  PROTECT(
-    auto outputs__ = torch::float_power(*self, *exponent);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_float_power_tensor_scalar_out(tensor *out__, tensor out, tensor self, scalar exponent) {
-  PROTECT(
-    auto outputs__ = torch::float_power_out(*out, *self, *exponent);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_float_power_tensor_tensor_out(tensor *out__, tensor out, tensor self, tensor exponent) {
-  PROTECT(
-    auto outputs__ = torch::float_power_out(*out, *self, *exponent);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_floor(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::floor(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_floor_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::floor_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_floor_divide(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::floor_divide(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_floor_divide_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->floor_divide_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_floor_divide_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::floor_divide_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_floor_divide_scalar(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::floor_divide(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_floor_divide_scalar_(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = self->floor_divide_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_floor_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::floor_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fmax(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::fmax(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fmax_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::fmax_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fmin(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::fmin(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fmin_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::fmin_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fmod(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::fmod(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fmod_(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = self->fmod_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fmod_scalar_out(tensor *out__, tensor out, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::fmod_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fmod_tensor(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::fmod(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fmod_tensor_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->fmod_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fmod_tensor_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::fmod_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_frac(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::frac(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_frac_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::frac_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_frac_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::frac_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fractional_max_pool2d(tensor *out__, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *output_size_data, int output_size_len, tensor random_samples) {
-  PROTECT(
-    auto outputs__ = torch::fractional_max_pool2d(*self, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(output_size_data, output_size_len), *random_samples);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_fractional_max_pool2d_backward(tensor *out__, tensor grad_output, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *output_size_data, int output_size_len, tensor indices) {
-  PROTECT(
-    auto outputs__ = torch::fractional_max_pool2d_backward(*grad_output, *self, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(output_size_data, output_size_len), *indices);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fractional_max_pool2d_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *output_size_data, int output_size_len, tensor indices) {
-  PROTECT(
-    auto outputs__ = torch::fractional_max_pool2d_backward_out(*grad_input, *grad_output, *self, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(output_size_data, output_size_len), *indices);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fractional_max_pool2d_output(tensor *out__, tensor output, tensor indices, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *output_size_data, int output_size_len, tensor random_samples) {
-  PROTECT(
-    auto outputs__ = torch::fractional_max_pool2d_out(*output, *indices, *self, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(output_size_data, output_size_len), *random_samples);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_fractional_max_pool3d(tensor *out__, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *output_size_data, int output_size_len, tensor random_samples) {
-  PROTECT(
-    auto outputs__ = torch::fractional_max_pool3d(*self, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(output_size_data, output_size_len), *random_samples);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_fractional_max_pool3d_backward(tensor *out__, tensor grad_output, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *output_size_data, int output_size_len, tensor indices) {
-  PROTECT(
-    auto outputs__ = torch::fractional_max_pool3d_backward(*grad_output, *self, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(output_size_data, output_size_len), *indices);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fractional_max_pool3d_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *output_size_data, int output_size_len, tensor indices) {
-  PROTECT(
-    auto outputs__ = torch::fractional_max_pool3d_backward_out(*grad_input, *grad_output, *self, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(output_size_data, output_size_len), *indices);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fractional_max_pool3d_output(tensor *out__, tensor output, tensor indices, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *output_size_data, int output_size_len, tensor random_samples) {
-  PROTECT(
-    auto outputs__ = torch::fractional_max_pool3d_out(*output, *indices, *self, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(output_size_data, output_size_len), *random_samples);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_frexp(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::frexp(*self);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_frexp_tensor_out(tensor *out__, tensor mantissa, tensor exponent, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::frexp_out(*mantissa, *exponent, *self);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_frobenius_norm(tensor *out__, tensor self, int64_t *dim_data, int dim_len, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::frobenius_norm(*self, torch::IntArrayRef(dim_data, dim_len), (bool)keepdim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_frobenius_norm_out(tensor *out__, tensor out, tensor self, int64_t *dim_data, int dim_len, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::frobenius_norm_out(*out, *self, torch::IntArrayRef(dim_data, dim_len), (bool)keepdim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_from_file(tensor *out__, char * filename, int shared, int64_t size_v, int size_null, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::from_file(std::string(filename), (bool)shared, size_null ? c10::nullopt : c10::optional<int64_t>(size_v), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_from_file_out(tensor *out__, tensor out, char * filename, int shared, int64_t size_v, int size_null) {
-  PROTECT(
-    auto outputs__ = torch::from_file_out(*out, std::string(filename), (bool)shared, size_null ? c10::nullopt : c10::optional<int64_t>(size_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_full(tensor *out__, int64_t *size_data, int size_len, scalar fill_value, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::full(torch::IntArrayRef(size_data, size_len), *fill_value, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_full_like(tensor *out__, tensor self, scalar fill_value) {
-  PROTECT(
-    auto outputs__ = torch::full_like(*self, *fill_value);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_full_like_out(tensor *out__, tensor out, tensor self, scalar fill_value) {
-  PROTECT(
-    auto outputs__ = torch::full_like_out(*out, *self, *fill_value);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_full_out(tensor *out__, tensor out, int64_t *size_data, int size_len, scalar fill_value) {
-  PROTECT(
-    auto outputs__ = torch::full_out(*out, torch::IntArrayRef(size_data, size_len), *fill_value);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_fused_moving_avg_obs_fake_quant(tensor *out__, tensor self, tensor observer_on, tensor fake_quant_on, tensor running_min, tensor running_max, tensor scale, tensor zero_point, double averaging_const, int64_t quant_min, int64_t quant_max, int64_t ch_axis, int per_row_fake_quant, int symmetric_quant) {
-  PROTECT(
-    auto outputs__ = torch::fused_moving_avg_obs_fake_quant(*self, *observer_on, *fake_quant_on, *running_min, *running_max, *scale, *zero_point, averaging_const, quant_min, quant_max, ch_axis, (bool)per_row_fake_quant, (bool)symmetric_quant);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_gather(tensor *out__, tensor self, int64_t dim, tensor index, int sparse_grad) {
-  PROTECT(
-    auto outputs__ = torch::gather(*self, dim, *index, (bool)sparse_grad);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_gather_backward(tensor *out__, tensor grad, tensor self, int64_t dim, tensor index, int sparse_grad) {
-  PROTECT(
-    auto outputs__ = torch::gather_backward(*grad, *self, dim, *index, (bool)sparse_grad);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_gather_out(tensor *out__, tensor out, tensor self, int64_t dim, tensor index, int sparse_grad) {
-  PROTECT(
-    auto outputs__ = torch::gather_out(*out, *self, dim, *index, (bool)sparse_grad);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_gcd(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::gcd(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_gcd_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::gcd_(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_gcd_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::gcd_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_ge(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::ge(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_ge_(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = self->ge_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_ge_scalar_out(tensor *out__, tensor out, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::ge_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_ge_tensor(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::ge(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_ge_tensor_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->ge_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_ge_tensor_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::ge_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_gelu(tensor *out__, tensor self, char * approximate) {
-  PROTECT(
-    auto outputs__ = torch::gelu(*self, std::string(approximate));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_gelu_(tensor *out__, tensor self, char * approximate) {
-  PROTECT(
-    auto outputs__ = torch::gelu_(*self, std::string(approximate));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_gelu_backward(tensor *out__, tensor grad_output, tensor self, char * approximate) {
-  PROTECT(
-    auto outputs__ = torch::gelu_backward(*grad_output, *self, std::string(approximate));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_gelu_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, tensor self, char * approximate) {
-  PROTECT(
-    auto outputs__ = torch::gelu_backward_out(*grad_input, *grad_output, *self, std::string(approximate));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_gelu_out(tensor *out__, tensor out, tensor self, char * approximate) {
-  PROTECT(
-    auto outputs__ = torch::gelu_out(*out, *self, std::string(approximate));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_geometric(tensor *out__, tensor self, double p) {
-  PROTECT(
-    auto outputs__ = torch::geometric(*self, p);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_geometric_(tensor *out__, tensor self, double p) {
-  PROTECT(
-    auto outputs__ = self->geometric_(p);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_geometric_out(tensor *out__, tensor out, tensor self, double p) {
-  PROTECT(
-    auto outputs__ = torch::geometric_out(*out, *self, p);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_geqrf(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::geqrf(*self);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_geqrf_a(tensor *out__, tensor a, tensor tau, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::geqrf_out(*a, *tau, *self);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_ger(tensor *out__, tensor self, tensor vec2) {
-  PROTECT(
-    auto outputs__ = torch::ger(*self, *vec2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_ger_out(tensor *out__, tensor out, tensor self, tensor vec2) {
-  PROTECT(
-    auto outputs__ = torch::ger_out(*out, *self, *vec2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_glu(tensor *out__, tensor self, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::glu(*self, dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_glu_backward(tensor *out__, tensor grad_output, tensor self, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::glu_backward(*grad_output, *self, dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_glu_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, tensor self, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::glu_backward_out(*grad_input, *grad_output, *self, dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_glu_backward_jvp(tensor *out__, tensor grad_x, tensor grad_glu, tensor x, tensor dgrad_glu, tensor dx, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::glu_backward_jvp(*grad_x, *grad_glu, *x, *dgrad_glu, *dx, dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_glu_backward_jvp_out(tensor *out__, tensor out, tensor grad_x, tensor grad_glu, tensor x, tensor dgrad_glu, tensor dx, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::glu_backward_jvp_out(*out, *grad_x, *grad_glu, *x, *dgrad_glu, *dx, dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_glu_jvp(tensor *out__, tensor glu, tensor x, tensor dx, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::glu_jvp(*glu, *x, *dx, dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_glu_jvp_out(tensor *out__, tensor out, tensor glu, tensor x, tensor dx, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::glu_jvp_out(*out, *glu, *x, *dx, dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_glu_out(tensor *out__, tensor out, tensor self, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::glu_out(*out, *self, dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_grad(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = self->grad();
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_greater(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::greater(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_greater_(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = self->greater_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_greater_equal(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::greater_equal(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_greater_equal_(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = self->greater_equal_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_greater_equal_scalar_out(tensor *out__, tensor out, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::greater_equal_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_greater_equal_tensor(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::greater_equal(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_greater_equal_tensor_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->greater_equal_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_greater_equal_tensor_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::greater_equal_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_greater_scalar_out(tensor *out__, tensor out, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::greater_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_greater_tensor(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::greater(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_greater_tensor_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->greater_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_greater_tensor_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::greater_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_grid_sampler(tensor *out__, tensor input, tensor grid, int64_t interpolation_mode, int64_t padding_mode, int align_corners) {
-  PROTECT(
-    auto outputs__ = torch::grid_sampler(*input, *grid, interpolation_mode, padding_mode, (bool)align_corners);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_grid_sampler_2d(tensor *out__, tensor input, tensor grid, int64_t interpolation_mode, int64_t padding_mode, int align_corners) {
-  PROTECT(
-    auto outputs__ = torch::grid_sampler_2d(*input, *grid, interpolation_mode, padding_mode, (bool)align_corners);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_grid_sampler_2d_out(tensor *out__, tensor out, tensor input, tensor grid, int64_t interpolation_mode, int64_t padding_mode, int align_corners) {
-  PROTECT(
-    auto outputs__ = torch::grid_sampler_2d_out(*out, *input, *grid, interpolation_mode, padding_mode, (bool)align_corners);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_grid_sampler_3d(tensor *out__, tensor input, tensor grid, int64_t interpolation_mode, int64_t padding_mode, int align_corners) {
-  PROTECT(
-    auto outputs__ = torch::grid_sampler_3d(*input, *grid, interpolation_mode, padding_mode, (bool)align_corners);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_grid_sampler_3d_out(tensor *out__, tensor out, tensor input, tensor grid, int64_t interpolation_mode, int64_t padding_mode, int align_corners) {
-  PROTECT(
-    auto outputs__ = torch::grid_sampler_3d_out(*out, *input, *grid, interpolation_mode, padding_mode, (bool)align_corners);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_group_norm(tensor *out__, tensor input, int64_t num_groups, tensor weight, tensor bias, double eps, int cudnn_enabled) {
-  PROTECT(
-    auto outputs__ = torch::group_norm(*input, num_groups, (weight ? *weight : torch::Tensor()), (bias ? *bias : torch::Tensor()), eps, (bool)cudnn_enabled);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_gru(tensor *out__, tensor input, tensor hx, tensor *params_data, int params_len, int has_biases, int64_t num_layers, double dropout, int train, int bidirectional, int batch_first) {
-  PROTECT(
-    auto outputs__ = torch::gru(*input, *hx, of_carray_tensor(params_data, params_len), (bool)has_biases, num_layers, dropout, (bool)train, (bool)bidirectional, (bool)batch_first);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_gru_cell(tensor *out__, tensor input, tensor hx, tensor w_ih, tensor w_hh, tensor b_ih, tensor b_hh) {
-  PROTECT(
-    auto outputs__ = torch::gru_cell(*input, *hx, *w_ih, *w_hh, (b_ih ? *b_ih : torch::Tensor()), (b_hh ? *b_hh : torch::Tensor()));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_gru_data(tensor *out__, tensor data, tensor batch_sizes, tensor hx, tensor *params_data, int params_len, int has_biases, int64_t num_layers, double dropout, int train, int bidirectional) {
-  PROTECT(
-    auto outputs__ = torch::gru(*data, *batch_sizes, *hx, of_carray_tensor(params_data, params_len), (bool)has_biases, num_layers, dropout, (bool)train, (bool)bidirectional);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_gt(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::gt(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_gt_(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = self->gt_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_gt_scalar_out(tensor *out__, tensor out, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::gt_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_gt_tensor(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::gt(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_gt_tensor_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->gt_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_gt_tensor_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::gt_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_hamming_window(tensor *out__, int64_t window_length, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::hamming_window(window_length, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_hamming_window_out(tensor *out__, tensor out, int64_t window_length) {
-  PROTECT(
-    auto outputs__ = torch::hamming_window_out(*out, window_length);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_hamming_window_periodic(tensor *out__, int64_t window_length, int periodic, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::hamming_window(window_length, (bool)periodic, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_hamming_window_periodic_alpha(tensor *out__, int64_t window_length, int periodic, double alpha, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::hamming_window(window_length, (bool)periodic, alpha, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_hamming_window_periodic_alpha_beta(tensor *out__, int64_t window_length, int periodic, double alpha, double beta, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::hamming_window(window_length, (bool)periodic, alpha, beta, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_hamming_window_periodic_alpha_beta_out(tensor *out__, tensor out, int64_t window_length, int periodic, double alpha, double beta) {
-  PROTECT(
-    auto outputs__ = torch::hamming_window_out(*out, window_length, (bool)periodic, alpha, beta);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_hamming_window_periodic_alpha_out(tensor *out__, tensor out, int64_t window_length, int periodic, double alpha) {
-  PROTECT(
-    auto outputs__ = torch::hamming_window_out(*out, window_length, (bool)periodic, alpha);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_hamming_window_periodic_out(tensor *out__, tensor out, int64_t window_length, int periodic) {
-  PROTECT(
-    auto outputs__ = torch::hamming_window_out(*out, window_length, (bool)periodic);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_hann_window(tensor *out__, int64_t window_length, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::hann_window(window_length, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_hann_window_out(tensor *out__, tensor out, int64_t window_length) {
-  PROTECT(
-    auto outputs__ = torch::hann_window_out(*out, window_length);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_hann_window_periodic(tensor *out__, int64_t window_length, int periodic, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::hann_window(window_length, (bool)periodic, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_hann_window_periodic_out(tensor *out__, tensor out, int64_t window_length, int periodic) {
-  PROTECT(
-    auto outputs__ = torch::hann_window_out(*out, window_length, (bool)periodic);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_hardshrink(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::hardshrink(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_hardshrink_backward(tensor *out__, tensor grad_out, tensor self, scalar lambd) {
-  PROTECT(
-    auto outputs__ = torch::hardshrink_backward(*grad_out, *self, *lambd);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_hardshrink_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_out, tensor self, scalar lambd) {
-  PROTECT(
-    auto outputs__ = torch::hardshrink_backward_out(*grad_input, *grad_out, *self, *lambd);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_hardshrink_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::hardshrink_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_hardsigmoid(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::hardsigmoid(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_hardsigmoid_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::hardsigmoid_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_hardsigmoid_backward(tensor *out__, tensor grad_output, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::hardsigmoid_backward(*grad_output, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_hardsigmoid_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::hardsigmoid_backward_out(*grad_input, *grad_output, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_hardsigmoid_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::hardsigmoid_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_hardswish(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::hardswish(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_hardswish_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::hardswish_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_hardswish_backward(tensor *out__, tensor grad_output, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::hardswish_backward(*grad_output, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_hardswish_backward_out(tensor *out__, tensor out, tensor grad_output, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::hardswish_backward_out(*out, *grad_output, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_hardswish_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::hardswish_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_hardtanh(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::hardtanh(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_hardtanh_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::hardtanh_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_hardtanh_backward(tensor *out__, tensor grad_output, tensor self, scalar min_val, scalar max_val) {
-  PROTECT(
-    auto outputs__ = torch::hardtanh_backward(*grad_output, *self, *min_val, *max_val);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_hardtanh_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, tensor self, scalar min_val, scalar max_val) {
-  PROTECT(
-    auto outputs__ = torch::hardtanh_backward_out(*grad_input, *grad_output, *self, *min_val, *max_val);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_hardtanh_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::hardtanh_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_heaviside(tensor *out__, tensor self, tensor values) {
-  PROTECT(
-    auto outputs__ = torch::heaviside(*self, *values);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_heaviside_(tensor *out__, tensor self, tensor values) {
-  PROTECT(
-    auto outputs__ = self->heaviside_(*values);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_heaviside_out(tensor *out__, tensor out, tensor self, tensor values) {
-  PROTECT(
-    auto outputs__ = torch::heaviside_out(*out, *self, *values);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_hinge_embedding_loss(tensor *out__, tensor self, tensor target, double margin, int64_t reduction) {
-  PROTECT(
-    auto outputs__ = torch::hinge_embedding_loss(*self, *target, margin, reduction);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_histc(tensor *out__, tensor self, int64_t bins) {
-  PROTECT(
-    auto outputs__ = torch::histc(*self, bins);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_histc_out(tensor *out__, tensor out, tensor self, int64_t bins) {
-  PROTECT(
-    auto outputs__ = torch::histc_out(*out, *self, bins);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-tensor *atg_hsplit(tensor self, int64_t sections) {
-  PROTECT(
-    auto outputs__ = torch::hsplit(*self, sections);
-    int sz = outputs__.size();
-    torch::Tensor **out__ = (torch::Tensor**)malloc((sz + 1) * sizeof(torch::Tensor*));
-    for (int i = 0; i < sz; ++i)
-      out__[i] = new torch::Tensor(outputs__[i]);
-    out__[sz] = nullptr;
-    return out__;
-  )
-}
-
-tensor *atg_hsplit_array(tensor self, int64_t *indices_data, int indices_len) {
-  PROTECT(
-    auto outputs__ = torch::hsplit(*self, torch::IntArrayRef(indices_data, indices_len));
-    int sz = outputs__.size();
-    torch::Tensor **out__ = (torch::Tensor**)malloc((sz + 1) * sizeof(torch::Tensor*));
-    for (int i = 0; i < sz; ++i)
-      out__[i] = new torch::Tensor(outputs__[i]);
-    out__[sz] = nullptr;
-    return out__;
-  )
-}
-
-void atg_hspmm(tensor *out__, tensor mat1, tensor mat2) {
-  PROTECT(
-    auto outputs__ = torch::hspmm(*mat1, *mat2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_hspmm_out(tensor *out__, tensor out, tensor mat1, tensor mat2) {
-  PROTECT(
-    auto outputs__ = torch::hspmm_out(*out, *mat1, *mat2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_hstack(tensor *out__, tensor *tensors_data, int tensors_len) {
-  PROTECT(
-    auto outputs__ = torch::hstack(of_carray_tensor(tensors_data, tensors_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_hstack_out(tensor *out__, tensor out, tensor *tensors_data, int tensors_len) {
-  PROTECT(
-    auto outputs__ = torch::hstack_out(*out, of_carray_tensor(tensors_data, tensors_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_huber_loss(tensor *out__, tensor self, tensor target, int64_t reduction, double delta) {
-  PROTECT(
-    auto outputs__ = torch::huber_loss(*self, *target, reduction, delta);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_huber_loss_backward(tensor *out__, tensor grad_output, tensor self, tensor target, int64_t reduction, double delta) {
-  PROTECT(
-    auto outputs__ = torch::huber_loss_backward(*grad_output, *self, *target, reduction, delta);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_huber_loss_backward_out(tensor *out__, tensor grad_input, tensor grad_output, tensor self, tensor target, int64_t reduction, double delta) {
-  PROTECT(
-    auto outputs__ = torch::huber_loss_backward_out(*grad_input, *grad_output, *self, *target, reduction, delta);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_huber_loss_out(tensor *out__, tensor out, tensor self, tensor target, int64_t reduction, double delta) {
-  PROTECT(
-    auto outputs__ = torch::huber_loss_out(*out, *self, *target, reduction, delta);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_hypot(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::hypot(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_hypot_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->hypot_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_hypot_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::hypot_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_i0(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::i0(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_i0_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::i0_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_i0_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::i0_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_igamma(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::igamma(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_igamma_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->igamma_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_igamma_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::igamma_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_igammac(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::igammac(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_igammac_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->igammac_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_igammac_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::igammac_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_im2col(tensor *out__, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *dilation_data, int dilation_len, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len) {
-  PROTECT(
-    auto outputs__ = torch::im2col(*self, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(dilation_data, dilation_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(stride_data, stride_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_im2col_out(tensor *out__, tensor out, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *dilation_data, int dilation_len, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len) {
-  PROTECT(
-    auto outputs__ = torch::im2col_out(*out, *self, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(dilation_data, dilation_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(stride_data, stride_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_imag(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::imag(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_index(tensor *out__, tensor self, tensor *indices_data, int indices_len) {
-  PROTECT(
-    auto outputs__ = torch::index(*self, of_carray_tensor_opt(indices_data, indices_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_index_add(tensor *out__, tensor self, int64_t dim, tensor index, tensor source) {
-  PROTECT(
-    auto outputs__ = torch::index_add(*self, dim, *index, *source);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_index_add_(tensor *out__, tensor self, int64_t dim, tensor index, tensor source) {
-  PROTECT(
-    auto outputs__ = self->index_add_(dim, *index, *source);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_index_add_out(tensor *out__, tensor out, tensor self, int64_t dim, tensor index, tensor source) {
-  PROTECT(
-    auto outputs__ = torch::index_add_out(*out, *self, dim, *index, *source);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_index_copy(tensor *out__, tensor self, int64_t dim, tensor index, tensor source) {
-  PROTECT(
-    auto outputs__ = torch::index_copy(*self, dim, *index, *source);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_index_copy_(tensor *out__, tensor self, int64_t dim, tensor index, tensor source) {
-  PROTECT(
-    auto outputs__ = self->index_copy_(dim, *index, *source);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_index_copy_out(tensor *out__, tensor out, tensor self, int64_t dim, tensor index, tensor source) {
-  PROTECT(
-    auto outputs__ = torch::index_copy_out(*out, *self, dim, *index, *source);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_index_fill(tensor *out__, tensor self, int64_t dim, tensor index, scalar value) {
-  PROTECT(
-    auto outputs__ = torch::index_fill(*self, dim, *index, *value);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_index_fill_(tensor *out__, tensor self, int64_t dim, tensor index, scalar value) {
-  PROTECT(
-    auto outputs__ = self->index_fill_(dim, *index, *value);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_index_fill_int_scalar_out(tensor *out__, tensor out, tensor self, int64_t dim, tensor index, scalar value) {
-  PROTECT(
-    auto outputs__ = torch::index_fill_out(*out, *self, dim, *index, *value);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_index_fill_int_tensor(tensor *out__, tensor self, int64_t dim, tensor index, tensor value) {
-  PROTECT(
-    auto outputs__ = torch::index_fill(*self, dim, *index, *value);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_index_fill_int_tensor_(tensor *out__, tensor self, int64_t dim, tensor index, tensor value) {
-  PROTECT(
-    auto outputs__ = self->index_fill_(dim, *index, *value);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_index_fill_int_tensor_out(tensor *out__, tensor out, tensor self, int64_t dim, tensor index, tensor value) {
-  PROTECT(
-    auto outputs__ = torch::index_fill_out(*out, *self, dim, *index, *value);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_index_put(tensor *out__, tensor self, tensor *indices_data, int indices_len, tensor values, int accumulate) {
-  PROTECT(
-    auto outputs__ = torch::index_put(*self, of_carray_tensor_opt(indices_data, indices_len), *values, (bool)accumulate);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_index_put_(tensor *out__, tensor self, tensor *indices_data, int indices_len, tensor values, int accumulate) {
-  PROTECT(
-    auto outputs__ = torch::index_put_(*self, of_carray_tensor_opt(indices_data, indices_len), *values, (bool)accumulate);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_index_put_out(tensor *out__, tensor out, tensor self, tensor *indices_data, int indices_len, tensor values, int accumulate) {
-  PROTECT(
-    auto outputs__ = torch::index_put_out(*out, *self, of_carray_tensor_opt(indices_data, indices_len), *values, (bool)accumulate);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_index_reduce(tensor *out__, tensor self, int64_t dim, tensor index, tensor source, char * reduce, int include_self) {
-  PROTECT(
-    auto outputs__ = torch::index_reduce(*self, dim, *index, *source, std::string(reduce), (bool)include_self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_index_reduce_(tensor *out__, tensor self, int64_t dim, tensor index, tensor source, char * reduce, int include_self) {
-  PROTECT(
-    auto outputs__ = self->index_reduce_(dim, *index, *source, std::string(reduce), (bool)include_self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_index_reduce_out(tensor *out__, tensor out, tensor self, int64_t dim, tensor index, tensor source, char * reduce, int include_self) {
-  PROTECT(
-    auto outputs__ = torch::index_reduce_out(*out, *self, dim, *index, *source, std::string(reduce), (bool)include_self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_index_select(tensor *out__, tensor self, int64_t dim, tensor index) {
-  PROTECT(
-    auto outputs__ = torch::index_select(*self, dim, *index);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_index_select_backward(tensor *out__, tensor grad, int64_t *self_sizes_data, int self_sizes_len, int64_t dim, tensor index) {
-  PROTECT(
-    auto outputs__ = torch::index_select_backward(*grad, torch::IntArrayRef(self_sizes_data, self_sizes_len), dim, *index);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_index_select_out(tensor *out__, tensor out, tensor self, int64_t dim, tensor index) {
-  PROTECT(
-    auto outputs__ = torch::index_select_out(*out, *self, dim, *index);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_index_tensor_out(tensor *out__, tensor out, tensor self, tensor *indices_data, int indices_len) {
-  PROTECT(
-    auto outputs__ = torch::index_out(*out, *self, of_carray_tensor_opt(indices_data, indices_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_indices(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = self->indices();
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_indices_copy(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::indices_copy(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_indices_copy_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::indices_copy_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_infinitely_differentiable_gelu_backward(tensor *out__, tensor grad, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::infinitely_differentiable_gelu_backward(*grad, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_inner(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::inner(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_inner_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::inner_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_instance_norm(tensor *out__, tensor input, tensor weight, tensor bias, tensor running_mean, tensor running_var, int use_input_stats, double momentum, double eps, int cudnn_enabled) {
-  PROTECT(
-    auto outputs__ = torch::instance_norm(*input, (weight ? *weight : torch::Tensor()), (bias ? *bias : torch::Tensor()), (running_mean ? *running_mean : torch::Tensor()), (running_var ? *running_var : torch::Tensor()), (bool)use_input_stats, momentum, eps, (bool)cudnn_enabled);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_int_repr(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::int_repr(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_int_repr_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::int_repr_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_inverse(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::inverse(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_inverse_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::inverse_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-int atg_is_coalesced(tensor self) {
-  PROTECT(
-    return self->is_coalesced();
-  )
-  return 0;
-}
-
-int atg_is_complex(tensor self) {
-  PROTECT(
-    return torch::is_complex(*self);
-  )
-  return 0;
-}
-
-int atg_is_conj(tensor self) {
-  PROTECT(
-    return torch::is_conj(*self);
-  )
-  return 0;
-}
-
-int atg_is_distributed(tensor self) {
-  PROTECT(
-    return torch::is_distributed(*self);
-  )
-  return 0;
-}
-
-int atg_is_floating_point(tensor self) {
-  PROTECT(
-    return torch::is_floating_point(*self);
-  )
-  return 0;
-}
-
-int atg_is_inference(tensor self) {
-  PROTECT(
-    return torch::is_inference(*self);
-  )
-  return 0;
-}
-
-int atg_is_leaf(tensor self) {
-  PROTECT(
-    return self->is_leaf();
-  )
-  return 0;
-}
-
-int atg_is_neg(tensor self) {
-  PROTECT(
-    return torch::is_neg(*self);
-  )
-  return 0;
-}
-
-int atg_is_nonzero(tensor self) {
-  PROTECT(
-    return torch::is_nonzero(*self);
-  )
-  return 0;
-}
-
-int atg_is_pinned(tensor self, int device) {
-  PROTECT(
-    return self->is_pinned(device_of_int(device));
-  )
-  return 0;
-}
-
-int atg_is_same_size(tensor self, tensor other) {
-  PROTECT(
-    return torch::is_same_size(*self, *other);
-  )
-  return 0;
-}
-
-int atg_is_set_to(tensor self, tensor tensor) {
-  PROTECT(
-    return self->is_set_to(*tensor);
-  )
-  return 0;
-}
-
-int atg_is_signed(tensor self) {
-  PROTECT(
-    return torch::is_signed(*self);
-  )
-  return 0;
-}
-
-int atg_is_vulkan_available() {
-  PROTECT(
-    return torch::is_vulkan_available();
-  )
-  return 0;
-}
-
-void atg_isclose(tensor *out__, tensor self, tensor other, double rtol, double atol, int equal_nan) {
-  PROTECT(
-    auto outputs__ = torch::isclose(*self, *other, rtol, atol, (bool)equal_nan);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_isfinite(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::isfinite(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_isin(tensor *out__, tensor elements, tensor test_elements, int assume_unique, int invert) {
-  PROTECT(
-    auto outputs__ = torch::isin(*elements, *test_elements, (bool)assume_unique, (bool)invert);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_isin_scalar_tensor(tensor *out__, scalar element, tensor test_elements, int assume_unique, int invert) {
-  PROTECT(
-    auto outputs__ = torch::isin(*element, *test_elements, (bool)assume_unique, (bool)invert);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_isin_scalar_tensor_out(tensor *out__, tensor out, scalar element, tensor test_elements, int assume_unique, int invert) {
-  PROTECT(
-    auto outputs__ = torch::isin_out(*out, *element, *test_elements, (bool)assume_unique, (bool)invert);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_isin_tensor_scalar(tensor *out__, tensor elements, scalar test_element, int assume_unique, int invert) {
-  PROTECT(
-    auto outputs__ = torch::isin(*elements, *test_element, (bool)assume_unique, (bool)invert);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_isin_tensor_scalar_out(tensor *out__, tensor out, tensor elements, scalar test_element, int assume_unique, int invert) {
-  PROTECT(
-    auto outputs__ = torch::isin_out(*out, *elements, *test_element, (bool)assume_unique, (bool)invert);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_isin_tensor_tensor_out(tensor *out__, tensor out, tensor elements, tensor test_elements, int assume_unique, int invert) {
-  PROTECT(
-    auto outputs__ = torch::isin_out(*out, *elements, *test_elements, (bool)assume_unique, (bool)invert);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_isinf(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::isinf(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_isinf_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::isinf_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_isnan(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::isnan(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_isnan_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::isnan_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_isneginf(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::isneginf(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_isneginf_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::isneginf_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_isposinf(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::isposinf(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_isposinf_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::isposinf_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_isreal(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::isreal(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_istft(tensor *out__, tensor self, int64_t n_fft, int64_t hop_length_v, int hop_length_null, int64_t win_length_v, int win_length_null, tensor window, int center, int normalized, int onesided, int64_t length_v, int length_null, int return_complex) {
-  PROTECT(
-    auto outputs__ = torch::istft(*self, n_fft, hop_length_null ? c10::nullopt : c10::optional<int64_t>(hop_length_v), win_length_null ? c10::nullopt : c10::optional<int64_t>(win_length_v), (window ? *window : torch::Tensor()), (bool)center, (bool)normalized, (bool)onesided, length_null ? c10::nullopt : c10::optional<int64_t>(length_v), (bool)return_complex);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_kaiser_window(tensor *out__, int64_t window_length, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::kaiser_window(window_length, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_kaiser_window_beta(tensor *out__, int64_t window_length, int periodic, double beta, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::kaiser_window(window_length, (bool)periodic, beta, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_kaiser_window_beta_out(tensor *out__, tensor out, int64_t window_length, int periodic, double beta) {
-  PROTECT(
-    auto outputs__ = torch::kaiser_window_out(*out, window_length, (bool)periodic, beta);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_kaiser_window_out(tensor *out__, tensor out, int64_t window_length) {
-  PROTECT(
-    auto outputs__ = torch::kaiser_window_out(*out, window_length);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_kaiser_window_periodic(tensor *out__, int64_t window_length, int periodic, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::kaiser_window(window_length, (bool)periodic, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_kaiser_window_periodic_out(tensor *out__, tensor out, int64_t window_length, int periodic) {
-  PROTECT(
-    auto outputs__ = torch::kaiser_window_out(*out, window_length, (bool)periodic);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_kl_div(tensor *out__, tensor self, tensor target, int64_t reduction, int log_target) {
-  PROTECT(
-    auto outputs__ = torch::kl_div(*self, *target, reduction, (bool)log_target);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_kron(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::kron(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_kron_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::kron_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_kthvalue(tensor *out__, tensor self, int64_t k, int64_t dim, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::kthvalue(*self, k, dim, (bool)keepdim);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_kthvalue_values(tensor *out__, tensor values, tensor indices, tensor self, int64_t k, int64_t dim, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::kthvalue_out(*values, *indices, *self, k, dim, (bool)keepdim);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_l1_loss(tensor *out__, tensor self, tensor target, int64_t reduction) {
-  PROTECT(
-    auto outputs__ = torch::l1_loss(*self, *target, reduction);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_layer_norm(tensor *out__, tensor input, int64_t *normalized_shape_data, int normalized_shape_len, tensor weight, tensor bias, double eps, int cudnn_enable) {
-  PROTECT(
-    auto outputs__ = torch::layer_norm(*input, torch::IntArrayRef(normalized_shape_data, normalized_shape_len), (weight ? *weight : torch::Tensor()), (bias ? *bias : torch::Tensor()), eps, (bool)cudnn_enable);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_lcm(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::lcm(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_lcm_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::lcm_(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_lcm_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::lcm_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_ldexp(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::ldexp(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_ldexp_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::ldexp_(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_ldexp_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::ldexp_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_le(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::le(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_le_(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = self->le_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_le_scalar_out(tensor *out__, tensor out, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::le_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_le_tensor(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::le(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_le_tensor_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->le_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_le_tensor_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::le_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_leaky_relu(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::leaky_relu(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_leaky_relu_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::leaky_relu_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_leaky_relu_backward(tensor *out__, tensor grad_output, tensor self, scalar negative_slope, int self_is_result) {
-  PROTECT(
-    auto outputs__ = torch::leaky_relu_backward(*grad_output, *self, *negative_slope, (bool)self_is_result);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_leaky_relu_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, tensor self, scalar negative_slope, int self_is_result) {
-  PROTECT(
-    auto outputs__ = torch::leaky_relu_backward_out(*grad_input, *grad_output, *self, *negative_slope, (bool)self_is_result);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_leaky_relu_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::leaky_relu_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_lerp(tensor *out__, tensor self, tensor end, scalar weight) {
-  PROTECT(
-    auto outputs__ = torch::lerp(*self, *end, *weight);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_lerp_(tensor *out__, tensor self, tensor end, scalar weight) {
-  PROTECT(
-    auto outputs__ = self->lerp_(*end, *weight);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_lerp_scalar_out(tensor *out__, tensor out, tensor self, tensor end, scalar weight) {
-  PROTECT(
-    auto outputs__ = torch::lerp_out(*out, *self, *end, *weight);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_lerp_tensor(tensor *out__, tensor self, tensor end, tensor weight) {
-  PROTECT(
-    auto outputs__ = torch::lerp(*self, *end, *weight);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_lerp_tensor_(tensor *out__, tensor self, tensor end, tensor weight) {
-  PROTECT(
-    auto outputs__ = self->lerp_(*end, *weight);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_lerp_tensor_out(tensor *out__, tensor out, tensor self, tensor end, tensor weight) {
-  PROTECT(
-    auto outputs__ = torch::lerp_out(*out, *self, *end, *weight);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_less(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::less(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_less_(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = self->less_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_less_equal(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::less_equal(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_less_equal_(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = self->less_equal_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_less_equal_scalar_out(tensor *out__, tensor out, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::less_equal_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_less_equal_tensor(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::less_equal(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_less_equal_tensor_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->less_equal_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_less_equal_tensor_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::less_equal_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_less_scalar_out(tensor *out__, tensor out, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::less_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_less_tensor(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::less(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_less_tensor_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->less_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_less_tensor_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::less_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_lgamma(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::lgamma(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_lgamma_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = self->lgamma_();
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_lgamma_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::lgamma_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_lift(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::lift(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_lift_fresh(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::lift_fresh(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_lift_fresh_copy(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::lift_fresh_copy(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_lift_fresh_copy_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::lift_fresh_copy_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_lift_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::lift_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_cholesky(tensor *out__, tensor self, int upper) {
-  PROTECT(
-    auto outputs__ = torch::linalg_cholesky(*self, (bool)upper);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_cholesky_ex(tensor *out__, tensor self, int upper, int check_errors) {
-  PROTECT(
-    auto outputs__ = torch::linalg_cholesky_ex(*self, (bool)upper, (bool)check_errors);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_linalg_cholesky_ex_l(tensor *out__, tensor L, tensor info, tensor self, int upper, int check_errors) {
-  PROTECT(
-    auto outputs__ = torch::linalg_cholesky_ex_out(*L, *info, *self, (bool)upper, (bool)check_errors);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_linalg_cholesky_out(tensor *out__, tensor out, tensor self, int upper) {
-  PROTECT(
-    auto outputs__ = torch::linalg_cholesky_out(*out, *self, (bool)upper);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_cond(tensor *out__, tensor self, scalar p) {
-  PROTECT(
-    auto outputs__ = torch::linalg_cond(*self, *p);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_cond_out(tensor *out__, tensor out, tensor self, scalar p) {
-  PROTECT(
-    auto outputs__ = torch::linalg_cond_out(*out, *self, *p);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_cond_p_str(tensor *out__, tensor self, char * p) {
-  PROTECT(
-    auto outputs__ = torch::linalg_cond(*self, std::string(p));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_cond_p_str_out(tensor *out__, tensor out, tensor self, char * p) {
-  PROTECT(
-    auto outputs__ = torch::linalg_cond_out(*out, *self, std::string(p));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_cross(tensor *out__, tensor self, tensor other, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::linalg_cross(*self, *other, dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_cross_out(tensor *out__, tensor out, tensor self, tensor other, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::linalg_cross_out(*out, *self, *other, dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_det(tensor *out__, tensor A) {
-  PROTECT(
-    auto outputs__ = torch::linalg_det(*A);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_det_out(tensor *out__, tensor out, tensor A) {
-  PROTECT(
-    auto outputs__ = torch::linalg_det_out(*out, *A);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_diagonal(tensor *out__, tensor A, int64_t offset, int64_t dim1, int64_t dim2) {
-  PROTECT(
-    auto outputs__ = torch::linalg_diagonal(*A, offset, dim1, dim2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_eig(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::linalg_eig(*self);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_linalg_eig_out(tensor *out__, tensor eigenvalues, tensor eigenvectors, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::linalg_eig_out(*eigenvalues, *eigenvectors, *self);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_linalg_eigh(tensor *out__, tensor self, char * UPLO) {
-  PROTECT(
-    auto outputs__ = torch::linalg_eigh(*self, std::string(UPLO));
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_linalg_eigh_eigvals(tensor *out__, tensor eigvals, tensor eigvecs, tensor self, char * UPLO) {
-  PROTECT(
-    auto outputs__ = torch::linalg_eigh_out(*eigvals, *eigvecs, *self, std::string(UPLO));
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_linalg_eigvals(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::linalg_eigvals(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_eigvals_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::linalg_eigvals_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_eigvalsh(tensor *out__, tensor self, char * UPLO) {
-  PROTECT(
-    auto outputs__ = torch::linalg_eigvalsh(*self, std::string(UPLO));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_eigvalsh_out(tensor *out__, tensor out, tensor self, char * UPLO) {
-  PROTECT(
-    auto outputs__ = torch::linalg_eigvalsh_out(*out, *self, std::string(UPLO));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_householder_product(tensor *out__, tensor input, tensor tau) {
-  PROTECT(
-    auto outputs__ = torch::linalg_householder_product(*input, *tau);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_householder_product_out(tensor *out__, tensor out, tensor input, tensor tau) {
-  PROTECT(
-    auto outputs__ = torch::linalg_householder_product_out(*out, *input, *tau);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_inv(tensor *out__, tensor A) {
-  PROTECT(
-    auto outputs__ = torch::linalg_inv(*A);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_inv_ex(tensor *out__, tensor A, int check_errors) {
-  PROTECT(
-    auto outputs__ = torch::linalg_inv_ex(*A, (bool)check_errors);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_linalg_inv_ex_inverse(tensor *out__, tensor inverse, tensor info, tensor A, int check_errors) {
-  PROTECT(
-    auto outputs__ = torch::linalg_inv_ex_out(*inverse, *info, *A, (bool)check_errors);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_linalg_inv_out(tensor *out__, tensor out, tensor A) {
-  PROTECT(
-    auto outputs__ = torch::linalg_inv_out(*out, *A);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_ldl_factor(tensor *out__, tensor self, int hermitian) {
-  PROTECT(
-    auto outputs__ = torch::linalg_ldl_factor(*self, (bool)hermitian);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_linalg_ldl_factor_ex(tensor *out__, tensor self, int hermitian, int check_errors) {
-  PROTECT(
-    auto outputs__ = torch::linalg_ldl_factor_ex(*self, (bool)hermitian, (bool)check_errors);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg_linalg_ldl_factor_ex_out(tensor *out__, tensor LD, tensor pivots, tensor info, tensor self, int hermitian, int check_errors) {
-  PROTECT(
-    auto outputs__ = torch::linalg_ldl_factor_ex_out(*LD, *pivots, *info, *self, (bool)hermitian, (bool)check_errors);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg_linalg_ldl_factor_out(tensor *out__, tensor LD, tensor pivots, tensor self, int hermitian) {
-  PROTECT(
-    auto outputs__ = torch::linalg_ldl_factor_out(*LD, *pivots, *self, (bool)hermitian);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_linalg_ldl_solve(tensor *out__, tensor LD, tensor pivots, tensor B, int hermitian) {
-  PROTECT(
-    auto outputs__ = torch::linalg_ldl_solve(*LD, *pivots, *B, (bool)hermitian);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_ldl_solve_out(tensor *out__, tensor out, tensor LD, tensor pivots, tensor B, int hermitian) {
-  PROTECT(
-    auto outputs__ = torch::linalg_ldl_solve_out(*out, *LD, *pivots, *B, (bool)hermitian);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_lstsq(tensor *out__, tensor self, tensor b, double rcond_v, int rcond_null, char * driver) {
-  PROTECT(
-    auto outputs__ = torch::linalg_lstsq(*self, *b, rcond_null ? c10::nullopt : c10::optional<double>(rcond_v), std::string(driver));
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-    out__[3] = new torch::Tensor(std::get<3>(outputs__));
-  )
-}
-
-void atg_linalg_lstsq_out(tensor *out__, tensor solution, tensor residuals, tensor rank, tensor singular_values, tensor self, tensor b, double rcond_v, int rcond_null, char * driver) {
-  PROTECT(
-    auto outputs__ = torch::linalg_lstsq_out(*solution, *residuals, *rank, *singular_values, *self, *b, rcond_null ? c10::nullopt : c10::optional<double>(rcond_v), std::string(driver));
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-    out__[3] = new torch::Tensor(std::get<3>(outputs__));
-  )
-}
-
-void atg_linalg_lu(tensor *out__, tensor A, int pivot) {
-  PROTECT(
-    auto outputs__ = torch::linalg_lu(*A, (bool)pivot);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg_linalg_lu_factor(tensor *out__, tensor A, int pivot) {
-  PROTECT(
-    auto outputs__ = torch::linalg_lu_factor(*A, (bool)pivot);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_linalg_lu_factor_ex(tensor *out__, tensor A, int pivot, int check_errors) {
-  PROTECT(
-    auto outputs__ = torch::linalg_lu_factor_ex(*A, (bool)pivot, (bool)check_errors);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg_linalg_lu_factor_ex_out(tensor *out__, tensor LU, tensor pivots, tensor info, tensor A, int pivot, int check_errors) {
-  PROTECT(
-    auto outputs__ = torch::linalg_lu_factor_ex_out(*LU, *pivots, *info, *A, (bool)pivot, (bool)check_errors);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg_linalg_lu_factor_out(tensor *out__, tensor LU, tensor pivots, tensor A, int pivot) {
-  PROTECT(
-    auto outputs__ = torch::linalg_lu_factor_out(*LU, *pivots, *A, (bool)pivot);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_linalg_lu_out(tensor *out__, tensor P, tensor L, tensor U, tensor A, int pivot) {
-  PROTECT(
-    auto outputs__ = torch::linalg_lu_out(*P, *L, *U, *A, (bool)pivot);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg_linalg_lu_solve(tensor *out__, tensor LU, tensor pivots, tensor B, int left, int adjoint) {
-  PROTECT(
-    auto outputs__ = torch::linalg_lu_solve(*LU, *pivots, *B, (bool)left, (bool)adjoint);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_lu_solve_out(tensor *out__, tensor out, tensor LU, tensor pivots, tensor B, int left, int adjoint) {
-  PROTECT(
-    auto outputs__ = torch::linalg_lu_solve_out(*out, *LU, *pivots, *B, (bool)left, (bool)adjoint);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_matmul(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::linalg_matmul(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_matmul_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::linalg_matmul_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_matrix_exp(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::linalg_matrix_exp(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_matrix_exp_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::linalg_matrix_exp_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_matrix_power(tensor *out__, tensor self, int64_t n) {
-  PROTECT(
-    auto outputs__ = torch::linalg_matrix_power(*self, n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_matrix_power_out(tensor *out__, tensor out, tensor self, int64_t n) {
-  PROTECT(
-    auto outputs__ = torch::linalg_matrix_power_out(*out, *self, n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_matrix_rank(tensor *out__, tensor self, double tol, int hermitian) {
-  PROTECT(
-    auto outputs__ = torch::linalg_matrix_rank(*self, tol, (bool)hermitian);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_matrix_rank_atol_rtol_float(tensor *out__, tensor self, double atol_v, int atol_null, double rtol_v, int rtol_null, int hermitian) {
-  PROTECT(
-    auto outputs__ = torch::linalg_matrix_rank(*self, atol_null ? c10::nullopt : c10::optional<double>(atol_v), rtol_null ? c10::nullopt : c10::optional<double>(rtol_v), (bool)hermitian);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_matrix_rank_atol_rtol_float_out(tensor *out__, tensor out, tensor self, double atol_v, int atol_null, double rtol_v, int rtol_null, int hermitian) {
-  PROTECT(
-    auto outputs__ = torch::linalg_matrix_rank_out(*out, *self, atol_null ? c10::nullopt : c10::optional<double>(atol_v), rtol_null ? c10::nullopt : c10::optional<double>(rtol_v), (bool)hermitian);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_matrix_rank_atol_rtol_tensor(tensor *out__, tensor input, tensor atol, tensor rtol, int hermitian) {
-  PROTECT(
-    auto outputs__ = torch::linalg_matrix_rank(*input, (atol ? *atol : torch::Tensor()), (rtol ? *rtol : torch::Tensor()), (bool)hermitian);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_matrix_rank_atol_rtol_tensor_out(tensor *out__, tensor out, tensor input, tensor atol, tensor rtol, int hermitian) {
-  PROTECT(
-    auto outputs__ = torch::linalg_matrix_rank_out(*out, *input, (atol ? *atol : torch::Tensor()), (rtol ? *rtol : torch::Tensor()), (bool)hermitian);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_matrix_rank_out(tensor *out__, tensor out, tensor self, double tol, int hermitian) {
-  PROTECT(
-    auto outputs__ = torch::linalg_matrix_rank_out(*out, *self, tol, (bool)hermitian);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_matrix_rank_out_tol_tensor(tensor *out__, tensor out, tensor input, tensor tol, int hermitian) {
-  PROTECT(
-    auto outputs__ = torch::linalg_matrix_rank_out(*out, *input, *tol, (bool)hermitian);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_matrix_rank_tol_tensor(tensor *out__, tensor input, tensor tol, int hermitian) {
-  PROTECT(
-    auto outputs__ = torch::linalg_matrix_rank(*input, *tol, (bool)hermitian);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_multi_dot(tensor *out__, tensor *tensors_data, int tensors_len) {
-  PROTECT(
-    auto outputs__ = torch::linalg_multi_dot(of_carray_tensor(tensors_data, tensors_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_multi_dot_out(tensor *out__, tensor out, tensor *tensors_data, int tensors_len) {
-  PROTECT(
-    auto outputs__ = torch::linalg_multi_dot_out(*out, of_carray_tensor(tensors_data, tensors_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_pinv(tensor *out__, tensor self, double rcond, int hermitian) {
-  PROTECT(
-    auto outputs__ = torch::linalg_pinv(*self, rcond, (bool)hermitian);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_pinv_atol_rtol_float(tensor *out__, tensor self, double atol_v, int atol_null, double rtol_v, int rtol_null, int hermitian) {
-  PROTECT(
-    auto outputs__ = torch::linalg_pinv(*self, atol_null ? c10::nullopt : c10::optional<double>(atol_v), rtol_null ? c10::nullopt : c10::optional<double>(rtol_v), (bool)hermitian);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_pinv_atol_rtol_float_out(tensor *out__, tensor out, tensor self, double atol_v, int atol_null, double rtol_v, int rtol_null, int hermitian) {
-  PROTECT(
-    auto outputs__ = torch::linalg_pinv_out(*out, *self, atol_null ? c10::nullopt : c10::optional<double>(atol_v), rtol_null ? c10::nullopt : c10::optional<double>(rtol_v), (bool)hermitian);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_pinv_atol_rtol_tensor(tensor *out__, tensor self, tensor atol, tensor rtol, int hermitian) {
-  PROTECT(
-    auto outputs__ = torch::linalg_pinv(*self, (atol ? *atol : torch::Tensor()), (rtol ? *rtol : torch::Tensor()), (bool)hermitian);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_pinv_atol_rtol_tensor_out(tensor *out__, tensor out, tensor self, tensor atol, tensor rtol, int hermitian) {
-  PROTECT(
-    auto outputs__ = torch::linalg_pinv_out(*out, *self, (atol ? *atol : torch::Tensor()), (rtol ? *rtol : torch::Tensor()), (bool)hermitian);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_pinv_out(tensor *out__, tensor out, tensor self, double rcond, int hermitian) {
-  PROTECT(
-    auto outputs__ = torch::linalg_pinv_out(*out, *self, rcond, (bool)hermitian);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_pinv_out_rcond_tensor(tensor *out__, tensor out, tensor self, tensor rcond, int hermitian) {
-  PROTECT(
-    auto outputs__ = torch::linalg_pinv_out(*out, *self, *rcond, (bool)hermitian);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_pinv_rcond_tensor(tensor *out__, tensor self, tensor rcond, int hermitian) {
-  PROTECT(
-    auto outputs__ = torch::linalg_pinv(*self, *rcond, (bool)hermitian);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_qr(tensor *out__, tensor A, char * mode) {
-  PROTECT(
-    auto outputs__ = torch::linalg_qr(*A, std::string(mode));
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_linalg_qr_out(tensor *out__, tensor Q, tensor R, tensor A, char * mode) {
-  PROTECT(
-    auto outputs__ = torch::linalg_qr_out(*Q, *R, *A, std::string(mode));
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_linalg_slogdet(tensor *out__, tensor A) {
-  PROTECT(
-    auto outputs__ = torch::linalg_slogdet(*A);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_linalg_slogdet_out(tensor *out__, tensor sign, tensor logabsdet, tensor A) {
-  PROTECT(
-    auto outputs__ = torch::linalg_slogdet_out(*sign, *logabsdet, *A);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_linalg_solve(tensor *out__, tensor A, tensor B, int left) {
-  PROTECT(
-    auto outputs__ = torch::linalg_solve(*A, *B, (bool)left);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_solve_ex(tensor *out__, tensor A, tensor B, int left, int check_errors) {
-  PROTECT(
-    auto outputs__ = torch::linalg_solve_ex(*A, *B, (bool)left, (bool)check_errors);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_linalg_solve_ex_out(tensor *out__, tensor result, tensor info, tensor A, tensor B, int left, int check_errors) {
-  PROTECT(
-    auto outputs__ = torch::linalg_solve_ex_out(*result, *info, *A, *B, (bool)left, (bool)check_errors);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_linalg_solve_out(tensor *out__, tensor out, tensor A, tensor B, int left) {
-  PROTECT(
-    auto outputs__ = torch::linalg_solve_out(*out, *A, *B, (bool)left);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_solve_triangular(tensor *out__, tensor self, tensor B, int upper, int left, int unitriangular) {
-  PROTECT(
-    auto outputs__ = torch::linalg_solve_triangular(*self, *B, (bool)upper, (bool)left, (bool)unitriangular);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_solve_triangular_out(tensor *out__, tensor out, tensor self, tensor B, int upper, int left, int unitriangular) {
-  PROTECT(
-    auto outputs__ = torch::linalg_solve_triangular_out(*out, *self, *B, (bool)upper, (bool)left, (bool)unitriangular);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_svd(tensor *out__, tensor A, int full_matrices, char * driver) {
-  PROTECT(
-    auto outputs__ = torch::linalg_svd(*A, (bool)full_matrices, std::string(driver));
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg_linalg_svd_u(tensor *out__, tensor U, tensor S, tensor Vh, tensor A, int full_matrices, char * driver) {
-  PROTECT(
-    auto outputs__ = torch::linalg_svd_out(*U, *S, *Vh, *A, (bool)full_matrices, std::string(driver));
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg_linalg_svdvals(tensor *out__, tensor A, char * driver) {
-  PROTECT(
-    auto outputs__ = torch::linalg_svdvals(*A, std::string(driver));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_svdvals_out(tensor *out__, tensor out, tensor A, char * driver) {
-  PROTECT(
-    auto outputs__ = torch::linalg_svdvals_out(*out, *A, std::string(driver));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_tensorinv(tensor *out__, tensor self, int64_t ind) {
-  PROTECT(
-    auto outputs__ = torch::linalg_tensorinv(*self, ind);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_tensorinv_out(tensor *out__, tensor out, tensor self, int64_t ind) {
-  PROTECT(
-    auto outputs__ = torch::linalg_tensorinv_out(*out, *self, ind);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_tensorsolve(tensor *out__, tensor self, tensor other, int64_t *dims_data, int dims_len) {
-  PROTECT(
-    auto outputs__ = torch::linalg_tensorsolve(*self, *other, dims_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dims_data, dims_len)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_tensorsolve_out(tensor *out__, tensor out, tensor self, tensor other, int64_t *dims_data, int dims_len) {
-  PROTECT(
-    auto outputs__ = torch::linalg_tensorsolve_out(*out, *self, *other, dims_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dims_data, dims_len)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_vander(tensor *out__, tensor x, int64_t n_v, int n_null) {
-  PROTECT(
-    auto outputs__ = torch::linalg_vander(*x, n_null ? c10::nullopt : c10::optional<int64_t>(n_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_vecdot(tensor *out__, tensor x, tensor y, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::linalg_vecdot(*x, *y, dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linalg_vecdot_out(tensor *out__, tensor out, tensor x, tensor y, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::linalg_vecdot_out(*out, *x, *y, dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linear(tensor *out__, tensor input, tensor weight, tensor bias) {
-  PROTECT(
-    auto outputs__ = torch::linear(*input, *weight, (bias ? *bias : torch::Tensor()));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linear_out(tensor *out__, tensor out, tensor input, tensor weight, tensor bias) {
-  PROTECT(
-    auto outputs__ = torch::linear_out(*out, *input, *weight, (bias ? *bias : torch::Tensor()));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linspace(tensor *out__, scalar start, scalar end, int64_t steps, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::linspace(*start, *end, steps, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_linspace_out(tensor *out__, tensor out, scalar start, scalar end, int64_t steps) {
-  PROTECT(
-    auto outputs__ = torch::linspace_out(*out, *start, *end, steps);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_log(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::log(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_log10(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::log10(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_log10_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::log10_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_log10_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::log10_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_log1p(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::log1p(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_log1p_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::log1p_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_log1p_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::log1p_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_log2(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::log2(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_log2_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::log2_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_log2_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::log2_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_log_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::log_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_log_normal(tensor *out__, tensor self, double mean, double std) {
-  PROTECT(
-    auto outputs__ = torch::log_normal(*self, mean, std);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_log_normal_(tensor *out__, tensor self, double mean, double std) {
-  PROTECT(
-    auto outputs__ = self->log_normal_(mean, std);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_log_normal_out(tensor *out__, tensor out, tensor self, double mean, double std) {
-  PROTECT(
-    auto outputs__ = torch::log_normal_out(*out, *self, mean, std);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_log_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::log_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_log_sigmoid(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::log_sigmoid(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_log_sigmoid_backward(tensor *out__, tensor grad_output, tensor self, tensor buffer) {
-  PROTECT(
-    auto outputs__ = torch::log_sigmoid_backward(*grad_output, *self, *buffer);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_log_sigmoid_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, tensor self, tensor buffer) {
-  PROTECT(
-    auto outputs__ = torch::log_sigmoid_backward_out(*grad_input, *grad_output, *self, *buffer);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_log_sigmoid_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::log_sigmoid_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_log_softmax(tensor *out__, tensor self, int64_t dim, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::log_softmax(*self, dim, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_log_softmax_int_out(tensor *out__, tensor out, tensor self, int64_t dim, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::log_softmax_out(*out, *self, dim, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_logaddexp(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::logaddexp(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_logaddexp2(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::logaddexp2(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_logaddexp2_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::logaddexp2_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_logaddexp_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::logaddexp_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_logcumsumexp(tensor *out__, tensor self, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::logcumsumexp(*self, dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_logcumsumexp_out(tensor *out__, tensor out, tensor self, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::logcumsumexp_out(*out, *self, dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_logdet(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::logdet(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_logical_and(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::logical_and(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_logical_and_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->logical_and_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_logical_and_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::logical_and_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_logical_not(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::logical_not(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_logical_not_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = self->logical_not_();
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_logical_not_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::logical_not_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_logical_or(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::logical_or(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_logical_or_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->logical_or_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_logical_or_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::logical_or_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_logical_xor(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::logical_xor(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_logical_xor_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->logical_xor_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_logical_xor_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::logical_xor_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_logit(tensor *out__, tensor self, double eps_v, int eps_null) {
-  PROTECT(
-    auto outputs__ = torch::logit(*self, eps_null ? c10::nullopt : c10::optional<double>(eps_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_logit_(tensor *out__, tensor self, double eps_v, int eps_null) {
-  PROTECT(
-    auto outputs__ = torch::logit_(*self, eps_null ? c10::nullopt : c10::optional<double>(eps_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_logit_backward(tensor *out__, tensor grad_output, tensor self, double eps_v, int eps_null) {
-  PROTECT(
-    auto outputs__ = torch::logit_backward(*grad_output, *self, eps_null ? c10::nullopt : c10::optional<double>(eps_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_logit_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, tensor self, double eps_v, int eps_null) {
-  PROTECT(
-    auto outputs__ = torch::logit_backward_out(*grad_input, *grad_output, *self, eps_null ? c10::nullopt : c10::optional<double>(eps_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_logit_out(tensor *out__, tensor out, tensor self, double eps_v, int eps_null) {
-  PROTECT(
-    auto outputs__ = torch::logit_out(*out, *self, eps_null ? c10::nullopt : c10::optional<double>(eps_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_logspace(tensor *out__, scalar start, scalar end, int64_t steps, double base, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::logspace(*start, *end, steps, base, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_logspace_out(tensor *out__, tensor out, scalar start, scalar end, int64_t steps, double base) {
-  PROTECT(
-    auto outputs__ = torch::logspace_out(*out, *start, *end, steps, base);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_logsumexp(tensor *out__, tensor self, int64_t *dim_data, int dim_len, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::logsumexp(*self, torch::IntArrayRef(dim_data, dim_len), (bool)keepdim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_logsumexp_out(tensor *out__, tensor out, tensor self, int64_t *dim_data, int dim_len, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::logsumexp_out(*out, *self, torch::IntArrayRef(dim_data, dim_len), (bool)keepdim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_lstm(tensor *out__, tensor input, tensor *hx_data, int hx_len, tensor *params_data, int params_len, int has_biases, int64_t num_layers, double dropout, int train, int bidirectional, int batch_first) {
-  PROTECT(
-    auto outputs__ = torch::lstm(*input, of_carray_tensor(hx_data, hx_len), of_carray_tensor(params_data, params_len), (bool)has_biases, num_layers, dropout, (bool)train, (bool)bidirectional, (bool)batch_first);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg_lstm_cell(tensor *out__, tensor input, tensor *hx_data, int hx_len, tensor w_ih, tensor w_hh, tensor b_ih, tensor b_hh) {
-  PROTECT(
-    auto outputs__ = torch::lstm_cell(*input, of_carray_tensor(hx_data, hx_len), *w_ih, *w_hh, (b_ih ? *b_ih : torch::Tensor()), (b_hh ? *b_hh : torch::Tensor()));
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_lstm_data(tensor *out__, tensor data, tensor batch_sizes, tensor *hx_data, int hx_len, tensor *params_data, int params_len, int has_biases, int64_t num_layers, double dropout, int train, int bidirectional) {
-  PROTECT(
-    auto outputs__ = torch::lstm(*data, *batch_sizes, of_carray_tensor(hx_data, hx_len), of_carray_tensor(params_data, params_len), (bool)has_biases, num_layers, dropout, (bool)train, (bool)bidirectional);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg_lstm_mps_backward(tensor out0, tensor *out1_data, int out1_len, tensor *out2_data, int out2_len, tensor grad_y, tensor grad_hy, tensor grad_cy, tensor z_state, tensor cell_state_fwd, tensor input, tensor layersOutputs, tensor *hx_data, int hx_len, tensor *params_data, int params_len, int has_biases, int64_t num_layers, double dropout, int train, int bidirectional, int batch_first) {
-  PROTECT(
-    torch::lstm_mps_backward_out(*out0, of_carray_tensor(out1_data, out1_len), of_carray_tensor(out2_data, out2_len), (grad_y ? *grad_y : torch::Tensor()), (grad_hy ? *grad_hy : torch::Tensor()), (grad_cy ? *grad_cy : torch::Tensor()), *z_state, *cell_state_fwd, *input, *layersOutputs, of_carray_tensor(hx_data, hx_len), of_carray_tensor(params_data, params_len), (bool)has_biases, num_layers, dropout, (bool)train, (bool)bidirectional, (bool)batch_first);
-  )
-}
-
-void atg_lt(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::lt(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_lt_(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = self->lt_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_lt_scalar_out(tensor *out__, tensor out, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::lt_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_lt_tensor(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::lt(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_lt_tensor_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->lt_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_lt_tensor_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::lt_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_lu_solve(tensor *out__, tensor self, tensor LU_data, tensor LU_pivots) {
-  PROTECT(
-    auto outputs__ = torch::lu_solve(*self, *LU_data, *LU_pivots);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_lu_solve_out(tensor *out__, tensor out, tensor self, tensor LU_data, tensor LU_pivots) {
-  PROTECT(
-    auto outputs__ = torch::lu_solve_out(*out, *self, *LU_data, *LU_pivots);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_lu_unpack(tensor *out__, tensor LU_data, tensor LU_pivots, int unpack_data, int unpack_pivots) {
-  PROTECT(
-    auto outputs__ = torch::lu_unpack(*LU_data, *LU_pivots, (bool)unpack_data, (bool)unpack_pivots);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg_lu_unpack_out(tensor *out__, tensor P, tensor L, tensor U, tensor LU_data, tensor LU_pivots, int unpack_data, int unpack_pivots) {
-  PROTECT(
-    auto outputs__ = torch::lu_unpack_out(*P, *L, *U, *LU_data, *LU_pivots, (bool)unpack_data, (bool)unpack_pivots);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg_margin_ranking_loss(tensor *out__, tensor input1, tensor input2, tensor target, double margin, int64_t reduction) {
-  PROTECT(
-    auto outputs__ = torch::margin_ranking_loss(*input1, *input2, *target, margin, reduction);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_masked_fill(tensor *out__, tensor self, tensor mask, scalar value) {
-  PROTECT(
-    auto outputs__ = torch::masked_fill(*self, *mask, *value);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_masked_fill_(tensor *out__, tensor self, tensor mask, scalar value) {
-  PROTECT(
-    auto outputs__ = self->masked_fill_(*mask, *value);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_masked_fill_scalar_out(tensor *out__, tensor out, tensor self, tensor mask, scalar value) {
-  PROTECT(
-    auto outputs__ = torch::masked_fill_out(*out, *self, *mask, *value);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_masked_fill_tensor(tensor *out__, tensor self, tensor mask, tensor value) {
-  PROTECT(
-    auto outputs__ = torch::masked_fill(*self, *mask, *value);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_masked_fill_tensor_(tensor *out__, tensor self, tensor mask, tensor value) {
-  PROTECT(
-    auto outputs__ = self->masked_fill_(*mask, *value);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_masked_fill_tensor_out(tensor *out__, tensor out, tensor self, tensor mask, tensor value) {
-  PROTECT(
-    auto outputs__ = torch::masked_fill_out(*out, *self, *mask, *value);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_masked_scatter(tensor *out__, tensor self, tensor mask, tensor source) {
-  PROTECT(
-    auto outputs__ = torch::masked_scatter(*self, *mask, *source);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_masked_scatter_(tensor *out__, tensor self, tensor mask, tensor source) {
-  PROTECT(
-    auto outputs__ = self->masked_scatter_(*mask, *source);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_masked_scatter_out(tensor *out__, tensor out, tensor self, tensor mask, tensor source) {
-  PROTECT(
-    auto outputs__ = torch::masked_scatter_out(*out, *self, *mask, *source);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_masked_select(tensor *out__, tensor self, tensor mask) {
-  PROTECT(
-    auto outputs__ = torch::masked_select(*self, *mask);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_masked_select_backward(tensor *out__, tensor grad, tensor input, tensor mask) {
-  PROTECT(
-    auto outputs__ = torch::masked_select_backward(*grad, *input, *mask);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_masked_select_out(tensor *out__, tensor out, tensor self, tensor mask) {
-  PROTECT(
-    auto outputs__ = torch::masked_select_out(*out, *self, *mask);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_matmul(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::matmul(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_matmul_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::matmul_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_matrix_exp(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::matrix_exp(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_matrix_exp_backward(tensor *out__, tensor self, tensor grad) {
-  PROTECT(
-    auto outputs__ = torch::matrix_exp_backward(*self, *grad);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_matrix_h(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = self->matrix_H();
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_matrix_power(tensor *out__, tensor self, int64_t n) {
-  PROTECT(
-    auto outputs__ = torch::matrix_power(*self, n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_matrix_power_out(tensor *out__, tensor out, tensor self, int64_t n) {
-  PROTECT(
-    auto outputs__ = torch::matrix_power_out(*out, *self, n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_max(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::max(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_max_dim(tensor *out__, tensor self, int64_t dim, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::max(*self, dim, (bool)keepdim);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_max_dim_max(tensor *out__, tensor max, tensor max_values, tensor self, int64_t dim, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::max_out(*max, *max_values, *self, dim, (bool)keepdim);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_max_other(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::max(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_max_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::max_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_max_pool1d(tensor *out__, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
-  PROTECT(
-    auto outputs__ = torch::max_pool1d(*self, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_max_pool1d_with_indices(tensor *out__, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
-  PROTECT(
-    auto outputs__ = torch::max_pool1d_with_indices(*self, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_max_pool2d(tensor *out__, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
-  PROTECT(
-    auto outputs__ = torch::max_pool2d(*self, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_max_pool2d_backward(tensor *out__, tensor grad_output, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
-  PROTECT(
-    auto outputs__ = torch::max_pool2d_backward(*grad_output, *self, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_max_pool2d_backward_out(tensor *out__, tensor out, tensor grad_output, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
-  PROTECT(
-    auto outputs__ = torch::max_pool2d_backward_out(*out, *grad_output, *self, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_max_pool2d_with_indices(tensor *out__, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
-  PROTECT(
-    auto outputs__ = torch::max_pool2d_with_indices(*self, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_max_pool2d_with_indices_backward(tensor *out__, tensor grad_output, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode, tensor indices) {
-  PROTECT(
-    auto outputs__ = torch::max_pool2d_with_indices_backward(*grad_output, *self, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode, *indices);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_max_pool2d_with_indices_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode, tensor indices) {
-  PROTECT(
-    auto outputs__ = torch::max_pool2d_with_indices_backward_out(*grad_input, *grad_output, *self, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode, *indices);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_max_pool2d_with_indices_out(tensor *out__, tensor out, tensor indices, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
-  PROTECT(
-    auto outputs__ = torch::max_pool2d_with_indices_out(*out, *indices, *self, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_max_pool3d(tensor *out__, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
-  PROTECT(
-    auto outputs__ = torch::max_pool3d(*self, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_max_pool3d_with_indices(tensor *out__, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
-  PROTECT(
-    auto outputs__ = torch::max_pool3d_with_indices(*self, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_max_pool3d_with_indices_backward(tensor *out__, tensor grad_output, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode, tensor indices) {
-  PROTECT(
-    auto outputs__ = torch::max_pool3d_with_indices_backward(*grad_output, *self, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode, *indices);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_max_pool3d_with_indices_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode, tensor indices) {
-  PROTECT(
-    auto outputs__ = torch::max_pool3d_with_indices_backward_out(*grad_input, *grad_output, *self, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode, *indices);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_max_pool3d_with_indices_out(tensor *out__, tensor out, tensor indices, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
-  PROTECT(
-    auto outputs__ = torch::max_pool3d_with_indices_out(*out, *indices, *self, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_max_unary_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::max_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_max_unpool2d(tensor *out__, tensor self, tensor indices, int64_t *output_size_data, int output_size_len) {
-  PROTECT(
-    auto outputs__ = torch::max_unpool2d(*self, *indices, torch::IntArrayRef(output_size_data, output_size_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_max_unpool2d_out(tensor *out__, tensor out, tensor self, tensor indices, int64_t *output_size_data, int output_size_len) {
-  PROTECT(
-    auto outputs__ = torch::max_unpool2d_out(*out, *self, *indices, torch::IntArrayRef(output_size_data, output_size_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_max_unpool3d(tensor *out__, tensor self, tensor indices, int64_t *output_size_data, int output_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len) {
-  PROTECT(
-    auto outputs__ = torch::max_unpool3d(*self, *indices, torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_max_unpool3d_out(tensor *out__, tensor out, tensor self, tensor indices, int64_t *output_size_data, int output_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len) {
-  PROTECT(
-    auto outputs__ = torch::max_unpool3d_out(*out, *self, *indices, torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_maximum(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::maximum(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_maximum_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::maximum_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mean(tensor *out__, tensor self, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::mean(*self, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mean_dim(tensor *out__, tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::mean(*self, dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), (bool)keepdim, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mean_out(tensor *out__, tensor out, tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::mean_out(*out, *self, dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), (bool)keepdim, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_median(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::median(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_median_dim(tensor *out__, tensor self, int64_t dim, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::median(*self, dim, (bool)keepdim);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_median_dim_values(tensor *out__, tensor values, tensor indices, tensor self, int64_t dim, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::median_out(*values, *indices, *self, dim, (bool)keepdim);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_median_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::median_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-tensor *atg_meshgrid(tensor *tensors_data, int tensors_len) {
-  PROTECT(
-    auto outputs__ = torch::meshgrid(of_carray_tensor(tensors_data, tensors_len));
-    int sz = outputs__.size();
-    torch::Tensor **out__ = (torch::Tensor**)malloc((sz + 1) * sizeof(torch::Tensor*));
-    for (int i = 0; i < sz; ++i)
-      out__[i] = new torch::Tensor(outputs__[i]);
-    out__[sz] = nullptr;
-    return out__;
-  )
-}
-
-tensor *atg_meshgrid_indexing(tensor *tensors_data, int tensors_len, char * indexing) {
-  PROTECT(
-    auto outputs__ = torch::meshgrid(of_carray_tensor(tensors_data, tensors_len), std::string(indexing));
-    int sz = outputs__.size();
-    torch::Tensor **out__ = (torch::Tensor**)malloc((sz + 1) * sizeof(torch::Tensor*));
-    for (int i = 0; i < sz; ++i)
-      out__[i] = new torch::Tensor(outputs__[i]);
-    out__[sz] = nullptr;
-    return out__;
-  )
-}
-
-void atg_mh(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = self->mH();
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_min(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::min(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_min_dim(tensor *out__, tensor self, int64_t dim, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::min(*self, dim, (bool)keepdim);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_min_dim_min(tensor *out__, tensor min, tensor min_indices, tensor self, int64_t dim, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::min_out(*min, *min_indices, *self, dim, (bool)keepdim);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_min_other(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::min(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_min_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::min_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_min_unary_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::min_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_minimum(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::minimum(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_minimum_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::minimum_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_miopen_batch_norm(tensor *out__, tensor input, tensor weight, tensor bias, tensor running_mean, tensor running_var, int training, double exponential_average_factor, double epsilon) {
-  PROTECT(
-    auto outputs__ = torch::miopen_batch_norm(*input, *weight, (bias ? *bias : torch::Tensor()), (running_mean ? *running_mean : torch::Tensor()), (running_var ? *running_var : torch::Tensor()), (bool)training, exponential_average_factor, epsilon);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg_miopen_batch_norm_backward(tensor *out__, tensor input, tensor grad_output, tensor weight, tensor running_mean, tensor running_var, tensor save_mean, tensor save_var, double epsilon) {
-  PROTECT(
-    auto outputs__ = torch::miopen_batch_norm_backward(*input, *grad_output, *weight, (running_mean ? *running_mean : torch::Tensor()), (running_var ? *running_var : torch::Tensor()), (save_mean ? *save_mean : torch::Tensor()), (save_var ? *save_var : torch::Tensor()), epsilon);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg_miopen_batch_norm_backward_out(tensor *out__, tensor out0, tensor out1, tensor out2, tensor input, tensor grad_output, tensor weight, tensor running_mean, tensor running_var, tensor save_mean, tensor save_var, double epsilon) {
-  PROTECT(
-    auto outputs__ = torch::miopen_batch_norm_backward_out(*out0, *out1, *out2, *input, *grad_output, *weight, (running_mean ? *running_mean : torch::Tensor()), (running_var ? *running_var : torch::Tensor()), (save_mean ? *save_mean : torch::Tensor()), (save_var ? *save_var : torch::Tensor()), epsilon);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg_miopen_batch_norm_out(tensor *out__, tensor out0, tensor out1, tensor out2, tensor input, tensor weight, tensor bias, tensor running_mean, tensor running_var, int training, double exponential_average_factor, double epsilon) {
-  PROTECT(
-    auto outputs__ = torch::miopen_batch_norm_out(*out0, *out1, *out2, *input, *weight, (bias ? *bias : torch::Tensor()), (running_mean ? *running_mean : torch::Tensor()), (running_var ? *running_var : torch::Tensor()), (bool)training, exponential_average_factor, epsilon);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg_miopen_convolution(tensor *out__, tensor self, tensor weight, tensor bias, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int benchmark, int deterministic) {
-  PROTECT(
-    auto outputs__ = torch::miopen_convolution(*self, *weight, (bias ? *bias : torch::Tensor()), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(dilation_data, dilation_len), groups, (bool)benchmark, (bool)deterministic);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_miopen_convolution_add_relu(tensor *out__, tensor self, tensor weight, tensor z, scalar alpha, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int64_t groups) {
-  PROTECT(
-    auto outputs__ = torch::miopen_convolution_add_relu(*self, *weight, *z, *alpha, (bias ? *bias : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), groups);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_miopen_convolution_out(tensor *out__, tensor out, tensor self, tensor weight, tensor bias, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int benchmark, int deterministic) {
-  PROTECT(
-    auto outputs__ = torch::miopen_convolution_out(*out, *self, *weight, (bias ? *bias : torch::Tensor()), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(dilation_data, dilation_len), groups, (bool)benchmark, (bool)deterministic);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_miopen_convolution_relu(tensor *out__, tensor self, tensor weight, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int64_t groups) {
-  PROTECT(
-    auto outputs__ = torch::miopen_convolution_relu(*self, *weight, (bias ? *bias : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), groups);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_miopen_convolution_transpose(tensor *out__, tensor self, tensor weight, tensor bias, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int benchmark, int deterministic) {
-  PROTECT(
-    auto outputs__ = torch::miopen_convolution_transpose(*self, *weight, (bias ? *bias : torch::Tensor()), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(output_padding_data, output_padding_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(dilation_data, dilation_len), groups, (bool)benchmark, (bool)deterministic);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_miopen_convolution_transpose_out(tensor *out__, tensor out, tensor self, tensor weight, tensor bias, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int benchmark, int deterministic) {
-  PROTECT(
-    auto outputs__ = torch::miopen_convolution_transpose_out(*out, *self, *weight, (bias ? *bias : torch::Tensor()), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(output_padding_data, output_padding_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(dilation_data, dilation_len), groups, (bool)benchmark, (bool)deterministic);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_miopen_depthwise_convolution(tensor *out__, tensor self, tensor weight, tensor bias, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int benchmark, int deterministic) {
-  PROTECT(
-    auto outputs__ = torch::miopen_depthwise_convolution(*self, *weight, (bias ? *bias : torch::Tensor()), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(dilation_data, dilation_len), groups, (bool)benchmark, (bool)deterministic);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_miopen_depthwise_convolution_out(tensor *out__, tensor out, tensor self, tensor weight, tensor bias, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int benchmark, int deterministic) {
-  PROTECT(
-    auto outputs__ = torch::miopen_depthwise_convolution_out(*out, *self, *weight, (bias ? *bias : torch::Tensor()), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(dilation_data, dilation_len), groups, (bool)benchmark, (bool)deterministic);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_miopen_rnn(tensor *out__, tensor input, tensor *weight_data, int weight_len, int64_t weight_stride0, tensor hx, tensor cx, int64_t mode, int64_t hidden_size, int64_t num_layers, int batch_first, double dropout, int train, int bidirectional, int64_t *batch_sizes_data, int batch_sizes_len, tensor dropout_state) {
-  PROTECT(
-    auto outputs__ = torch::miopen_rnn(*input, of_carray_tensor(weight_data, weight_len), weight_stride0, *hx, (cx ? *cx : torch::Tensor()), mode, hidden_size, num_layers, (bool)batch_first, dropout, (bool)train, (bool)bidirectional, torch::IntArrayRef(batch_sizes_data, batch_sizes_len), (dropout_state ? *dropout_state : torch::Tensor()));
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-    out__[3] = new torch::Tensor(std::get<3>(outputs__));
-    out__[4] = new torch::Tensor(std::get<4>(outputs__));
-  )
-}
-
-void atg_miopen_rnn_out(tensor *out__, tensor out0, tensor out1, tensor out2, tensor out3, tensor out4, tensor input, tensor *weight_data, int weight_len, int64_t weight_stride0, tensor hx, tensor cx, int64_t mode, int64_t hidden_size, int64_t num_layers, int batch_first, double dropout, int train, int bidirectional, int64_t *batch_sizes_data, int batch_sizes_len, tensor dropout_state) {
-  PROTECT(
-    auto outputs__ = torch::miopen_rnn_out(*out0, *out1, *out2, *out3, *out4, *input, of_carray_tensor(weight_data, weight_len), weight_stride0, *hx, (cx ? *cx : torch::Tensor()), mode, hidden_size, num_layers, (bool)batch_first, dropout, (bool)train, (bool)bidirectional, torch::IntArrayRef(batch_sizes_data, batch_sizes_len), (dropout_state ? *dropout_state : torch::Tensor()));
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-    out__[3] = new torch::Tensor(std::get<3>(outputs__));
-    out__[4] = new torch::Tensor(std::get<4>(outputs__));
-  )
-}
-
-void atg_mish(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::mish(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mish_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::mish_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mish_backward(tensor *out__, tensor grad_output, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::mish_backward(*grad_output, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mish_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::mish_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mkldnn_adaptive_avg_pool2d(tensor *out__, tensor self, int64_t *output_size_data, int output_size_len) {
-  PROTECT(
-    auto outputs__ = torch::mkldnn_adaptive_avg_pool2d(*self, torch::IntArrayRef(output_size_data, output_size_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mkldnn_adaptive_avg_pool2d_backward(tensor *out__, tensor grad_output, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::mkldnn_adaptive_avg_pool2d_backward(*grad_output, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mkldnn_adaptive_avg_pool2d_backward_out(tensor *out__, tensor out, tensor grad_output, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::mkldnn_adaptive_avg_pool2d_backward_out(*out, *grad_output, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mkldnn_adaptive_avg_pool2d_out(tensor *out__, tensor out, tensor self, int64_t *output_size_data, int output_size_len) {
-  PROTECT(
-    auto outputs__ = torch::mkldnn_adaptive_avg_pool2d_out(*out, *self, torch::IntArrayRef(output_size_data, output_size_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mkldnn_convolution(tensor *out__, tensor self, tensor weight, tensor bias, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups) {
-  PROTECT(
-    auto outputs__ = torch::mkldnn_convolution(*self, *weight, (bias ? *bias : torch::Tensor()), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(dilation_data, dilation_len), groups);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mkldnn_convolution_out(tensor *out__, tensor out, tensor self, tensor weight, tensor bias, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups) {
-  PROTECT(
-    auto outputs__ = torch::mkldnn_convolution_out(*out, *self, *weight, (bias ? *bias : torch::Tensor()), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(dilation_data, dilation_len), groups);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mkldnn_linear(tensor *out__, tensor self, tensor weight, tensor bias) {
-  PROTECT(
-    auto outputs__ = torch::mkldnn_linear(*self, *weight, (bias ? *bias : torch::Tensor()));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mkldnn_linear_backward_input(tensor *out__, int64_t *input_size_data, int input_size_len, tensor grad_output, tensor weight) {
-  PROTECT(
-    auto outputs__ = torch::mkldnn_linear_backward_input(torch::IntArrayRef(input_size_data, input_size_len), *grad_output, *weight);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mkldnn_linear_backward_input_out(tensor *out__, tensor out, int64_t *input_size_data, int input_size_len, tensor grad_output, tensor weight) {
-  PROTECT(
-    auto outputs__ = torch::mkldnn_linear_backward_input_out(*out, torch::IntArrayRef(input_size_data, input_size_len), *grad_output, *weight);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mkldnn_linear_backward_weights(tensor *out__, tensor grad_output, tensor input, tensor weight, int bias_defined) {
-  PROTECT(
-    auto outputs__ = torch::mkldnn_linear_backward_weights(*grad_output, *input, *weight, (bool)bias_defined);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_mkldnn_linear_backward_weights_out(tensor *out__, tensor out0, tensor out1, tensor grad_output, tensor input, tensor weight, int bias_defined) {
-  PROTECT(
-    auto outputs__ = torch::mkldnn_linear_backward_weights_out(*out0, *out1, *grad_output, *input, *weight, (bool)bias_defined);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_mkldnn_linear_out(tensor *out__, tensor out, tensor self, tensor weight, tensor bias) {
-  PROTECT(
-    auto outputs__ = torch::mkldnn_linear_out(*out, *self, *weight, (bias ? *bias : torch::Tensor()));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mkldnn_max_pool2d(tensor *out__, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
-  PROTECT(
-    auto outputs__ = torch::mkldnn_max_pool2d(*self, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mkldnn_max_pool2d_backward(tensor *out__, tensor grad_output, tensor output, tensor input, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
-  PROTECT(
-    auto outputs__ = torch::mkldnn_max_pool2d_backward(*grad_output, *output, *input, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mkldnn_max_pool2d_backward_out(tensor *out__, tensor out, tensor grad_output, tensor output, tensor input, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
-  PROTECT(
-    auto outputs__ = torch::mkldnn_max_pool2d_backward_out(*out, *grad_output, *output, *input, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mkldnn_max_pool2d_out(tensor *out__, tensor out, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
-  PROTECT(
-    auto outputs__ = torch::mkldnn_max_pool2d_out(*out, *self, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mkldnn_max_pool3d(tensor *out__, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
-  PROTECT(
-    auto outputs__ = torch::mkldnn_max_pool3d(*self, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mkldnn_max_pool3d_backward(tensor *out__, tensor grad_output, tensor output, tensor input, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
-  PROTECT(
-    auto outputs__ = torch::mkldnn_max_pool3d_backward(*grad_output, *output, *input, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mkldnn_max_pool3d_backward_out(tensor *out__, tensor out, tensor grad_output, tensor output, tensor input, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
-  PROTECT(
-    auto outputs__ = torch::mkldnn_max_pool3d_backward_out(*out, *grad_output, *output, *input, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mkldnn_max_pool3d_out(tensor *out__, tensor out, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
-  PROTECT(
-    auto outputs__ = torch::mkldnn_max_pool3d_out(*out, *self, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mkldnn_reorder_conv2d_weight(tensor *out__, tensor self, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int64_t *input_size_data, int input_size_len) {
-  PROTECT(
-    auto outputs__ = torch::mkldnn_reorder_conv2d_weight(*self, torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(dilation_data, dilation_len), groups, input_size_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(input_size_data, input_size_len)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mkldnn_reorder_conv2d_weight_out(tensor *out__, tensor out, tensor self, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int64_t *input_size_data, int input_size_len) {
-  PROTECT(
-    auto outputs__ = torch::mkldnn_reorder_conv2d_weight_out(*out, *self, torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(dilation_data, dilation_len), groups, input_size_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(input_size_data, input_size_len)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mkldnn_reorder_conv3d_weight(tensor *out__, tensor self, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups) {
-  PROTECT(
-    auto outputs__ = torch::mkldnn_reorder_conv3d_weight(*self, torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(dilation_data, dilation_len), groups);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mkldnn_reorder_conv3d_weight_out(tensor *out__, tensor out, tensor self, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups) {
-  PROTECT(
-    auto outputs__ = torch::mkldnn_reorder_conv3d_weight_out(*out, *self, torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(dilation_data, dilation_len), groups);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mkldnn_rnn_layer(tensor *out__, tensor input, tensor weight0, tensor weight1, tensor weight2, tensor weight3, tensor hx_, tensor cx_, int reverse, int64_t *batch_sizes_data, int batch_sizes_len, int64_t mode, int64_t hidden_size, int64_t num_layers, int has_biases, int bidirectional, int batch_first, int train) {
-  PROTECT(
-    auto outputs__ = torch::mkldnn_rnn_layer(*input, *weight0, *weight1, *weight2, *weight3, *hx_, *cx_, (bool)reverse, torch::IntArrayRef(batch_sizes_data, batch_sizes_len), mode, hidden_size, num_layers, (bool)has_biases, (bool)bidirectional, (bool)batch_first, (bool)train);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-    out__[3] = new torch::Tensor(std::get<3>(outputs__));
-  )
-}
-
-void atg_mkldnn_rnn_layer_backward(tensor *out__, tensor input, tensor weight1, tensor weight2, tensor weight3, tensor weight4, tensor hx_, tensor cx_tmp, tensor output, tensor hy_, tensor cy_, tensor grad_output, tensor grad_hy, tensor grad_cy, int reverse, int64_t mode, int64_t hidden_size, int64_t num_layers, int has_biases, int train, int bidirectional, int64_t *batch_sizes_data, int batch_sizes_len, int batch_first, tensor workspace) {
-  PROTECT(
-    auto outputs__ = torch::mkldnn_rnn_layer_backward(*input, *weight1, *weight2, *weight3, *weight4, *hx_, *cx_tmp, *output, *hy_, *cy_, (grad_output ? *grad_output : torch::Tensor()), (grad_hy ? *grad_hy : torch::Tensor()), (grad_cy ? *grad_cy : torch::Tensor()), (bool)reverse, mode, hidden_size, num_layers, (bool)has_biases, (bool)train, (bool)bidirectional, torch::IntArrayRef(batch_sizes_data, batch_sizes_len), (bool)batch_first, *workspace);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-    out__[3] = new torch::Tensor(std::get<3>(outputs__));
-    out__[4] = new torch::Tensor(std::get<4>(outputs__));
-    out__[5] = new torch::Tensor(std::get<5>(outputs__));
-    out__[6] = new torch::Tensor(std::get<6>(outputs__));
-  )
-}
-
-void atg_mkldnn_rnn_layer_backward_out(tensor *out__, tensor out0, tensor out1, tensor out2, tensor out3, tensor out4, tensor out5, tensor out6, tensor input, tensor weight1, tensor weight2, tensor weight3, tensor weight4, tensor hx_, tensor cx_tmp, tensor output, tensor hy_, tensor cy_, tensor grad_output, tensor grad_hy, tensor grad_cy, int reverse, int64_t mode, int64_t hidden_size, int64_t num_layers, int has_biases, int train, int bidirectional, int64_t *batch_sizes_data, int batch_sizes_len, int batch_first, tensor workspace) {
-  PROTECT(
-    auto outputs__ = torch::mkldnn_rnn_layer_backward_out(*out0, *out1, *out2, *out3, *out4, *out5, *out6, *input, *weight1, *weight2, *weight3, *weight4, *hx_, *cx_tmp, *output, *hy_, *cy_, (grad_output ? *grad_output : torch::Tensor()), (grad_hy ? *grad_hy : torch::Tensor()), (grad_cy ? *grad_cy : torch::Tensor()), (bool)reverse, mode, hidden_size, num_layers, (bool)has_biases, (bool)train, (bool)bidirectional, torch::IntArrayRef(batch_sizes_data, batch_sizes_len), (bool)batch_first, *workspace);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-    out__[3] = new torch::Tensor(std::get<3>(outputs__));
-    out__[4] = new torch::Tensor(std::get<4>(outputs__));
-    out__[5] = new torch::Tensor(std::get<5>(outputs__));
-    out__[6] = new torch::Tensor(std::get<6>(outputs__));
-  )
-}
-
-void atg_mkldnn_rnn_layer_out(tensor *out__, tensor out0, tensor out1, tensor out2, tensor out3, tensor input, tensor weight0, tensor weight1, tensor weight2, tensor weight3, tensor hx_, tensor cx_, int reverse, int64_t *batch_sizes_data, int batch_sizes_len, int64_t mode, int64_t hidden_size, int64_t num_layers, int has_biases, int bidirectional, int batch_first, int train) {
-  PROTECT(
-    auto outputs__ = torch::mkldnn_rnn_layer_out(*out0, *out1, *out2, *out3, *input, *weight0, *weight1, *weight2, *weight3, *hx_, *cx_, (bool)reverse, torch::IntArrayRef(batch_sizes_data, batch_sizes_len), mode, hidden_size, num_layers, (bool)has_biases, (bool)bidirectional, (bool)batch_first, (bool)train);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-    out__[3] = new torch::Tensor(std::get<3>(outputs__));
-  )
-}
-
-void atg_mm(tensor *out__, tensor self, tensor mat2) {
-  PROTECT(
-    auto outputs__ = torch::mm(*self, *mat2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mm_out(tensor *out__, tensor out, tensor self, tensor mat2) {
-  PROTECT(
-    auto outputs__ = torch::mm_out(*out, *self, *mat2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mode(tensor *out__, tensor self, int64_t dim, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::mode(*self, dim, (bool)keepdim);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_mode_values(tensor *out__, tensor values, tensor indices, tensor self, int64_t dim, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::mode_out(*values, *indices, *self, dim, (bool)keepdim);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_moveaxis(tensor *out__, tensor self, int64_t *source_data, int source_len, int64_t *destination_data, int destination_len) {
-  PROTECT(
-    auto outputs__ = torch::moveaxis(*self, torch::IntArrayRef(source_data, source_len), torch::IntArrayRef(destination_data, destination_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_moveaxis_int(tensor *out__, tensor self, int64_t source, int64_t destination) {
-  PROTECT(
-    auto outputs__ = torch::moveaxis(*self, source, destination);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_movedim(tensor *out__, tensor self, int64_t *source_data, int source_len, int64_t *destination_data, int destination_len) {
-  PROTECT(
-    auto outputs__ = torch::movedim(*self, torch::IntArrayRef(source_data, source_len), torch::IntArrayRef(destination_data, destination_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_movedim_int(tensor *out__, tensor self, int64_t source, int64_t destination) {
-  PROTECT(
-    auto outputs__ = torch::movedim(*self, source, destination);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mse_loss(tensor *out__, tensor self, tensor target, int64_t reduction) {
-  PROTECT(
-    auto outputs__ = torch::mse_loss(*self, *target, reduction);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mse_loss_backward(tensor *out__, tensor grad_output, tensor self, tensor target, int64_t reduction) {
-  PROTECT(
-    auto outputs__ = torch::mse_loss_backward(*grad_output, *self, *target, reduction);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mse_loss_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, tensor self, tensor target, int64_t reduction) {
-  PROTECT(
-    auto outputs__ = torch::mse_loss_backward_out(*grad_input, *grad_output, *self, *target, reduction);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mse_loss_out(tensor *out__, tensor out, tensor self, tensor target, int64_t reduction) {
-  PROTECT(
-    auto outputs__ = torch::mse_loss_out(*out, *self, *target, reduction);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_msort(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::msort(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_msort_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::msort_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mt(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = self->mT();
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mul(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::mul(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mul_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->mul_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mul_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::mul_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mul_scalar(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::mul(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mul_scalar_(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = self->mul_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mul_scalar_out(tensor *out__, tensor out, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::mul_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_multi_margin_loss_backward(tensor *out__, tensor grad_output, tensor self, tensor target, scalar p, scalar margin, tensor weight, int64_t reduction) {
-  PROTECT(
-    auto outputs__ = torch::multi_margin_loss_backward(*grad_output, *self, *target, *p, *margin, (weight ? *weight : torch::Tensor()), reduction);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_multi_margin_loss_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, tensor self, tensor target, scalar p, scalar margin, tensor weight, int64_t reduction) {
-  PROTECT(
-    auto outputs__ = torch::multi_margin_loss_backward_out(*grad_input, *grad_output, *self, *target, *p, *margin, (weight ? *weight : torch::Tensor()), reduction);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_multilabel_margin_loss(tensor *out__, tensor self, tensor target, int64_t reduction) {
-  PROTECT(
-    auto outputs__ = torch::multilabel_margin_loss(*self, *target, reduction);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_multilabel_margin_loss_backward(tensor *out__, tensor grad_output, tensor self, tensor target, int64_t reduction, tensor is_target) {
-  PROTECT(
-    auto outputs__ = torch::multilabel_margin_loss_backward(*grad_output, *self, *target, reduction, *is_target);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_multilabel_margin_loss_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, tensor self, tensor target, int64_t reduction, tensor is_target) {
-  PROTECT(
-    auto outputs__ = torch::multilabel_margin_loss_backward_out(*grad_input, *grad_output, *self, *target, reduction, *is_target);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_multilabel_margin_loss_out(tensor *out__, tensor out, tensor self, tensor target, int64_t reduction) {
-  PROTECT(
-    auto outputs__ = torch::multilabel_margin_loss_out(*out, *self, *target, reduction);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_multinomial(tensor *out__, tensor self, int64_t num_samples, int replacement) {
-  PROTECT(
-    auto outputs__ = torch::multinomial(*self, num_samples, (bool)replacement);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_multinomial_out(tensor *out__, tensor out, tensor self, int64_t num_samples, int replacement) {
-  PROTECT(
-    auto outputs__ = torch::multinomial_out(*out, *self, num_samples, (bool)replacement);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_multiply(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::multiply(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_multiply_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->multiply_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_multiply_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::multiply_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_multiply_scalar(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::multiply(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_multiply_scalar_(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = self->multiply_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mv(tensor *out__, tensor self, tensor vec) {
-  PROTECT(
-    auto outputs__ = torch::mv(*self, *vec);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mv_out(tensor *out__, tensor out, tensor self, tensor vec) {
-  PROTECT(
-    auto outputs__ = torch::mv_out(*out, *self, *vec);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mvlgamma(tensor *out__, tensor self, int64_t p) {
-  PROTECT(
-    auto outputs__ = torch::mvlgamma(*self, p);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mvlgamma_(tensor *out__, tensor self, int64_t p) {
-  PROTECT(
-    auto outputs__ = self->mvlgamma_(p);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_mvlgamma_out(tensor *out__, tensor out, tensor self, int64_t p) {
-  PROTECT(
-    auto outputs__ = torch::mvlgamma_out(*out, *self, p);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_nan_to_num(tensor *out__, tensor self, double nan_v, int nan_null, double posinf_v, int posinf_null, double neginf_v, int neginf_null) {
-  PROTECT(
-    auto outputs__ = torch::nan_to_num(*self, nan_null ? c10::nullopt : c10::optional<double>(nan_v), posinf_null ? c10::nullopt : c10::optional<double>(posinf_v), neginf_null ? c10::nullopt : c10::optional<double>(neginf_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_nan_to_num_(tensor *out__, tensor self, double nan_v, int nan_null, double posinf_v, int posinf_null, double neginf_v, int neginf_null) {
-  PROTECT(
-    auto outputs__ = torch::nan_to_num_(*self, nan_null ? c10::nullopt : c10::optional<double>(nan_v), posinf_null ? c10::nullopt : c10::optional<double>(posinf_v), neginf_null ? c10::nullopt : c10::optional<double>(neginf_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_nan_to_num_out(tensor *out__, tensor out, tensor self, double nan_v, int nan_null, double posinf_v, int posinf_null, double neginf_v, int neginf_null) {
-  PROTECT(
-    auto outputs__ = torch::nan_to_num_out(*out, *self, nan_null ? c10::nullopt : c10::optional<double>(nan_v), posinf_null ? c10::nullopt : c10::optional<double>(posinf_v), neginf_null ? c10::nullopt : c10::optional<double>(neginf_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_nanmean(tensor *out__, tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::nanmean(*self, dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), (bool)keepdim, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_nanmean_out(tensor *out__, tensor out, tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::nanmean_out(*out, *self, dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), (bool)keepdim, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_nanmedian(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::nanmedian(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_nanmedian_dim(tensor *out__, tensor self, int64_t dim, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::nanmedian(*self, dim, (bool)keepdim);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_nanmedian_dim_values(tensor *out__, tensor values, tensor indices, tensor self, int64_t dim, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::nanmedian_out(*values, *indices, *self, dim, (bool)keepdim);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_nanmedian_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::nanmedian_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_nanquantile(tensor *out__, tensor self, tensor q, int64_t dim_v, int dim_null, int keepdim, char * interpolation) {
-  PROTECT(
-    auto outputs__ = torch::nanquantile(*self, *q, dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v), (bool)keepdim, std::string(interpolation));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_nanquantile_out(tensor *out__, tensor out, tensor self, tensor q, int64_t dim_v, int dim_null, int keepdim, char * interpolation) {
-  PROTECT(
-    auto outputs__ = torch::nanquantile_out(*out, *self, *q, dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v), (bool)keepdim, std::string(interpolation));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_nanquantile_scalar(tensor *out__, tensor self, double q, int64_t dim_v, int dim_null, int keepdim, char * interpolation) {
-  PROTECT(
-    auto outputs__ = torch::nanquantile(*self, q, dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v), (bool)keepdim, std::string(interpolation));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_nanquantile_scalar_out(tensor *out__, tensor out, tensor self, double q, int64_t dim_v, int dim_null, int keepdim, char * interpolation) {
-  PROTECT(
-    auto outputs__ = torch::nanquantile_out(*out, *self, q, dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v), (bool)keepdim, std::string(interpolation));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_nansum(tensor *out__, tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::nansum(*self, dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), (bool)keepdim, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_nansum_out(tensor *out__, tensor out, tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::nansum_out(*out, *self, dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), (bool)keepdim, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_narrow(tensor *out__, tensor self, int64_t dim, int64_t start, int64_t length) {
-  PROTECT(
-    auto outputs__ = torch::narrow(*self, dim, start, length);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_narrow_copy(tensor *out__, tensor self, int64_t dim, int64_t start, int64_t length) {
-  PROTECT(
-    auto outputs__ = torch::narrow_copy(*self, dim, start, length);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_narrow_copy_out(tensor *out__, tensor out, tensor self, int64_t dim, int64_t start, int64_t length) {
-  PROTECT(
-    auto outputs__ = torch::narrow_copy_out(*out, *self, dim, start, length);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_narrow_tensor(tensor *out__, tensor self, int64_t dim, tensor start, int64_t length) {
-  PROTECT(
-    auto outputs__ = torch::narrow(*self, dim, *start, length);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_native_batch_norm(tensor *out__, tensor input, tensor weight, tensor bias, tensor running_mean, tensor running_var, int training, double momentum, double eps) {
-  PROTECT(
-    auto outputs__ = torch::native_batch_norm(*input, (weight ? *weight : torch::Tensor()), (bias ? *bias : torch::Tensor()), (running_mean ? *running_mean : torch::Tensor()), (running_var ? *running_var : torch::Tensor()), (bool)training, momentum, eps);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg_native_batch_norm_out(tensor *out__, tensor out, tensor save_mean, tensor save_invstd, tensor input, tensor weight, tensor bias, tensor running_mean, tensor running_var, int training, double momentum, double eps) {
-  PROTECT(
-    auto outputs__ = torch::native_batch_norm_out(*out, *save_mean, *save_invstd, *input, (weight ? *weight : torch::Tensor()), (bias ? *bias : torch::Tensor()), (running_mean ? *running_mean : torch::Tensor()), (running_var ? *running_var : torch::Tensor()), (bool)training, momentum, eps);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg_native_channel_shuffle(tensor *out__, tensor self, int64_t groups) {
-  PROTECT(
-    auto outputs__ = torch::native_channel_shuffle(*self, groups);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_native_dropout(tensor *out__, tensor input, double p, int train) {
-  PROTECT(
-    auto outputs__ = torch::native_dropout(*input, p, (bool)train);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_native_dropout_backward(tensor *out__, tensor grad_output, tensor mask, double scale) {
-  PROTECT(
-    auto outputs__ = torch::native_dropout_backward(*grad_output, *mask, scale);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_native_dropout_backward_out(tensor *out__, tensor out, tensor grad_output, tensor mask, double scale) {
-  PROTECT(
-    auto outputs__ = torch::native_dropout_backward_out(*out, *grad_output, *mask, scale);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_native_dropout_out(tensor *out__, tensor out0, tensor out1, tensor input, double p, int train) {
-  PROTECT(
-    auto outputs__ = torch::native_dropout_out(*out0, *out1, *input, p, (bool)train);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_native_group_norm(tensor *out__, tensor input, tensor weight, tensor bias, int64_t n, int64_t C, int64_t HxW, int64_t group, double eps) {
-  PROTECT(
-    auto outputs__ = torch::native_group_norm(*input, (weight ? *weight : torch::Tensor()), (bias ? *bias : torch::Tensor()), n, C, HxW, group, eps);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg_native_group_norm_out(tensor *out__, tensor out0, tensor out1, tensor out2, tensor input, tensor weight, tensor bias, int64_t n, int64_t C, int64_t HxW, int64_t group, double eps) {
-  PROTECT(
-    auto outputs__ = torch::native_group_norm_out(*out0, *out1, *out2, *input, (weight ? *weight : torch::Tensor()), (bias ? *bias : torch::Tensor()), n, C, HxW, group, eps);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg_native_layer_norm(tensor *out__, tensor input, int64_t *normalized_shape_data, int normalized_shape_len, tensor weight, tensor bias, double eps) {
-  PROTECT(
-    auto outputs__ = torch::native_layer_norm(*input, torch::IntArrayRef(normalized_shape_data, normalized_shape_len), (weight ? *weight : torch::Tensor()), (bias ? *bias : torch::Tensor()), eps);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg_native_layer_norm_out(tensor *out__, tensor out0, tensor out1, tensor out2, tensor input, int64_t *normalized_shape_data, int normalized_shape_len, tensor weight, tensor bias, double eps) {
-  PROTECT(
-    auto outputs__ = torch::native_layer_norm_out(*out0, *out1, *out2, *input, torch::IntArrayRef(normalized_shape_data, normalized_shape_len), (weight ? *weight : torch::Tensor()), (bias ? *bias : torch::Tensor()), eps);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg_native_norm(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::native_norm(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_native_norm_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::native_norm_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_native_norm_scalaropt_dim_dtype(tensor *out__, tensor self, scalar p, int64_t *dim_data, int dim_len, int keepdim, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::native_norm(*self, *p, torch::IntArrayRef(dim_data, dim_len), (bool)keepdim, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_native_norm_scalaropt_dim_dtype_out(tensor *out__, tensor out, tensor self, scalar p, int64_t *dim_data, int dim_len, int keepdim, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::native_norm_out(*out, *self, *p, torch::IntArrayRef(dim_data, dim_len), (bool)keepdim, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_ne(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::ne(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_ne_(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = self->ne_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_ne_scalar_out(tensor *out__, tensor out, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::ne_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_ne_tensor(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::ne(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_ne_tensor_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->ne_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_ne_tensor_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::ne_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_neg(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::neg(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_neg_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::neg_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_neg_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::neg_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_negative(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::negative(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_negative_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::negative_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_negative_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::negative_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_nested_to_padded_tensor(tensor *out__, tensor self, double padding, int64_t *output_size_data, int output_size_len) {
-  PROTECT(
-    auto outputs__ = torch::nested_to_padded_tensor(*self, padding, output_size_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(output_size_data, output_size_len)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_new_empty(tensor *out__, tensor self, int64_t *size_data, int size_len, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = self->new_empty(torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_new_empty_out(tensor *out__, tensor out, tensor self, int64_t *size_data, int size_len) {
-  PROTECT(
-    auto outputs__ = torch::new_empty_out(*out, *self, torch::IntArrayRef(size_data, size_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_new_empty_strided(tensor *out__, tensor self, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = self->new_empty_strided(torch::IntArrayRef(size_data, size_len), torch::IntArrayRef(stride_data, stride_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_new_empty_strided_out(tensor *out__, tensor out, tensor self, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len) {
-  PROTECT(
-    auto outputs__ = torch::new_empty_strided_out(*out, *self, torch::IntArrayRef(size_data, size_len), torch::IntArrayRef(stride_data, stride_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_new_full(tensor *out__, tensor self, int64_t *size_data, int size_len, scalar fill_value, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = self->new_full(torch::IntArrayRef(size_data, size_len), *fill_value, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_new_full_out(tensor *out__, tensor out, tensor self, int64_t *size_data, int size_len, scalar fill_value) {
-  PROTECT(
-    auto outputs__ = torch::new_full_out(*out, *self, torch::IntArrayRef(size_data, size_len), *fill_value);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_new_ones(tensor *out__, tensor self, int64_t *size_data, int size_len, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = self->new_ones(torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_new_ones_out(tensor *out__, tensor out, tensor self, int64_t *size_data, int size_len) {
-  PROTECT(
-    auto outputs__ = torch::new_ones_out(*out, *self, torch::IntArrayRef(size_data, size_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_new_zeros(tensor *out__, tensor self, int64_t *size_data, int size_len, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = self->new_zeros(torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_new_zeros_out(tensor *out__, tensor out, tensor self, int64_t *size_data, int size_len) {
-  PROTECT(
-    auto outputs__ = torch::new_zeros_out(*out, *self, torch::IntArrayRef(size_data, size_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_nextafter(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::nextafter(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_nextafter_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->nextafter_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_nextafter_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::nextafter_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_nll_loss(tensor *out__, tensor self, tensor target, tensor weight, int64_t reduction, int64_t ignore_index) {
-  PROTECT(
-    auto outputs__ = torch::nll_loss(*self, *target, (weight ? *weight : torch::Tensor()), reduction, ignore_index);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_nll_loss2d(tensor *out__, tensor self, tensor target, tensor weight, int64_t reduction, int64_t ignore_index) {
-  PROTECT(
-    auto outputs__ = torch::nll_loss2d(*self, *target, (weight ? *weight : torch::Tensor()), reduction, ignore_index);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_nll_loss2d_backward(tensor *out__, tensor grad_output, tensor self, tensor target, tensor weight, int64_t reduction, int64_t ignore_index, tensor total_weight) {
-  PROTECT(
-    auto outputs__ = torch::nll_loss2d_backward(*grad_output, *self, *target, (weight ? *weight : torch::Tensor()), reduction, ignore_index, *total_weight);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_nll_loss2d_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, tensor self, tensor target, tensor weight, int64_t reduction, int64_t ignore_index, tensor total_weight) {
-  PROTECT(
-    auto outputs__ = torch::nll_loss2d_backward_out(*grad_input, *grad_output, *self, *target, (weight ? *weight : torch::Tensor()), reduction, ignore_index, *total_weight);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_nll_loss2d_out(tensor *out__, tensor out, tensor self, tensor target, tensor weight, int64_t reduction, int64_t ignore_index) {
-  PROTECT(
-    auto outputs__ = torch::nll_loss2d_out(*out, *self, *target, (weight ? *weight : torch::Tensor()), reduction, ignore_index);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_nll_loss_backward(tensor *out__, tensor grad_output, tensor self, tensor target, tensor weight, int64_t reduction, int64_t ignore_index, tensor total_weight) {
-  PROTECT(
-    auto outputs__ = torch::nll_loss_backward(*grad_output, *self, *target, (weight ? *weight : torch::Tensor()), reduction, ignore_index, *total_weight);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_nll_loss_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, tensor self, tensor target, tensor weight, int64_t reduction, int64_t ignore_index, tensor total_weight) {
-  PROTECT(
-    auto outputs__ = torch::nll_loss_backward_out(*grad_input, *grad_output, *self, *target, (weight ? *weight : torch::Tensor()), reduction, ignore_index, *total_weight);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_nll_loss_nd(tensor *out__, tensor self, tensor target, tensor weight, int64_t reduction, int64_t ignore_index) {
-  PROTECT(
-    auto outputs__ = torch::nll_loss_nd(*self, *target, (weight ? *weight : torch::Tensor()), reduction, ignore_index);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_nll_loss_out(tensor *out__, tensor out, tensor self, tensor target, tensor weight, int64_t reduction, int64_t ignore_index) {
-  PROTECT(
-    auto outputs__ = torch::nll_loss_out(*out, *self, *target, (weight ? *weight : torch::Tensor()), reduction, ignore_index);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_nonzero(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::nonzero(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-tensor *atg_nonzero_numpy(tensor self) {
-  PROTECT(
-    auto outputs__ = torch::nonzero_numpy(*self);
-    int sz = outputs__.size();
-    torch::Tensor **out__ = (torch::Tensor**)malloc((sz + 1) * sizeof(torch::Tensor*));
-    for (int i = 0; i < sz; ++i)
-      out__[i] = new torch::Tensor(outputs__[i]);
-    out__[sz] = nullptr;
-    return out__;
-  )
-}
-
-void atg_nonzero_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::nonzero_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_nonzero_static(tensor *out__, tensor self, int64_t size, int64_t fill_value) {
-  PROTECT(
-    auto outputs__ = torch::nonzero_static(*self, size, fill_value);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_nonzero_static_out(tensor *out__, tensor out, tensor self, int64_t size, int64_t fill_value) {
-  PROTECT(
-    auto outputs__ = torch::nonzero_static_out(*out, *self, size, fill_value);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_norm(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::norm(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_norm_dtype_out(tensor *out__, tensor out, tensor self, scalar p, int64_t *dim_data, int dim_len, int keepdim, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::norm_out(*out, *self, *p, torch::IntArrayRef(dim_data, dim_len), (bool)keepdim, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_norm_except_dim(tensor *out__, tensor v, int64_t pow, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::norm_except_dim(*v, pow, dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_norm_out(tensor *out__, tensor out, tensor self, scalar p, int64_t *dim_data, int dim_len, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::norm_out(*out, *self, *p, torch::IntArrayRef(dim_data, dim_len), (bool)keepdim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_norm_scalar_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::norm_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_norm_scalaropt_dim(tensor *out__, tensor self, scalar p, int64_t *dim_data, int dim_len, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::norm(*self, *p, torch::IntArrayRef(dim_data, dim_len), (bool)keepdim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_norm_scalaropt_dim_dtype(tensor *out__, tensor self, scalar p, int64_t *dim_data, int dim_len, int keepdim, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::norm(*self, *p, torch::IntArrayRef(dim_data, dim_len), (bool)keepdim, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_norm_scalaropt_dtype(tensor *out__, tensor self, scalar p, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::norm(*self, *p, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_norm_scalaropt_dtype_out(tensor *out__, tensor out, tensor self, scalar p, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::norm_out(*out, *self, *p, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_normal_(tensor *out__, tensor self, double mean, double std) {
-  PROTECT(
-    auto outputs__ = self->normal_(mean, std);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_normal_functional(tensor *out__, tensor self, double mean, double std) {
-  PROTECT(
-    auto outputs__ = torch::normal_functional(*self, mean, std);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_not_equal(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::not_equal(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_not_equal_(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = self->not_equal_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_not_equal_scalar_out(tensor *out__, tensor out, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::not_equal_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_not_equal_tensor(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::not_equal(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_not_equal_tensor_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->not_equal_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_not_equal_tensor_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::not_equal_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_nuclear_norm(tensor *out__, tensor self, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::nuclear_norm(*self, (bool)keepdim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_nuclear_norm_dim(tensor *out__, tensor self, int64_t *dim_data, int dim_len, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::nuclear_norm(*self, torch::IntArrayRef(dim_data, dim_len), (bool)keepdim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_nuclear_norm_dim_out(tensor *out__, tensor out, tensor self, int64_t *dim_data, int dim_len, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::nuclear_norm_out(*out, *self, torch::IntArrayRef(dim_data, dim_len), (bool)keepdim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_nuclear_norm_out(tensor *out__, tensor out, tensor self, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::nuclear_norm_out(*out, *self, (bool)keepdim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_numpy_t(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = self->numpy_T();
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_one_hot(tensor *out__, tensor self, int64_t num_classes) {
-  PROTECT(
-    auto outputs__ = torch::one_hot(*self, num_classes);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_ones(tensor *out__, int64_t *size_data, int size_len, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::ones(torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_ones_like(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::ones_like(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_ones_like_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::ones_like_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_ones_out(tensor *out__, tensor out, int64_t *size_data, int size_len) {
-  PROTECT(
-    auto outputs__ = torch::ones_out(*out, torch::IntArrayRef(size_data, size_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_orgqr(tensor *out__, tensor self, tensor input2) {
-  PROTECT(
-    auto outputs__ = torch::orgqr(*self, *input2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_orgqr_out(tensor *out__, tensor out, tensor self, tensor input2) {
-  PROTECT(
-    auto outputs__ = torch::orgqr_out(*out, *self, *input2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_ormqr(tensor *out__, tensor self, tensor input2, tensor input3, int left, int transpose) {
-  PROTECT(
-    auto outputs__ = torch::ormqr(*self, *input2, *input3, (bool)left, (bool)transpose);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_ormqr_out(tensor *out__, tensor out, tensor self, tensor input2, tensor input3, int left, int transpose) {
-  PROTECT(
-    auto outputs__ = torch::ormqr_out(*out, *self, *input2, *input3, (bool)left, (bool)transpose);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_outer(tensor *out__, tensor self, tensor vec2) {
-  PROTECT(
-    auto outputs__ = torch::outer(*self, *vec2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_outer_out(tensor *out__, tensor out, tensor self, tensor vec2) {
-  PROTECT(
-    auto outputs__ = torch::outer_out(*out, *self, *vec2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-int64_t atg_output_nr(tensor self) {
-  PROTECT(
-    return self->output_nr();
-  )
-  return 0;
-}
-
-void atg_pad(tensor *out__, tensor self, int64_t *pad_data, int pad_len, char * mode, double value_v, int value_null) {
-  PROTECT(
-    auto outputs__ = torch::pad(*self, torch::IntArrayRef(pad_data, pad_len), std::string(mode), value_null ? c10::nullopt : c10::optional<double>(value_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_pad_sequence(tensor *out__, tensor *sequences_data, int sequences_len, int batch_first, double padding_value) {
-  PROTECT(
-    auto outputs__ = torch::pad_sequence(of_carray_tensor(sequences_data, sequences_len), (bool)batch_first, padding_value);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_pairwise_distance(tensor *out__, tensor x1, tensor x2, double p, double eps, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::pairwise_distance(*x1, *x2, p, eps, (bool)keepdim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_pdist(tensor *out__, tensor self, double p) {
-  PROTECT(
-    auto outputs__ = torch::pdist(*self, p);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_permute(tensor *out__, tensor self, int64_t *dims_data, int dims_len) {
-  PROTECT(
-    auto outputs__ = torch::permute(*self, torch::IntArrayRef(dims_data, dims_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_permute_copy(tensor *out__, tensor self, int64_t *dims_data, int dims_len) {
-  PROTECT(
-    auto outputs__ = torch::permute_copy(*self, torch::IntArrayRef(dims_data, dims_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_permute_copy_out(tensor *out__, tensor out, tensor self, int64_t *dims_data, int dims_len) {
-  PROTECT(
-    auto outputs__ = torch::permute_copy_out(*out, *self, torch::IntArrayRef(dims_data, dims_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_pin_memory(tensor *out__, tensor self, int device) {
-  PROTECT(
-    auto outputs__ = self->pin_memory(device_of_int(device));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_pinverse(tensor *out__, tensor self, double rcond) {
-  PROTECT(
-    auto outputs__ = torch::pinverse(*self, rcond);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_pixel_shuffle(tensor *out__, tensor self, int64_t upscale_factor) {
-  PROTECT(
-    auto outputs__ = torch::pixel_shuffle(*self, upscale_factor);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_pixel_shuffle_out(tensor *out__, tensor out, tensor self, int64_t upscale_factor) {
-  PROTECT(
-    auto outputs__ = torch::pixel_shuffle_out(*out, *self, upscale_factor);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_pixel_unshuffle(tensor *out__, tensor self, int64_t downscale_factor) {
-  PROTECT(
-    auto outputs__ = torch::pixel_unshuffle(*self, downscale_factor);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_pixel_unshuffle_out(tensor *out__, tensor out, tensor self, int64_t downscale_factor) {
-  PROTECT(
-    auto outputs__ = torch::pixel_unshuffle_out(*out, *self, downscale_factor);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_poisson(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::poisson(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_poisson_nll_loss(tensor *out__, tensor input, tensor target, int log_input, int full, double eps, int64_t reduction) {
-  PROTECT(
-    auto outputs__ = torch::poisson_nll_loss(*input, *target, (bool)log_input, (bool)full, eps, reduction);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_poisson_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::poisson_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_polar(tensor *out__, tensor abs, tensor angle) {
-  PROTECT(
-    auto outputs__ = torch::polar(*abs, *angle);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_polar_out(tensor *out__, tensor out, tensor abs, tensor angle) {
-  PROTECT(
-    auto outputs__ = torch::polar_out(*out, *abs, *angle);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_polygamma(tensor *out__, int64_t n, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::polygamma(n, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_polygamma_(tensor *out__, tensor self, int64_t n) {
-  PROTECT(
-    auto outputs__ = self->polygamma_(n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_polygamma_out(tensor *out__, tensor out, int64_t n, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::polygamma_out(*out, n, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_positive(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::positive(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_pow(tensor *out__, tensor self, tensor exponent) {
-  PROTECT(
-    auto outputs__ = torch::pow(*self, *exponent);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_pow_(tensor *out__, tensor self, scalar exponent) {
-  PROTECT(
-    auto outputs__ = self->pow_(*exponent);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_pow_scalar(tensor *out__, scalar self, tensor exponent) {
-  PROTECT(
-    auto outputs__ = torch::pow(*self, *exponent);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_pow_scalar_out(tensor *out__, tensor out, scalar self, tensor exponent) {
-  PROTECT(
-    auto outputs__ = torch::pow_out(*out, *self, *exponent);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_pow_tensor_(tensor *out__, tensor self, tensor exponent) {
-  PROTECT(
-    auto outputs__ = self->pow_(*exponent);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_pow_tensor_scalar(tensor *out__, tensor self, scalar exponent) {
-  PROTECT(
-    auto outputs__ = torch::pow(*self, *exponent);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_pow_tensor_scalar_out(tensor *out__, tensor out, tensor self, scalar exponent) {
-  PROTECT(
-    auto outputs__ = torch::pow_out(*out, *self, *exponent);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_pow_tensor_tensor_out(tensor *out__, tensor out, tensor self, tensor exponent) {
-  PROTECT(
-    auto outputs__ = torch::pow_out(*out, *self, *exponent);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_prelu(tensor *out__, tensor self, tensor weight) {
-  PROTECT(
-    auto outputs__ = torch::prelu(*self, *weight);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_prod(tensor *out__, tensor self, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::prod(*self, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_prod_dim_int(tensor *out__, tensor self, int64_t dim, int keepdim, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::prod(*self, dim, (bool)keepdim, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_prod_int_out(tensor *out__, tensor out, tensor self, int64_t dim, int keepdim, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::prod_out(*out, *self, dim, (bool)keepdim, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_prod_out(tensor *out__, tensor out, tensor self, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::prod_out(*out, *self, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_put(tensor *out__, tensor self, tensor index, tensor source, int accumulate) {
-  PROTECT(
-    auto outputs__ = torch::put(*self, *index, *source, (bool)accumulate);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_put_(tensor *out__, tensor self, tensor index, tensor source, int accumulate) {
-  PROTECT(
-    auto outputs__ = self->put_(*index, *source, (bool)accumulate);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_put_out(tensor *out__, tensor out, tensor self, tensor index, tensor source, int accumulate) {
-  PROTECT(
-    auto outputs__ = torch::put_out(*out, *self, *index, *source, (bool)accumulate);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-int64_t atg_q_per_channel_axis(tensor self) {
-  PROTECT(
-    return torch::q_per_channel_axis(*self);
-  )
-  return 0;
-}
-
-void atg_q_per_channel_scales(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::q_per_channel_scales(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_q_per_channel_scales_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::q_per_channel_scales_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_q_per_channel_zero_points(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::q_per_channel_zero_points(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_q_per_channel_zero_points_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::q_per_channel_zero_points_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-double atg_q_scale(tensor self) {
-  PROTECT(
-    return torch::q_scale(*self);
-  )
-  return 0;
-}
-
-int64_t atg_q_zero_point(tensor self) {
-  PROTECT(
-    return torch::q_zero_point(*self);
-  )
-  return 0;
-}
-
-void atg_qr(tensor *out__, tensor self, int some) {
-  PROTECT(
-    auto outputs__ = torch::qr(*self, (bool)some);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_qr_q(tensor *out__, tensor Q, tensor R, tensor self, int some) {
-  PROTECT(
-    auto outputs__ = torch::qr_out(*Q, *R, *self, (bool)some);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_quantile(tensor *out__, tensor self, tensor q, int64_t dim_v, int dim_null, int keepdim, char * interpolation) {
-  PROTECT(
-    auto outputs__ = torch::quantile(*self, *q, dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v), (bool)keepdim, std::string(interpolation));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_quantile_out(tensor *out__, tensor out, tensor self, tensor q, int64_t dim_v, int dim_null, int keepdim, char * interpolation) {
-  PROTECT(
-    auto outputs__ = torch::quantile_out(*out, *self, *q, dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v), (bool)keepdim, std::string(interpolation));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_quantile_scalar(tensor *out__, tensor self, double q, int64_t dim_v, int dim_null, int keepdim, char * interpolation) {
-  PROTECT(
-    auto outputs__ = torch::quantile(*self, q, dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v), (bool)keepdim, std::string(interpolation));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_quantile_scalar_out(tensor *out__, tensor out, tensor self, double q, int64_t dim_v, int dim_null, int keepdim, char * interpolation) {
-  PROTECT(
-    auto outputs__ = torch::quantile_out(*out, *self, q, dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v), (bool)keepdim, std::string(interpolation));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_quantize_per_channel(tensor *out__, tensor self, tensor scales, tensor zero_points, int64_t axis, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::quantize_per_channel(*self, *scales, *zero_points, axis, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_quantize_per_channel_out(tensor *out__, tensor out, tensor self, tensor scales, tensor zero_points, int64_t axis, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::quantize_per_channel_out(*out, *self, *scales, *zero_points, axis, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_quantize_per_tensor(tensor *out__, tensor self, double scale, int64_t zero_point, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::quantize_per_tensor(*self, scale, zero_point, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_quantize_per_tensor_dynamic(tensor *out__, tensor self, int dtype, int reduce_range) {
-  PROTECT(
-    auto outputs__ = torch::quantize_per_tensor_dynamic(*self, torch::ScalarType(dtype), (bool)reduce_range);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_quantize_per_tensor_dynamic_out(tensor *out__, tensor out, tensor self, int dtype, int reduce_range) {
-  PROTECT(
-    auto outputs__ = torch::quantize_per_tensor_dynamic_out(*out, *self, torch::ScalarType(dtype), (bool)reduce_range);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_quantize_per_tensor_out(tensor *out__, tensor out, tensor self, double scale, int64_t zero_point, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::quantize_per_tensor_out(*out, *self, scale, zero_point, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_quantize_per_tensor_tensor_qparams(tensor *out__, tensor self, tensor scale, tensor zero_point, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::quantize_per_tensor(*self, *scale, *zero_point, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_quantize_per_tensor_tensor_qparams_out(tensor *out__, tensor out, tensor self, tensor scale, tensor zero_point, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::quantize_per_tensor_out(*out, *self, *scale, *zero_point, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-tensor *atg_quantize_per_tensor_tensors(tensor *tensors_data, int tensors_len, tensor scales, tensor zero_points, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::quantize_per_tensor(of_carray_tensor(tensors_data, tensors_len), *scales, *zero_points, torch::ScalarType(dtype));
-    int sz = outputs__.size();
-    torch::Tensor **out__ = (torch::Tensor**)malloc((sz + 1) * sizeof(torch::Tensor*));
-    for (int i = 0; i < sz; ++i)
-      out__[i] = new torch::Tensor(outputs__[i]);
-    out__[sz] = nullptr;
-    return out__;
-  )
-}
-
-void atg_quantize_per_tensor_tensors_out(tensor *out_data, int out_len, tensor *tensors_data, int tensors_len, tensor scales, tensor zero_points, int dtype) {
-  PROTECT(
-    torch::quantize_per_tensor_out(of_carray_tensor(out_data, out_len), of_carray_tensor(tensors_data, tensors_len), *scales, *zero_points, torch::ScalarType(dtype));
-  )
-}
-
-void atg_quantized_batch_norm(tensor *out__, tensor input, tensor weight, tensor bias, tensor mean, tensor var, double eps, double output_scale, int64_t output_zero_point) {
-  PROTECT(
-    auto outputs__ = torch::quantized_batch_norm(*input, (weight ? *weight : torch::Tensor()), (bias ? *bias : torch::Tensor()), *mean, *var, eps, output_scale, output_zero_point);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_quantized_batch_norm_out(tensor *out__, tensor out, tensor input, tensor weight, tensor bias, tensor mean, tensor var, double eps, double output_scale, int64_t output_zero_point) {
-  PROTECT(
-    auto outputs__ = torch::quantized_batch_norm_out(*out, *input, (weight ? *weight : torch::Tensor()), (bias ? *bias : torch::Tensor()), *mean, *var, eps, output_scale, output_zero_point);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_quantized_gru_cell(tensor *out__, tensor input, tensor hx, tensor w_ih, tensor w_hh, tensor b_ih, tensor b_hh, tensor packed_ih, tensor packed_hh, tensor col_offsets_ih, tensor col_offsets_hh, scalar scale_ih, scalar scale_hh, scalar zero_point_ih, scalar zero_point_hh) {
-  PROTECT(
-    auto outputs__ = torch::quantized_gru_cell(*input, *hx, *w_ih, *w_hh, *b_ih, *b_hh, *packed_ih, *packed_hh, *col_offsets_ih, *col_offsets_hh, *scale_ih, *scale_hh, *zero_point_ih, *zero_point_hh);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_quantized_lstm_cell(tensor *out__, tensor input, tensor *hx_data, int hx_len, tensor w_ih, tensor w_hh, tensor b_ih, tensor b_hh, tensor packed_ih, tensor packed_hh, tensor col_offsets_ih, tensor col_offsets_hh, scalar scale_ih, scalar scale_hh, scalar zero_point_ih, scalar zero_point_hh) {
-  PROTECT(
-    auto outputs__ = torch::quantized_lstm_cell(*input, of_carray_tensor(hx_data, hx_len), *w_ih, *w_hh, *b_ih, *b_hh, *packed_ih, *packed_hh, *col_offsets_ih, *col_offsets_hh, *scale_ih, *scale_hh, *zero_point_ih, *zero_point_hh);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_quantized_max_pool1d(tensor *out__, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
-  PROTECT(
-    auto outputs__ = torch::quantized_max_pool1d(*self, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_quantized_max_pool1d_out(tensor *out__, tensor out, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
-  PROTECT(
-    auto outputs__ = torch::quantized_max_pool1d_out(*out, *self, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_quantized_max_pool2d(tensor *out__, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
-  PROTECT(
-    auto outputs__ = torch::quantized_max_pool2d(*self, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_quantized_max_pool2d_out(tensor *out__, tensor out, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
-  PROTECT(
-    auto outputs__ = torch::quantized_max_pool2d_out(*out, *self, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_quantized_max_pool3d(tensor *out__, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
-  PROTECT(
-    auto outputs__ = torch::quantized_max_pool3d(*self, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_quantized_max_pool3d_out(tensor *out__, tensor out, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode) {
-  PROTECT(
-    auto outputs__ = torch::quantized_max_pool3d_out(*out, *self, torch::IntArrayRef(kernel_size_data, kernel_size_len), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len), (bool)ceil_mode);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_quantized_rnn_relu_cell(tensor *out__, tensor input, tensor hx, tensor w_ih, tensor w_hh, tensor b_ih, tensor b_hh, tensor packed_ih, tensor packed_hh, tensor col_offsets_ih, tensor col_offsets_hh, scalar scale_ih, scalar scale_hh, scalar zero_point_ih, scalar zero_point_hh) {
-  PROTECT(
-    auto outputs__ = torch::quantized_rnn_relu_cell(*input, *hx, *w_ih, *w_hh, *b_ih, *b_hh, *packed_ih, *packed_hh, *col_offsets_ih, *col_offsets_hh, *scale_ih, *scale_hh, *zero_point_ih, *zero_point_hh);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_quantized_rnn_tanh_cell(tensor *out__, tensor input, tensor hx, tensor w_ih, tensor w_hh, tensor b_ih, tensor b_hh, tensor packed_ih, tensor packed_hh, tensor col_offsets_ih, tensor col_offsets_hh, scalar scale_ih, scalar scale_hh, scalar zero_point_ih, scalar zero_point_hh) {
-  PROTECT(
-    auto outputs__ = torch::quantized_rnn_tanh_cell(*input, *hx, *w_ih, *w_hh, *b_ih, *b_hh, *packed_ih, *packed_hh, *col_offsets_ih, *col_offsets_hh, *scale_ih, *scale_hh, *zero_point_ih, *zero_point_hh);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_rad2deg(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::rad2deg(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_rad2deg_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::rad2deg_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_rad2deg_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::rad2deg_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_rand(tensor *out__, int64_t *size_data, int size_len, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::rand(torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_rand_like(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::rand_like(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_rand_like_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::rand_like_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_rand_out(tensor *out__, tensor out, int64_t *size_data, int size_len) {
-  PROTECT(
-    auto outputs__ = torch::rand_out(*out, torch::IntArrayRef(size_data, size_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_randint(tensor *out__, int64_t high, int64_t *size_data, int size_len, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::randint(high, torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_randint_like(tensor *out__, tensor self, int64_t high) {
-  PROTECT(
-    auto outputs__ = torch::randint_like(*self, high);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_randint_like_low_dtype(tensor *out__, tensor self, int64_t low, int64_t high) {
-  PROTECT(
-    auto outputs__ = torch::randint_like(*self, low, high);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_randint_like_low_dtype_out(tensor *out__, tensor out, tensor self, int64_t low, int64_t high) {
-  PROTECT(
-    auto outputs__ = torch::randint_like_out(*out, *self, low, high);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_randint_like_out(tensor *out__, tensor out, tensor self, int64_t high) {
-  PROTECT(
-    auto outputs__ = torch::randint_like_out(*out, *self, high);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_randint_low(tensor *out__, int64_t low, int64_t high, int64_t *size_data, int size_len, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::randint(low, high, torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_randint_low_out(tensor *out__, tensor out, int64_t low, int64_t high, int64_t *size_data, int size_len) {
-  PROTECT(
-    auto outputs__ = torch::randint_out(*out, low, high, torch::IntArrayRef(size_data, size_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_randint_out(tensor *out__, tensor out, int64_t high, int64_t *size_data, int size_len) {
-  PROTECT(
-    auto outputs__ = torch::randint_out(*out, high, torch::IntArrayRef(size_data, size_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_randn(tensor *out__, int64_t *size_data, int size_len, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::randn(torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_randn_like(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::randn_like(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_randn_like_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::randn_like_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_randn_out(tensor *out__, tensor out, int64_t *size_data, int size_len) {
-  PROTECT(
-    auto outputs__ = torch::randn_out(*out, torch::IntArrayRef(size_data, size_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_random(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::random(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_random_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = self->random_();
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_random_from(tensor *out__, tensor self, int64_t from, int64_t to_v, int to_null) {
-  PROTECT(
-    auto outputs__ = torch::random(*self, from, to_null ? c10::nullopt : c10::optional<int64_t>(to_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_random_from_(tensor *out__, tensor self, int64_t from, int64_t to_v, int to_null) {
-  PROTECT(
-    auto outputs__ = self->random_(from, to_null ? c10::nullopt : c10::optional<int64_t>(to_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_random_from_out(tensor *out__, tensor out, tensor self, int64_t from, int64_t to_v, int to_null) {
-  PROTECT(
-    auto outputs__ = torch::random_out(*out, *self, from, to_null ? c10::nullopt : c10::optional<int64_t>(to_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_random_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::random_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_random_to(tensor *out__, tensor self, int64_t to) {
-  PROTECT(
-    auto outputs__ = torch::random(*self, to);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_random_to_(tensor *out__, tensor self, int64_t to) {
-  PROTECT(
-    auto outputs__ = self->random_(to);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_random_to_out(tensor *out__, tensor out, tensor self, int64_t to) {
-  PROTECT(
-    auto outputs__ = torch::random_out(*out, *self, to);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_randperm(tensor *out__, int64_t n, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::randperm(n, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_randperm_out(tensor *out__, tensor out, int64_t n) {
-  PROTECT(
-    auto outputs__ = torch::randperm_out(*out, n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_range(tensor *out__, scalar start, scalar end, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::range(*start, *end, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_range_out(tensor *out__, tensor out, scalar start, scalar end) {
-  PROTECT(
-    auto outputs__ = torch::range_out(*out, *start, *end);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_range_out_(tensor *out__, tensor out, scalar start, scalar end) {
-  PROTECT(
-    auto outputs__ = torch::range_out(*out, *start, *end);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_range_step(tensor *out__, scalar start, scalar end, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::range(*start, *end, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_ravel(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::ravel(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_real(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::real(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_reciprocal(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::reciprocal(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_reciprocal_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::reciprocal_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_reciprocal_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::reciprocal_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_reflection_pad1d(tensor *out__, tensor self, int64_t *padding_data, int padding_len) {
-  PROTECT(
-    auto outputs__ = torch::reflection_pad1d(*self, torch::IntArrayRef(padding_data, padding_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_reflection_pad1d_backward(tensor *out__, tensor grad_output, tensor self, int64_t *padding_data, int padding_len) {
-  PROTECT(
-    auto outputs__ = torch::reflection_pad1d_backward(*grad_output, *self, torch::IntArrayRef(padding_data, padding_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_reflection_pad1d_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, tensor self, int64_t *padding_data, int padding_len) {
-  PROTECT(
-    auto outputs__ = torch::reflection_pad1d_backward_out(*grad_input, *grad_output, *self, torch::IntArrayRef(padding_data, padding_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_reflection_pad1d_out(tensor *out__, tensor out, tensor self, int64_t *padding_data, int padding_len) {
-  PROTECT(
-    auto outputs__ = torch::reflection_pad1d_out(*out, *self, torch::IntArrayRef(padding_data, padding_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_reflection_pad2d(tensor *out__, tensor self, int64_t *padding_data, int padding_len) {
-  PROTECT(
-    auto outputs__ = torch::reflection_pad2d(*self, torch::IntArrayRef(padding_data, padding_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_reflection_pad2d_backward(tensor *out__, tensor grad_output, tensor self, int64_t *padding_data, int padding_len) {
-  PROTECT(
-    auto outputs__ = torch::reflection_pad2d_backward(*grad_output, *self, torch::IntArrayRef(padding_data, padding_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_reflection_pad2d_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, tensor self, int64_t *padding_data, int padding_len) {
-  PROTECT(
-    auto outputs__ = torch::reflection_pad2d_backward_out(*grad_input, *grad_output, *self, torch::IntArrayRef(padding_data, padding_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_reflection_pad2d_out(tensor *out__, tensor out, tensor self, int64_t *padding_data, int padding_len) {
-  PROTECT(
-    auto outputs__ = torch::reflection_pad2d_out(*out, *self, torch::IntArrayRef(padding_data, padding_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_reflection_pad3d(tensor *out__, tensor self, int64_t *padding_data, int padding_len) {
-  PROTECT(
-    auto outputs__ = torch::reflection_pad3d(*self, torch::IntArrayRef(padding_data, padding_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_reflection_pad3d_backward(tensor *out__, tensor grad_output, tensor self, int64_t *padding_data, int padding_len) {
-  PROTECT(
-    auto outputs__ = torch::reflection_pad3d_backward(*grad_output, *self, torch::IntArrayRef(padding_data, padding_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_reflection_pad3d_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, tensor self, int64_t *padding_data, int padding_len) {
-  PROTECT(
-    auto outputs__ = torch::reflection_pad3d_backward_out(*grad_input, *grad_output, *self, torch::IntArrayRef(padding_data, padding_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_reflection_pad3d_out(tensor *out__, tensor out, tensor self, int64_t *padding_data, int padding_len) {
-  PROTECT(
-    auto outputs__ = torch::reflection_pad3d_out(*out, *self, torch::IntArrayRef(padding_data, padding_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_relu(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::relu(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_relu6(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::relu6(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_relu6_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::relu6_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_relu_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::relu_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_relu_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::relu_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_remainder(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::remainder(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_remainder_(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = self->remainder_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_remainder_scalar_out(tensor *out__, tensor out, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::remainder_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_remainder_scalar_tensor(tensor *out__, scalar self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::remainder(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_remainder_scalar_tensor_out(tensor *out__, tensor out, scalar self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::remainder_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_remainder_tensor(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::remainder(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_remainder_tensor_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->remainder_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_remainder_tensor_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::remainder_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_renorm(tensor *out__, tensor self, scalar p, int64_t dim, scalar maxnorm) {
-  PROTECT(
-    auto outputs__ = torch::renorm(*self, *p, dim, *maxnorm);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_renorm_(tensor *out__, tensor self, scalar p, int64_t dim, scalar maxnorm) {
-  PROTECT(
-    auto outputs__ = self->renorm_(*p, dim, *maxnorm);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_renorm_out(tensor *out__, tensor out, tensor self, scalar p, int64_t dim, scalar maxnorm) {
-  PROTECT(
-    auto outputs__ = torch::renorm_out(*out, *self, *p, dim, *maxnorm);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_repeat(tensor *out__, tensor self, int64_t *repeats_data, int repeats_len) {
-  PROTECT(
-    auto outputs__ = self->repeat(torch::IntArrayRef(repeats_data, repeats_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_repeat_interleave(tensor *out__, tensor repeats, int64_t output_size_v, int output_size_null) {
-  PROTECT(
-    auto outputs__ = torch::repeat_interleave(*repeats, output_size_null ? c10::nullopt : c10::optional<int64_t>(output_size_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_repeat_interleave_self_int(tensor *out__, tensor self, int64_t repeats, int64_t dim_v, int dim_null, int64_t output_size_v, int output_size_null) {
-  PROTECT(
-    auto outputs__ = torch::repeat_interleave(*self, repeats, dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v), output_size_null ? c10::nullopt : c10::optional<int64_t>(output_size_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_repeat_interleave_self_tensor(tensor *out__, tensor self, tensor repeats, int64_t dim_v, int dim_null, int64_t output_size_v, int output_size_null) {
-  PROTECT(
-    auto outputs__ = torch::repeat_interleave(*self, *repeats, dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v), output_size_null ? c10::nullopt : c10::optional<int64_t>(output_size_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_repeat_interleave_tensor_out(tensor *out__, tensor out, tensor repeats, int64_t output_size_v, int output_size_null) {
-  PROTECT(
-    auto outputs__ = torch::repeat_interleave_out(*out, *repeats, output_size_null ? c10::nullopt : c10::optional<int64_t>(output_size_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_repeat_out(tensor *out__, tensor out, tensor self, int64_t *repeats_data, int repeats_len) {
-  PROTECT(
-    auto outputs__ = torch::repeat_out(*out, *self, torch::IntArrayRef(repeats_data, repeats_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_replication_pad1d(tensor *out__, tensor self, int64_t *padding_data, int padding_len) {
-  PROTECT(
-    auto outputs__ = torch::replication_pad1d(*self, torch::IntArrayRef(padding_data, padding_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_replication_pad1d_backward(tensor *out__, tensor grad_output, tensor self, int64_t *padding_data, int padding_len) {
-  PROTECT(
-    auto outputs__ = torch::replication_pad1d_backward(*grad_output, *self, torch::IntArrayRef(padding_data, padding_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_replication_pad1d_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, tensor self, int64_t *padding_data, int padding_len) {
-  PROTECT(
-    auto outputs__ = torch::replication_pad1d_backward_out(*grad_input, *grad_output, *self, torch::IntArrayRef(padding_data, padding_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_replication_pad1d_out(tensor *out__, tensor out, tensor self, int64_t *padding_data, int padding_len) {
-  PROTECT(
-    auto outputs__ = torch::replication_pad1d_out(*out, *self, torch::IntArrayRef(padding_data, padding_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_replication_pad2d(tensor *out__, tensor self, int64_t *padding_data, int padding_len) {
-  PROTECT(
-    auto outputs__ = torch::replication_pad2d(*self, torch::IntArrayRef(padding_data, padding_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_replication_pad2d_backward(tensor *out__, tensor grad_output, tensor self, int64_t *padding_data, int padding_len) {
-  PROTECT(
-    auto outputs__ = torch::replication_pad2d_backward(*grad_output, *self, torch::IntArrayRef(padding_data, padding_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_replication_pad2d_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, tensor self, int64_t *padding_data, int padding_len) {
-  PROTECT(
-    auto outputs__ = torch::replication_pad2d_backward_out(*grad_input, *grad_output, *self, torch::IntArrayRef(padding_data, padding_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_replication_pad2d_out(tensor *out__, tensor out, tensor self, int64_t *padding_data, int padding_len) {
-  PROTECT(
-    auto outputs__ = torch::replication_pad2d_out(*out, *self, torch::IntArrayRef(padding_data, padding_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_replication_pad3d(tensor *out__, tensor self, int64_t *padding_data, int padding_len) {
-  PROTECT(
-    auto outputs__ = torch::replication_pad3d(*self, torch::IntArrayRef(padding_data, padding_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_replication_pad3d_backward(tensor *out__, tensor grad_output, tensor self, int64_t *padding_data, int padding_len) {
-  PROTECT(
-    auto outputs__ = torch::replication_pad3d_backward(*grad_output, *self, torch::IntArrayRef(padding_data, padding_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_replication_pad3d_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, tensor self, int64_t *padding_data, int padding_len) {
-  PROTECT(
-    auto outputs__ = torch::replication_pad3d_backward_out(*grad_input, *grad_output, *self, torch::IntArrayRef(padding_data, padding_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_replication_pad3d_out(tensor *out__, tensor out, tensor self, int64_t *padding_data, int padding_len) {
-  PROTECT(
-    auto outputs__ = torch::replication_pad3d_out(*out, *self, torch::IntArrayRef(padding_data, padding_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_requires_grad_(tensor *out__, tensor self, int requires_grad) {
-  PROTECT(
-    auto outputs__ = self->requires_grad_((bool)requires_grad);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_reshape(tensor *out__, tensor self, int64_t *shape_data, int shape_len) {
-  PROTECT(
-    auto outputs__ = torch::reshape(*self, torch::IntArrayRef(shape_data, shape_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_reshape_as(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->reshape_as(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_resize(tensor *out__, tensor self, int64_t *size_data, int size_len) {
-  PROTECT(
-    auto outputs__ = torch::resize(*self, torch::IntArrayRef(size_data, size_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_resize_(tensor *out__, tensor self, int64_t *size_data, int size_len) {
-  PROTECT(
-    auto outputs__ = self->resize_(torch::IntArrayRef(size_data, size_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_resize_as(tensor *out__, tensor self, tensor the_template) {
-  PROTECT(
-    auto outputs__ = torch::resize_as(*self, *the_template);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_resize_as_(tensor *out__, tensor self, tensor the_template) {
-  PROTECT(
-    auto outputs__ = torch::resize_as_(*self, *the_template);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_resize_as_out(tensor *out__, tensor out, tensor self, tensor the_template) {
-  PROTECT(
-    auto outputs__ = torch::resize_as_out(*out, *self, *the_template);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_resize_as_sparse(tensor *out__, tensor self, tensor the_template) {
-  PROTECT(
-    auto outputs__ = torch::resize_as_sparse(*self, *the_template);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_resize_as_sparse_(tensor *out__, tensor self, tensor the_template) {
-  PROTECT(
-    auto outputs__ = torch::resize_as_sparse_(*self, *the_template);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_resize_as_sparse_out(tensor *out__, tensor out, tensor self, tensor the_template) {
-  PROTECT(
-    auto outputs__ = torch::resize_as_sparse_out(*out, *self, *the_template);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_resize_out(tensor *out__, tensor out, tensor self, int64_t *size_data, int size_len) {
-  PROTECT(
-    auto outputs__ = torch::resize_out(*out, *self, torch::IntArrayRef(size_data, size_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_resolve_conj(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::resolve_conj(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_resolve_neg(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::resolve_neg(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-int atg_retains_grad(tensor self) {
-  PROTECT(
-    return self->retains_grad();
-  )
-  return 0;
-}
-
-void atg_rnn_relu(tensor *out__, tensor input, tensor hx, tensor *params_data, int params_len, int has_biases, int64_t num_layers, double dropout, int train, int bidirectional, int batch_first) {
-  PROTECT(
-    auto outputs__ = torch::rnn_relu(*input, *hx, of_carray_tensor(params_data, params_len), (bool)has_biases, num_layers, dropout, (bool)train, (bool)bidirectional, (bool)batch_first);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_rnn_relu_cell(tensor *out__, tensor input, tensor hx, tensor w_ih, tensor w_hh, tensor b_ih, tensor b_hh) {
-  PROTECT(
-    auto outputs__ = torch::rnn_relu_cell(*input, *hx, *w_ih, *w_hh, (b_ih ? *b_ih : torch::Tensor()), (b_hh ? *b_hh : torch::Tensor()));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_rnn_relu_data(tensor *out__, tensor data, tensor batch_sizes, tensor hx, tensor *params_data, int params_len, int has_biases, int64_t num_layers, double dropout, int train, int bidirectional) {
-  PROTECT(
-    auto outputs__ = torch::rnn_relu(*data, *batch_sizes, *hx, of_carray_tensor(params_data, params_len), (bool)has_biases, num_layers, dropout, (bool)train, (bool)bidirectional);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_rnn_tanh(tensor *out__, tensor input, tensor hx, tensor *params_data, int params_len, int has_biases, int64_t num_layers, double dropout, int train, int bidirectional, int batch_first) {
-  PROTECT(
-    auto outputs__ = torch::rnn_tanh(*input, *hx, of_carray_tensor(params_data, params_len), (bool)has_biases, num_layers, dropout, (bool)train, (bool)bidirectional, (bool)batch_first);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_rnn_tanh_cell(tensor *out__, tensor input, tensor hx, tensor w_ih, tensor w_hh, tensor b_ih, tensor b_hh) {
-  PROTECT(
-    auto outputs__ = torch::rnn_tanh_cell(*input, *hx, *w_ih, *w_hh, (b_ih ? *b_ih : torch::Tensor()), (b_hh ? *b_hh : torch::Tensor()));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_rnn_tanh_data(tensor *out__, tensor data, tensor batch_sizes, tensor hx, tensor *params_data, int params_len, int has_biases, int64_t num_layers, double dropout, int train, int bidirectional) {
-  PROTECT(
-    auto outputs__ = torch::rnn_tanh(*data, *batch_sizes, *hx, of_carray_tensor(params_data, params_len), (bool)has_biases, num_layers, dropout, (bool)train, (bool)bidirectional);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_roll(tensor *out__, tensor self, int64_t *shifts_data, int shifts_len, int64_t *dims_data, int dims_len) {
-  PROTECT(
-    auto outputs__ = torch::roll(*self, torch::IntArrayRef(shifts_data, shifts_len), torch::IntArrayRef(dims_data, dims_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_roll_out(tensor *out__, tensor out, tensor self, int64_t *shifts_data, int shifts_len, int64_t *dims_data, int dims_len) {
-  PROTECT(
-    auto outputs__ = torch::roll_out(*out, *self, torch::IntArrayRef(shifts_data, shifts_len), torch::IntArrayRef(dims_data, dims_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_rot90(tensor *out__, tensor self, int64_t k, int64_t *dims_data, int dims_len) {
-  PROTECT(
-    auto outputs__ = torch::rot90(*self, k, torch::IntArrayRef(dims_data, dims_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_rot90_out(tensor *out__, tensor out, tensor self, int64_t k, int64_t *dims_data, int dims_len) {
-  PROTECT(
-    auto outputs__ = torch::rot90_out(*out, *self, k, torch::IntArrayRef(dims_data, dims_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_round(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::round(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_round_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::round_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_round_decimals(tensor *out__, tensor self, int64_t decimals) {
-  PROTECT(
-    auto outputs__ = torch::round(*self, decimals);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_round_decimals_(tensor *out__, tensor self, int64_t decimals) {
-  PROTECT(
-    auto outputs__ = torch::round_(*self, decimals);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_round_decimals_out(tensor *out__, tensor out, tensor self, int64_t decimals) {
-  PROTECT(
-    auto outputs__ = torch::round_out(*out, *self, decimals);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_round_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::round_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_row_indices(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = self->row_indices();
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_row_indices_copy(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::row_indices_copy(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_row_indices_copy_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::row_indices_copy_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_row_stack(tensor *out__, tensor *tensors_data, int tensors_len) {
-  PROTECT(
-    auto outputs__ = torch::row_stack(of_carray_tensor(tensors_data, tensors_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_row_stack_out(tensor *out__, tensor out, tensor *tensors_data, int tensors_len) {
-  PROTECT(
-    auto outputs__ = torch::row_stack_out(*out, of_carray_tensor(tensors_data, tensors_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_rrelu(tensor *out__, tensor self, int training) {
-  PROTECT(
-    auto outputs__ = torch::rrelu(*self, (bool)training);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_rrelu_(tensor *out__, tensor self, int training) {
-  PROTECT(
-    auto outputs__ = torch::rrelu_(*self, (bool)training);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_rrelu_with_noise(tensor *out__, tensor self, tensor noise, int training) {
-  PROTECT(
-    auto outputs__ = torch::rrelu_with_noise(*self, *noise, (bool)training);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_rrelu_with_noise_(tensor *out__, tensor self, tensor noise, int training) {
-  PROTECT(
-    auto outputs__ = torch::rrelu_with_noise_(*self, *noise, (bool)training);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_rrelu_with_noise_backward(tensor *out__, tensor grad_output, tensor self, tensor noise, scalar lower, scalar upper, int training, int self_is_result) {
-  PROTECT(
-    auto outputs__ = torch::rrelu_with_noise_backward(*grad_output, *self, *noise, *lower, *upper, (bool)training, (bool)self_is_result);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_rrelu_with_noise_backward_out(tensor *out__, tensor out, tensor grad_output, tensor self, tensor noise, scalar lower, scalar upper, int training, int self_is_result) {
-  PROTECT(
-    auto outputs__ = torch::rrelu_with_noise_backward_out(*out, *grad_output, *self, *noise, *lower, *upper, (bool)training, (bool)self_is_result);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_rrelu_with_noise_out(tensor *out__, tensor out, tensor self, tensor noise, int training) {
-  PROTECT(
-    auto outputs__ = torch::rrelu_with_noise_out(*out, *self, *noise, (bool)training);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_rsqrt(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::rsqrt(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_rsqrt_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::rsqrt_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_rsqrt_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::rsqrt_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_rsub(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::rsub(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_rsub_scalar(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::rsub(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_rsub_scalar_out(tensor *out__, tensor out, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::rsub_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_rsub_tensor_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::rsub_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_scalar_tensor(tensor *out__, scalar s, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::scalar_tensor(*s, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_scalar_tensor_out(tensor *out__, tensor out, scalar s) {
-  PROTECT(
-    auto outputs__ = torch::scalar_tensor_out(*out, *s);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_scaled_dot_product_attention(tensor *out__, tensor query, tensor key, tensor value, tensor attn_mask, double dropout_p, int is_causal, double scale_v, int scale_null) {
-  PROTECT(
-    auto outputs__ = torch::scaled_dot_product_attention(*query, *key, *value, (attn_mask ? *attn_mask : torch::Tensor()), dropout_p, (bool)is_causal, scale_null ? c10::nullopt : c10::optional<double>(scale_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_scatter(tensor *out__, tensor self, int64_t dim, tensor index, tensor src) {
-  PROTECT(
-    auto outputs__ = torch::scatter(*self, dim, *index, *src);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_scatter_(tensor *out__, tensor self, int64_t dim, tensor index, tensor src) {
-  PROTECT(
-    auto outputs__ = self->scatter_(dim, *index, *src);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_scatter_add(tensor *out__, tensor self, int64_t dim, tensor index, tensor src) {
-  PROTECT(
-    auto outputs__ = torch::scatter_add(*self, dim, *index, *src);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_scatter_add_(tensor *out__, tensor self, int64_t dim, tensor index, tensor src) {
-  PROTECT(
-    auto outputs__ = self->scatter_add_(dim, *index, *src);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_scatter_add_out(tensor *out__, tensor out, tensor self, int64_t dim, tensor index, tensor src) {
-  PROTECT(
-    auto outputs__ = torch::scatter_add_out(*out, *self, dim, *index, *src);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_scatter_reduce(tensor *out__, tensor self, int64_t dim, tensor index, tensor src, char * reduce) {
-  PROTECT(
-    auto outputs__ = torch::scatter(*self, dim, *index, *src, std::string(reduce));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_scatter_reduce_(tensor *out__, tensor self, int64_t dim, tensor index, tensor src, char * reduce) {
-  PROTECT(
-    auto outputs__ = self->scatter_(dim, *index, *src, std::string(reduce));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_scatter_reduce_out(tensor *out__, tensor out, tensor self, int64_t dim, tensor index, tensor src, char * reduce) {
-  PROTECT(
-    auto outputs__ = torch::scatter_out(*out, *self, dim, *index, *src, std::string(reduce));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_scatter_src_out(tensor *out__, tensor out, tensor self, int64_t dim, tensor index, tensor src) {
-  PROTECT(
-    auto outputs__ = torch::scatter_out(*out, *self, dim, *index, *src);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_scatter_value(tensor *out__, tensor self, int64_t dim, tensor index, scalar value) {
-  PROTECT(
-    auto outputs__ = torch::scatter(*self, dim, *index, *value);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_scatter_value_(tensor *out__, tensor self, int64_t dim, tensor index, scalar value) {
-  PROTECT(
-    auto outputs__ = self->scatter_(dim, *index, *value);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_scatter_value_out(tensor *out__, tensor out, tensor self, int64_t dim, tensor index, scalar value) {
-  PROTECT(
-    auto outputs__ = torch::scatter_out(*out, *self, dim, *index, *value);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_scatter_value_reduce(tensor *out__, tensor self, int64_t dim, tensor index, scalar value, char * reduce) {
-  PROTECT(
-    auto outputs__ = torch::scatter(*self, dim, *index, *value, std::string(reduce));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_scatter_value_reduce_(tensor *out__, tensor self, int64_t dim, tensor index, scalar value, char * reduce) {
-  PROTECT(
-    auto outputs__ = self->scatter_(dim, *index, *value, std::string(reduce));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_scatter_value_reduce_out(tensor *out__, tensor out, tensor self, int64_t dim, tensor index, scalar value, char * reduce) {
-  PROTECT(
-    auto outputs__ = torch::scatter_out(*out, *self, dim, *index, *value, std::string(reduce));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_searchsorted(tensor *out__, tensor sorted_sequence, tensor self, int out_int32, int right, char * side, tensor sorter) {
-  PROTECT(
-    auto outputs__ = torch::searchsorted(*sorted_sequence, *self, (bool)out_int32, (bool)right, std::string(side), (sorter ? *sorter : torch::Tensor()));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_searchsorted_scalar(tensor *out__, tensor sorted_sequence, scalar self, int out_int32, int right, char * side, tensor sorter) {
-  PROTECT(
-    auto outputs__ = torch::searchsorted(*sorted_sequence, *self, (bool)out_int32, (bool)right, std::string(side), (sorter ? *sorter : torch::Tensor()));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_searchsorted_scalar_out(tensor *out__, tensor out, tensor sorted_sequence, scalar self, int out_int32, int right, char * side, tensor sorter) {
-  PROTECT(
-    auto outputs__ = torch::searchsorted_out(*out, *sorted_sequence, *self, (bool)out_int32, (bool)right, std::string(side), (sorter ? *sorter : torch::Tensor()));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_searchsorted_tensor_out(tensor *out__, tensor out, tensor sorted_sequence, tensor self, int out_int32, int right, char * side, tensor sorter) {
-  PROTECT(
-    auto outputs__ = torch::searchsorted_out(*out, *sorted_sequence, *self, (bool)out_int32, (bool)right, std::string(side), (sorter ? *sorter : torch::Tensor()));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_segment_reduce(tensor *out__, tensor data, char * reduce, tensor lengths, tensor indices, tensor offsets, int64_t axis, int unsafe, scalar initial) {
-  PROTECT(
-    auto outputs__ = torch::segment_reduce(*data, std::string(reduce), (lengths ? *lengths : torch::Tensor()), (indices ? *indices : torch::Tensor()), (offsets ? *offsets : torch::Tensor()), axis, (bool)unsafe, *initial);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_segment_reduce_out(tensor *out__, tensor out, tensor data, char * reduce, tensor lengths, tensor indices, tensor offsets, int64_t axis, int unsafe, scalar initial) {
-  PROTECT(
-    auto outputs__ = torch::segment_reduce_out(*out, *data, std::string(reduce), (lengths ? *lengths : torch::Tensor()), (indices ? *indices : torch::Tensor()), (offsets ? *offsets : torch::Tensor()), axis, (bool)unsafe, *initial);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_select(tensor *out__, tensor self, int64_t dim, int64_t index) {
-  PROTECT(
-    auto outputs__ = torch::select(*self, dim, index);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_select_backward(tensor *out__, tensor grad_output, int64_t *input_sizes_data, int input_sizes_len, int64_t dim, int64_t index) {
-  PROTECT(
-    auto outputs__ = torch::select_backward(*grad_output, torch::IntArrayRef(input_sizes_data, input_sizes_len), dim, index);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_select_backward_out(tensor *out__, tensor out, tensor grad_output, int64_t *input_sizes_data, int input_sizes_len, int64_t dim, int64_t index) {
-  PROTECT(
-    auto outputs__ = torch::select_backward_out(*out, *grad_output, torch::IntArrayRef(input_sizes_data, input_sizes_len), dim, index);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_select_copy(tensor *out__, tensor self, int64_t dim, int64_t index) {
-  PROTECT(
-    auto outputs__ = torch::select_copy(*self, dim, index);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_select_copy_int_out(tensor *out__, tensor out, tensor self, int64_t dim, int64_t index) {
-  PROTECT(
-    auto outputs__ = torch::select_copy_out(*out, *self, dim, index);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_select_scatter(tensor *out__, tensor self, tensor src, int64_t dim, int64_t index) {
-  PROTECT(
-    auto outputs__ = torch::select_scatter(*self, *src, dim, index);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_select_scatter_out(tensor *out__, tensor out, tensor self, tensor src, int64_t dim, int64_t index) {
-  PROTECT(
-    auto outputs__ = torch::select_scatter_out(*out, *self, *src, dim, index);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_selu(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::selu(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_selu_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::selu_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_set(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::set(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_set_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = self->set_();
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_set_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::set_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_set_requires_grad(tensor *out__, tensor self, int r) {
-  PROTECT(
-    auto outputs__ = self->set_requires_grad((bool)r);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_set_source_tensor(tensor *out__, tensor self, tensor source) {
-  PROTECT(
-    auto outputs__ = torch::set(*self, *source);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_set_source_tensor_(tensor *out__, tensor self, tensor source) {
-  PROTECT(
-    auto outputs__ = self->set_(*source);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_set_source_tensor_out(tensor *out__, tensor out, tensor self, tensor source) {
-  PROTECT(
-    auto outputs__ = torch::set_out(*out, *self, *source);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_set_source_tensor_storage_offset_(tensor *out__, tensor self, tensor source, int64_t storage_offset, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len) {
-  PROTECT(
-    auto outputs__ = self->set_(*source, storage_offset, torch::IntArrayRef(size_data, size_len), torch::IntArrayRef(stride_data, stride_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sgn(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::sgn(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sgn_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = self->sgn_();
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sgn_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::sgn_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sigmoid(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::sigmoid(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sigmoid_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::sigmoid_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sigmoid_backward(tensor *out__, tensor grad_output, tensor output) {
-  PROTECT(
-    auto outputs__ = torch::sigmoid_backward(*grad_output, *output);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sigmoid_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, tensor output) {
-  PROTECT(
-    auto outputs__ = torch::sigmoid_backward_out(*grad_input, *grad_output, *output);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sigmoid_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::sigmoid_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sign(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::sign(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sign_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = self->sign_();
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sign_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::sign_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_signbit(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::signbit(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_signbit_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::signbit_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_silu(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::silu(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_silu_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::silu_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_silu_backward(tensor *out__, tensor grad_output, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::silu_backward(*grad_output, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_silu_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::silu_backward_out(*grad_input, *grad_output, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_silu_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::silu_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sin(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::sin(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sin_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::sin_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sin_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::sin_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sinc(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::sinc(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sinc_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::sinc_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sinc_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::sinc_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sinh(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::sinh(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sinh_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::sinh_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sinh_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::sinh_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-int64_t atg_size(tensor self, int64_t dim) {
-  PROTECT(
-    return torch::size(*self, dim);
-  )
-  return 0;
-}
-
-void atg_slice(tensor *out__, tensor self, int64_t dim, int64_t start_v, int start_null, int64_t end_v, int end_null, int64_t step) {
-  PROTECT(
-    auto outputs__ = torch::slice(*self, dim, start_null ? c10::nullopt : c10::optional<int64_t>(start_v), end_null ? c10::nullopt : c10::optional<int64_t>(end_v), step);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_slice_backward(tensor *out__, tensor grad_output, int64_t *input_sizes_data, int input_sizes_len, int64_t dim, int64_t start, int64_t end, int64_t step) {
-  PROTECT(
-    auto outputs__ = torch::slice_backward(*grad_output, torch::IntArrayRef(input_sizes_data, input_sizes_len), dim, start, end, step);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_slice_backward_out(tensor *out__, tensor out, tensor grad_output, int64_t *input_sizes_data, int input_sizes_len, int64_t dim, int64_t start, int64_t end, int64_t step) {
-  PROTECT(
-    auto outputs__ = torch::slice_backward_out(*out, *grad_output, torch::IntArrayRef(input_sizes_data, input_sizes_len), dim, start, end, step);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_slice_copy(tensor *out__, tensor self, int64_t dim, int64_t start_v, int start_null, int64_t end_v, int end_null, int64_t step) {
-  PROTECT(
-    auto outputs__ = torch::slice_copy(*self, dim, start_null ? c10::nullopt : c10::optional<int64_t>(start_v), end_null ? c10::nullopt : c10::optional<int64_t>(end_v), step);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_slice_copy_tensor_out(tensor *out__, tensor out, tensor self, int64_t dim, int64_t start_v, int start_null, int64_t end_v, int end_null, int64_t step) {
-  PROTECT(
-    auto outputs__ = torch::slice_copy_out(*out, *self, dim, start_null ? c10::nullopt : c10::optional<int64_t>(start_v), end_null ? c10::nullopt : c10::optional<int64_t>(end_v), step);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_slice_scatter(tensor *out__, tensor self, tensor src, int64_t dim, int64_t start_v, int start_null, int64_t end_v, int end_null, int64_t step) {
-  PROTECT(
-    auto outputs__ = torch::slice_scatter(*self, *src, dim, start_null ? c10::nullopt : c10::optional<int64_t>(start_v), end_null ? c10::nullopt : c10::optional<int64_t>(end_v), step);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_slice_scatter_out(tensor *out__, tensor out, tensor self, tensor src, int64_t dim, int64_t start_v, int start_null, int64_t end_v, int end_null, int64_t step) {
-  PROTECT(
-    auto outputs__ = torch::slice_scatter_out(*out, *self, *src, dim, start_null ? c10::nullopt : c10::optional<int64_t>(start_v), end_null ? c10::nullopt : c10::optional<int64_t>(end_v), step);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_slogdet(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::slogdet(*self);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_slogdet_out(tensor *out__, tensor sign, tensor logabsdet, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::slogdet_out(*sign, *logabsdet, *self);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_slow_conv3d(tensor *out__, tensor self, tensor weight, int64_t *kernel_size_data, int kernel_size_len, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len) {
-  PROTECT(
-    auto outputs__ = torch::slow_conv3d(*self, *weight, torch::IntArrayRef(kernel_size_data, kernel_size_len), (bias ? *bias : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_slow_conv3d_out(tensor *out__, tensor out, tensor self, tensor weight, int64_t *kernel_size_data, int kernel_size_len, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len) {
-  PROTECT(
-    auto outputs__ = torch::slow_conv3d_out(*out, *self, *weight, torch::IntArrayRef(kernel_size_data, kernel_size_len), (bias ? *bias : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_slow_conv_dilated2d(tensor *out__, tensor self, tensor weight, int64_t *kernel_size_data, int kernel_size_len, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len) {
-  PROTECT(
-    auto outputs__ = torch::slow_conv_dilated2d(*self, *weight, torch::IntArrayRef(kernel_size_data, kernel_size_len), (bias ? *bias : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_slow_conv_dilated2d_out(tensor *out__, tensor out, tensor self, tensor weight, int64_t *kernel_size_data, int kernel_size_len, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len) {
-  PROTECT(
-    auto outputs__ = torch::slow_conv_dilated2d_out(*out, *self, *weight, torch::IntArrayRef(kernel_size_data, kernel_size_len), (bias ? *bias : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_slow_conv_dilated3d(tensor *out__, tensor self, tensor weight, int64_t *kernel_size_data, int kernel_size_len, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len) {
-  PROTECT(
-    auto outputs__ = torch::slow_conv_dilated3d(*self, *weight, torch::IntArrayRef(kernel_size_data, kernel_size_len), (bias ? *bias : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_slow_conv_dilated3d_out(tensor *out__, tensor out, tensor self, tensor weight, int64_t *kernel_size_data, int kernel_size_len, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len) {
-  PROTECT(
-    auto outputs__ = torch::slow_conv_dilated3d_out(*out, *self, *weight, torch::IntArrayRef(kernel_size_data, kernel_size_len), (bias ? *bias : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(dilation_data, dilation_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_slow_conv_transpose2d(tensor *out__, tensor self, tensor weight, int64_t *kernel_size_data, int kernel_size_len, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t *dilation_data, int dilation_len) {
-  PROTECT(
-    auto outputs__ = torch::slow_conv_transpose2d(*self, *weight, torch::IntArrayRef(kernel_size_data, kernel_size_len), (bias ? *bias : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(output_padding_data, output_padding_len), torch::IntArrayRef(dilation_data, dilation_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_slow_conv_transpose2d_out(tensor *out__, tensor out, tensor self, tensor weight, int64_t *kernel_size_data, int kernel_size_len, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t *dilation_data, int dilation_len) {
-  PROTECT(
-    auto outputs__ = torch::slow_conv_transpose2d_out(*out, *self, *weight, torch::IntArrayRef(kernel_size_data, kernel_size_len), (bias ? *bias : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(output_padding_data, output_padding_len), torch::IntArrayRef(dilation_data, dilation_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_slow_conv_transpose3d(tensor *out__, tensor self, tensor weight, int64_t *kernel_size_data, int kernel_size_len, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t *dilation_data, int dilation_len) {
-  PROTECT(
-    auto outputs__ = torch::slow_conv_transpose3d(*self, *weight, torch::IntArrayRef(kernel_size_data, kernel_size_len), (bias ? *bias : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(output_padding_data, output_padding_len), torch::IntArrayRef(dilation_data, dilation_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_slow_conv_transpose3d_out(tensor *out__, tensor out, tensor self, tensor weight, int64_t *kernel_size_data, int kernel_size_len, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t *dilation_data, int dilation_len) {
-  PROTECT(
-    auto outputs__ = torch::slow_conv_transpose3d_out(*out, *self, *weight, torch::IntArrayRef(kernel_size_data, kernel_size_len), (bias ? *bias : torch::Tensor()), torch::IntArrayRef(stride_data, stride_len), torch::IntArrayRef(padding_data, padding_len), torch::IntArrayRef(output_padding_data, output_padding_len), torch::IntArrayRef(dilation_data, dilation_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_smm(tensor *out__, tensor self, tensor mat2) {
-  PROTECT(
-    auto outputs__ = torch::smm(*self, *mat2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_smooth_l1_loss(tensor *out__, tensor self, tensor target, int64_t reduction, double beta) {
-  PROTECT(
-    auto outputs__ = torch::smooth_l1_loss(*self, *target, reduction, beta);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_smooth_l1_loss_backward(tensor *out__, tensor grad_output, tensor self, tensor target, int64_t reduction, double beta) {
-  PROTECT(
-    auto outputs__ = torch::smooth_l1_loss_backward(*grad_output, *self, *target, reduction, beta);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_smooth_l1_loss_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, tensor self, tensor target, int64_t reduction, double beta) {
-  PROTECT(
-    auto outputs__ = torch::smooth_l1_loss_backward_out(*grad_input, *grad_output, *self, *target, reduction, beta);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_smooth_l1_loss_out(tensor *out__, tensor out, tensor self, tensor target, int64_t reduction, double beta) {
-  PROTECT(
-    auto outputs__ = torch::smooth_l1_loss_out(*out, *self, *target, reduction, beta);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_soft_margin_loss(tensor *out__, tensor self, tensor target, int64_t reduction) {
-  PROTECT(
-    auto outputs__ = torch::soft_margin_loss(*self, *target, reduction);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_soft_margin_loss_backward(tensor *out__, tensor grad_output, tensor self, tensor target, int64_t reduction) {
-  PROTECT(
-    auto outputs__ = torch::soft_margin_loss_backward(*grad_output, *self, *target, reduction);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_soft_margin_loss_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, tensor self, tensor target, int64_t reduction) {
-  PROTECT(
-    auto outputs__ = torch::soft_margin_loss_backward_out(*grad_input, *grad_output, *self, *target, reduction);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_soft_margin_loss_out(tensor *out__, tensor out, tensor self, tensor target, int64_t reduction) {
-  PROTECT(
-    auto outputs__ = torch::soft_margin_loss_out(*out, *self, *target, reduction);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_softmax(tensor *out__, tensor self, int64_t dim, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::softmax(*self, dim, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_softmax_int_out(tensor *out__, tensor out, tensor self, int64_t dim, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::softmax_out(*out, *self, dim, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_softplus(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::softplus(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_softplus_backward(tensor *out__, tensor grad_output, tensor self, scalar beta, scalar threshold) {
-  PROTECT(
-    auto outputs__ = torch::softplus_backward(*grad_output, *self, *beta, *threshold);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_softplus_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, tensor self, scalar beta, scalar threshold) {
-  PROTECT(
-    auto outputs__ = torch::softplus_backward_out(*grad_input, *grad_output, *self, *beta, *threshold);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_softplus_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::softplus_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_softshrink(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::softshrink(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_softshrink_backward(tensor *out__, tensor grad_output, tensor self, scalar lambd) {
-  PROTECT(
-    auto outputs__ = torch::softshrink_backward(*grad_output, *self, *lambd);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_softshrink_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, tensor self, scalar lambd) {
-  PROTECT(
-    auto outputs__ = torch::softshrink_backward_out(*grad_input, *grad_output, *self, *lambd);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_softshrink_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::softshrink_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sort(tensor *out__, tensor self, int64_t dim, int descending) {
-  PROTECT(
-    auto outputs__ = torch::sort(*self, dim, (bool)descending);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_sort_stable(tensor *out__, tensor self, int stable, int64_t dim, int descending) {
-  PROTECT(
-    auto outputs__ = torch::sort(*self, (bool)stable, dim, (bool)descending);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_sort_values(tensor *out__, tensor values, tensor indices, tensor self, int64_t dim, int descending) {
-  PROTECT(
-    auto outputs__ = torch::sort_out(*values, *indices, *self, dim, (bool)descending);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_sort_values_stable(tensor *out__, tensor values, tensor indices, tensor self, int stable, int64_t dim, int descending) {
-  PROTECT(
-    auto outputs__ = torch::sort_out(*values, *indices, *self, (bool)stable, dim, (bool)descending);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_sparse_bsc_tensor(tensor *out__, tensor ccol_indices, tensor row_indices, tensor values, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::sparse_bsc_tensor(*ccol_indices, *row_indices, *values, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sparse_bsc_tensor_ccol_row_value_size(tensor *out__, tensor ccol_indices, tensor row_indices, tensor values, int64_t *size_data, int size_len, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::sparse_bsc_tensor(*ccol_indices, *row_indices, *values, torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sparse_bsr_tensor(tensor *out__, tensor crow_indices, tensor col_indices, tensor values, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::sparse_bsr_tensor(*crow_indices, *col_indices, *values, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sparse_bsr_tensor_crow_col_value_size(tensor *out__, tensor crow_indices, tensor col_indices, tensor values, int64_t *size_data, int size_len, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::sparse_bsr_tensor(*crow_indices, *col_indices, *values, torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sparse_compressed_tensor(tensor *out__, tensor compressed_indices, tensor plain_indices, tensor values, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::sparse_compressed_tensor(*compressed_indices, *plain_indices, *values, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sparse_compressed_tensor_comp_plain_value_size(tensor *out__, tensor compressed_indices, tensor plain_indices, tensor values, int64_t *size_data, int size_len, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::sparse_compressed_tensor(*compressed_indices, *plain_indices, *values, torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sparse_coo_tensor(tensor *out__, int64_t *size_data, int size_len, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::sparse_coo_tensor(torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sparse_coo_tensor_indices(tensor *out__, tensor indices, tensor values, int options_kind, int options_device, int is_coalesced) {
-  PROTECT(
-    auto outputs__ = torch::sparse_coo_tensor(*indices, *values, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)), (bool)is_coalesced);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sparse_coo_tensor_indices_size(tensor *out__, tensor indices, tensor values, int64_t *size_data, int size_len, int options_kind, int options_device, int is_coalesced) {
-  PROTECT(
-    auto outputs__ = torch::sparse_coo_tensor(*indices, *values, torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)), (bool)is_coalesced);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sparse_coo_tensor_size_out(tensor *out__, tensor out, int64_t *size_data, int size_len) {
-  PROTECT(
-    auto outputs__ = torch::sparse_coo_tensor_out(*out, torch::IntArrayRef(size_data, size_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sparse_csc_tensor(tensor *out__, tensor ccol_indices, tensor row_indices, tensor values, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::sparse_csc_tensor(*ccol_indices, *row_indices, *values, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sparse_csc_tensor_ccol_row_value_size(tensor *out__, tensor ccol_indices, tensor row_indices, tensor values, int64_t *size_data, int size_len, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::sparse_csc_tensor(*ccol_indices, *row_indices, *values, torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sparse_csr_tensor(tensor *out__, tensor crow_indices, tensor col_indices, tensor values, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::sparse_csr_tensor(*crow_indices, *col_indices, *values, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sparse_csr_tensor_crow_col_value_size(tensor *out__, tensor crow_indices, tensor col_indices, tensor values, int64_t *size_data, int size_len, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::sparse_csr_tensor(*crow_indices, *col_indices, *values, torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-int64_t atg_sparse_dim(tensor self) {
-  PROTECT(
-    return self->sparse_dim();
-  )
-  return 0;
-}
-
-void atg_sparse_mask(tensor *out__, tensor self, tensor mask) {
-  PROTECT(
-    auto outputs__ = self->sparse_mask(*mask);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sparse_mask_out(tensor *out__, tensor out, tensor self, tensor mask) {
-  PROTECT(
-    auto outputs__ = torch::sparse_mask_out(*out, *self, *mask);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sparse_resize(tensor *out__, tensor self, int64_t *size_data, int size_len, int64_t sparse_dim, int64_t dense_dim) {
-  PROTECT(
-    auto outputs__ = torch::sparse_resize(*self, torch::IntArrayRef(size_data, size_len), sparse_dim, dense_dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sparse_resize_(tensor *out__, tensor self, int64_t *size_data, int size_len, int64_t sparse_dim, int64_t dense_dim) {
-  PROTECT(
-    auto outputs__ = self->sparse_resize_(torch::IntArrayRef(size_data, size_len), sparse_dim, dense_dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sparse_resize_and_clear(tensor *out__, tensor self, int64_t *size_data, int size_len, int64_t sparse_dim, int64_t dense_dim) {
-  PROTECT(
-    auto outputs__ = torch::sparse_resize_and_clear(*self, torch::IntArrayRef(size_data, size_len), sparse_dim, dense_dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sparse_resize_and_clear_(tensor *out__, tensor self, int64_t *size_data, int size_len, int64_t sparse_dim, int64_t dense_dim) {
-  PROTECT(
-    auto outputs__ = self->sparse_resize_and_clear_(torch::IntArrayRef(size_data, size_len), sparse_dim, dense_dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sparse_resize_and_clear_out(tensor *out__, tensor out, tensor self, int64_t *size_data, int size_len, int64_t sparse_dim, int64_t dense_dim) {
-  PROTECT(
-    auto outputs__ = torch::sparse_resize_and_clear_out(*out, *self, torch::IntArrayRef(size_data, size_len), sparse_dim, dense_dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sparse_resize_out(tensor *out__, tensor out, tensor self, int64_t *size_data, int size_len, int64_t sparse_dim, int64_t dense_dim) {
-  PROTECT(
-    auto outputs__ = torch::sparse_resize_out(*out, *self, torch::IntArrayRef(size_data, size_len), sparse_dim, dense_dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sparse_sampled_addmm(tensor *out__, tensor self, tensor mat1, tensor mat2) {
-  PROTECT(
-    auto outputs__ = torch::sparse_sampled_addmm(*self, *mat1, *mat2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sparse_sampled_addmm_out(tensor *out__, tensor out, tensor self, tensor mat1, tensor mat2) {
-  PROTECT(
-    auto outputs__ = torch::sparse_sampled_addmm_out(*out, *self, *mat1, *mat2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_airy_ai(tensor *out__, tensor x) {
-  PROTECT(
-    auto outputs__ = torch::special_airy_ai(*x);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_airy_ai_out(tensor *out__, tensor out, tensor x) {
-  PROTECT(
-    auto outputs__ = torch::special_airy_ai_out(*out, *x);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_bessel_j0(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_bessel_j0(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_bessel_j0_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_bessel_j0_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_bessel_j1(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_bessel_j1(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_bessel_j1_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_bessel_j1_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_bessel_y0(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_bessel_y0(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_bessel_y0_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_bessel_y0_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_bessel_y1(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_bessel_y1(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_bessel_y1_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_bessel_y1_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_chebyshev_polynomial_t(tensor *out__, tensor x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_chebyshev_polynomial_t(*x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_chebyshev_polynomial_t_n_scalar(tensor *out__, tensor x, scalar n) {
-  PROTECT(
-    auto outputs__ = torch::special_chebyshev_polynomial_t(*x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_chebyshev_polynomial_t_n_scalar_out(tensor *out__, tensor out, tensor x, scalar n) {
-  PROTECT(
-    auto outputs__ = torch::special_chebyshev_polynomial_t_out(*out, *x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_chebyshev_polynomial_t_out(tensor *out__, tensor out, tensor x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_chebyshev_polynomial_t_out(*out, *x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_chebyshev_polynomial_t_x_scalar(tensor *out__, scalar x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_chebyshev_polynomial_t(*x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_chebyshev_polynomial_t_x_scalar_out(tensor *out__, tensor out, scalar x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_chebyshev_polynomial_t_out(*out, *x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_chebyshev_polynomial_u(tensor *out__, tensor x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_chebyshev_polynomial_u(*x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_chebyshev_polynomial_u_n_scalar(tensor *out__, tensor x, scalar n) {
-  PROTECT(
-    auto outputs__ = torch::special_chebyshev_polynomial_u(*x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_chebyshev_polynomial_u_n_scalar_out(tensor *out__, tensor out, tensor x, scalar n) {
-  PROTECT(
-    auto outputs__ = torch::special_chebyshev_polynomial_u_out(*out, *x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_chebyshev_polynomial_u_out(tensor *out__, tensor out, tensor x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_chebyshev_polynomial_u_out(*out, *x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_chebyshev_polynomial_u_x_scalar(tensor *out__, scalar x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_chebyshev_polynomial_u(*x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_chebyshev_polynomial_u_x_scalar_out(tensor *out__, tensor out, scalar x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_chebyshev_polynomial_u_out(*out, *x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_chebyshev_polynomial_v(tensor *out__, tensor x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_chebyshev_polynomial_v(*x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_chebyshev_polynomial_v_n_scalar(tensor *out__, tensor x, scalar n) {
-  PROTECT(
-    auto outputs__ = torch::special_chebyshev_polynomial_v(*x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_chebyshev_polynomial_v_n_scalar_out(tensor *out__, tensor out, tensor x, scalar n) {
-  PROTECT(
-    auto outputs__ = torch::special_chebyshev_polynomial_v_out(*out, *x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_chebyshev_polynomial_v_out(tensor *out__, tensor out, tensor x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_chebyshev_polynomial_v_out(*out, *x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_chebyshev_polynomial_v_x_scalar(tensor *out__, scalar x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_chebyshev_polynomial_v(*x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_chebyshev_polynomial_v_x_scalar_out(tensor *out__, tensor out, scalar x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_chebyshev_polynomial_v_out(*out, *x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_chebyshev_polynomial_w(tensor *out__, tensor x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_chebyshev_polynomial_w(*x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_chebyshev_polynomial_w_n_scalar(tensor *out__, tensor x, scalar n) {
-  PROTECT(
-    auto outputs__ = torch::special_chebyshev_polynomial_w(*x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_chebyshev_polynomial_w_n_scalar_out(tensor *out__, tensor out, tensor x, scalar n) {
-  PROTECT(
-    auto outputs__ = torch::special_chebyshev_polynomial_w_out(*out, *x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_chebyshev_polynomial_w_out(tensor *out__, tensor out, tensor x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_chebyshev_polynomial_w_out(*out, *x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_chebyshev_polynomial_w_x_scalar(tensor *out__, scalar x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_chebyshev_polynomial_w(*x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_chebyshev_polynomial_w_x_scalar_out(tensor *out__, tensor out, scalar x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_chebyshev_polynomial_w_out(*out, *x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_digamma(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_digamma(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_digamma_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_digamma_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_entr(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_entr(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_entr_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_entr_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_erf(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_erf(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_erf_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_erf_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_erfc(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_erfc(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_erfc_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_erfc_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_erfcx(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_erfcx(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_erfcx_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_erfcx_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_erfinv(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_erfinv(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_erfinv_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_erfinv_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_exp2(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_exp2(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_exp2_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_exp2_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_expit(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_expit(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_expit_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_expit_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_expm1(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_expm1(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_expm1_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_expm1_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_gammainc(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::special_gammainc(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_gammainc_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::special_gammainc_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_gammaincc(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::special_gammaincc(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_gammaincc_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::special_gammaincc_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_gammaln(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_gammaln(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_gammaln_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_gammaln_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_hermite_polynomial_h(tensor *out__, tensor x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_hermite_polynomial_h(*x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_hermite_polynomial_h_n_scalar(tensor *out__, tensor x, scalar n) {
-  PROTECT(
-    auto outputs__ = torch::special_hermite_polynomial_h(*x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_hermite_polynomial_h_n_scalar_out(tensor *out__, tensor out, tensor x, scalar n) {
-  PROTECT(
-    auto outputs__ = torch::special_hermite_polynomial_h_out(*out, *x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_hermite_polynomial_h_out(tensor *out__, tensor out, tensor x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_hermite_polynomial_h_out(*out, *x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_hermite_polynomial_h_x_scalar(tensor *out__, scalar x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_hermite_polynomial_h(*x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_hermite_polynomial_h_x_scalar_out(tensor *out__, tensor out, scalar x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_hermite_polynomial_h_out(*out, *x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_hermite_polynomial_he(tensor *out__, tensor x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_hermite_polynomial_he(*x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_hermite_polynomial_he_n_scalar(tensor *out__, tensor x, scalar n) {
-  PROTECT(
-    auto outputs__ = torch::special_hermite_polynomial_he(*x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_hermite_polynomial_he_n_scalar_out(tensor *out__, tensor out, tensor x, scalar n) {
-  PROTECT(
-    auto outputs__ = torch::special_hermite_polynomial_he_out(*out, *x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_hermite_polynomial_he_out(tensor *out__, tensor out, tensor x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_hermite_polynomial_he_out(*out, *x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_hermite_polynomial_he_x_scalar(tensor *out__, scalar x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_hermite_polynomial_he(*x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_hermite_polynomial_he_x_scalar_out(tensor *out__, tensor out, scalar x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_hermite_polynomial_he_out(*out, *x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_i0(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_i0(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_i0_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_i0_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_i0e(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_i0e(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_i0e_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_i0e_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_i1(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_i1(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_i1_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_i1_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_i1e(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_i1e(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_i1e_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_i1e_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_laguerre_polynomial_l(tensor *out__, tensor x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_laguerre_polynomial_l(*x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_laguerre_polynomial_l_n_scalar(tensor *out__, tensor x, scalar n) {
-  PROTECT(
-    auto outputs__ = torch::special_laguerre_polynomial_l(*x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_laguerre_polynomial_l_n_scalar_out(tensor *out__, tensor out, tensor x, scalar n) {
-  PROTECT(
-    auto outputs__ = torch::special_laguerre_polynomial_l_out(*out, *x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_laguerre_polynomial_l_out(tensor *out__, tensor out, tensor x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_laguerre_polynomial_l_out(*out, *x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_laguerre_polynomial_l_x_scalar(tensor *out__, scalar x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_laguerre_polynomial_l(*x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_laguerre_polynomial_l_x_scalar_out(tensor *out__, tensor out, scalar x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_laguerre_polynomial_l_out(*out, *x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_legendre_polynomial_p(tensor *out__, tensor x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_legendre_polynomial_p(*x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_legendre_polynomial_p_n_scalar(tensor *out__, tensor x, scalar n) {
-  PROTECT(
-    auto outputs__ = torch::special_legendre_polynomial_p(*x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_legendre_polynomial_p_n_scalar_out(tensor *out__, tensor out, tensor x, scalar n) {
-  PROTECT(
-    auto outputs__ = torch::special_legendre_polynomial_p_out(*out, *x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_legendre_polynomial_p_out(tensor *out__, tensor out, tensor x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_legendre_polynomial_p_out(*out, *x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_legendre_polynomial_p_x_scalar(tensor *out__, scalar x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_legendre_polynomial_p(*x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_legendre_polynomial_p_x_scalar_out(tensor *out__, tensor out, scalar x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_legendre_polynomial_p_out(*out, *x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_log1p(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_log1p(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_log1p_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_log1p_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_log_ndtr(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_log_ndtr(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_log_ndtr_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_log_ndtr_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_log_softmax(tensor *out__, tensor self, int64_t dim, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::special_log_softmax(*self, dim, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_logit(tensor *out__, tensor self, double eps_v, int eps_null) {
-  PROTECT(
-    auto outputs__ = torch::special_logit(*self, eps_null ? c10::nullopt : c10::optional<double>(eps_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_logit_out(tensor *out__, tensor out, tensor self, double eps_v, int eps_null) {
-  PROTECT(
-    auto outputs__ = torch::special_logit_out(*out, *self, eps_null ? c10::nullopt : c10::optional<double>(eps_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_logsumexp(tensor *out__, tensor self, int64_t *dim_data, int dim_len, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::special_logsumexp(*self, torch::IntArrayRef(dim_data, dim_len), (bool)keepdim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_logsumexp_out(tensor *out__, tensor out, tensor self, int64_t *dim_data, int dim_len, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::special_logsumexp_out(*out, *self, torch::IntArrayRef(dim_data, dim_len), (bool)keepdim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_modified_bessel_i0(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_modified_bessel_i0(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_modified_bessel_i0_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_modified_bessel_i0_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_modified_bessel_i1(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_modified_bessel_i1(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_modified_bessel_i1_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_modified_bessel_i1_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_modified_bessel_k0(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_modified_bessel_k0(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_modified_bessel_k0_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_modified_bessel_k0_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_modified_bessel_k1(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_modified_bessel_k1(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_modified_bessel_k1_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_modified_bessel_k1_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_multigammaln(tensor *out__, tensor self, int64_t p) {
-  PROTECT(
-    auto outputs__ = torch::special_multigammaln(*self, p);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_multigammaln_out(tensor *out__, tensor out, tensor self, int64_t p) {
-  PROTECT(
-    auto outputs__ = torch::special_multigammaln_out(*out, *self, p);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_ndtr(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_ndtr(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_ndtr_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_ndtr_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_ndtri(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_ndtri(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_ndtri_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_ndtri_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_polygamma(tensor *out__, int64_t n, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_polygamma(n, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_polygamma_out(tensor *out__, tensor out, int64_t n, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_polygamma_out(*out, n, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_psi(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_psi(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_psi_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_psi_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_round(tensor *out__, tensor self, int64_t decimals) {
-  PROTECT(
-    auto outputs__ = torch::special_round(*self, decimals);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_round_out(tensor *out__, tensor out, tensor self, int64_t decimals) {
-  PROTECT(
-    auto outputs__ = torch::special_round_out(*out, *self, decimals);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_scaled_modified_bessel_k0(tensor *out__, tensor x) {
-  PROTECT(
-    auto outputs__ = torch::special_scaled_modified_bessel_k0(*x);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_scaled_modified_bessel_k0_out(tensor *out__, tensor out, tensor x) {
-  PROTECT(
-    auto outputs__ = torch::special_scaled_modified_bessel_k0_out(*out, *x);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_scaled_modified_bessel_k1(tensor *out__, tensor x) {
-  PROTECT(
-    auto outputs__ = torch::special_scaled_modified_bessel_k1(*x);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_scaled_modified_bessel_k1_out(tensor *out__, tensor out, tensor x) {
-  PROTECT(
-    auto outputs__ = torch::special_scaled_modified_bessel_k1_out(*out, *x);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_shifted_chebyshev_polynomial_t(tensor *out__, tensor x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_shifted_chebyshev_polynomial_t(*x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_shifted_chebyshev_polynomial_t_n_scalar(tensor *out__, tensor x, scalar n) {
-  PROTECT(
-    auto outputs__ = torch::special_shifted_chebyshev_polynomial_t(*x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_shifted_chebyshev_polynomial_t_n_scalar_out(tensor *out__, tensor out, tensor x, scalar n) {
-  PROTECT(
-    auto outputs__ = torch::special_shifted_chebyshev_polynomial_t_out(*out, *x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_shifted_chebyshev_polynomial_t_out(tensor *out__, tensor out, tensor x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_shifted_chebyshev_polynomial_t_out(*out, *x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_shifted_chebyshev_polynomial_t_x_scalar(tensor *out__, scalar x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_shifted_chebyshev_polynomial_t(*x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_shifted_chebyshev_polynomial_t_x_scalar_out(tensor *out__, tensor out, scalar x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_shifted_chebyshev_polynomial_t_out(*out, *x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_shifted_chebyshev_polynomial_u(tensor *out__, tensor x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_shifted_chebyshev_polynomial_u(*x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_shifted_chebyshev_polynomial_u_n_scalar(tensor *out__, tensor x, scalar n) {
-  PROTECT(
-    auto outputs__ = torch::special_shifted_chebyshev_polynomial_u(*x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_shifted_chebyshev_polynomial_u_n_scalar_out(tensor *out__, tensor out, tensor x, scalar n) {
-  PROTECT(
-    auto outputs__ = torch::special_shifted_chebyshev_polynomial_u_out(*out, *x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_shifted_chebyshev_polynomial_u_out(tensor *out__, tensor out, tensor x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_shifted_chebyshev_polynomial_u_out(*out, *x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_shifted_chebyshev_polynomial_u_x_scalar(tensor *out__, scalar x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_shifted_chebyshev_polynomial_u(*x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_shifted_chebyshev_polynomial_u_x_scalar_out(tensor *out__, tensor out, scalar x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_shifted_chebyshev_polynomial_u_out(*out, *x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_shifted_chebyshev_polynomial_v(tensor *out__, tensor x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_shifted_chebyshev_polynomial_v(*x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_shifted_chebyshev_polynomial_v_n_scalar(tensor *out__, tensor x, scalar n) {
-  PROTECT(
-    auto outputs__ = torch::special_shifted_chebyshev_polynomial_v(*x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_shifted_chebyshev_polynomial_v_n_scalar_out(tensor *out__, tensor out, tensor x, scalar n) {
-  PROTECT(
-    auto outputs__ = torch::special_shifted_chebyshev_polynomial_v_out(*out, *x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_shifted_chebyshev_polynomial_v_out(tensor *out__, tensor out, tensor x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_shifted_chebyshev_polynomial_v_out(*out, *x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_shifted_chebyshev_polynomial_v_x_scalar(tensor *out__, scalar x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_shifted_chebyshev_polynomial_v(*x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_shifted_chebyshev_polynomial_v_x_scalar_out(tensor *out__, tensor out, scalar x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_shifted_chebyshev_polynomial_v_out(*out, *x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_shifted_chebyshev_polynomial_w(tensor *out__, tensor x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_shifted_chebyshev_polynomial_w(*x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_shifted_chebyshev_polynomial_w_n_scalar(tensor *out__, tensor x, scalar n) {
-  PROTECT(
-    auto outputs__ = torch::special_shifted_chebyshev_polynomial_w(*x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_shifted_chebyshev_polynomial_w_n_scalar_out(tensor *out__, tensor out, tensor x, scalar n) {
-  PROTECT(
-    auto outputs__ = torch::special_shifted_chebyshev_polynomial_w_out(*out, *x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_shifted_chebyshev_polynomial_w_out(tensor *out__, tensor out, tensor x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_shifted_chebyshev_polynomial_w_out(*out, *x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_shifted_chebyshev_polynomial_w_x_scalar(tensor *out__, scalar x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_shifted_chebyshev_polynomial_w(*x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_shifted_chebyshev_polynomial_w_x_scalar_out(tensor *out__, tensor out, scalar x, tensor n) {
-  PROTECT(
-    auto outputs__ = torch::special_shifted_chebyshev_polynomial_w_out(*out, *x, *n);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_sinc(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_sinc(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_sinc_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::special_sinc_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_softmax(tensor *out__, tensor self, int64_t dim, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::special_softmax(*self, dim, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_spherical_bessel_j0(tensor *out__, tensor x) {
-  PROTECT(
-    auto outputs__ = torch::special_spherical_bessel_j0(*x);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_spherical_bessel_j0_out(tensor *out__, tensor out, tensor x) {
-  PROTECT(
-    auto outputs__ = torch::special_spherical_bessel_j0_out(*out, *x);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_xlog1py(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::special_xlog1py(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_xlog1py_other_scalar(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::special_xlog1py(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_xlog1py_other_scalar_out(tensor *out__, tensor out, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::special_xlog1py_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_xlog1py_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::special_xlog1py_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_xlog1py_self_scalar(tensor *out__, scalar self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::special_xlog1py(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_xlog1py_self_scalar_out(tensor *out__, tensor out, scalar self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::special_xlog1py_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_xlogy(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::special_xlogy(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_xlogy_other_scalar(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::special_xlogy(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_xlogy_other_scalar_out(tensor *out__, tensor out, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::special_xlogy_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_xlogy_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::special_xlogy_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_xlogy_self_scalar(tensor *out__, scalar self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::special_xlogy(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_xlogy_self_scalar_out(tensor *out__, tensor out, scalar self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::special_xlogy_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_zeta(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::special_zeta(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_zeta_other_scalar(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::special_zeta(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_zeta_other_scalar_out(tensor *out__, tensor out, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::special_zeta_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_zeta_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::special_zeta_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_zeta_self_scalar(tensor *out__, scalar self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::special_zeta(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_special_zeta_self_scalar_out(tensor *out__, tensor out, scalar self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::special_zeta_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-tensor *atg_split(tensor self, int64_t split_size, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::split(*self, split_size, dim);
-    int sz = outputs__.size();
-    torch::Tensor **out__ = (torch::Tensor**)malloc((sz + 1) * sizeof(torch::Tensor*));
-    for (int i = 0; i < sz; ++i)
-      out__[i] = new torch::Tensor(outputs__[i]);
-    out__[sz] = nullptr;
-    return out__;
-  )
-}
-
-tensor *atg_split_copy(tensor self, int64_t split_size, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::split_copy(*self, split_size, dim);
-    int sz = outputs__.size();
-    torch::Tensor **out__ = (torch::Tensor**)malloc((sz + 1) * sizeof(torch::Tensor*));
-    for (int i = 0; i < sz; ++i)
-      out__[i] = new torch::Tensor(outputs__[i]);
-    out__[sz] = nullptr;
-    return out__;
-  )
-}
-
-void atg_split_copy_tensor_out(tensor *out_data, int out_len, tensor self, int64_t split_size, int64_t dim) {
-  PROTECT(
-    torch::split_copy_out(of_carray_tensor(out_data, out_len), *self, split_size, dim);
-  )
-}
-
-tensor *atg_split_sizes(tensor self, int64_t *split_size_data, int split_size_len, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::split(*self, torch::IntArrayRef(split_size_data, split_size_len), dim);
-    int sz = outputs__.size();
-    torch::Tensor **out__ = (torch::Tensor**)malloc((sz + 1) * sizeof(torch::Tensor*));
-    for (int i = 0; i < sz; ++i)
-      out__[i] = new torch::Tensor(outputs__[i]);
-    out__[sz] = nullptr;
-    return out__;
-  )
-}
-
-tensor *atg_split_with_sizes(tensor self, int64_t *split_sizes_data, int split_sizes_len, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::split_with_sizes(*self, torch::IntArrayRef(split_sizes_data, split_sizes_len), dim);
-    int sz = outputs__.size();
-    torch::Tensor **out__ = (torch::Tensor**)malloc((sz + 1) * sizeof(torch::Tensor*));
-    for (int i = 0; i < sz; ++i)
-      out__[i] = new torch::Tensor(outputs__[i]);
-    out__[sz] = nullptr;
-    return out__;
-  )
-}
-
-tensor *atg_split_with_sizes_copy(tensor self, int64_t *split_sizes_data, int split_sizes_len, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::split_with_sizes_copy(*self, torch::IntArrayRef(split_sizes_data, split_sizes_len), dim);
-    int sz = outputs__.size();
-    torch::Tensor **out__ = (torch::Tensor**)malloc((sz + 1) * sizeof(torch::Tensor*));
-    for (int i = 0; i < sz; ++i)
-      out__[i] = new torch::Tensor(outputs__[i]);
-    out__[sz] = nullptr;
-    return out__;
-  )
-}
-
-void atg_split_with_sizes_copy_out(tensor *out_data, int out_len, tensor self, int64_t *split_sizes_data, int split_sizes_len, int64_t dim) {
-  PROTECT(
-    torch::split_with_sizes_copy_out(of_carray_tensor(out_data, out_len), *self, torch::IntArrayRef(split_sizes_data, split_sizes_len), dim);
-  )
-}
-
-void atg_sqrt(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::sqrt(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sqrt_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::sqrt_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sqrt_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::sqrt_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_square(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::square(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_square_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::square_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_square_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::square_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_squeeze(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::squeeze(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_squeeze_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = self->squeeze_();
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_squeeze_copy(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::squeeze_copy(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_squeeze_copy_dim(tensor *out__, tensor self, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::squeeze_copy(*self, dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_squeeze_copy_dim_out(tensor *out__, tensor out, tensor self, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::squeeze_copy_out(*out, *self, dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_squeeze_copy_dims(tensor *out__, tensor self, int64_t *dim_data, int dim_len) {
-  PROTECT(
-    auto outputs__ = torch::squeeze_copy(*self, torch::IntArrayRef(dim_data, dim_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_squeeze_copy_dims_out(tensor *out__, tensor out, tensor self, int64_t *dim_data, int dim_len) {
-  PROTECT(
-    auto outputs__ = torch::squeeze_copy_out(*out, *self, torch::IntArrayRef(dim_data, dim_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_squeeze_copy_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::squeeze_copy_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_squeeze_dim(tensor *out__, tensor self, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::squeeze(*self, dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_squeeze_dim_(tensor *out__, tensor self, int64_t dim) {
-  PROTECT(
-    auto outputs__ = self->squeeze_(dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_squeeze_dims(tensor *out__, tensor self, int64_t *dim_data, int dim_len) {
-  PROTECT(
-    auto outputs__ = torch::squeeze(*self, torch::IntArrayRef(dim_data, dim_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_squeeze_dims_(tensor *out__, tensor self, int64_t *dim_data, int dim_len) {
-  PROTECT(
-    auto outputs__ = self->squeeze_(torch::IntArrayRef(dim_data, dim_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sspaddmm(tensor *out__, tensor self, tensor mat1, tensor mat2) {
-  PROTECT(
-    auto outputs__ = torch::sspaddmm(*self, *mat1, *mat2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sspaddmm_out(tensor *out__, tensor out, tensor self, tensor mat1, tensor mat2) {
-  PROTECT(
-    auto outputs__ = torch::sspaddmm_out(*out, *self, *mat1, *mat2);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_stack(tensor *out__, tensor *tensors_data, int tensors_len, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::stack(of_carray_tensor(tensors_data, tensors_len), dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_stack_out(tensor *out__, tensor out, tensor *tensors_data, int tensors_len, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::stack_out(*out, of_carray_tensor(tensors_data, tensors_len), dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_std(tensor *out__, tensor self, int unbiased) {
-  PROTECT(
-    auto outputs__ = torch::std(*self, (bool)unbiased);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_std_correction(tensor *out__, tensor self, int64_t *dim_data, int dim_len, scalar correction, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::std(*self, dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), *correction, (bool)keepdim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_std_correction_out(tensor *out__, tensor out, tensor self, int64_t *dim_data, int dim_len, scalar correction, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::std_out(*out, *self, dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), *correction, (bool)keepdim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_std_dim(tensor *out__, tensor self, int64_t *dim_data, int dim_len, int unbiased, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::std(*self, dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), (bool)unbiased, (bool)keepdim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_std_mean(tensor *out__, tensor self, int unbiased) {
-  PROTECT(
-    auto outputs__ = torch::std_mean(*self, (bool)unbiased);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_std_mean_correction(tensor *out__, tensor self, int64_t *dim_data, int dim_len, scalar correction, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::std_mean(*self, dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), *correction, (bool)keepdim);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_std_mean_correction_out(tensor *out__, tensor out0, tensor out1, tensor self, int64_t *dim_data, int dim_len, scalar correction, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::std_mean_out(*out0, *out1, *self, dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), *correction, (bool)keepdim);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_std_mean_dim(tensor *out__, tensor self, int64_t *dim_data, int dim_len, int unbiased, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::std_mean(*self, dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), (bool)unbiased, (bool)keepdim);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_std_out(tensor *out__, tensor out, tensor self, int64_t *dim_data, int dim_len, int unbiased, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::std_out(*out, *self, dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), (bool)unbiased, (bool)keepdim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_stft(tensor *out__, tensor self, int64_t n_fft, int64_t hop_length_v, int hop_length_null, int64_t win_length_v, int win_length_null, tensor window, int normalized, int onesided, int return_complex) {
-  PROTECT(
-    auto outputs__ = torch::stft(*self, n_fft, hop_length_null ? c10::nullopt : c10::optional<int64_t>(hop_length_v), win_length_null ? c10::nullopt : c10::optional<int64_t>(win_length_v), (window ? *window : torch::Tensor()), (bool)normalized, (bool)onesided, (bool)return_complex);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_stft_center(tensor *out__, tensor self, int64_t n_fft, int64_t hop_length_v, int hop_length_null, int64_t win_length_v, int win_length_null, tensor window, int center, char * pad_mode, int normalized, int onesided, int return_complex) {
-  PROTECT(
-    auto outputs__ = torch::stft(*self, n_fft, hop_length_null ? c10::nullopt : c10::optional<int64_t>(hop_length_v), win_length_null ? c10::nullopt : c10::optional<int64_t>(win_length_v), (window ? *window : torch::Tensor()), (bool)center, std::string(pad_mode), (bool)normalized, (bool)onesided, (bool)return_complex);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-int64_t atg_stride(tensor self, int64_t dim) {
-  PROTECT(
-    return torch::stride(*self, dim);
-  )
-  return 0;
-}
-
-void atg_sub(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::sub(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sub_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->sub_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sub_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::sub_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sub_scalar(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::sub(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sub_scalar_(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = self->sub_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sub_scalar_out(tensor *out__, tensor out, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::sub_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_subtract(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::subtract(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_subtract_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->subtract_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_subtract_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::subtract_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_subtract_scalar(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::subtract(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_subtract_scalar_(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = self->subtract_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sum(tensor *out__, tensor self, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::sum(*self, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sum_dim_intlist(tensor *out__, tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::sum(*self, dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), (bool)keepdim, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sum_intlist_out(tensor *out__, tensor out, tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::sum_out(*out, *self, dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), (bool)keepdim, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sum_out(tensor *out__, tensor out, tensor self, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::sum_out(*out, *self, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_sum_to_size(tensor *out__, tensor self, int64_t *size_data, int size_len) {
-  PROTECT(
-    auto outputs__ = self->sum_to_size(torch::IntArrayRef(size_data, size_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_svd(tensor *out__, tensor self, int some, int compute_uv) {
-  PROTECT(
-    auto outputs__ = torch::svd(*self, (bool)some, (bool)compute_uv);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg_svd_u(tensor *out__, tensor U, tensor S, tensor V, tensor self, int some, int compute_uv) {
-  PROTECT(
-    auto outputs__ = torch::svd_out(*U, *S, *V, *self, (bool)some, (bool)compute_uv);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg_swapaxes(tensor *out__, tensor self, int64_t axis0, int64_t axis1) {
-  PROTECT(
-    auto outputs__ = torch::swapaxes(*self, axis0, axis1);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_swapaxes_(tensor *out__, tensor self, int64_t axis0, int64_t axis1) {
-  PROTECT(
-    auto outputs__ = self->swapaxes_(axis0, axis1);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_swapdims(tensor *out__, tensor self, int64_t dim0, int64_t dim1) {
-  PROTECT(
-    auto outputs__ = torch::swapdims(*self, dim0, dim1);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_swapdims_(tensor *out__, tensor self, int64_t dim0, int64_t dim1) {
-  PROTECT(
-    auto outputs__ = self->swapdims_(dim0, dim1);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_t(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::t(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_t_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = self->t_();
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_t_copy(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::t_copy(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_t_copy_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::t_copy_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_take(tensor *out__, tensor self, tensor index) {
-  PROTECT(
-    auto outputs__ = torch::take(*self, *index);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_take_along_dim(tensor *out__, tensor self, tensor indices, int64_t dim_v, int dim_null) {
-  PROTECT(
-    auto outputs__ = torch::take_along_dim(*self, *indices, dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_take_along_dim_out(tensor *out__, tensor out, tensor self, tensor indices, int64_t dim_v, int dim_null) {
-  PROTECT(
-    auto outputs__ = torch::take_along_dim_out(*out, *self, *indices, dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_take_out(tensor *out__, tensor out, tensor self, tensor index) {
-  PROTECT(
-    auto outputs__ = torch::take_out(*out, *self, *index);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_tan(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::tan(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_tan_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::tan_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_tan_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::tan_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_tanh(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::tanh(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_tanh_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::tanh_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_tanh_backward(tensor *out__, tensor grad_output, tensor output) {
-  PROTECT(
-    auto outputs__ = torch::tanh_backward(*grad_output, *output);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_tanh_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, tensor output) {
-  PROTECT(
-    auto outputs__ = torch::tanh_backward_out(*grad_input, *grad_output, *output);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_tanh_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::tanh_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-tensor *atg_tensor_split(tensor self, int64_t sections, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::tensor_split(*self, sections, dim);
-    int sz = outputs__.size();
-    torch::Tensor **out__ = (torch::Tensor**)malloc((sz + 1) * sizeof(torch::Tensor*));
-    for (int i = 0; i < sz; ++i)
-      out__[i] = new torch::Tensor(outputs__[i]);
-    out__[sz] = nullptr;
-    return out__;
-  )
-}
-
-tensor *atg_tensor_split_indices(tensor self, int64_t *indices_data, int indices_len, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::tensor_split(*self, torch::IntArrayRef(indices_data, indices_len), dim);
-    int sz = outputs__.size();
-    torch::Tensor **out__ = (torch::Tensor**)malloc((sz + 1) * sizeof(torch::Tensor*));
-    for (int i = 0; i < sz; ++i)
-      out__[i] = new torch::Tensor(outputs__[i]);
-    out__[sz] = nullptr;
-    return out__;
-  )
-}
-
-tensor *atg_tensor_split_tensor_indices_or_sections(tensor self, tensor tensor_indices_or_sections, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::tensor_split(*self, *tensor_indices_or_sections, dim);
-    int sz = outputs__.size();
-    torch::Tensor **out__ = (torch::Tensor**)malloc((sz + 1) * sizeof(torch::Tensor*));
-    for (int i = 0; i < sz; ++i)
-      out__[i] = new torch::Tensor(outputs__[i]);
-    out__[sz] = nullptr;
-    return out__;
-  )
-}
-
-void atg_tensordot(tensor *out__, tensor self, tensor other, int64_t *dims_self_data, int dims_self_len, int64_t *dims_other_data, int dims_other_len) {
-  PROTECT(
-    auto outputs__ = torch::tensordot(*self, *other, torch::IntArrayRef(dims_self_data, dims_self_len), torch::IntArrayRef(dims_other_data, dims_other_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_tensordot_out(tensor *out__, tensor out, tensor self, tensor other, int64_t *dims_self_data, int dims_self_len, int64_t *dims_other_data, int dims_other_len) {
-  PROTECT(
-    auto outputs__ = torch::tensordot_out(*out, *self, *other, torch::IntArrayRef(dims_self_data, dims_self_len), torch::IntArrayRef(dims_other_data, dims_other_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_threshold(tensor *out__, tensor self, scalar threshold, scalar value) {
-  PROTECT(
-    auto outputs__ = torch::threshold(*self, *threshold, *value);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_threshold_(tensor *out__, tensor self, scalar threshold, scalar value) {
-  PROTECT(
-    auto outputs__ = torch::threshold_(*self, *threshold, *value);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_threshold_backward(tensor *out__, tensor grad_output, tensor self, scalar threshold) {
-  PROTECT(
-    auto outputs__ = torch::threshold_backward(*grad_output, *self, *threshold);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_threshold_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, tensor self, scalar threshold) {
-  PROTECT(
-    auto outputs__ = torch::threshold_backward_out(*grad_input, *grad_output, *self, *threshold);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_threshold_out(tensor *out__, tensor out, tensor self, scalar threshold, scalar value) {
-  PROTECT(
-    auto outputs__ = torch::threshold_out(*out, *self, *threshold, *value);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_tile(tensor *out__, tensor self, int64_t *dims_data, int dims_len) {
-  PROTECT(
-    auto outputs__ = torch::tile(*self, torch::IntArrayRef(dims_data, dims_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_to(tensor *out__, tensor self, int device) {
-  PROTECT(
-    auto outputs__ = self->to(device_of_int(device));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_to_dense(tensor *out__, tensor self, int dtype, int masked_grad) {
-  PROTECT(
-    auto outputs__ = self->to_dense(torch::ScalarType(dtype), (bool)masked_grad);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_to_dense_backward(tensor *out__, tensor grad, tensor input, int masked_grad) {
-  PROTECT(
-    auto outputs__ = torch::to_dense_backward(*grad, *input, (bool)masked_grad);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_to_device(tensor *out__, tensor self, int device, int dtype, int non_blocking, int copy) {
-  PROTECT(
-    auto outputs__ = self->to(device_of_int(device), torch::ScalarType(dtype), (bool)non_blocking, (bool)copy);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_to_dtype(tensor *out__, tensor self, int dtype, int non_blocking, int copy) {
-  PROTECT(
-    auto outputs__ = self->to(torch::ScalarType(dtype), (bool)non_blocking, (bool)copy);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_to_dtype_layout(tensor *out__, tensor self, int options_kind, int options_device, int non_blocking, int copy) {
-  PROTECT(
-    auto outputs__ = self->to(at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)), (bool)non_blocking, (bool)copy);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_to_mkldnn(tensor *out__, tensor self, int dtype) {
-  PROTECT(
-    auto outputs__ = self->to_mkldnn(torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_to_mkldnn_backward(tensor *out__, tensor grad, tensor input) {
-  PROTECT(
-    auto outputs__ = torch::to_mkldnn_backward(*grad, *input);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_to_mkldnn_out(tensor *out__, tensor out, tensor self, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::to_mkldnn_out(*out, *self, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_to_other(tensor *out__, tensor self, tensor other, int non_blocking, int copy) {
-  PROTECT(
-    auto outputs__ = self->to(*other, (bool)non_blocking, (bool)copy);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_to_padded_tensor(tensor *out__, tensor self, double padding, int64_t *output_size_data, int output_size_len) {
-  PROTECT(
-    auto outputs__ = self->to_padded_tensor(padding, output_size_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(output_size_data, output_size_len)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_to_padded_tensor_out(tensor *out__, tensor out, tensor self, double padding, int64_t *output_size_data, int output_size_len) {
-  PROTECT(
-    auto outputs__ = torch::to_padded_tensor_out(*out, *self, padding, output_size_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(output_size_data, output_size_len)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_topk(tensor *out__, tensor self, int64_t k, int64_t dim, int largest, int sorted) {
-  PROTECT(
-    auto outputs__ = torch::topk(*self, k, dim, (bool)largest, (bool)sorted);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_topk_values(tensor *out__, tensor values, tensor indices, tensor self, int64_t k, int64_t dim, int largest, int sorted) {
-  PROTECT(
-    auto outputs__ = torch::topk_out(*values, *indices, *self, k, dim, (bool)largest, (bool)sorted);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_totype(tensor *out__, tensor self, int scalar_type) {
-  PROTECT(
-    auto outputs__ = self->toType(torch::ScalarType(scalar_type));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_trace(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::trace(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_trace_backward(tensor *out__, tensor grad, int64_t *sizes_data, int sizes_len) {
-  PROTECT(
-    auto outputs__ = torch::trace_backward(*grad, torch::IntArrayRef(sizes_data, sizes_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_trace_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::trace_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_transpose(tensor *out__, tensor self, int64_t dim0, int64_t dim1) {
-  PROTECT(
-    auto outputs__ = torch::transpose(*self, dim0, dim1);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_transpose_(tensor *out__, tensor self, int64_t dim0, int64_t dim1) {
-  PROTECT(
-    auto outputs__ = self->transpose_(dim0, dim1);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_transpose_copy(tensor *out__, tensor self, int64_t dim0, int64_t dim1) {
-  PROTECT(
-    auto outputs__ = torch::transpose_copy(*self, dim0, dim1);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_transpose_copy_int_out(tensor *out__, tensor out, tensor self, int64_t dim0, int64_t dim1) {
-  PROTECT(
-    auto outputs__ = torch::transpose_copy_out(*out, *self, dim0, dim1);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_trapezoid(tensor *out__, tensor y, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::trapezoid(*y, dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_trapezoid_x(tensor *out__, tensor y, tensor x, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::trapezoid(*y, *x, dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_trapz(tensor *out__, tensor y, tensor x, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::trapz(*y, *x, dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_trapz_dx(tensor *out__, tensor y, double dx, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::trapz(*y, dx, dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_triangular_solve(tensor *out__, tensor self, tensor A, int upper, int transpose, int unitriangular) {
-  PROTECT(
-    auto outputs__ = torch::triangular_solve(*self, *A, (bool)upper, (bool)transpose, (bool)unitriangular);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_triangular_solve_x(tensor *out__, tensor X, tensor M, tensor self, tensor A, int upper, int transpose, int unitriangular) {
-  PROTECT(
-    auto outputs__ = torch::triangular_solve_out(*X, *M, *self, *A, (bool)upper, (bool)transpose, (bool)unitriangular);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_tril(tensor *out__, tensor self, int64_t diagonal) {
-  PROTECT(
-    auto outputs__ = torch::tril(*self, diagonal);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_tril_(tensor *out__, tensor self, int64_t diagonal) {
-  PROTECT(
-    auto outputs__ = self->tril_(diagonal);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_tril_indices(tensor *out__, int64_t row, int64_t col, int64_t offset, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::tril_indices(row, col, offset, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_tril_indices_out(tensor *out__, tensor out, int64_t row, int64_t col, int64_t offset) {
-  PROTECT(
-    auto outputs__ = torch::tril_indices_out(*out, row, col, offset);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_tril_out(tensor *out__, tensor out, tensor self, int64_t diagonal) {
-  PROTECT(
-    auto outputs__ = torch::tril_out(*out, *self, diagonal);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_triplet_margin_loss(tensor *out__, tensor anchor, tensor positive, tensor negative, double margin, double p, double eps, int swap, int64_t reduction) {
-  PROTECT(
-    auto outputs__ = torch::triplet_margin_loss(*anchor, *positive, *negative, margin, p, eps, (bool)swap, reduction);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_triu(tensor *out__, tensor self, int64_t diagonal) {
-  PROTECT(
-    auto outputs__ = torch::triu(*self, diagonal);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_triu_(tensor *out__, tensor self, int64_t diagonal) {
-  PROTECT(
-    auto outputs__ = self->triu_(diagonal);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_triu_indices(tensor *out__, int64_t row, int64_t col, int64_t offset, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::triu_indices(row, col, offset, at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_triu_indices_out(tensor *out__, tensor out, int64_t row, int64_t col, int64_t offset) {
-  PROTECT(
-    auto outputs__ = torch::triu_indices_out(*out, row, col, offset);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_triu_out(tensor *out__, tensor out, tensor self, int64_t diagonal) {
-  PROTECT(
-    auto outputs__ = torch::triu_out(*out, *self, diagonal);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_true_divide(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::true_divide(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_true_divide_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->true_divide_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_true_divide_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::true_divide_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_true_divide_scalar(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::true_divide(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_true_divide_scalar_(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = self->true_divide_(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_trunc(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::trunc(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_trunc_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::trunc_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_trunc_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::trunc_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_type_as(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->type_as(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-tensor *atg_unbind(tensor self, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::unbind(*self, dim);
-    int sz = outputs__.size();
-    torch::Tensor **out__ = (torch::Tensor**)malloc((sz + 1) * sizeof(torch::Tensor*));
-    for (int i = 0; i < sz; ++i)
-      out__[i] = new torch::Tensor(outputs__[i]);
-    out__[sz] = nullptr;
-    return out__;
-  )
-}
-
-tensor *atg_unbind_copy(tensor self, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::unbind_copy(*self, dim);
-    int sz = outputs__.size();
-    torch::Tensor **out__ = (torch::Tensor**)malloc((sz + 1) * sizeof(torch::Tensor*));
-    for (int i = 0; i < sz; ++i)
-      out__[i] = new torch::Tensor(outputs__[i]);
-    out__[sz] = nullptr;
-    return out__;
-  )
-}
-
-void atg_unbind_copy_int_out(tensor *out_data, int out_len, tensor self, int64_t dim) {
-  PROTECT(
-    torch::unbind_copy_out(of_carray_tensor(out_data, out_len), *self, dim);
-  )
-}
-
-void atg_unflatten(tensor *out__, tensor self, int64_t dim, int64_t *sizes_data, int sizes_len) {
-  PROTECT(
-    auto outputs__ = torch::unflatten(*self, dim, torch::IntArrayRef(sizes_data, sizes_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-tensor *atg_unflatten_dense_tensors(tensor flat, tensor *tensors_data, int tensors_len) {
-  PROTECT(
-    auto outputs__ = torch::unflatten_dense_tensors(*flat, of_carray_tensor(tensors_data, tensors_len));
-    int sz = outputs__.size();
-    torch::Tensor **out__ = (torch::Tensor**)malloc((sz + 1) * sizeof(torch::Tensor*));
-    for (int i = 0; i < sz; ++i)
-      out__[i] = new torch::Tensor(outputs__[i]);
-    out__[sz] = nullptr;
-    return out__;
-  )
-}
-
-void atg_unfold(tensor *out__, tensor self, int64_t dimension, int64_t size, int64_t step) {
-  PROTECT(
-    auto outputs__ = self->unfold(dimension, size, step);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_unfold_backward(tensor *out__, tensor grad_in, int64_t *input_sizes_data, int input_sizes_len, int64_t dim, int64_t size, int64_t step) {
-  PROTECT(
-    auto outputs__ = torch::unfold_backward(*grad_in, torch::IntArrayRef(input_sizes_data, input_sizes_len), dim, size, step);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_unfold_backward_out(tensor *out__, tensor out, tensor grad_in, int64_t *input_sizes_data, int input_sizes_len, int64_t dim, int64_t size, int64_t step) {
-  PROTECT(
-    auto outputs__ = torch::unfold_backward_out(*out, *grad_in, torch::IntArrayRef(input_sizes_data, input_sizes_len), dim, size, step);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_unfold_copy(tensor *out__, tensor self, int64_t dimension, int64_t size, int64_t step) {
-  PROTECT(
-    auto outputs__ = torch::unfold_copy(*self, dimension, size, step);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_unfold_copy_out(tensor *out__, tensor out, tensor self, int64_t dimension, int64_t size, int64_t step) {
-  PROTECT(
-    auto outputs__ = torch::unfold_copy_out(*out, *self, dimension, size, step);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_uniform(tensor *out__, tensor self, double from, double to) {
-  PROTECT(
-    auto outputs__ = torch::uniform(*self, from, to);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_uniform_(tensor *out__, tensor self, double from, double to) {
-  PROTECT(
-    auto outputs__ = self->uniform_(from, to);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_uniform_out(tensor *out__, tensor out, tensor self, double from, double to) {
-  PROTECT(
-    auto outputs__ = torch::uniform_out(*out, *self, from, to);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_unique_consecutive(tensor *out__, tensor self, int return_inverse, int return_counts, int64_t dim_v, int dim_null) {
-  PROTECT(
-    auto outputs__ = torch::unique_consecutive(*self, (bool)return_inverse, (bool)return_counts, dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v));
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg_unique_consecutive_out(tensor *out__, tensor out0, tensor out1, tensor out2, tensor self, int return_inverse, int return_counts, int64_t dim_v, int dim_null) {
-  PROTECT(
-    auto outputs__ = torch::unique_consecutive_out(*out0, *out1, *out2, *self, (bool)return_inverse, (bool)return_counts, dim_null ? c10::nullopt : c10::optional<int64_t>(dim_v));
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg_unique_dim(tensor *out__, tensor self, int64_t dim, int sorted, int return_inverse, int return_counts) {
-  PROTECT(
-    auto outputs__ = torch::unique_dim(*self, dim, (bool)sorted, (bool)return_inverse, (bool)return_counts);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg_unique_dim_consecutive(tensor *out__, tensor self, int64_t dim, int return_inverse, int return_counts) {
-  PROTECT(
-    auto outputs__ = torch::unique_dim_consecutive(*self, dim, (bool)return_inverse, (bool)return_counts);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg_unique_dim_consecutive_out(tensor *out__, tensor out0, tensor out1, tensor out2, tensor self, int64_t dim, int return_inverse, int return_counts) {
-  PROTECT(
-    auto outputs__ = torch::unique_dim_consecutive_out(*out0, *out1, *out2, *self, dim, (bool)return_inverse, (bool)return_counts);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-void atg_unique_dim_out(tensor *out__, tensor out0, tensor out1, tensor out2, tensor self, int64_t dim, int sorted, int return_inverse, int return_counts) {
-  PROTECT(
-    auto outputs__ = torch::unique_dim_out(*out0, *out1, *out2, *self, dim, (bool)sorted, (bool)return_inverse, (bool)return_counts);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-    out__[2] = new torch::Tensor(std::get<2>(outputs__));
-  )
-}
-
-tensor *atg_unsafe_chunk(tensor self, int64_t chunks, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::unsafe_chunk(*self, chunks, dim);
-    int sz = outputs__.size();
-    torch::Tensor **out__ = (torch::Tensor**)malloc((sz + 1) * sizeof(torch::Tensor*));
-    for (int i = 0; i < sz; ++i)
-      out__[i] = new torch::Tensor(outputs__[i]);
-    out__[sz] = nullptr;
-    return out__;
-  )
-}
-
-tensor *atg_unsafe_split(tensor self, int64_t split_size, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::unsafe_split(*self, split_size, dim);
-    int sz = outputs__.size();
-    torch::Tensor **out__ = (torch::Tensor**)malloc((sz + 1) * sizeof(torch::Tensor*));
-    for (int i = 0; i < sz; ++i)
-      out__[i] = new torch::Tensor(outputs__[i]);
-    out__[sz] = nullptr;
-    return out__;
-  )
-}
-
-void atg_unsafe_split_tensor_out(tensor *out_data, int out_len, tensor self, int64_t split_size, int64_t dim) {
-  PROTECT(
-    torch::unsafe_split_out(of_carray_tensor(out_data, out_len), *self, split_size, dim);
-  )
-}
-
-tensor *atg_unsafe_split_with_sizes(tensor self, int64_t *split_sizes_data, int split_sizes_len, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::unsafe_split_with_sizes(*self, torch::IntArrayRef(split_sizes_data, split_sizes_len), dim);
-    int sz = outputs__.size();
-    torch::Tensor **out__ = (torch::Tensor**)malloc((sz + 1) * sizeof(torch::Tensor*));
-    for (int i = 0; i < sz; ++i)
-      out__[i] = new torch::Tensor(outputs__[i]);
-    out__[sz] = nullptr;
-    return out__;
-  )
-}
-
-void atg_unsafe_split_with_sizes_out(tensor *out_data, int out_len, tensor self, int64_t *split_sizes_data, int split_sizes_len, int64_t dim) {
-  PROTECT(
-    torch::unsafe_split_with_sizes_out(of_carray_tensor(out_data, out_len), *self, torch::IntArrayRef(split_sizes_data, split_sizes_len), dim);
-  )
-}
-
-void atg_unsqueeze(tensor *out__, tensor self, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::unsqueeze(*self, dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_unsqueeze_(tensor *out__, tensor self, int64_t dim) {
-  PROTECT(
-    auto outputs__ = self->unsqueeze_(dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_unsqueeze_copy(tensor *out__, tensor self, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::unsqueeze_copy(*self, dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_unsqueeze_copy_out(tensor *out__, tensor out, tensor self, int64_t dim) {
-  PROTECT(
-    auto outputs__ = torch::unsqueeze_copy_out(*out, *self, dim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_upsample_bicubic2d(tensor *out__, tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
-  PROTECT(
-    auto outputs__ = torch::upsample_bicubic2d(*self, torch::IntArrayRef(output_size_data, output_size_len), (bool)align_corners, scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_upsample_bicubic2d_backward(tensor *out__, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
-  PROTECT(
-    auto outputs__ = torch::upsample_bicubic2d_backward(*grad_output, torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), (bool)align_corners, scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_upsample_bicubic2d_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
-  PROTECT(
-    auto outputs__ = torch::upsample_bicubic2d_backward_out(*grad_input, *grad_output, torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), (bool)align_corners, scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_upsample_bicubic2d_out(tensor *out__, tensor out, tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
-  PROTECT(
-    auto outputs__ = torch::upsample_bicubic2d_out(*out, *self, torch::IntArrayRef(output_size_data, output_size_len), (bool)align_corners, scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_upsample_bicubic2d_vec(tensor *out__, tensor input, int64_t *output_size_data, int output_size_len, int align_corners, double *scale_factors_data, int scale_factors_len) {
-  PROTECT(
-    auto outputs__ = torch::upsample_bicubic2d(*input, output_size_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(output_size_data, output_size_len)), (bool)align_corners, at::ArrayRef<double>(scale_factors_data, scale_factors_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_upsample_bilinear2d(tensor *out__, tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
-  PROTECT(
-    auto outputs__ = torch::upsample_bilinear2d(*self, torch::IntArrayRef(output_size_data, output_size_len), (bool)align_corners, scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_upsample_bilinear2d_backward(tensor *out__, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
-  PROTECT(
-    auto outputs__ = torch::upsample_bilinear2d_backward(*grad_output, torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), (bool)align_corners, scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_upsample_bilinear2d_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
-  PROTECT(
-    auto outputs__ = torch::upsample_bilinear2d_backward_out(*grad_input, *grad_output, torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), (bool)align_corners, scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_upsample_bilinear2d_out(tensor *out__, tensor out, tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
-  PROTECT(
-    auto outputs__ = torch::upsample_bilinear2d_out(*out, *self, torch::IntArrayRef(output_size_data, output_size_len), (bool)align_corners, scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_upsample_bilinear2d_vec(tensor *out__, tensor input, int64_t *output_size_data, int output_size_len, int align_corners, double *scale_factors_data, int scale_factors_len) {
-  PROTECT(
-    auto outputs__ = torch::upsample_bilinear2d(*input, output_size_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(output_size_data, output_size_len)), (bool)align_corners, at::ArrayRef<double>(scale_factors_data, scale_factors_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_upsample_linear1d(tensor *out__, tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_v, int scales_null) {
-  PROTECT(
-    auto outputs__ = torch::upsample_linear1d(*self, torch::IntArrayRef(output_size_data, output_size_len), (bool)align_corners, scales_null ? c10::nullopt : c10::optional<double>(scales_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_upsample_linear1d_backward(tensor *out__, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_v, int scales_null) {
-  PROTECT(
-    auto outputs__ = torch::upsample_linear1d_backward(*grad_output, torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), (bool)align_corners, scales_null ? c10::nullopt : c10::optional<double>(scales_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_upsample_linear1d_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_v, int scales_null) {
-  PROTECT(
-    auto outputs__ = torch::upsample_linear1d_backward_out(*grad_input, *grad_output, torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), (bool)align_corners, scales_null ? c10::nullopt : c10::optional<double>(scales_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_upsample_linear1d_out(tensor *out__, tensor out, tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_v, int scales_null) {
-  PROTECT(
-    auto outputs__ = torch::upsample_linear1d_out(*out, *self, torch::IntArrayRef(output_size_data, output_size_len), (bool)align_corners, scales_null ? c10::nullopt : c10::optional<double>(scales_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_upsample_linear1d_vec(tensor *out__, tensor input, int64_t *output_size_data, int output_size_len, int align_corners, double *scale_factors_data, int scale_factors_len) {
-  PROTECT(
-    auto outputs__ = torch::upsample_linear1d(*input, output_size_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(output_size_data, output_size_len)), (bool)align_corners, at::ArrayRef<double>(scale_factors_data, scale_factors_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_upsample_nearest1d(tensor *out__, tensor self, int64_t *output_size_data, int output_size_len, double scales_v, int scales_null) {
-  PROTECT(
-    auto outputs__ = torch::upsample_nearest1d(*self, torch::IntArrayRef(output_size_data, output_size_len), scales_null ? c10::nullopt : c10::optional<double>(scales_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_upsample_nearest1d_backward(tensor *out__, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_v, int scales_null) {
-  PROTECT(
-    auto outputs__ = torch::upsample_nearest1d_backward(*grad_output, torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), scales_null ? c10::nullopt : c10::optional<double>(scales_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_upsample_nearest1d_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_v, int scales_null) {
-  PROTECT(
-    auto outputs__ = torch::upsample_nearest1d_backward_out(*grad_input, *grad_output, torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), scales_null ? c10::nullopt : c10::optional<double>(scales_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_upsample_nearest1d_out(tensor *out__, tensor out, tensor self, int64_t *output_size_data, int output_size_len, double scales_v, int scales_null) {
-  PROTECT(
-    auto outputs__ = torch::upsample_nearest1d_out(*out, *self, torch::IntArrayRef(output_size_data, output_size_len), scales_null ? c10::nullopt : c10::optional<double>(scales_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_upsample_nearest1d_vec(tensor *out__, tensor input, int64_t *output_size_data, int output_size_len, double *scale_factors_data, int scale_factors_len) {
-  PROTECT(
-    auto outputs__ = torch::upsample_nearest1d(*input, output_size_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(output_size_data, output_size_len)), at::ArrayRef<double>(scale_factors_data, scale_factors_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_upsample_nearest2d(tensor *out__, tensor self, int64_t *output_size_data, int output_size_len, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
-  PROTECT(
-    auto outputs__ = torch::upsample_nearest2d(*self, torch::IntArrayRef(output_size_data, output_size_len), scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_upsample_nearest2d_backward(tensor *out__, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
-  PROTECT(
-    auto outputs__ = torch::upsample_nearest2d_backward(*grad_output, torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_upsample_nearest2d_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
-  PROTECT(
-    auto outputs__ = torch::upsample_nearest2d_backward_out(*grad_input, *grad_output, torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_upsample_nearest2d_out(tensor *out__, tensor out, tensor self, int64_t *output_size_data, int output_size_len, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
-  PROTECT(
-    auto outputs__ = torch::upsample_nearest2d_out(*out, *self, torch::IntArrayRef(output_size_data, output_size_len), scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_upsample_nearest2d_vec(tensor *out__, tensor input, int64_t *output_size_data, int output_size_len, double *scale_factors_data, int scale_factors_len) {
-  PROTECT(
-    auto outputs__ = torch::upsample_nearest2d(*input, output_size_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(output_size_data, output_size_len)), at::ArrayRef<double>(scale_factors_data, scale_factors_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_upsample_nearest3d(tensor *out__, tensor self, int64_t *output_size_data, int output_size_len, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
-  PROTECT(
-    auto outputs__ = torch::upsample_nearest3d(*self, torch::IntArrayRef(output_size_data, output_size_len), scales_d_null ? c10::nullopt : c10::optional<double>(scales_d_v), scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_upsample_nearest3d_backward(tensor *out__, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
-  PROTECT(
-    auto outputs__ = torch::upsample_nearest3d_backward(*grad_output, torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), scales_d_null ? c10::nullopt : c10::optional<double>(scales_d_v), scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_upsample_nearest3d_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
-  PROTECT(
-    auto outputs__ = torch::upsample_nearest3d_backward_out(*grad_input, *grad_output, torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), scales_d_null ? c10::nullopt : c10::optional<double>(scales_d_v), scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_upsample_nearest3d_out(tensor *out__, tensor out, tensor self, int64_t *output_size_data, int output_size_len, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
-  PROTECT(
-    auto outputs__ = torch::upsample_nearest3d_out(*out, *self, torch::IntArrayRef(output_size_data, output_size_len), scales_d_null ? c10::nullopt : c10::optional<double>(scales_d_v), scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_upsample_nearest3d_vec(tensor *out__, tensor input, int64_t *output_size_data, int output_size_len, double *scale_factors_data, int scale_factors_len) {
-  PROTECT(
-    auto outputs__ = torch::upsample_nearest3d(*input, output_size_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(output_size_data, output_size_len)), at::ArrayRef<double>(scale_factors_data, scale_factors_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_upsample_trilinear3d(tensor *out__, tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
-  PROTECT(
-    auto outputs__ = torch::upsample_trilinear3d(*self, torch::IntArrayRef(output_size_data, output_size_len), (bool)align_corners, scales_d_null ? c10::nullopt : c10::optional<double>(scales_d_v), scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_upsample_trilinear3d_backward(tensor *out__, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
-  PROTECT(
-    auto outputs__ = torch::upsample_trilinear3d_backward(*grad_output, torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), (bool)align_corners, scales_d_null ? c10::nullopt : c10::optional<double>(scales_d_v), scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_upsample_trilinear3d_backward_grad_input(tensor *out__, tensor grad_input, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
-  PROTECT(
-    auto outputs__ = torch::upsample_trilinear3d_backward_out(*grad_input, *grad_output, torch::IntArrayRef(output_size_data, output_size_len), torch::IntArrayRef(input_size_data, input_size_len), (bool)align_corners, scales_d_null ? c10::nullopt : c10::optional<double>(scales_d_v), scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_upsample_trilinear3d_out(tensor *out__, tensor out, tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null) {
-  PROTECT(
-    auto outputs__ = torch::upsample_trilinear3d_out(*out, *self, torch::IntArrayRef(output_size_data, output_size_len), (bool)align_corners, scales_d_null ? c10::nullopt : c10::optional<double>(scales_d_v), scales_h_null ? c10::nullopt : c10::optional<double>(scales_h_v), scales_w_null ? c10::nullopt : c10::optional<double>(scales_w_v));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_upsample_trilinear3d_vec(tensor *out__, tensor input, int64_t *output_size_data, int output_size_len, int align_corners, double *scale_factors_data, int scale_factors_len) {
-  PROTECT(
-    auto outputs__ = torch::upsample_trilinear3d(*input, output_size_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(output_size_data, output_size_len)), (bool)align_corners, at::ArrayRef<double>(scale_factors_data, scale_factors_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_value_selecting_reduction_backward(tensor *out__, tensor grad, int64_t dim, tensor indices, int64_t *sizes_data, int sizes_len, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::value_selecting_reduction_backward(*grad, dim, *indices, torch::IntArrayRef(sizes_data, sizes_len), (bool)keepdim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_values(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = self->values();
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_values_copy(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::values_copy(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_values_copy_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::values_copy_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_vander(tensor *out__, tensor x, int64_t n_v, int n_null, int increasing) {
-  PROTECT(
-    auto outputs__ = torch::vander(*x, n_null ? c10::nullopt : c10::optional<int64_t>(n_v), (bool)increasing);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_var(tensor *out__, tensor self, int unbiased) {
-  PROTECT(
-    auto outputs__ = torch::var(*self, (bool)unbiased);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_var_correction(tensor *out__, tensor self, int64_t *dim_data, int dim_len, scalar correction, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::var(*self, dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), *correction, (bool)keepdim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_var_correction_out(tensor *out__, tensor out, tensor self, int64_t *dim_data, int dim_len, scalar correction, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::var_out(*out, *self, dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), *correction, (bool)keepdim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_var_dim(tensor *out__, tensor self, int64_t *dim_data, int dim_len, int unbiased, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::var(*self, dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), (bool)unbiased, (bool)keepdim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_var_mean(tensor *out__, tensor self, int unbiased) {
-  PROTECT(
-    auto outputs__ = torch::var_mean(*self, (bool)unbiased);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_var_mean_correction(tensor *out__, tensor self, int64_t *dim_data, int dim_len, scalar correction, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::var_mean(*self, dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), *correction, (bool)keepdim);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_var_mean_correction_out(tensor *out__, tensor out0, tensor out1, tensor self, int64_t *dim_data, int dim_len, scalar correction, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::var_mean_out(*out0, *out1, *self, dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), *correction, (bool)keepdim);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_var_mean_dim(tensor *out__, tensor self, int64_t *dim_data, int dim_len, int unbiased, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::var_mean(*self, dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), (bool)unbiased, (bool)keepdim);
-    out__[0] = new torch::Tensor(std::get<0>(outputs__));
-    out__[1] = new torch::Tensor(std::get<1>(outputs__));
-  )
-}
-
-void atg_var_out(tensor *out__, tensor out, tensor self, int64_t *dim_data, int dim_len, int unbiased, int keepdim) {
-  PROTECT(
-    auto outputs__ = torch::var_out(*out, *self, dim_data == nullptr ? c10::nullopt : c10::optional<torch::IntArrayRef>(torch::IntArrayRef(dim_data, dim_len)), (bool)unbiased, (bool)keepdim);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_vdot(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::vdot(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_vdot_out(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::vdot_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_view(tensor *out__, tensor self, int64_t *size_data, int size_len) {
-  PROTECT(
-    auto outputs__ = self->view(torch::IntArrayRef(size_data, size_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_view_as(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = self->view_as(*other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_view_as_complex(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::view_as_complex(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_view_as_complex_copy(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::view_as_complex_copy(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_view_as_complex_copy_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::view_as_complex_copy_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_view_as_real(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::view_as_real(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_view_as_real_copy(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::view_as_real_copy(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_view_as_real_copy_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::view_as_real_copy_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_view_copy(tensor *out__, tensor self, int64_t *size_data, int size_len) {
-  PROTECT(
-    auto outputs__ = torch::view_copy(*self, torch::IntArrayRef(size_data, size_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_view_copy_dtype(tensor *out__, tensor self, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::view_copy(*self, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_view_copy_dtype_out(tensor *out__, tensor out, tensor self, int dtype) {
-  PROTECT(
-    auto outputs__ = torch::view_copy_out(*out, *self, torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_view_copy_out(tensor *out__, tensor out, tensor self, int64_t *size_data, int size_len) {
-  PROTECT(
-    auto outputs__ = torch::view_copy_out(*out, *self, torch::IntArrayRef(size_data, size_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_view_dtype(tensor *out__, tensor self, int dtype) {
-  PROTECT(
-    auto outputs__ = self->view(torch::ScalarType(dtype));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-tensor *atg_vsplit(tensor self, int64_t sections) {
-  PROTECT(
-    auto outputs__ = torch::vsplit(*self, sections);
-    int sz = outputs__.size();
-    torch::Tensor **out__ = (torch::Tensor**)malloc((sz + 1) * sizeof(torch::Tensor*));
-    for (int i = 0; i < sz; ++i)
-      out__[i] = new torch::Tensor(outputs__[i]);
-    out__[sz] = nullptr;
-    return out__;
-  )
-}
-
-tensor *atg_vsplit_array(tensor self, int64_t *indices_data, int indices_len) {
-  PROTECT(
-    auto outputs__ = torch::vsplit(*self, torch::IntArrayRef(indices_data, indices_len));
-    int sz = outputs__.size();
-    torch::Tensor **out__ = (torch::Tensor**)malloc((sz + 1) * sizeof(torch::Tensor*));
-    for (int i = 0; i < sz; ++i)
-      out__[i] = new torch::Tensor(outputs__[i]);
-    out__[sz] = nullptr;
-    return out__;
-  )
-}
-
-void atg_vstack(tensor *out__, tensor *tensors_data, int tensors_len) {
-  PROTECT(
-    auto outputs__ = torch::vstack(of_carray_tensor(tensors_data, tensors_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_vstack_out(tensor *out__, tensor out, tensor *tensors_data, int tensors_len) {
-  PROTECT(
-    auto outputs__ = torch::vstack_out(*out, of_carray_tensor(tensors_data, tensors_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-tensor *atg_where(tensor condition) {
-  PROTECT(
-    auto outputs__ = torch::where(*condition);
-    int sz = outputs__.size();
-    torch::Tensor **out__ = (torch::Tensor**)malloc((sz + 1) * sizeof(torch::Tensor*));
-    for (int i = 0; i < sz; ++i)
-      out__[i] = new torch::Tensor(outputs__[i]);
-    out__[sz] = nullptr;
-    return out__;
-  )
-}
-
-void atg_where_scalar(tensor *out__, tensor condition, scalar self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::where(*condition, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_where_scalarother(tensor *out__, tensor condition, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::where(*condition, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_where_scalarself(tensor *out__, tensor condition, scalar self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::where(*condition, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_where_self(tensor *out__, tensor condition, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::where(*condition, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_where_self_out(tensor *out__, tensor out, tensor condition, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::where_out(*out, *condition, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_xlogy(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::xlogy(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_xlogy_(tensor *out__, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::xlogy_(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_xlogy_outscalar_other(tensor *out__, tensor out, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::xlogy_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_xlogy_outscalar_self(tensor *out__, tensor out, scalar self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::xlogy_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_xlogy_outtensor(tensor *out__, tensor out, tensor self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::xlogy_out(*out, *self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_xlogy_scalar_other(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::xlogy(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_xlogy_scalar_other_(tensor *out__, tensor self, scalar other) {
-  PROTECT(
-    auto outputs__ = torch::xlogy_(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_xlogy_scalar_self(tensor *out__, scalar self, tensor other) {
-  PROTECT(
-    auto outputs__ = torch::xlogy(*self, *other);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_zero(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::zero(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_zero_(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::zero_(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_zero_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::zero_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_zeros(tensor *out__, int64_t *size_data, int size_len, int options_kind, int options_device) {
-  PROTECT(
-    auto outputs__ = torch::zeros(torch::IntArrayRef(size_data, size_len), at::device(device_of_int(options_device)).dtype(at::ScalarType(options_kind)));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_zeros_like(tensor *out__, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::zeros_like(*self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_zeros_like_out(tensor *out__, tensor out, tensor self) {
-  PROTECT(
-    auto outputs__ = torch::zeros_like_out(*out, *self);
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
-void atg_zeros_out(tensor *out__, tensor out, int64_t *size_data, int size_len) {
-  PROTECT(
-    auto outputs__ = torch::zeros_out(*out, torch::IntArrayRef(size_data, size_len));
-    out__[0] = new torch::Tensor(outputs__);
-  )
-}
-
diff --git a/src/wrapper/torch_api_generated.h b/src/wrapper/torch_api_generated.h
index b63341d..bab5428 100644
--- a/src/wrapper/torch_api_generated.h
+++ b/src/wrapper/torch_api_generated.h
@@ -1,2551 +1,2551 @@
 // THIS FILE IS AUTOMATICALLY GENERATED, DO NOT EDIT BY HAND!
 
-void atg___and__(tensor *, tensor self, scalar other);
-void atg___and__tensor_(tensor *, tensor self, tensor other);
-void atg___iand__(tensor *, tensor self, scalar other);
-void atg___iand__tensor_(tensor *, tensor self, tensor other);
-void atg___ilshift__(tensor *, tensor self, scalar other);
-void atg___ilshift__tensor_(tensor *, tensor self, tensor other);
-void atg___ior__(tensor *, tensor self, scalar other);
-void atg___ior__tensor_(tensor *, tensor self, tensor other);
-void atg___irshift__(tensor *, tensor self, scalar other);
-void atg___irshift__tensor_(tensor *, tensor self, tensor other);
-void atg___ixor__(tensor *, tensor self, scalar other);
-void atg___ixor__tensor_(tensor *, tensor self, tensor other);
-void atg___lshift__(tensor *, tensor self, scalar other);
-void atg___lshift__scalar_out_(tensor *, tensor out, tensor self, scalar other);
-void atg___lshift__tensor_(tensor *, tensor self, tensor other);
-void atg___lshift__tensor_out_(tensor *, tensor out, tensor self, tensor other);
-void atg___or__(tensor *, tensor self, scalar other);
-void atg___or__tensor_(tensor *, tensor self, tensor other);
-void atg___rshift__(tensor *, tensor self, scalar other);
-void atg___rshift__scalar_out_(tensor *, tensor out, tensor self, scalar other);
-void atg___rshift__tensor_(tensor *, tensor self, tensor other);
-void atg___rshift__tensor_out_(tensor *, tensor out, tensor self, tensor other);
-void atg___xor__(tensor *, tensor self, scalar other);
-void atg___xor__tensor_(tensor *, tensor self, tensor other);
-void atg__adaptive_avg_pool2d(tensor *, tensor self, int64_t *output_size_data, int output_size_len);
-void atg__adaptive_avg_pool2d_backward(tensor *, tensor grad_output, tensor self);
-void atg__adaptive_avg_pool2d_backward_out(tensor *, tensor out, tensor grad_output, tensor self);
-void atg__adaptive_avg_pool2d_out(tensor *, tensor out, tensor self, int64_t *output_size_data, int output_size_len);
-void atg__adaptive_avg_pool3d(tensor *, tensor self, int64_t *output_size_data, int output_size_len);
-void atg__adaptive_avg_pool3d_backward(tensor *, tensor grad_output, tensor self);
-void atg__adaptive_avg_pool3d_backward_out(tensor *, tensor out, tensor grad_output, tensor self);
-void atg__adaptive_avg_pool3d_out(tensor *, tensor out, tensor self, int64_t *output_size_data, int output_size_len);
-void atg__add_batch_dim(tensor *, tensor self, int64_t batch_dim, int64_t level);
-void atg__add_relu(tensor *, tensor self, tensor other);
-void atg__add_relu_(tensor *, tensor self, tensor other);
-void atg__add_relu_out(tensor *, tensor out, tensor self, tensor other);
-void atg__add_relu_scalar(tensor *, tensor self, scalar other);
-void atg__add_relu_scalar_(tensor *, tensor self, scalar other);
-void atg__add_relu_scalar_out(tensor *, tensor out, tensor self, scalar other);
-void atg__addmm_activation(tensor *, tensor self, tensor mat1, tensor mat2, int use_gelu);
-void atg__addmm_activation_out(tensor *, tensor out, tensor self, tensor mat1, tensor mat2, int use_gelu);
-void atg__aminmax(tensor *, tensor self);
-void atg__aminmax_dim(tensor *, tensor self, int64_t dim, int keepdim);
-void atg__aminmax_dim_out(tensor *, tensor out0, tensor out1, tensor self, int64_t dim, int keepdim);
-void atg__aminmax_out(tensor *, tensor out0, tensor out1, tensor self);
-void atg__amp_update_scale(tensor *, tensor self, tensor growth_tracker, tensor found_inf, double scale_growth_factor, double scale_backoff_factor, int64_t growth_interval);
-void atg__amp_update_scale_(tensor *, tensor self, tensor growth_tracker, tensor found_inf, double scale_growth_factor, double scale_backoff_factor, int64_t growth_interval);
-void atg__amp_update_scale_out(tensor *, tensor out, tensor self, tensor growth_tracker, tensor found_inf, double scale_growth_factor, double scale_backoff_factor, int64_t growth_interval);
-void atg__assert_tensor_metadata(tensor a, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len, int dtype);
-void atg__autocast_to_full_precision(tensor *, tensor self, int cuda_enabled, int cpu_enabled);
-void atg__autocast_to_reduced_precision(tensor *, tensor self, int cuda_enabled, int cpu_enabled, int cuda_dtype, int cpu_dtype);
-void atg__cast_byte(tensor *, tensor self, int non_blocking);
-void atg__cast_char(tensor *, tensor self, int non_blocking);
-void atg__cast_double(tensor *, tensor self, int non_blocking);
-void atg__cast_float(tensor *, tensor self, int non_blocking);
-void atg__cast_half(tensor *, tensor self, int non_blocking);
-void atg__cast_int(tensor *, tensor self, int non_blocking);
-void atg__cast_long(tensor *, tensor self, int non_blocking);
-void atg__cast_short(tensor *, tensor self, int non_blocking);
-void atg__cdist_backward(tensor *, tensor grad, tensor x1, tensor x2, double p, tensor cdist);
-void atg__cdist_backward_out(tensor *, tensor out, tensor grad, tensor x1, tensor x2, double p, tensor cdist);
-void atg__cholesky_solve_helper(tensor *, tensor self, tensor A, int upper);
-void atg__cholesky_solve_helper_out(tensor *, tensor out, tensor self, tensor A, int upper);
-void atg__coalesce(tensor *, tensor self);
-void atg__coalesce_out(tensor *, tensor out, tensor self);
-void atg__coalesced(tensor *, tensor self, int coalesced);
-void atg__coalesced_(tensor *, tensor self, int coalesced);
-void atg__coalesced_out(tensor *, tensor out, tensor self, int coalesced);
-void atg__compute_linear_combination(tensor *, tensor input, tensor coefficients);
-void atg__compute_linear_combination_out(tensor *, tensor out, tensor input, tensor coefficients);
-void atg__conj(tensor *, tensor self);
-void atg__conj_copy(tensor *, tensor self);
-void atg__conj_copy_out(tensor *, tensor out, tensor self);
-void atg__conj_physical(tensor *, tensor self);
-void atg__conj_physical_out(tensor *, tensor out, tensor self);
-void atg__conv_depthwise2d(tensor *, tensor self, tensor weight, int64_t *kernel_size_data, int kernel_size_len, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len);
-void atg__conv_depthwise2d_out(tensor *, tensor out, tensor self, tensor weight, int64_t *kernel_size_data, int kernel_size_len, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len);
-void atg__convert_indices_from_coo_to_csr(tensor *, tensor self, int64_t size, int out_int32);
-void atg__convert_indices_from_coo_to_csr_out(tensor *, tensor out, tensor self, int64_t size, int out_int32);
-void atg__convert_indices_from_csr_to_coo(tensor *, tensor crow_indices, tensor col_indices, int out_int32, int transpose);
-void atg__convert_indices_from_csr_to_coo_out(tensor *, tensor out, tensor crow_indices, tensor col_indices, int out_int32, int transpose);
-void atg__convolution(tensor *, tensor input, tensor weight, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int transposed, int64_t *output_padding_data, int output_padding_len, int64_t groups, int benchmark, int deterministic, int cudnn_enabled, int allow_tf32);
-void atg__convolution_deprecated(tensor *, tensor input, tensor weight, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int transposed, int64_t *output_padding_data, int output_padding_len, int64_t groups, int benchmark, int deterministic, int cudnn_enabled);
-void atg__convolution_mode(tensor *, tensor input, tensor weight, tensor bias, int64_t *stride_data, int stride_len, char * padding, int64_t *dilation_data, int dilation_len, int64_t groups);
-void atg__convolution_out(tensor *, tensor out, tensor input, tensor weight, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int transposed, int64_t *output_padding_data, int output_padding_len, int64_t groups, int benchmark, int deterministic, int cudnn_enabled, int allow_tf32);
-void atg__copy_from(tensor *, tensor self, tensor dst, int non_blocking);
-void atg__copy_from_and_resize(tensor *, tensor self, tensor dst);
-void atg__copy_from_and_resize_out(tensor *, tensor out, tensor self, tensor dst);
-void atg__copy_from_out(tensor *, tensor out, tensor self, tensor dst, int non_blocking);
-void atg__cslt_compress(tensor *, tensor input);
-void atg__cslt_sparse_mm(tensor *, tensor compressed_A, tensor dense_B, tensor bias, int transpose_result);
-void atg__ctc_loss(tensor *, tensor log_probs, tensor targets, int64_t *input_lengths_data, int input_lengths_len, int64_t *target_lengths_data, int target_lengths_len, int64_t blank, int zero_infinity);
-void atg__ctc_loss_backward(tensor *, tensor grad, tensor log_probs, tensor targets, int64_t *input_lengths_data, int input_lengths_len, int64_t *target_lengths_data, int target_lengths_len, tensor neg_log_likelihood, tensor log_alpha, int64_t blank, int zero_infinity);
-void atg__ctc_loss_backward_out(tensor *, tensor out, tensor grad, tensor log_probs, tensor targets, int64_t *input_lengths_data, int input_lengths_len, int64_t *target_lengths_data, int target_lengths_len, tensor neg_log_likelihood, tensor log_alpha, int64_t blank, int zero_infinity);
-void atg__ctc_loss_backward_tensor(tensor *, tensor grad, tensor log_probs, tensor targets, tensor input_lengths, tensor target_lengths, tensor neg_log_likelihood, tensor log_alpha, int64_t blank, int zero_infinity);
-void atg__ctc_loss_out(tensor *, tensor out0, tensor out1, tensor log_probs, tensor targets, int64_t *input_lengths_data, int input_lengths_len, int64_t *target_lengths_data, int target_lengths_len, int64_t blank, int zero_infinity);
-void atg__ctc_loss_tensor(tensor *, tensor log_probs, tensor targets, tensor input_lengths, tensor target_lengths, int64_t blank, int zero_infinity);
-void atg__ctc_loss_tensor_out(tensor *, tensor out0, tensor out1, tensor log_probs, tensor targets, tensor input_lengths, tensor target_lengths, int64_t blank, int zero_infinity);
-void atg__cudnn_ctc_loss(tensor *, tensor log_probs, tensor targets, int64_t *input_lengths_data, int input_lengths_len, int64_t *target_lengths_data, int target_lengths_len, int64_t blank, int deterministic, int zero_infinity);
-void atg__cudnn_ctc_loss_out(tensor *, tensor out0, tensor out1, tensor log_probs, tensor targets, int64_t *input_lengths_data, int input_lengths_len, int64_t *target_lengths_data, int target_lengths_len, int64_t blank, int deterministic, int zero_infinity);
-void atg__cudnn_ctc_loss_tensor(tensor *, tensor log_probs, tensor targets, tensor input_lengths, tensor target_lengths, int64_t blank, int deterministic, int zero_infinity);
-void atg__cudnn_init_dropout_state(tensor *, double dropout, int train, int64_t dropout_seed, int options_kind, int options_device);
-void atg__cudnn_init_dropout_state_out(tensor *, tensor out, double dropout, int train, int64_t dropout_seed);
-void atg__cudnn_rnn(tensor *, tensor input, tensor *weight_data, int weight_len, int64_t weight_stride0, tensor weight_buf, tensor hx, tensor cx, int64_t mode, int64_t hidden_size, int64_t proj_size, int64_t num_layers, int batch_first, double dropout, int train, int bidirectional, int64_t *batch_sizes_data, int batch_sizes_len, tensor dropout_state);
-void atg__cudnn_rnn_flatten_weight(tensor *, tensor *weight_arr_data, int weight_arr_len, int64_t weight_stride0, int64_t input_size, int64_t mode, int64_t hidden_size, int64_t proj_size, int64_t num_layers, int batch_first, int bidirectional);
-void atg__cudnn_rnn_flatten_weight_out(tensor *, tensor out, tensor *weight_arr_data, int weight_arr_len, int64_t weight_stride0, int64_t input_size, int64_t mode, int64_t hidden_size, int64_t proj_size, int64_t num_layers, int batch_first, int bidirectional);
-void atg__cudnn_rnn_out(tensor *, tensor out0, tensor out1, tensor out2, tensor out3, tensor out4, tensor input, tensor *weight_data, int weight_len, int64_t weight_stride0, tensor weight_buf, tensor hx, tensor cx, int64_t mode, int64_t hidden_size, int64_t proj_size, int64_t num_layers, int batch_first, double dropout, int train, int bidirectional, int64_t *batch_sizes_data, int batch_sizes_len, tensor dropout_state);
-int64_t atg__debug_has_internal_overlap(tensor self);
-void atg__dim_arange(tensor *, tensor like, int64_t dim);
-int64_t atg__dimi(tensor self);
-int64_t atg__dimv(tensor self);
-void atg__dirichlet_grad(tensor *, tensor x, tensor alpha, tensor total);
-void atg__dirichlet_grad_out(tensor *, tensor out, tensor x, tensor alpha, tensor total);
-void atg__efficient_attention_backward(tensor *, tensor grad_out_, tensor query, tensor key, tensor value, tensor bias, tensor out, tensor cu_seqlens_q, tensor cu_seqlens_k, int64_t max_seqlen_k, int64_t max_seqlen_q, tensor logsumexp, double dropout_p, tensor philox_seed, tensor philox_offset, int64_t custom_mask_type, int bias_requires_grad, double scale_v, int scale_null, int64_t num_splits_key_v, int num_splits_key_null);
-void atg__efficientzerotensor(tensor *, int64_t *size_data, int size_len, int options_kind, int options_device);
-void atg__efficientzerotensor_out(tensor *, tensor out, int64_t *size_data, int size_len);
-void atg__embedding_bag(tensor *, tensor weight, tensor indices, tensor offsets, int scale_grad_by_freq, int64_t mode, int sparse, tensor per_sample_weights, int include_last_offset, int64_t padding_idx);
-void atg__embedding_bag_backward(tensor *, tensor grad, tensor indices, tensor offsets, tensor offset2bag, tensor bag_size, tensor maximum_indices, int64_t num_weights, int scale_grad_by_freq, int64_t mode, int sparse, tensor per_sample_weights, int64_t padding_idx);
-void atg__embedding_bag_dense_backward(tensor *, tensor grad, tensor indices, tensor offset2bag, tensor bag_size, tensor maximum_indices, int64_t num_weights, int scale_grad_by_freq, int64_t mode, tensor per_sample_weights, int64_t padding_idx);
-void atg__embedding_bag_dense_backward_out(tensor *, tensor out, tensor grad, tensor indices, tensor offset2bag, tensor bag_size, tensor maximum_indices, int64_t num_weights, int scale_grad_by_freq, int64_t mode, tensor per_sample_weights, int64_t padding_idx);
-void atg__embedding_bag_forward_only(tensor *, tensor weight, tensor indices, tensor offsets, int scale_grad_by_freq, int64_t mode, int sparse, tensor per_sample_weights, int include_last_offset, int64_t padding_idx);
-void atg__embedding_bag_forward_only_out(tensor *, tensor out0, tensor out1, tensor out2, tensor out3, tensor weight, tensor indices, tensor offsets, int scale_grad_by_freq, int64_t mode, int sparse, tensor per_sample_weights, int include_last_offset, int64_t padding_idx);
-void atg__embedding_bag_out(tensor *, tensor out0, tensor out1, tensor out2, tensor out3, tensor weight, tensor indices, tensor offsets, int scale_grad_by_freq, int64_t mode, int sparse, tensor per_sample_weights, int include_last_offset, int64_t padding_idx);
-void atg__embedding_bag_per_sample_weights_backward(tensor *, tensor grad, tensor weight, tensor indices, tensor offsets, tensor offset2bag, int64_t mode, int64_t padding_idx);
-void atg__embedding_bag_per_sample_weights_backward_out(tensor *, tensor out, tensor grad, tensor weight, tensor indices, tensor offsets, tensor offset2bag, int64_t mode, int64_t padding_idx);
-void atg__embedding_bag_sparse_backward(tensor *, tensor grad, tensor indices, tensor offsets, tensor offset2bag, tensor bag_size, int64_t num_weights, int scale_grad_by_freq, int64_t mode, tensor per_sample_weights, int64_t padding_idx);
-void atg__empty_affine_quantized(tensor *, int64_t *size_data, int size_len, int options_kind, int options_device, double scale, int64_t zero_point);
-void atg__empty_affine_quantized_out(tensor *, tensor out, int64_t *size_data, int size_len, double scale, int64_t zero_point);
-void atg__empty_per_channel_affine_quantized(tensor *, int64_t *size_data, int size_len, tensor scales, tensor zero_points, int64_t axis, int options_kind, int options_device);
-void atg__empty_per_channel_affine_quantized_out(tensor *, tensor out, int64_t *size_data, int size_len, tensor scales, tensor zero_points, int64_t axis);
-void atg__euclidean_dist(tensor *, tensor x1, tensor x2);
-void atg__euclidean_dist_out(tensor *, tensor out, tensor x1, tensor x2);
-void atg__fake_quantize_learnable_per_channel_affine(tensor *, tensor self, tensor scale, tensor zero_point, int64_t axis, int64_t quant_min, int64_t quant_max, double grad_factor);
-void atg__fake_quantize_learnable_per_channel_affine_backward(tensor *, tensor grad, tensor self, tensor scale, tensor zero_point, int64_t axis, int64_t quant_min, int64_t quant_max, double grad_factor);
-void atg__fake_quantize_learnable_per_channel_affine_out(tensor *, tensor out, tensor self, tensor scale, tensor zero_point, int64_t axis, int64_t quant_min, int64_t quant_max, double grad_factor);
-void atg__fake_quantize_learnable_per_tensor_affine(tensor *, tensor self, tensor scale, tensor zero_point, int64_t quant_min, int64_t quant_max, double grad_factor);
-void atg__fake_quantize_learnable_per_tensor_affine_backward(tensor *, tensor grad, tensor self, tensor scale, tensor zero_point, int64_t quant_min, int64_t quant_max, double grad_factor);
-void atg__fake_quantize_learnable_per_tensor_affine_out(tensor *, tensor out, tensor self, tensor scale, tensor zero_point, int64_t quant_min, int64_t quant_max, double grad_factor);
-void atg__fake_quantize_per_tensor_affine_cachemask_tensor_qparams(tensor *, tensor self, tensor scale, tensor zero_point, tensor fake_quant_enabled, int64_t quant_min, int64_t quant_max);
-void atg__fake_quantize_per_tensor_affine_cachemask_tensor_qparams_out(tensor *, tensor out0, tensor out1, tensor self, tensor scale, tensor zero_point, tensor fake_quant_enabled, int64_t quant_min, int64_t quant_max);
-void atg__fft_c2c(tensor *, tensor self, int64_t *dim_data, int dim_len, int64_t normalization, int forward);
-void atg__fft_c2c_out(tensor *, tensor out, tensor self, int64_t *dim_data, int dim_len, int64_t normalization, int forward);
-void atg__fft_c2r(tensor *, tensor self, int64_t *dim_data, int dim_len, int64_t normalization, int64_t last_dim_size);
-void atg__fft_c2r_out(tensor *, tensor out, tensor self, int64_t *dim_data, int dim_len, int64_t normalization, int64_t last_dim_size);
-void atg__fft_r2c(tensor *, tensor self, int64_t *dim_data, int dim_len, int64_t normalization, int onesided);
-void atg__fft_r2c_out(tensor *, tensor out, tensor self, int64_t *dim_data, int dim_len, int64_t normalization, int onesided);
-void atg__fill_mem_eff_dropout_mask_(tensor *, tensor self, double dropout_p, int64_t seed, int64_t offset);
-void atg__flash_attention_backward(tensor *, tensor grad_out, tensor query, tensor key, tensor value, tensor out, tensor logsumexp, tensor cum_seq_q, tensor cum_seq_k, int64_t max_q, int64_t max_k, double dropout_p, int is_causal, tensor philox_seed, tensor philox_offset, double scale_v, int scale_null);
-void atg__foobar(tensor *, tensor self, int arg1, int arg2, int arg3);
-void atg__foobar_out(tensor *, tensor out, tensor self, int arg1, int arg2, int arg3);
-void atg__functional_assert_async(tensor *, tensor self, char * assert_msg, tensor dep_token);
-void atg__functional_sym_constrain_range(tensor *, scalar size, int64_t min_v, int min_null, int64_t max_v, int max_null, tensor dep_token);
-void atg__functional_sym_constrain_range_for_size(tensor *, scalar size, int64_t min_v, int min_null, int64_t max_v, int max_null, tensor dep_token);
-void atg__fused_adam(tensor *out_data, int out_len, tensor *self_data, int self_len, tensor *grads_data, int grads_len, tensor *exp_avgs_data, int exp_avgs_len, tensor *exp_avg_sqs_data, int exp_avg_sqs_len, tensor *max_exp_avg_sqs_data, int max_exp_avg_sqs_len, tensor *state_steps_data, int state_steps_len, double lr, double beta1, double beta2, double weight_decay, double eps, int amsgrad, int maximize, tensor grad_scale, tensor found_inf);
-void atg__fused_adam_(tensor *self_data, int self_len, tensor *grads_data, int grads_len, tensor *exp_avgs_data, int exp_avgs_len, tensor *exp_avg_sqs_data, int exp_avg_sqs_len, tensor *max_exp_avg_sqs_data, int max_exp_avg_sqs_len, tensor *state_steps_data, int state_steps_len, double lr, double beta1, double beta2, double weight_decay, double eps, int amsgrad, int maximize, tensor grad_scale, tensor found_inf);
-void atg__fused_adam_tensor_lr_(tensor *self_data, int self_len, tensor *grads_data, int grads_len, tensor *exp_avgs_data, int exp_avgs_len, tensor *exp_avg_sqs_data, int exp_avg_sqs_len, tensor *max_exp_avg_sqs_data, int max_exp_avg_sqs_len, tensor *state_steps_data, int state_steps_len, tensor lr, double beta1, double beta2, double weight_decay, double eps, int amsgrad, int maximize, tensor grad_scale, tensor found_inf);
-void atg__fused_adam_tensor_lr_out(tensor *out_data, int out_len, tensor *self_data, int self_len, tensor *grads_data, int grads_len, tensor *exp_avgs_data, int exp_avgs_len, tensor *exp_avg_sqs_data, int exp_avg_sqs_len, tensor *max_exp_avg_sqs_data, int max_exp_avg_sqs_len, tensor *state_steps_data, int state_steps_len, tensor lr, double beta1, double beta2, double weight_decay, double eps, int amsgrad, int maximize, tensor grad_scale, tensor found_inf);
-void atg__fused_adamw(tensor *out_data, int out_len, tensor *self_data, int self_len, tensor *grads_data, int grads_len, tensor *exp_avgs_data, int exp_avgs_len, tensor *exp_avg_sqs_data, int exp_avg_sqs_len, tensor *max_exp_avg_sqs_data, int max_exp_avg_sqs_len, tensor *state_steps_data, int state_steps_len, double lr, double beta1, double beta2, double weight_decay, double eps, int amsgrad, int maximize, tensor grad_scale, tensor found_inf);
-void atg__fused_adamw_(tensor *self_data, int self_len, tensor *grads_data, int grads_len, tensor *exp_avgs_data, int exp_avgs_len, tensor *exp_avg_sqs_data, int exp_avg_sqs_len, tensor *max_exp_avg_sqs_data, int max_exp_avg_sqs_len, tensor *state_steps_data, int state_steps_len, double lr, double beta1, double beta2, double weight_decay, double eps, int amsgrad, int maximize, tensor grad_scale, tensor found_inf);
-void atg__fused_adamw_tensor_lr_(tensor *self_data, int self_len, tensor *grads_data, int grads_len, tensor *exp_avgs_data, int exp_avgs_len, tensor *exp_avg_sqs_data, int exp_avg_sqs_len, tensor *max_exp_avg_sqs_data, int max_exp_avg_sqs_len, tensor *state_steps_data, int state_steps_len, tensor lr, double beta1, double beta2, double weight_decay, double eps, int amsgrad, int maximize, tensor grad_scale, tensor found_inf);
-void atg__fused_adamw_tensor_lr_out(tensor *out_data, int out_len, tensor *self_data, int self_len, tensor *grads_data, int grads_len, tensor *exp_avgs_data, int exp_avgs_len, tensor *exp_avg_sqs_data, int exp_avg_sqs_len, tensor *max_exp_avg_sqs_data, int max_exp_avg_sqs_len, tensor *state_steps_data, int state_steps_len, tensor lr, double beta1, double beta2, double weight_decay, double eps, int amsgrad, int maximize, tensor grad_scale, tensor found_inf);
-void atg__fused_dropout(tensor *, tensor self, double p);
-void atg__fused_dropout_out(tensor *, tensor out0, tensor out1, tensor self, double p);
-void atg__fused_moving_avg_obs_fq_helper(tensor *, tensor self, tensor observer_on, tensor fake_quant_on, tensor running_min, tensor running_max, tensor scale, tensor zero_point, double averaging_const, int64_t quant_min, int64_t quant_max, int64_t ch_axis, int per_row_fake_quant, int symmetric_quant);
-void atg__fused_moving_avg_obs_fq_helper_functional(tensor *, tensor self, tensor observer_on, tensor fake_quant_on, tensor running_min, tensor running_max, tensor scale, tensor zero_point, double averaging_const, int64_t quant_min, int64_t quant_max, int64_t ch_axis, int per_row_fake_quant, int symmetric_quant);
-void atg__fused_moving_avg_obs_fq_helper_out(tensor *, tensor out0, tensor out1, tensor self, tensor observer_on, tensor fake_quant_on, tensor running_min, tensor running_max, tensor scale, tensor zero_point, double averaging_const, int64_t quant_min, int64_t quant_max, int64_t ch_axis, int per_row_fake_quant, int symmetric_quant);
-int64_t atg__fused_sdp_choice(tensor query, tensor key, tensor value, tensor attn_mask, double dropout_p, int is_causal, double scale_v, int scale_null);
-void atg__fw_primal(tensor *, tensor self, int64_t level);
-void atg__fw_primal_copy(tensor *, tensor self, int64_t level);
-void atg__fw_primal_copy_out(tensor *, tensor out, tensor self, int64_t level);
-void atg__gather_sparse_backward(tensor *, tensor self, int64_t dim, tensor index, tensor grad);
-void atg__grid_sampler_2d_cpu_fallback(tensor *, tensor input, tensor grid, int64_t interpolation_mode, int64_t padding_mode, int align_corners);
-void atg__grid_sampler_2d_cpu_fallback_backward(tensor *, tensor grad_output, tensor input, tensor grid, int64_t interpolation_mode, int64_t padding_mode, int align_corners);
-void atg__grid_sampler_2d_cpu_fallback_out(tensor *, tensor out, tensor input, tensor grid, int64_t interpolation_mode, int64_t padding_mode, int align_corners);
-int atg__has_compatible_shallow_copy_type(tensor self, tensor from);
-int atg__has_same_storage_numel(tensor self, tensor other);
-tensor *atg__histogramdd_bin_edges(tensor self, int64_t *bins_data, int bins_len, double *range_data, int range_len, tensor weight, int density);
-void atg__histogramdd_bin_edges_out(tensor *out_data, int out_len, tensor self, int64_t *bins_data, int bins_len, double *range_data, int range_len, tensor weight, int density);
-void atg__histogramdd_from_bin_cts(tensor *, tensor self, int64_t *bins_data, int bins_len, double *range_data, int range_len, tensor weight, int density);
-void atg__histogramdd_from_bin_cts_out(tensor *, tensor out, tensor self, int64_t *bins_data, int bins_len, double *range_data, int range_len, tensor weight, int density);
-void atg__histogramdd_from_bin_tensors(tensor *, tensor self, tensor *bins_data, int bins_len, tensor weight, int density);
-void atg__histogramdd_from_bin_tensors_out(tensor *, tensor out, tensor self, tensor *bins_data, int bins_len, tensor weight, int density);
-void atg__index_put_impl(tensor *, tensor self, tensor *indices_data, int indices_len, tensor values, int accumulate, int unsafe);
-void atg__index_put_impl_(tensor *, tensor self, tensor *indices_data, int indices_len, tensor values, int accumulate, int unsafe);
-void atg__index_put_impl_out(tensor *, tensor out, tensor self, tensor *indices_data, int indices_len, tensor values, int accumulate, int unsafe);
-void atg__indices(tensor *, tensor self);
-void atg__indices_copy(tensor *, tensor self);
-void atg__indices_copy_out(tensor *, tensor out, tensor self);
-void atg__int_mm(tensor *, tensor self, tensor mat2);
-void atg__int_mm_out(tensor *, tensor out, tensor self, tensor mat2);
-void atg__is_all_true(tensor *, tensor self);
-void atg__is_any_true(tensor *, tensor self);
-int atg__is_zerotensor(tensor self);
-void atg__linalg_check_errors(tensor info, char * api_name, int is_matrix);
-void atg__linalg_det(tensor *, tensor A);
-void atg__linalg_det_result(tensor *, tensor result, tensor LU, tensor pivots, tensor A);
-void atg__linalg_eigh(tensor *, tensor A, char * UPLO, int compute_v);
-void atg__linalg_eigh_eigenvalues(tensor *, tensor eigenvalues, tensor eigenvectors, tensor A, char * UPLO, int compute_v);
-void atg__linalg_slogdet(tensor *, tensor A);
-void atg__linalg_slogdet_sign(tensor *, tensor sign, tensor logabsdet, tensor LU, tensor pivots, tensor A);
-void atg__linalg_solve_ex(tensor *, tensor A, tensor B, int left, int check_errors);
-void atg__linalg_solve_ex_result(tensor *, tensor result, tensor LU, tensor pivots, tensor info, tensor A, tensor B, int left, int check_errors);
-void atg__linalg_svd(tensor *, tensor A, int full_matrices, int compute_uv, char * driver);
-void atg__linalg_svd_u(tensor *, tensor U, tensor S, tensor Vh, tensor A, int full_matrices, int compute_uv, char * driver);
-void atg__log_softmax(tensor *, tensor self, int64_t dim, int half_to_float);
-void atg__log_softmax_backward_data(tensor *, tensor grad_output, tensor output, int64_t dim, int input_dtype);
-void atg__log_softmax_backward_data_out(tensor *, tensor out, tensor grad_output, tensor output, int64_t dim, int input_dtype);
-void atg__log_softmax_out(tensor *, tensor out, tensor self, int64_t dim, int half_to_float);
-void atg__logcumsumexp(tensor *, tensor self, int64_t dim);
-void atg__logcumsumexp_out(tensor *, tensor out, tensor self, int64_t dim);
-void atg__lstm_mps(tensor *, tensor input, tensor *hx_data, int hx_len, tensor *params_data, int params_len, int has_biases, int64_t num_layers, double dropout, int train, int bidirectional, int batch_first);
-void atg__lstm_mps_out(tensor *, tensor out0, tensor out1, tensor out2, tensor out3, tensor out4, tensor out5, tensor input, tensor *hx_data, int hx_len, tensor *params_data, int params_len, int has_biases, int64_t num_layers, double dropout, int train, int bidirectional, int batch_first);
-void atg__lu_with_info(tensor *, tensor self, int pivot, int check_errors);
-void atg__make_dep_token(tensor *, int options_kind, int options_device);
-void atg__make_dual(tensor *, tensor primal, tensor tangent, int64_t level);
-void atg__make_dual_copy(tensor *, tensor primal, tensor tangent, int64_t level);
-void atg__make_dual_copy_out(tensor *, tensor out, tensor primal, tensor tangent, int64_t level);
-void atg__make_per_channel_quantized_tensor(tensor *, tensor self, tensor scale, tensor zero_point, int64_t axis);
-void atg__make_per_channel_quantized_tensor_out(tensor *, tensor out, tensor self, tensor scale, tensor zero_point, int64_t axis);
-void atg__make_per_tensor_quantized_tensor(tensor *, tensor self, double scale, int64_t zero_point);
-void atg__make_per_tensor_quantized_tensor_out(tensor *, tensor out, tensor self, double scale, int64_t zero_point);
-void atg__masked_scale(tensor *, tensor self, tensor mask, double scale);
-void atg__masked_scale_out(tensor *, tensor out, tensor self, tensor mask, double scale);
-void atg__masked_softmax(tensor *, tensor self, tensor mask, int64_t dim_v, int dim_null, int64_t mask_type_v, int mask_type_null);
-void atg__masked_softmax_backward(tensor *, tensor grad_output, tensor output, tensor mask, int64_t dim_v, int dim_null);
-void atg__masked_softmax_backward_out(tensor *, tensor out, tensor grad_output, tensor output, tensor mask, int64_t dim_v, int dim_null);
-void atg__masked_softmax_out(tensor *, tensor out, tensor self, tensor mask, int64_t dim_v, int dim_null, int64_t mask_type_v, int mask_type_null);
-void atg__mkldnn_reshape(tensor *, tensor self, int64_t *shape_data, int shape_len);
-void atg__mkldnn_reshape_out(tensor *, tensor out, tensor self, int64_t *shape_data, int shape_len);
-void atg__mkldnn_transpose(tensor *, tensor self, int64_t dim0, int64_t dim1);
-void atg__mkldnn_transpose_(tensor *, tensor self, int64_t dim0, int64_t dim1);
-void atg__mkldnn_transpose_out(tensor *, tensor out, tensor self, int64_t dim0, int64_t dim1);
-void atg__mps_convolution(tensor *, tensor self, tensor weight, tensor bias, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups);
-void atg__mps_convolution_out(tensor *, tensor out, tensor self, tensor weight, tensor bias, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups);
-void atg__mps_convolution_transpose(tensor *, tensor self, tensor weight, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups);
-void atg__mps_convolution_transpose_out(tensor *, tensor out, tensor self, tensor weight, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups);
-void atg__native_batch_norm_legit(tensor *, tensor input, tensor weight, tensor bias, tensor running_mean, tensor running_var, int training, double momentum, double eps);
-void atg__native_batch_norm_legit_functional(tensor *, tensor input, tensor weight, tensor bias, tensor running_mean, tensor running_var, int training, double momentum, double eps);
-void atg__native_batch_norm_legit_no_stats(tensor *, tensor input, tensor weight, tensor bias, int training, double momentum, double eps);
-void atg__native_batch_norm_legit_no_stats_out(tensor *, tensor out, tensor save_mean, tensor save_invstd, tensor input, tensor weight, tensor bias, int training, double momentum, double eps);
-void atg__native_batch_norm_legit_no_training(tensor *, tensor input, tensor weight, tensor bias, tensor running_mean, tensor running_var, double momentum, double eps);
-void atg__native_batch_norm_legit_no_training_out(tensor *, tensor out0, tensor out1, tensor out2, tensor input, tensor weight, tensor bias, tensor running_mean, tensor running_var, double momentum, double eps);
-void atg__native_batch_norm_legit_out(tensor *, tensor out, tensor save_mean, tensor save_invstd, tensor input, tensor weight, tensor bias, tensor running_mean, tensor running_var, int training, double momentum, double eps);
-void atg__native_multi_head_attention(tensor *, tensor query, tensor key, tensor value, int64_t embed_dim, int64_t num_head, tensor qkv_weight, tensor qkv_bias, tensor proj_weight, tensor proj_bias, tensor mask, int need_weights, int average_attn_weights, int64_t mask_type_v, int mask_type_null);
-void atg__native_multi_head_attention_out(tensor *, tensor out0, tensor out1, tensor query, tensor key, tensor value, int64_t embed_dim, int64_t num_head, tensor qkv_weight, tensor qkv_bias, tensor proj_weight, tensor proj_bias, tensor mask, int need_weights, int average_attn_weights, int64_t mask_type_v, int mask_type_null);
-void atg__neg_view(tensor *, tensor self);
-void atg__neg_view_copy(tensor *, tensor self);
-void atg__neg_view_copy_out(tensor *, tensor out, tensor self);
-void atg__nested_from_padded(tensor *, tensor padded, tensor cpu_nested_shape_example, int fuse_transform_0213);
-void atg__nested_from_padded_and_nested_example(tensor *, tensor padded, tensor nt_example);
-void atg__nested_from_padded_and_nested_example_out(tensor *, tensor out, tensor padded, tensor nt_example);
-void atg__nested_from_padded_out(tensor *, tensor out, tensor padded, tensor cpu_nested_shape_example, int fuse_transform_0213);
-void atg__nested_select_backward(tensor *, tensor grad_output, tensor self, int64_t dim, int64_t index);
-void atg__nested_sum_backward(tensor *, tensor grad, tensor self, int64_t *dim_data, int dim_len, int keepdim);
-void atg__nested_view_from_buffer(tensor *, tensor self, tensor nested_size, tensor nested_strides, tensor offsets);
-void atg__nested_view_from_buffer_copy(tensor *, tensor self, tensor nested_size, tensor nested_strides, tensor offsets);
-void atg__nested_view_from_buffer_copy_out(tensor *, tensor out, tensor self, tensor nested_size, tensor nested_strides, tensor offsets);
-void atg__new_zeros_with_same_feature_meta(tensor *, tensor self, tensor other, int64_t self_num_batch_dims);
-void atg__new_zeros_with_same_feature_meta_out(tensor *, tensor out, tensor self, tensor other, int64_t self_num_batch_dims);
+raw_tensor atg___and__(gc_tensor self, scalar other);
+raw_tensor atg___and__tensor_(gc_tensor self, gc_tensor other);
+raw_tensor atg___iand__(gc_tensor self, scalar other);
+raw_tensor atg___iand__tensor_(gc_tensor self, gc_tensor other);
+raw_tensor atg___ilshift__(gc_tensor self, scalar other);
+raw_tensor atg___ilshift__tensor_(gc_tensor self, gc_tensor other);
+raw_tensor atg___ior__(gc_tensor self, scalar other);
+raw_tensor atg___ior__tensor_(gc_tensor self, gc_tensor other);
+raw_tensor atg___irshift__(gc_tensor self, scalar other);
+raw_tensor atg___irshift__tensor_(gc_tensor self, gc_tensor other);
+raw_tensor atg___ixor__(gc_tensor self, scalar other);
+raw_tensor atg___ixor__tensor_(gc_tensor self, gc_tensor other);
+raw_tensor atg___lshift__(gc_tensor self, scalar other);
+raw_tensor atg___lshift__scalar_out_(gc_tensor out, gc_tensor self, scalar other);
+raw_tensor atg___lshift__tensor_(gc_tensor self, gc_tensor other);
+raw_tensor atg___lshift__tensor_out_(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg___or__(gc_tensor self, scalar other);
+raw_tensor atg___or__tensor_(gc_tensor self, gc_tensor other);
+raw_tensor atg___rshift__(gc_tensor self, scalar other);
+raw_tensor atg___rshift__scalar_out_(gc_tensor out, gc_tensor self, scalar other);
+raw_tensor atg___rshift__tensor_(gc_tensor self, gc_tensor other);
+raw_tensor atg___rshift__tensor_out_(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg___xor__(gc_tensor self, scalar other);
+raw_tensor atg___xor__tensor_(gc_tensor self, gc_tensor other);
+raw_tensor atg__adaptive_avg_pool2d(gc_tensor self, int64_t *output_size_data, int output_size_len);
+raw_tensor atg__adaptive_avg_pool2d_backward(gc_tensor grad_output, gc_tensor self);
+raw_tensor atg__adaptive_avg_pool2d_backward_out(gc_tensor out, gc_tensor grad_output, gc_tensor self);
+raw_tensor atg__adaptive_avg_pool2d_out(gc_tensor out, gc_tensor self, int64_t *output_size_data, int output_size_len);
+raw_tensor atg__adaptive_avg_pool3d(gc_tensor self, int64_t *output_size_data, int output_size_len);
+raw_tensor atg__adaptive_avg_pool3d_backward(gc_tensor grad_output, gc_tensor self);
+raw_tensor atg__adaptive_avg_pool3d_backward_out(gc_tensor out, gc_tensor grad_output, gc_tensor self);
+raw_tensor atg__adaptive_avg_pool3d_out(gc_tensor out, gc_tensor self, int64_t *output_size_data, int output_size_len);
+raw_tensor atg__add_batch_dim(gc_tensor self, int64_t batch_dim, int64_t level);
+raw_tensor atg__add_relu(gc_tensor self, gc_tensor other);
+raw_tensor atg__add_relu_(gc_tensor self, gc_tensor other);
+raw_tensor atg__add_relu_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg__add_relu_scalar(gc_tensor self, scalar other);
+raw_tensor atg__add_relu_scalar_(gc_tensor self, scalar other);
+raw_tensor atg__add_relu_scalar_out(gc_tensor out, gc_tensor self, scalar other);
+raw_tensor atg__addmm_activation(gc_tensor self, gc_tensor mat1, gc_tensor mat2, int use_gelu);
+raw_tensor atg__addmm_activation_out(gc_tensor out, gc_tensor self, gc_tensor mat1, gc_tensor mat2, int use_gelu);
+void atg__aminmax(raw_tensor *, gc_tensor self);
+void atg__aminmax_dim(raw_tensor *, gc_tensor self, int64_t dim, int keepdim);
+void atg__aminmax_dim_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor self, int64_t dim, int keepdim);
+void atg__aminmax_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor self);
+void atg__amp_update_scale(raw_tensor *, gc_tensor self, gc_tensor growth_tracker, gc_tensor found_inf, double scale_growth_factor, double scale_backoff_factor, int64_t growth_interval);
+raw_tensor atg__amp_update_scale_(gc_tensor self, gc_tensor growth_tracker, gc_tensor found_inf, double scale_growth_factor, double scale_backoff_factor, int64_t growth_interval);
+raw_tensor atg__amp_update_scale_out(gc_tensor out, gc_tensor self, gc_tensor growth_tracker, gc_tensor found_inf, double scale_growth_factor, double scale_backoff_factor, int64_t growth_interval);
+void atg__assert_tensor_metadata(gc_tensor a, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len, int dtype);
+raw_tensor atg__autocast_to_full_precision(gc_tensor self, int cuda_enabled, int cpu_enabled);
+raw_tensor atg__autocast_to_reduced_precision(gc_tensor self, int cuda_enabled, int cpu_enabled, int cuda_dtype, int cpu_dtype);
+raw_tensor atg__cast_byte(gc_tensor self, int non_blocking);
+raw_tensor atg__cast_char(gc_tensor self, int non_blocking);
+raw_tensor atg__cast_double(gc_tensor self, int non_blocking);
+raw_tensor atg__cast_float(gc_tensor self, int non_blocking);
+raw_tensor atg__cast_half(gc_tensor self, int non_blocking);
+raw_tensor atg__cast_int(gc_tensor self, int non_blocking);
+raw_tensor atg__cast_long(gc_tensor self, int non_blocking);
+raw_tensor atg__cast_short(gc_tensor self, int non_blocking);
+raw_tensor atg__cdist_backward(gc_tensor grad, gc_tensor x1, gc_tensor x2, double p, gc_tensor cdist);
+raw_tensor atg__cdist_backward_out(gc_tensor out, gc_tensor grad, gc_tensor x1, gc_tensor x2, double p, gc_tensor cdist);
+raw_tensor atg__cholesky_solve_helper(gc_tensor self, gc_tensor A, int upper);
+raw_tensor atg__cholesky_solve_helper_out(gc_tensor out, gc_tensor self, gc_tensor A, int upper);
+raw_tensor atg__coalesce(gc_tensor self);
+raw_tensor atg__coalesce_out(gc_tensor out, gc_tensor self);
+raw_tensor atg__coalesced(gc_tensor self, int coalesced);
+raw_tensor atg__coalesced_(gc_tensor self, int coalesced);
+raw_tensor atg__coalesced_out(gc_tensor out, gc_tensor self, int coalesced);
+raw_tensor atg__compute_linear_combination(gc_tensor input, gc_tensor coefficients);
+raw_tensor atg__compute_linear_combination_out(gc_tensor out, gc_tensor input, gc_tensor coefficients);
+raw_tensor atg__conj(gc_tensor self);
+raw_tensor atg__conj_copy(gc_tensor self);
+raw_tensor atg__conj_copy_out(gc_tensor out, gc_tensor self);
+raw_tensor atg__conj_physical(gc_tensor self);
+raw_tensor atg__conj_physical_out(gc_tensor out, gc_tensor self);
+raw_tensor atg__conv_depthwise2d(gc_tensor self, gc_tensor weight, int64_t *kernel_size_data, int kernel_size_len, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len);
+raw_tensor atg__conv_depthwise2d_out(gc_tensor out, gc_tensor self, gc_tensor weight, int64_t *kernel_size_data, int kernel_size_len, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len);
+raw_tensor atg__convert_indices_from_coo_to_csr(gc_tensor self, int64_t size, int out_int32);
+raw_tensor atg__convert_indices_from_coo_to_csr_out(gc_tensor out, gc_tensor self, int64_t size, int out_int32);
+raw_tensor atg__convert_indices_from_csr_to_coo(gc_tensor crow_indices, gc_tensor col_indices, int out_int32, int transpose);
+raw_tensor atg__convert_indices_from_csr_to_coo_out(gc_tensor out, gc_tensor crow_indices, gc_tensor col_indices, int out_int32, int transpose);
+raw_tensor atg__convolution(gc_tensor input, gc_tensor weight, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int transposed, int64_t *output_padding_data, int output_padding_len, int64_t groups, int benchmark, int deterministic, int cudnn_enabled, int allow_tf32);
+raw_tensor atg__convolution_deprecated(gc_tensor input, gc_tensor weight, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int transposed, int64_t *output_padding_data, int output_padding_len, int64_t groups, int benchmark, int deterministic, int cudnn_enabled);
+raw_tensor atg__convolution_mode(gc_tensor input, gc_tensor weight, gc_tensor bias, int64_t *stride_data, int stride_len, char * padding, int64_t *dilation_data, int dilation_len, int64_t groups);
+raw_tensor atg__convolution_out(gc_tensor out, gc_tensor input, gc_tensor weight, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int transposed, int64_t *output_padding_data, int output_padding_len, int64_t groups, int benchmark, int deterministic, int cudnn_enabled, int allow_tf32);
+raw_tensor atg__copy_from(gc_tensor self, gc_tensor dst, int non_blocking);
+raw_tensor atg__copy_from_and_resize(gc_tensor self, gc_tensor dst);
+raw_tensor atg__copy_from_and_resize_out(gc_tensor out, gc_tensor self, gc_tensor dst);
+raw_tensor atg__copy_from_out(gc_tensor out, gc_tensor self, gc_tensor dst, int non_blocking);
+raw_tensor atg__cslt_compress(gc_tensor input);
+raw_tensor atg__cslt_sparse_mm(gc_tensor compressed_A, gc_tensor dense_B, gc_tensor bias, int transpose_result);
+void atg__ctc_loss(raw_tensor *, gc_tensor log_probs, gc_tensor targets, int64_t *input_lengths_data, int input_lengths_len, int64_t *target_lengths_data, int target_lengths_len, int64_t blank, int zero_infinity);
+raw_tensor atg__ctc_loss_backward(gc_tensor grad, gc_tensor log_probs, gc_tensor targets, int64_t *input_lengths_data, int input_lengths_len, int64_t *target_lengths_data, int target_lengths_len, gc_tensor neg_log_likelihood, gc_tensor log_alpha, int64_t blank, int zero_infinity);
+raw_tensor atg__ctc_loss_backward_out(gc_tensor out, gc_tensor grad, gc_tensor log_probs, gc_tensor targets, int64_t *input_lengths_data, int input_lengths_len, int64_t *target_lengths_data, int target_lengths_len, gc_tensor neg_log_likelihood, gc_tensor log_alpha, int64_t blank, int zero_infinity);
+raw_tensor atg__ctc_loss_backward_tensor(gc_tensor grad, gc_tensor log_probs, gc_tensor targets, gc_tensor input_lengths, gc_tensor target_lengths, gc_tensor neg_log_likelihood, gc_tensor log_alpha, int64_t blank, int zero_infinity);
+void atg__ctc_loss_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor log_probs, gc_tensor targets, int64_t *input_lengths_data, int input_lengths_len, int64_t *target_lengths_data, int target_lengths_len, int64_t blank, int zero_infinity);
+void atg__ctc_loss_tensor(raw_tensor *, gc_tensor log_probs, gc_tensor targets, gc_tensor input_lengths, gc_tensor target_lengths, int64_t blank, int zero_infinity);
+void atg__ctc_loss_tensor_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor log_probs, gc_tensor targets, gc_tensor input_lengths, gc_tensor target_lengths, int64_t blank, int zero_infinity);
+void atg__cudnn_ctc_loss(raw_tensor *, gc_tensor log_probs, gc_tensor targets, int64_t *input_lengths_data, int input_lengths_len, int64_t *target_lengths_data, int target_lengths_len, int64_t blank, int deterministic, int zero_infinity);
+void atg__cudnn_ctc_loss_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor log_probs, gc_tensor targets, int64_t *input_lengths_data, int input_lengths_len, int64_t *target_lengths_data, int target_lengths_len, int64_t blank, int deterministic, int zero_infinity);
+void atg__cudnn_ctc_loss_tensor(raw_tensor *, gc_tensor log_probs, gc_tensor targets, gc_tensor input_lengths, gc_tensor target_lengths, int64_t blank, int deterministic, int zero_infinity);
+raw_tensor atg__cudnn_init_dropout_state(double dropout, int train, int64_t dropout_seed, int options_kind, int options_device);
+raw_tensor atg__cudnn_init_dropout_state_out(gc_tensor out, double dropout, int train, int64_t dropout_seed);
+void atg__cudnn_rnn(raw_tensor *, gc_tensor input, gc_tensor *weight_data, int weight_len, int64_t weight_stride0, gc_tensor weight_buf, gc_tensor hx, gc_tensor cx, int64_t mode, int64_t hidden_size, int64_t proj_size, int64_t num_layers, int batch_first, double dropout, int train, int bidirectional, int64_t *batch_sizes_data, int batch_sizes_len, gc_tensor dropout_state);
+raw_tensor atg__cudnn_rnn_flatten_weight(gc_tensor *weight_arr_data, int weight_arr_len, int64_t weight_stride0, int64_t input_size, int64_t mode, int64_t hidden_size, int64_t proj_size, int64_t num_layers, int batch_first, int bidirectional);
+raw_tensor atg__cudnn_rnn_flatten_weight_out(gc_tensor out, gc_tensor *weight_arr_data, int weight_arr_len, int64_t weight_stride0, int64_t input_size, int64_t mode, int64_t hidden_size, int64_t proj_size, int64_t num_layers, int batch_first, int bidirectional);
+void atg__cudnn_rnn_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor out3, gc_tensor out4, gc_tensor input, gc_tensor *weight_data, int weight_len, int64_t weight_stride0, gc_tensor weight_buf, gc_tensor hx, gc_tensor cx, int64_t mode, int64_t hidden_size, int64_t proj_size, int64_t num_layers, int batch_first, double dropout, int train, int bidirectional, int64_t *batch_sizes_data, int batch_sizes_len, gc_tensor dropout_state);
+int64_t atg__debug_has_internal_overlap(gc_tensor self);
+raw_tensor atg__dim_arange(gc_tensor like, int64_t dim);
+int64_t atg__dimi(gc_tensor self);
+int64_t atg__dimv(gc_tensor self);
+raw_tensor atg__dirichlet_grad(gc_tensor x, gc_tensor alpha, gc_tensor total);
+raw_tensor atg__dirichlet_grad_out(gc_tensor out, gc_tensor x, gc_tensor alpha, gc_tensor total);
+void atg__efficient_attention_backward(raw_tensor *, gc_tensor grad_out_, gc_tensor query, gc_tensor key, gc_tensor value, gc_tensor bias, gc_tensor out, gc_tensor cu_seqlens_q, gc_tensor cu_seqlens_k, int64_t max_seqlen_k, int64_t max_seqlen_q, gc_tensor logsumexp, double dropout_p, gc_tensor philox_seed, gc_tensor philox_offset, int64_t custom_mask_type, int bias_requires_grad, double scale_v, int scale_null, int64_t num_splits_key_v, int num_splits_key_null);
+raw_tensor atg__efficientzerotensor(int64_t *size_data, int size_len, int options_kind, int options_device);
+raw_tensor atg__efficientzerotensor_out(gc_tensor out, int64_t *size_data, int size_len);
+void atg__embedding_bag(raw_tensor *, gc_tensor weight, gc_tensor indices, gc_tensor offsets, int scale_grad_by_freq, int64_t mode, int sparse, gc_tensor per_sample_weights, int include_last_offset, int64_t padding_idx);
+raw_tensor atg__embedding_bag_backward(gc_tensor grad, gc_tensor indices, gc_tensor offsets, gc_tensor offset2bag, gc_tensor bag_size, gc_tensor maximum_indices, int64_t num_weights, int scale_grad_by_freq, int64_t mode, int sparse, gc_tensor per_sample_weights, int64_t padding_idx);
+raw_tensor atg__embedding_bag_dense_backward(gc_tensor grad, gc_tensor indices, gc_tensor offset2bag, gc_tensor bag_size, gc_tensor maximum_indices, int64_t num_weights, int scale_grad_by_freq, int64_t mode, gc_tensor per_sample_weights, int64_t padding_idx);
+raw_tensor atg__embedding_bag_dense_backward_out(gc_tensor out, gc_tensor grad, gc_tensor indices, gc_tensor offset2bag, gc_tensor bag_size, gc_tensor maximum_indices, int64_t num_weights, int scale_grad_by_freq, int64_t mode, gc_tensor per_sample_weights, int64_t padding_idx);
+void atg__embedding_bag_forward_only(raw_tensor *, gc_tensor weight, gc_tensor indices, gc_tensor offsets, int scale_grad_by_freq, int64_t mode, int sparse, gc_tensor per_sample_weights, int include_last_offset, int64_t padding_idx);
+void atg__embedding_bag_forward_only_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor out3, gc_tensor weight, gc_tensor indices, gc_tensor offsets, int scale_grad_by_freq, int64_t mode, int sparse, gc_tensor per_sample_weights, int include_last_offset, int64_t padding_idx);
+void atg__embedding_bag_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor out3, gc_tensor weight, gc_tensor indices, gc_tensor offsets, int scale_grad_by_freq, int64_t mode, int sparse, gc_tensor per_sample_weights, int include_last_offset, int64_t padding_idx);
+raw_tensor atg__embedding_bag_per_sample_weights_backward(gc_tensor grad, gc_tensor weight, gc_tensor indices, gc_tensor offsets, gc_tensor offset2bag, int64_t mode, int64_t padding_idx);
+raw_tensor atg__embedding_bag_per_sample_weights_backward_out(gc_tensor out, gc_tensor grad, gc_tensor weight, gc_tensor indices, gc_tensor offsets, gc_tensor offset2bag, int64_t mode, int64_t padding_idx);
+raw_tensor atg__embedding_bag_sparse_backward(gc_tensor grad, gc_tensor indices, gc_tensor offsets, gc_tensor offset2bag, gc_tensor bag_size, int64_t num_weights, int scale_grad_by_freq, int64_t mode, gc_tensor per_sample_weights, int64_t padding_idx);
+raw_tensor atg__empty_affine_quantized(int64_t *size_data, int size_len, int options_kind, int options_device, double scale, int64_t zero_point);
+raw_tensor atg__empty_affine_quantized_out(gc_tensor out, int64_t *size_data, int size_len, double scale, int64_t zero_point);
+raw_tensor atg__empty_per_channel_affine_quantized(int64_t *size_data, int size_len, gc_tensor scales, gc_tensor zero_points, int64_t axis, int options_kind, int options_device);
+raw_tensor atg__empty_per_channel_affine_quantized_out(gc_tensor out, int64_t *size_data, int size_len, gc_tensor scales, gc_tensor zero_points, int64_t axis);
+raw_tensor atg__euclidean_dist(gc_tensor x1, gc_tensor x2);
+raw_tensor atg__euclidean_dist_out(gc_tensor out, gc_tensor x1, gc_tensor x2);
+raw_tensor atg__fake_quantize_learnable_per_channel_affine(gc_tensor self, gc_tensor scale, gc_tensor zero_point, int64_t axis, int64_t quant_min, int64_t quant_max, double grad_factor);
+void atg__fake_quantize_learnable_per_channel_affine_backward(raw_tensor *, gc_tensor grad, gc_tensor self, gc_tensor scale, gc_tensor zero_point, int64_t axis, int64_t quant_min, int64_t quant_max, double grad_factor);
+raw_tensor atg__fake_quantize_learnable_per_channel_affine_out(gc_tensor out, gc_tensor self, gc_tensor scale, gc_tensor zero_point, int64_t axis, int64_t quant_min, int64_t quant_max, double grad_factor);
+raw_tensor atg__fake_quantize_learnable_per_tensor_affine(gc_tensor self, gc_tensor scale, gc_tensor zero_point, int64_t quant_min, int64_t quant_max, double grad_factor);
+void atg__fake_quantize_learnable_per_tensor_affine_backward(raw_tensor *, gc_tensor grad, gc_tensor self, gc_tensor scale, gc_tensor zero_point, int64_t quant_min, int64_t quant_max, double grad_factor);
+raw_tensor atg__fake_quantize_learnable_per_tensor_affine_out(gc_tensor out, gc_tensor self, gc_tensor scale, gc_tensor zero_point, int64_t quant_min, int64_t quant_max, double grad_factor);
+void atg__fake_quantize_per_tensor_affine_cachemask_tensor_qparams(raw_tensor *, gc_tensor self, gc_tensor scale, gc_tensor zero_point, gc_tensor fake_quant_enabled, int64_t quant_min, int64_t quant_max);
+void atg__fake_quantize_per_tensor_affine_cachemask_tensor_qparams_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor self, gc_tensor scale, gc_tensor zero_point, gc_tensor fake_quant_enabled, int64_t quant_min, int64_t quant_max);
+raw_tensor atg__fft_c2c(gc_tensor self, int64_t *dim_data, int dim_len, int64_t normalization, int forward);
+raw_tensor atg__fft_c2c_out(gc_tensor out, gc_tensor self, int64_t *dim_data, int dim_len, int64_t normalization, int forward);
+raw_tensor atg__fft_c2r(gc_tensor self, int64_t *dim_data, int dim_len, int64_t normalization, int64_t last_dim_size);
+raw_tensor atg__fft_c2r_out(gc_tensor out, gc_tensor self, int64_t *dim_data, int dim_len, int64_t normalization, int64_t last_dim_size);
+raw_tensor atg__fft_r2c(gc_tensor self, int64_t *dim_data, int dim_len, int64_t normalization, int onesided);
+raw_tensor atg__fft_r2c_out(gc_tensor out, gc_tensor self, int64_t *dim_data, int dim_len, int64_t normalization, int onesided);
+raw_tensor atg__fill_mem_eff_dropout_mask_(gc_tensor self, double dropout_p, int64_t seed, int64_t offset);
+void atg__flash_attention_backward(raw_tensor *, gc_tensor grad_out, gc_tensor query, gc_tensor key, gc_tensor value, gc_tensor out, gc_tensor logsumexp, gc_tensor cum_seq_q, gc_tensor cum_seq_k, int64_t max_q, int64_t max_k, double dropout_p, int is_causal, gc_tensor philox_seed, gc_tensor philox_offset, double scale_v, int scale_null);
+raw_tensor atg__foobar(gc_tensor self, int arg1, int arg2, int arg3);
+raw_tensor atg__foobar_out(gc_tensor out, gc_tensor self, int arg1, int arg2, int arg3);
+raw_tensor atg__functional_assert_async(gc_tensor self, char * assert_msg, gc_tensor dep_token);
+raw_tensor atg__functional_sym_constrain_range(scalar size, int64_t min_v, int min_null, int64_t max_v, int max_null, gc_tensor dep_token);
+raw_tensor atg__functional_sym_constrain_range_for_size(scalar size, int64_t min_v, int min_null, int64_t max_v, int max_null, gc_tensor dep_token);
+void atg__fused_adam(gc_tensor *out_data, int out_len, gc_tensor *self_data, int self_len, gc_tensor *grads_data, int grads_len, gc_tensor *exp_avgs_data, int exp_avgs_len, gc_tensor *exp_avg_sqs_data, int exp_avg_sqs_len, gc_tensor *max_exp_avg_sqs_data, int max_exp_avg_sqs_len, gc_tensor *state_steps_data, int state_steps_len, double lr, double beta1, double beta2, double weight_decay, double eps, int amsgrad, int maximize, gc_tensor grad_scale, gc_tensor found_inf);
+void atg__fused_adam_(gc_tensor *self_data, int self_len, gc_tensor *grads_data, int grads_len, gc_tensor *exp_avgs_data, int exp_avgs_len, gc_tensor *exp_avg_sqs_data, int exp_avg_sqs_len, gc_tensor *max_exp_avg_sqs_data, int max_exp_avg_sqs_len, gc_tensor *state_steps_data, int state_steps_len, double lr, double beta1, double beta2, double weight_decay, double eps, int amsgrad, int maximize, gc_tensor grad_scale, gc_tensor found_inf);
+void atg__fused_adam_tensor_lr_(gc_tensor *self_data, int self_len, gc_tensor *grads_data, int grads_len, gc_tensor *exp_avgs_data, int exp_avgs_len, gc_tensor *exp_avg_sqs_data, int exp_avg_sqs_len, gc_tensor *max_exp_avg_sqs_data, int max_exp_avg_sqs_len, gc_tensor *state_steps_data, int state_steps_len, gc_tensor lr, double beta1, double beta2, double weight_decay, double eps, int amsgrad, int maximize, gc_tensor grad_scale, gc_tensor found_inf);
+void atg__fused_adam_tensor_lr_out(gc_tensor *out_data, int out_len, gc_tensor *self_data, int self_len, gc_tensor *grads_data, int grads_len, gc_tensor *exp_avgs_data, int exp_avgs_len, gc_tensor *exp_avg_sqs_data, int exp_avg_sqs_len, gc_tensor *max_exp_avg_sqs_data, int max_exp_avg_sqs_len, gc_tensor *state_steps_data, int state_steps_len, gc_tensor lr, double beta1, double beta2, double weight_decay, double eps, int amsgrad, int maximize, gc_tensor grad_scale, gc_tensor found_inf);
+void atg__fused_adamw(gc_tensor *out_data, int out_len, gc_tensor *self_data, int self_len, gc_tensor *grads_data, int grads_len, gc_tensor *exp_avgs_data, int exp_avgs_len, gc_tensor *exp_avg_sqs_data, int exp_avg_sqs_len, gc_tensor *max_exp_avg_sqs_data, int max_exp_avg_sqs_len, gc_tensor *state_steps_data, int state_steps_len, double lr, double beta1, double beta2, double weight_decay, double eps, int amsgrad, int maximize, gc_tensor grad_scale, gc_tensor found_inf);
+void atg__fused_adamw_(gc_tensor *self_data, int self_len, gc_tensor *grads_data, int grads_len, gc_tensor *exp_avgs_data, int exp_avgs_len, gc_tensor *exp_avg_sqs_data, int exp_avg_sqs_len, gc_tensor *max_exp_avg_sqs_data, int max_exp_avg_sqs_len, gc_tensor *state_steps_data, int state_steps_len, double lr, double beta1, double beta2, double weight_decay, double eps, int amsgrad, int maximize, gc_tensor grad_scale, gc_tensor found_inf);
+void atg__fused_adamw_tensor_lr_(gc_tensor *self_data, int self_len, gc_tensor *grads_data, int grads_len, gc_tensor *exp_avgs_data, int exp_avgs_len, gc_tensor *exp_avg_sqs_data, int exp_avg_sqs_len, gc_tensor *max_exp_avg_sqs_data, int max_exp_avg_sqs_len, gc_tensor *state_steps_data, int state_steps_len, gc_tensor lr, double beta1, double beta2, double weight_decay, double eps, int amsgrad, int maximize, gc_tensor grad_scale, gc_tensor found_inf);
+void atg__fused_adamw_tensor_lr_out(gc_tensor *out_data, int out_len, gc_tensor *self_data, int self_len, gc_tensor *grads_data, int grads_len, gc_tensor *exp_avgs_data, int exp_avgs_len, gc_tensor *exp_avg_sqs_data, int exp_avg_sqs_len, gc_tensor *max_exp_avg_sqs_data, int max_exp_avg_sqs_len, gc_tensor *state_steps_data, int state_steps_len, gc_tensor lr, double beta1, double beta2, double weight_decay, double eps, int amsgrad, int maximize, gc_tensor grad_scale, gc_tensor found_inf);
+void atg__fused_dropout(raw_tensor *, gc_tensor self, double p);
+void atg__fused_dropout_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor self, double p);
+void atg__fused_moving_avg_obs_fq_helper(raw_tensor *, gc_tensor self, gc_tensor observer_on, gc_tensor fake_quant_on, gc_tensor running_min, gc_tensor running_max, gc_tensor scale, gc_tensor zero_point, double averaging_const, int64_t quant_min, int64_t quant_max, int64_t ch_axis, int per_row_fake_quant, int symmetric_quant);
+void atg__fused_moving_avg_obs_fq_helper_functional(raw_tensor *, gc_tensor self, gc_tensor observer_on, gc_tensor fake_quant_on, gc_tensor running_min, gc_tensor running_max, gc_tensor scale, gc_tensor zero_point, double averaging_const, int64_t quant_min, int64_t quant_max, int64_t ch_axis, int per_row_fake_quant, int symmetric_quant);
+void atg__fused_moving_avg_obs_fq_helper_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor self, gc_tensor observer_on, gc_tensor fake_quant_on, gc_tensor running_min, gc_tensor running_max, gc_tensor scale, gc_tensor zero_point, double averaging_const, int64_t quant_min, int64_t quant_max, int64_t ch_axis, int per_row_fake_quant, int symmetric_quant);
+int64_t atg__fused_sdp_choice(gc_tensor query, gc_tensor key, gc_tensor value, gc_tensor attn_mask, double dropout_p, int is_causal, double scale_v, int scale_null);
+raw_tensor atg__fw_primal(gc_tensor self, int64_t level);
+raw_tensor atg__fw_primal_copy(gc_tensor self, int64_t level);
+raw_tensor atg__fw_primal_copy_out(gc_tensor out, gc_tensor self, int64_t level);
+raw_tensor atg__gather_sparse_backward(gc_tensor self, int64_t dim, gc_tensor index, gc_tensor grad);
+raw_tensor atg__grid_sampler_2d_cpu_fallback(gc_tensor input, gc_tensor grid, int64_t interpolation_mode, int64_t padding_mode, int align_corners);
+void atg__grid_sampler_2d_cpu_fallback_backward(raw_tensor *, gc_tensor grad_output, gc_tensor input, gc_tensor grid, int64_t interpolation_mode, int64_t padding_mode, int align_corners);
+raw_tensor atg__grid_sampler_2d_cpu_fallback_out(gc_tensor out, gc_tensor input, gc_tensor grid, int64_t interpolation_mode, int64_t padding_mode, int align_corners);
+int atg__has_compatible_shallow_copy_type(gc_tensor self, gc_tensor from);
+int atg__has_same_storage_numel(gc_tensor self, gc_tensor other);
+raw_tensor *atg__histogramdd_bin_edges(gc_tensor self, int64_t *bins_data, int bins_len, double *range_data, int range_len, gc_tensor weight, int density);
+void atg__histogramdd_bin_edges_out(gc_tensor *out_data, int out_len, gc_tensor self, int64_t *bins_data, int bins_len, double *range_data, int range_len, gc_tensor weight, int density);
+raw_tensor atg__histogramdd_from_bin_cts(gc_tensor self, int64_t *bins_data, int bins_len, double *range_data, int range_len, gc_tensor weight, int density);
+raw_tensor atg__histogramdd_from_bin_cts_out(gc_tensor out, gc_tensor self, int64_t *bins_data, int bins_len, double *range_data, int range_len, gc_tensor weight, int density);
+raw_tensor atg__histogramdd_from_bin_tensors(gc_tensor self, gc_tensor *bins_data, int bins_len, gc_tensor weight, int density);
+raw_tensor atg__histogramdd_from_bin_tensors_out(gc_tensor out, gc_tensor self, gc_tensor *bins_data, int bins_len, gc_tensor weight, int density);
+raw_tensor atg__index_put_impl(gc_tensor self, gc_tensor *indices_data, int indices_len, gc_tensor values, int accumulate, int unsafe);
+raw_tensor atg__index_put_impl_(gc_tensor self, gc_tensor *indices_data, int indices_len, gc_tensor values, int accumulate, int unsafe);
+raw_tensor atg__index_put_impl_out(gc_tensor out, gc_tensor self, gc_tensor *indices_data, int indices_len, gc_tensor values, int accumulate, int unsafe);
+raw_tensor atg__indices(gc_tensor self);
+raw_tensor atg__indices_copy(gc_tensor self);
+raw_tensor atg__indices_copy_out(gc_tensor out, gc_tensor self);
+raw_tensor atg__int_mm(gc_tensor self, gc_tensor mat2);
+raw_tensor atg__int_mm_out(gc_tensor out, gc_tensor self, gc_tensor mat2);
+raw_tensor atg__is_all_true(gc_tensor self);
+raw_tensor atg__is_any_true(gc_tensor self);
+int atg__is_zerotensor(gc_tensor self);
+void atg__linalg_check_errors(gc_tensor info, char * api_name, int is_matrix);
+void atg__linalg_det(raw_tensor *, gc_tensor A);
+void atg__linalg_det_result(raw_tensor *, gc_tensor result, gc_tensor LU, gc_tensor pivots, gc_tensor A);
+void atg__linalg_eigh(raw_tensor *, gc_tensor A, char * UPLO, int compute_v);
+void atg__linalg_eigh_eigenvalues(raw_tensor *, gc_tensor eigenvalues, gc_tensor eigenvectors, gc_tensor A, char * UPLO, int compute_v);
+void atg__linalg_slogdet(raw_tensor *, gc_tensor A);
+void atg__linalg_slogdet_sign(raw_tensor *, gc_tensor sign, gc_tensor logabsdet, gc_tensor LU, gc_tensor pivots, gc_tensor A);
+void atg__linalg_solve_ex(raw_tensor *, gc_tensor A, gc_tensor B, int left, int check_errors);
+void atg__linalg_solve_ex_result(raw_tensor *, gc_tensor result, gc_tensor LU, gc_tensor pivots, gc_tensor info, gc_tensor A, gc_tensor B, int left, int check_errors);
+void atg__linalg_svd(raw_tensor *, gc_tensor A, int full_matrices, int compute_uv, char * driver);
+void atg__linalg_svd_u(raw_tensor *, gc_tensor U, gc_tensor S, gc_tensor Vh, gc_tensor A, int full_matrices, int compute_uv, char * driver);
+raw_tensor atg__log_softmax(gc_tensor self, int64_t dim, int half_to_float);
+raw_tensor atg__log_softmax_backward_data(gc_tensor grad_output, gc_tensor output, int64_t dim, int input_dtype);
+raw_tensor atg__log_softmax_backward_data_out(gc_tensor out, gc_tensor grad_output, gc_tensor output, int64_t dim, int input_dtype);
+raw_tensor atg__log_softmax_out(gc_tensor out, gc_tensor self, int64_t dim, int half_to_float);
+raw_tensor atg__logcumsumexp(gc_tensor self, int64_t dim);
+raw_tensor atg__logcumsumexp_out(gc_tensor out, gc_tensor self, int64_t dim);
+void atg__lstm_mps(raw_tensor *, gc_tensor input, gc_tensor *hx_data, int hx_len, gc_tensor *params_data, int params_len, int has_biases, int64_t num_layers, double dropout, int train, int bidirectional, int batch_first);
+void atg__lstm_mps_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor out3, gc_tensor out4, gc_tensor out5, gc_tensor input, gc_tensor *hx_data, int hx_len, gc_tensor *params_data, int params_len, int has_biases, int64_t num_layers, double dropout, int train, int bidirectional, int batch_first);
+void atg__lu_with_info(raw_tensor *, gc_tensor self, int pivot, int check_errors);
+raw_tensor atg__make_dep_token(int options_kind, int options_device);
+raw_tensor atg__make_dual(gc_tensor primal, gc_tensor tangent, int64_t level);
+raw_tensor atg__make_dual_copy(gc_tensor primal, gc_tensor tangent, int64_t level);
+raw_tensor atg__make_dual_copy_out(gc_tensor out, gc_tensor primal, gc_tensor tangent, int64_t level);
+raw_tensor atg__make_per_channel_quantized_tensor(gc_tensor self, gc_tensor scale, gc_tensor zero_point, int64_t axis);
+raw_tensor atg__make_per_channel_quantized_tensor_out(gc_tensor out, gc_tensor self, gc_tensor scale, gc_tensor zero_point, int64_t axis);
+raw_tensor atg__make_per_tensor_quantized_tensor(gc_tensor self, double scale, int64_t zero_point);
+raw_tensor atg__make_per_tensor_quantized_tensor_out(gc_tensor out, gc_tensor self, double scale, int64_t zero_point);
+raw_tensor atg__masked_scale(gc_tensor self, gc_tensor mask, double scale);
+raw_tensor atg__masked_scale_out(gc_tensor out, gc_tensor self, gc_tensor mask, double scale);
+raw_tensor atg__masked_softmax(gc_tensor self, gc_tensor mask, int64_t dim_v, int dim_null, int64_t mask_type_v, int mask_type_null);
+raw_tensor atg__masked_softmax_backward(gc_tensor grad_output, gc_tensor output, gc_tensor mask, int64_t dim_v, int dim_null);
+raw_tensor atg__masked_softmax_backward_out(gc_tensor out, gc_tensor grad_output, gc_tensor output, gc_tensor mask, int64_t dim_v, int dim_null);
+raw_tensor atg__masked_softmax_out(gc_tensor out, gc_tensor self, gc_tensor mask, int64_t dim_v, int dim_null, int64_t mask_type_v, int mask_type_null);
+raw_tensor atg__mkldnn_reshape(gc_tensor self, int64_t *shape_data, int shape_len);
+raw_tensor atg__mkldnn_reshape_out(gc_tensor out, gc_tensor self, int64_t *shape_data, int shape_len);
+raw_tensor atg__mkldnn_transpose(gc_tensor self, int64_t dim0, int64_t dim1);
+raw_tensor atg__mkldnn_transpose_(gc_tensor self, int64_t dim0, int64_t dim1);
+raw_tensor atg__mkldnn_transpose_out(gc_tensor out, gc_tensor self, int64_t dim0, int64_t dim1);
+raw_tensor atg__mps_convolution(gc_tensor self, gc_tensor weight, gc_tensor bias, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups);
+raw_tensor atg__mps_convolution_out(gc_tensor out, gc_tensor self, gc_tensor weight, gc_tensor bias, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups);
+raw_tensor atg__mps_convolution_transpose(gc_tensor self, gc_tensor weight, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups);
+raw_tensor atg__mps_convolution_transpose_out(gc_tensor out, gc_tensor self, gc_tensor weight, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups);
+void atg__native_batch_norm_legit(raw_tensor *, gc_tensor input, gc_tensor weight, gc_tensor bias, gc_tensor running_mean, gc_tensor running_var, int training, double momentum, double eps);
+void atg__native_batch_norm_legit_functional(raw_tensor *, gc_tensor input, gc_tensor weight, gc_tensor bias, gc_tensor running_mean, gc_tensor running_var, int training, double momentum, double eps);
+void atg__native_batch_norm_legit_no_stats(raw_tensor *, gc_tensor input, gc_tensor weight, gc_tensor bias, int training, double momentum, double eps);
+void atg__native_batch_norm_legit_no_stats_out(raw_tensor *, gc_tensor out, gc_tensor save_mean, gc_tensor save_invstd, gc_tensor input, gc_tensor weight, gc_tensor bias, int training, double momentum, double eps);
+void atg__native_batch_norm_legit_no_training(raw_tensor *, gc_tensor input, gc_tensor weight, gc_tensor bias, gc_tensor running_mean, gc_tensor running_var, double momentum, double eps);
+void atg__native_batch_norm_legit_no_training_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor input, gc_tensor weight, gc_tensor bias, gc_tensor running_mean, gc_tensor running_var, double momentum, double eps);
+void atg__native_batch_norm_legit_out(raw_tensor *, gc_tensor out, gc_tensor save_mean, gc_tensor save_invstd, gc_tensor input, gc_tensor weight, gc_tensor bias, gc_tensor running_mean, gc_tensor running_var, int training, double momentum, double eps);
+void atg__native_multi_head_attention(raw_tensor *, gc_tensor query, gc_tensor key, gc_tensor value, int64_t embed_dim, int64_t num_head, gc_tensor qkv_weight, gc_tensor qkv_bias, gc_tensor proj_weight, gc_tensor proj_bias, gc_tensor mask, int need_weights, int average_attn_weights, int64_t mask_type_v, int mask_type_null);
+void atg__native_multi_head_attention_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor query, gc_tensor key, gc_tensor value, int64_t embed_dim, int64_t num_head, gc_tensor qkv_weight, gc_tensor qkv_bias, gc_tensor proj_weight, gc_tensor proj_bias, gc_tensor mask, int need_weights, int average_attn_weights, int64_t mask_type_v, int mask_type_null);
+raw_tensor atg__neg_view(gc_tensor self);
+raw_tensor atg__neg_view_copy(gc_tensor self);
+raw_tensor atg__neg_view_copy_out(gc_tensor out, gc_tensor self);
+raw_tensor atg__nested_from_padded(gc_tensor padded, gc_tensor cpu_nested_shape_example, int fuse_transform_0213);
+raw_tensor atg__nested_from_padded_and_nested_example(gc_tensor padded, gc_tensor nt_example);
+raw_tensor atg__nested_from_padded_and_nested_example_out(gc_tensor out, gc_tensor padded, gc_tensor nt_example);
+raw_tensor atg__nested_from_padded_out(gc_tensor out, gc_tensor padded, gc_tensor cpu_nested_shape_example, int fuse_transform_0213);
+raw_tensor atg__nested_select_backward(gc_tensor grad_output, gc_tensor self, int64_t dim, int64_t index);
+raw_tensor atg__nested_sum_backward(gc_tensor grad, gc_tensor self, int64_t *dim_data, int dim_len, int keepdim);
+raw_tensor atg__nested_view_from_buffer(gc_tensor self, gc_tensor nested_size, gc_tensor nested_strides, gc_tensor offsets);
+raw_tensor atg__nested_view_from_buffer_copy(gc_tensor self, gc_tensor nested_size, gc_tensor nested_strides, gc_tensor offsets);
+raw_tensor atg__nested_view_from_buffer_copy_out(gc_tensor out, gc_tensor self, gc_tensor nested_size, gc_tensor nested_strides, gc_tensor offsets);
+raw_tensor atg__new_zeros_with_same_feature_meta(gc_tensor self, gc_tensor other, int64_t self_num_batch_dims);
+raw_tensor atg__new_zeros_with_same_feature_meta_out(gc_tensor out, gc_tensor self, gc_tensor other, int64_t self_num_batch_dims);
 int atg__nnpack_available();
-void atg__nnpack_spatial_convolution(tensor *, tensor input, tensor weight, tensor bias, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len);
-void atg__nnpack_spatial_convolution_out(tensor *, tensor out, tensor input, tensor weight, tensor bias, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len);
-int64_t atg__nnz(tensor self);
-void atg__pack_padded_sequence(tensor *, tensor input, tensor lengths, int batch_first);
-void atg__pack_padded_sequence_backward(tensor *, tensor grad, int64_t *input_size_data, int input_size_len, tensor batch_sizes, int batch_first);
-void atg__pack_padded_sequence_out(tensor *, tensor out0, tensor out1, tensor input, tensor lengths, int batch_first);
-void atg__pad_circular(tensor *, tensor self, int64_t *pad_data, int pad_len);
-void atg__pad_enum(tensor *, tensor self, int64_t *pad_data, int pad_len, int64_t mode, double value_v, int value_null);
-void atg__pad_packed_sequence(tensor *, tensor data, tensor batch_sizes, int batch_first, scalar padding_value, int64_t total_length);
-void atg__pdist_backward(tensor *, tensor grad, tensor self, double p, tensor pdist);
-void atg__pdist_backward_out(tensor *, tensor out, tensor grad, tensor self, double p, tensor pdist);
-void atg__pin_memory(tensor *, tensor self, int device);
-void atg__pin_memory_out(tensor *, tensor out, tensor self, int device);
-void atg__prelu_kernel(tensor *, tensor self, tensor weight);
-void atg__prelu_kernel_backward(tensor *, tensor grad_output, tensor self, tensor weight);
-void atg__propagate_xla_data(tensor input, tensor output);
-void atg__remove_batch_dim(tensor *, tensor self, int64_t level, int64_t batch_size, int64_t out_dim);
-void atg__reshape_alias(tensor *, tensor self, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len);
-void atg__reshape_alias_copy(tensor *, tensor self, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len);
-void atg__reshape_alias_copy_out(tensor *, tensor out, tensor self, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len);
-void atg__reshape_copy(tensor *, tensor self, int64_t *size_data, int size_len);
-void atg__reshape_from_tensor(tensor *, tensor self, tensor shape);
-void atg__resize_output(tensor *, tensor self, int64_t *size_data, int size_len, int device);
-void atg__resize_output_(tensor *, tensor self, int64_t *size_data, int size_len, int device);
-void atg__resize_output_out(tensor *, tensor out, tensor self, int64_t *size_data, int size_len, int device);
-void atg__rowwise_prune(tensor *, tensor weight, tensor mask, int compressed_indices_dtype);
-void atg__sample_dirichlet(tensor *, tensor self);
-void atg__sample_dirichlet_out(tensor *, tensor out, tensor self);
-void atg__saturate_weight_to_fp16(tensor *, tensor weight);
-void atg__scaled_dot_product_attention_math(tensor *, tensor query, tensor key, tensor value, tensor attn_mask, double dropout_p, int is_causal, tensor dropout_mask, double scale_v, int scale_null);
-void atg__scaled_dot_product_efficient_attention(tensor *, tensor query, tensor key, tensor value, tensor attn_bias, int compute_log_sumexp, double dropout_p, int is_causal, double scale_v, int scale_null);
-void atg__scaled_dot_product_flash_attention_backward(tensor *, tensor grad_out, tensor query, tensor key, tensor value, tensor out, tensor logsumexp, tensor cum_seq_q, tensor cum_seq_k, int64_t max_q, int64_t max_k, double dropout_p, int is_causal, tensor philox_seed, tensor philox_offset, double scale_v, int scale_null);
-void atg__scaled_mm(tensor *, tensor self, tensor mat2, tensor bias, int out_dtype, tensor scale_a, tensor scale_b, tensor scale_result);
-void atg__scaled_mm_out(tensor *, tensor out, tensor out_amax, tensor self, tensor mat2, tensor bias, int out_dtype, tensor scale_a, tensor scale_b, tensor scale_result);
-void atg__scatter_reduce(tensor *, tensor self, int64_t dim, tensor index, tensor src, char * reduce, int include_self);
-void atg__scatter_reduce_(tensor *, tensor self, int64_t dim, tensor index, tensor src, char * reduce, int include_self);
-void atg__scatter_reduce_two_out(tensor *, tensor out, tensor self, int64_t dim, tensor index, tensor src, char * reduce, int include_self);
-void atg__segment_reduce_backward(tensor *, tensor grad, tensor output, tensor data, char * reduce, tensor lengths, tensor offsets, int64_t axis, scalar initial);
-void atg__segment_reduce_backward_out(tensor *, tensor out, tensor grad, tensor output, tensor data, char * reduce, tensor lengths, tensor offsets, int64_t axis, scalar initial);
-void atg__shape_as_tensor(tensor *, tensor self);
-void atg__slow_conv2d_backward(tensor *, tensor grad_input, tensor grad_weight, tensor grad_bias, tensor grad_output, tensor self, tensor weight, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len);
-void atg__sobol_engine_draw(tensor *, tensor quasi, int64_t n, tensor sobolstate, int64_t dimension, int64_t num_generated, int dtype);
-void atg__sobol_engine_ff_(tensor *, tensor self, int64_t n, tensor sobolstate, int64_t dimension, int64_t num_generated);
-void atg__sobol_engine_initialize_state_(tensor *, tensor self, int64_t dimension);
-void atg__sobol_engine_scramble_(tensor *, tensor self, tensor ltm, int64_t dimension);
-void atg__softmax(tensor *, tensor self, int64_t dim, int half_to_float);
-void atg__softmax_backward_data(tensor *, tensor grad_output, tensor output, int64_t dim, int input_dtype);
-void atg__softmax_backward_data_out(tensor *, tensor grad_input, tensor grad_output, tensor output, int64_t dim, int input_dtype);
-void atg__softmax_out(tensor *, tensor out, tensor self, int64_t dim, int half_to_float);
-void atg__sparse_addmm(tensor *, tensor self, tensor mat1, tensor mat2);
-void atg__sparse_addmm_out(tensor *, tensor out, tensor self, tensor mat1, tensor mat2);
-void atg__sparse_broadcast_to(tensor *, tensor self, int64_t *size_data, int size_len);
-void atg__sparse_broadcast_to_copy(tensor *, tensor self, int64_t *size_data, int size_len);
-void atg__sparse_broadcast_to_copy_out(tensor *, tensor out, tensor self, int64_t *size_data, int size_len);
-void atg__sparse_bsc_tensor_unsafe(tensor *, tensor ccol_indices, tensor row_indices, tensor values, int64_t *size_data, int size_len, int options_kind, int options_device);
-void atg__sparse_bsr_tensor_unsafe(tensor *, tensor crow_indices, tensor col_indices, tensor values, int64_t *size_data, int size_len, int options_kind, int options_device);
-void atg__sparse_compressed_tensor_unsafe(tensor *, tensor compressed_indices, tensor plain_indices, tensor values, int64_t *size_data, int size_len, int options_kind, int options_device);
-void atg__sparse_coo_tensor_unsafe(tensor *, tensor indices, tensor values, int64_t *size_data, int size_len, int options_kind, int options_device, int is_coalesced);
-void atg__sparse_coo_tensor_with_dims(tensor *, int64_t sparse_dim, int64_t dense_dim, int64_t *size_data, int size_len, int options_kind, int options_device);
-void atg__sparse_coo_tensor_with_dims_and_tensors(tensor *, int64_t sparse_dim, int64_t dense_dim, int64_t *size_data, int size_len, tensor indices, tensor values, int options_kind, int options_device, int is_coalesced);
-void atg__sparse_coo_tensor_with_dims_and_tensors_out(tensor *, tensor out, int64_t sparse_dim, int64_t dense_dim, int64_t *size_data, int size_len, tensor indices, tensor values, int is_coalesced);
-void atg__sparse_coo_tensor_with_dims_out(tensor *, tensor out, int64_t sparse_dim, int64_t dense_dim, int64_t *size_data, int size_len);
-void atg__sparse_csc_tensor_unsafe(tensor *, tensor ccol_indices, tensor row_indices, tensor values, int64_t *size_data, int size_len, int options_kind, int options_device);
-void atg__sparse_csr_prod(tensor *, tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype);
-void atg__sparse_csr_prod_dim_dtype_out(tensor *, tensor out, tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype);
-void atg__sparse_csr_sum(tensor *, tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype);
-void atg__sparse_csr_sum_dim_dtype_out(tensor *, tensor out, tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype);
-void atg__sparse_csr_tensor_unsafe(tensor *, tensor crow_indices, tensor col_indices, tensor values, int64_t *size_data, int size_len, int options_kind, int options_device);
-void atg__sparse_log_softmax(tensor *, tensor self, int64_t dim, int half_to_float);
-void atg__sparse_log_softmax_backward_data(tensor *, tensor grad_output, tensor output, int64_t dim, tensor self);
-void atg__sparse_log_softmax_backward_data_out(tensor *, tensor out, tensor grad_output, tensor output, int64_t dim, tensor self);
-void atg__sparse_log_softmax_int(tensor *, tensor self, int64_t dim, int dtype);
-void atg__sparse_log_softmax_out(tensor *, tensor out, tensor self, int64_t dim, int half_to_float);
-void atg__sparse_mask_projection(tensor *, tensor self, tensor mask, int accumulate_matches);
-void atg__sparse_mask_projection_out(tensor *, tensor out, tensor self, tensor mask, int accumulate_matches);
-void atg__sparse_mm(tensor *, tensor sparse, tensor dense);
-void atg__sparse_mm_reduce(tensor *, tensor sparse, tensor dense, char * reduce);
-void atg__sparse_mm_reduce_impl(tensor *, tensor self, tensor other, char * reduce);
-void atg__sparse_semi_structured_linear(tensor *, tensor input, tensor weight, tensor meta, tensor bias, char * activation);
-void atg__sparse_softmax(tensor *, tensor self, int64_t dim, int half_to_float);
-void atg__sparse_softmax_backward_data(tensor *, tensor grad_output, tensor output, int64_t dim, tensor self);
-void atg__sparse_softmax_backward_data_out(tensor *, tensor out, tensor grad_output, tensor output, int64_t dim, tensor self);
-void atg__sparse_softmax_int(tensor *, tensor self, int64_t dim, int dtype);
-void atg__sparse_softmax_out(tensor *, tensor out, tensor self, int64_t dim, int half_to_float);
-void atg__sparse_sparse_matmul(tensor *, tensor self, tensor other);
-void atg__sparse_sparse_matmul_out(tensor *, tensor out, tensor self, tensor other);
-void atg__sparse_sum(tensor *, tensor self);
-void atg__sparse_sum_backward(tensor *, tensor grad, tensor self, int64_t *dim_data, int dim_len);
-void atg__sparse_sum_backward_out(tensor *, tensor out, tensor grad, tensor self, int64_t *dim_data, int dim_len);
-void atg__sparse_sum_dim(tensor *, tensor self, int64_t *dim_data, int dim_len);
-void atg__sparse_sum_dim_dtype(tensor *, tensor self, int64_t *dim_data, int dim_len, int dtype);
-void atg__sparse_sum_dim_out(tensor *, tensor out, tensor self, int64_t *dim_data, int dim_len);
-void atg__sparse_sum_dtype(tensor *, tensor self, int dtype);
-void atg__spdiags(tensor *, tensor diagonals, tensor offsets, int64_t *shape_data, int shape_len);
-void atg__spdiags_out(tensor *, tensor out, tensor diagonals, tensor offsets, int64_t *shape_data, int shape_len);
-void atg__stack(tensor *, tensor *tensors_data, int tensors_len, int64_t dim);
-void atg__stack_out(tensor *, tensor out, tensor *tensors_data, int tensors_len, int64_t dim);
-void atg__standard_gamma(tensor *, tensor self);
-void atg__standard_gamma_grad(tensor *, tensor self, tensor output);
-void atg__standard_gamma_grad_out(tensor *, tensor out, tensor self, tensor output);
-void atg__standard_gamma_out(tensor *, tensor out, tensor self);
-void atg__test_ambiguous_defaults(tensor *, tensor dummy, int64_t a, int64_t b);
-void atg__test_ambiguous_defaults_b(tensor *, tensor dummy, int64_t a, char * b);
-void atg__test_autograd_multiple_dispatch(tensor *, tensor self);
-void atg__test_autograd_multiple_dispatch_fullcoverage_out(tensor *, tensor out, tensor self);
-void atg__test_autograd_multiple_dispatch_ntonly(tensor *, tensor self, int b);
-void atg__test_autograd_multiple_dispatch_view(tensor *, tensor self);
-void atg__test_autograd_multiple_dispatch_view_copy(tensor *, tensor self);
-void atg__test_autograd_multiple_dispatch_view_copy_out(tensor *, tensor out, tensor self);
-void atg__test_check_tensor(tensor *, tensor self);
-void atg__test_functorch_fallback(tensor *, tensor self, tensor other);
-void atg__test_functorch_fallback_out(tensor *, tensor out, tensor self, tensor other);
-void atg__test_optional_filled_intlist(tensor *, tensor values, int64_t *addends_data, int addends_len);
-void atg__test_optional_filled_intlist_out(tensor *, tensor out, tensor values, int64_t *addends_data, int addends_len);
-void atg__test_optional_floatlist(tensor *, tensor values, double *addends_data, int addends_len);
-void atg__test_optional_floatlist_out(tensor *, tensor out, tensor values, double *addends_data, int addends_len);
-void atg__test_optional_intlist(tensor *, tensor values, int64_t *addends_data, int addends_len);
-void atg__test_optional_intlist_out(tensor *, tensor out, tensor values, int64_t *addends_data, int addends_len);
-void atg__test_serialization_subcmul(tensor *, tensor self, tensor other);
-void atg__test_string_default(tensor *, tensor dummy, char * a, char * b);
-void atg__test_warn_in_autograd(tensor *, tensor self);
-void atg__test_warn_in_autograd_out(tensor *, tensor out, tensor self);
-void atg__thnn_differentiable_gru_cell_backward(tensor *, tensor grad_hy, tensor input_gates, tensor hidden_gates, tensor hx, tensor input_bias, tensor hidden_bias);
-void atg__thnn_differentiable_lstm_cell_backward(tensor *, tensor grad_hy, tensor grad_cy, tensor input_gates, tensor hidden_gates, tensor input_bias, tensor hidden_bias, tensor cx, tensor cy);
-void atg__thnn_fused_gru_cell(tensor *, tensor input_gates, tensor hidden_gates, tensor hx, tensor input_bias, tensor hidden_bias);
-void atg__thnn_fused_gru_cell_backward(tensor *, tensor grad_hy, tensor workspace, int has_bias);
-void atg__thnn_fused_gru_cell_backward_out(tensor *, tensor out0, tensor out1, tensor out2, tensor out3, tensor out4, tensor grad_hy, tensor workspace, int has_bias);
-void atg__thnn_fused_gru_cell_out(tensor *, tensor out0, tensor out1, tensor input_gates, tensor hidden_gates, tensor hx, tensor input_bias, tensor hidden_bias);
-void atg__thnn_fused_lstm_cell(tensor *, tensor input_gates, tensor hidden_gates, tensor cx, tensor input_bias, tensor hidden_bias);
-void atg__thnn_fused_lstm_cell_backward(tensor *, tensor grad_hy, tensor grad_cy, tensor cx, tensor cy, tensor workspace, int has_bias);
-void atg__thnn_fused_lstm_cell_backward_impl(tensor *, tensor grad_hy, tensor grad_cy, tensor cx, tensor cy, tensor workspace, int has_bias);
-void atg__thnn_fused_lstm_cell_backward_impl_out(tensor *, tensor out0, tensor out1, tensor out2, tensor grad_hy, tensor grad_cy, tensor cx, tensor cy, tensor workspace, int has_bias);
-void atg__thnn_fused_lstm_cell_out(tensor *, tensor out0, tensor out1, tensor out2, tensor input_gates, tensor hidden_gates, tensor cx, tensor input_bias, tensor hidden_bias);
-void atg__to_copy(tensor *, tensor self, int options_kind, int options_device, int non_blocking);
-void atg__to_copy_out(tensor *, tensor out, tensor self, int non_blocking);
-tensor *atg__to_cpu(tensor *tensors_data, int tensors_len);
-void atg__to_dense(tensor *, tensor self, int dtype, int masked_grad);
-void atg__to_dense_out(tensor *, tensor out, tensor self, int dtype, int masked_grad);
-void atg__to_sparse_bsc(tensor *, tensor self, int64_t *blocksize_data, int blocksize_len, int64_t dense_dim_v, int dense_dim_null);
-void atg__to_sparse_bsc_out(tensor *, tensor out, tensor self, int64_t *blocksize_data, int blocksize_len, int64_t dense_dim_v, int dense_dim_null);
-void atg__to_sparse_bsr(tensor *, tensor self, int64_t *blocksize_data, int blocksize_len, int64_t dense_dim_v, int dense_dim_null);
-void atg__to_sparse_bsr_out(tensor *, tensor out, tensor self, int64_t *blocksize_data, int blocksize_len, int64_t dense_dim_v, int dense_dim_null);
-void atg__to_sparse_csc(tensor *, tensor self, int64_t dense_dim_v, int dense_dim_null);
-void atg__to_sparse_csc_out(tensor *, tensor out, tensor self, int64_t dense_dim_v, int dense_dim_null);
-void atg__to_sparse_csr(tensor *, tensor self, int64_t dense_dim_v, int dense_dim_null);
-void atg__to_sparse_csr_out(tensor *, tensor out, tensor self, int64_t dense_dim_v, int dense_dim_null);
-void atg__to_sparse_semi_structured(tensor *, tensor dense);
-void atg__transform_bias_rescale_qkv(tensor *, tensor qkv, tensor qkv_bias, int64_t num_heads);
-void atg__transform_bias_rescale_qkv_out(tensor *, tensor out0, tensor out1, tensor out2, tensor qkv, tensor qkv_bias, int64_t num_heads);
-void atg__transformer_encoder_layer_fwd(tensor *, tensor src, int64_t embed_dim, int64_t num_heads, tensor qkv_weight, tensor qkv_bias, tensor proj_weight, tensor proj_bias, int use_gelu, int norm_first, double eps, tensor norm_weight_1, tensor norm_bias_1, tensor norm_weight_2, tensor norm_bias_2, tensor ffn_weight_1, tensor ffn_bias_1, tensor ffn_weight_2, tensor ffn_bias_2, tensor mask, int64_t mask_type_v, int mask_type_null);
-void atg__transformer_encoder_layer_fwd_out(tensor *, tensor out, tensor src, int64_t embed_dim, int64_t num_heads, tensor qkv_weight, tensor qkv_bias, tensor proj_weight, tensor proj_bias, int use_gelu, int norm_first, double eps, tensor norm_weight_1, tensor norm_bias_1, tensor norm_weight_2, tensor norm_bias_2, tensor ffn_weight_1, tensor ffn_bias_1, tensor ffn_weight_2, tensor ffn_bias_2, tensor mask, int64_t mask_type_v, int mask_type_null);
-void atg__trilinear(tensor *, tensor i1, tensor i2, tensor i3, int64_t *expand1_data, int expand1_len, int64_t *expand2_data, int expand2_len, int64_t *expand3_data, int expand3_len, int64_t *sumdim_data, int sumdim_len, int64_t unroll_dim);
-void atg__trilinear_out(tensor *, tensor out, tensor i1, tensor i2, tensor i3, int64_t *expand1_data, int expand1_len, int64_t *expand2_data, int expand2_len, int64_t *expand3_data, int expand3_len, int64_t *sumdim_data, int sumdim_len, int64_t unroll_dim);
-void atg__triton_multi_head_attention(tensor *, tensor query, tensor key, tensor value, int64_t embed_dim, int64_t num_head, tensor qkv_weight, tensor qkv_bias, tensor proj_weight, tensor proj_bias, tensor mask);
-void atg__triton_multi_head_attention_out(tensor *, tensor out, tensor query, tensor key, tensor value, int64_t embed_dim, int64_t num_head, tensor qkv_weight, tensor qkv_bias, tensor proj_weight, tensor proj_bias, tensor mask);
-void atg__triton_scaled_dot_attention(tensor *, tensor q, tensor k, tensor v, double dropout_p);
-void atg__triton_scaled_dot_attention_out(tensor *, tensor out, tensor q, tensor k, tensor v, double dropout_p);
-void atg__unique(tensor *, tensor self, int sorted, int return_inverse);
-void atg__unique2(tensor *, tensor self, int sorted, int return_inverse, int return_counts);
-void atg__unique2_out(tensor *, tensor out0, tensor out1, tensor out2, tensor self, int sorted, int return_inverse, int return_counts);
-void atg__unique_out(tensor *, tensor out0, tensor out1, tensor self, int sorted, int return_inverse);
-void atg__unpack_dual(tensor *, tensor dual, int64_t level);
-void atg__unsafe_index(tensor *, tensor self, tensor *indices_data, int indices_len);
-void atg__unsafe_index_put(tensor *, tensor self, tensor *indices_data, int indices_len, tensor values, int accumulate);
-void atg__unsafe_view(tensor *, tensor self, int64_t *size_data, int size_len);
-void atg__unsafe_view_out(tensor *, tensor out, tensor self, int64_t *size_data, int size_len);
-void atg__upsample_bicubic2d_aa(tensor *, tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
-void atg__upsample_bicubic2d_aa_backward(tensor *, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
-void atg__upsample_bicubic2d_aa_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
-void atg__upsample_bicubic2d_aa_out(tensor *, tensor out, tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
-void atg__upsample_bicubic2d_aa_vec(tensor *, tensor input, int64_t *output_size_data, int output_size_len, int align_corners, double *scale_factors_data, int scale_factors_len);
-void atg__upsample_bilinear2d_aa(tensor *, tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
-void atg__upsample_bilinear2d_aa_backward(tensor *, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
-void atg__upsample_bilinear2d_aa_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
-void atg__upsample_bilinear2d_aa_out(tensor *, tensor out, tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
-void atg__upsample_bilinear2d_aa_vec(tensor *, tensor input, int64_t *output_size_data, int output_size_len, int align_corners, double *scale_factors_data, int scale_factors_len);
-void atg__upsample_nearest_exact1d(tensor *, tensor self, int64_t *output_size_data, int output_size_len, double scales_v, int scales_null);
-void atg__upsample_nearest_exact1d_backward(tensor *, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_v, int scales_null);
-void atg__upsample_nearest_exact1d_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_v, int scales_null);
-void atg__upsample_nearest_exact1d_out(tensor *, tensor out, tensor self, int64_t *output_size_data, int output_size_len, double scales_v, int scales_null);
-void atg__upsample_nearest_exact1d_vec(tensor *, tensor input, int64_t *output_size_data, int output_size_len, double *scale_factors_data, int scale_factors_len);
-void atg__upsample_nearest_exact2d(tensor *, tensor self, int64_t *output_size_data, int output_size_len, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
-void atg__upsample_nearest_exact2d_backward(tensor *, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
-void atg__upsample_nearest_exact2d_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
-void atg__upsample_nearest_exact2d_out(tensor *, tensor out, tensor self, int64_t *output_size_data, int output_size_len, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
-void atg__upsample_nearest_exact2d_vec(tensor *, tensor input, int64_t *output_size_data, int output_size_len, double *scale_factors_data, int scale_factors_len);
-void atg__upsample_nearest_exact3d(tensor *, tensor self, int64_t *output_size_data, int output_size_len, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
-void atg__upsample_nearest_exact3d_backward(tensor *, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
-void atg__upsample_nearest_exact3d_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
-void atg__upsample_nearest_exact3d_out(tensor *, tensor out, tensor self, int64_t *output_size_data, int output_size_len, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
-void atg__upsample_nearest_exact3d_vec(tensor *, tensor input, int64_t *output_size_data, int output_size_len, double *scale_factors_data, int scale_factors_len);
-int atg__use_cudnn_ctc_loss(tensor log_probs, tensor targets, int64_t *input_lengths_data, int input_lengths_len, int64_t *target_lengths_data, int target_lengths_len, int64_t blank);
-int atg__use_cudnn_ctc_loss_tensor(tensor log_probs, tensor targets, tensor input_lengths, tensor target_lengths, int64_t blank);
+raw_tensor atg__nnpack_spatial_convolution(gc_tensor input, gc_tensor weight, gc_tensor bias, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len);
+raw_tensor atg__nnpack_spatial_convolution_out(gc_tensor out, gc_tensor input, gc_tensor weight, gc_tensor bias, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len);
+int64_t atg__nnz(gc_tensor self);
+void atg__pack_padded_sequence(raw_tensor *, gc_tensor input, gc_tensor lengths, int batch_first);
+raw_tensor atg__pack_padded_sequence_backward(gc_tensor grad, int64_t *input_size_data, int input_size_len, gc_tensor batch_sizes, int batch_first);
+void atg__pack_padded_sequence_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor input, gc_tensor lengths, int batch_first);
+raw_tensor atg__pad_circular(gc_tensor self, int64_t *pad_data, int pad_len);
+raw_tensor atg__pad_enum(gc_tensor self, int64_t *pad_data, int pad_len, int64_t mode, double value_v, int value_null);
+void atg__pad_packed_sequence(raw_tensor *, gc_tensor data, gc_tensor batch_sizes, int batch_first, scalar padding_value, int64_t total_length);
+raw_tensor atg__pdist_backward(gc_tensor grad, gc_tensor self, double p, gc_tensor pdist);
+raw_tensor atg__pdist_backward_out(gc_tensor out, gc_tensor grad, gc_tensor self, double p, gc_tensor pdist);
+raw_tensor atg__pin_memory(gc_tensor self, int device);
+raw_tensor atg__pin_memory_out(gc_tensor out, gc_tensor self, int device);
+raw_tensor atg__prelu_kernel(gc_tensor self, gc_tensor weight);
+void atg__prelu_kernel_backward(raw_tensor *, gc_tensor grad_output, gc_tensor self, gc_tensor weight);
+void atg__propagate_xla_data(gc_tensor input, gc_tensor output);
+raw_tensor atg__remove_batch_dim(gc_tensor self, int64_t level, int64_t batch_size, int64_t out_dim);
+raw_tensor atg__reshape_alias(gc_tensor self, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len);
+raw_tensor atg__reshape_alias_copy(gc_tensor self, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len);
+raw_tensor atg__reshape_alias_copy_out(gc_tensor out, gc_tensor self, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len);
+raw_tensor atg__reshape_copy(gc_tensor self, int64_t *size_data, int size_len);
+raw_tensor atg__reshape_from_tensor(gc_tensor self, gc_tensor shape);
+raw_tensor atg__resize_output(gc_tensor self, int64_t *size_data, int size_len, int device);
+raw_tensor atg__resize_output_(gc_tensor self, int64_t *size_data, int size_len, int device);
+raw_tensor atg__resize_output_out(gc_tensor out, gc_tensor self, int64_t *size_data, int size_len, int device);
+void atg__rowwise_prune(raw_tensor *, gc_tensor weight, gc_tensor mask, int compressed_indices_dtype);
+raw_tensor atg__sample_dirichlet(gc_tensor self);
+raw_tensor atg__sample_dirichlet_out(gc_tensor out, gc_tensor self);
+raw_tensor atg__saturate_weight_to_fp16(gc_tensor weight);
+void atg__scaled_dot_product_attention_math(raw_tensor *, gc_tensor query, gc_tensor key, gc_tensor value, gc_tensor attn_mask, double dropout_p, int is_causal, gc_tensor dropout_mask, double scale_v, int scale_null);
+void atg__scaled_dot_product_efficient_attention(raw_tensor *, gc_tensor query, gc_tensor key, gc_tensor value, gc_tensor attn_bias, int compute_log_sumexp, double dropout_p, int is_causal, double scale_v, int scale_null);
+void atg__scaled_dot_product_flash_attention_backward(raw_tensor *, gc_tensor grad_out, gc_tensor query, gc_tensor key, gc_tensor value, gc_tensor out, gc_tensor logsumexp, gc_tensor cum_seq_q, gc_tensor cum_seq_k, int64_t max_q, int64_t max_k, double dropout_p, int is_causal, gc_tensor philox_seed, gc_tensor philox_offset, double scale_v, int scale_null);
+void atg__scaled_mm(raw_tensor *, gc_tensor self, gc_tensor mat2, gc_tensor bias, int out_dtype, gc_tensor scale_a, gc_tensor scale_b, gc_tensor scale_result);
+void atg__scaled_mm_out(raw_tensor *, gc_tensor out, gc_tensor out_amax, gc_tensor self, gc_tensor mat2, gc_tensor bias, int out_dtype, gc_tensor scale_a, gc_tensor scale_b, gc_tensor scale_result);
+raw_tensor atg__scatter_reduce(gc_tensor self, int64_t dim, gc_tensor index, gc_tensor src, char * reduce, int include_self);
+raw_tensor atg__scatter_reduce_(gc_tensor self, int64_t dim, gc_tensor index, gc_tensor src, char * reduce, int include_self);
+raw_tensor atg__scatter_reduce_two_out(gc_tensor out, gc_tensor self, int64_t dim, gc_tensor index, gc_tensor src, char * reduce, int include_self);
+raw_tensor atg__segment_reduce_backward(gc_tensor grad, gc_tensor output, gc_tensor data, char * reduce, gc_tensor lengths, gc_tensor offsets, int64_t axis, scalar initial);
+raw_tensor atg__segment_reduce_backward_out(gc_tensor out, gc_tensor grad, gc_tensor output, gc_tensor data, char * reduce, gc_tensor lengths, gc_tensor offsets, int64_t axis, scalar initial);
+raw_tensor atg__shape_as_tensor(gc_tensor self);
+void atg__slow_conv2d_backward(raw_tensor *, gc_tensor grad_input, gc_tensor grad_weight, gc_tensor grad_bias, gc_tensor grad_output, gc_tensor self, gc_tensor weight, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len);
+void atg__sobol_engine_draw(raw_tensor *, gc_tensor quasi, int64_t n, gc_tensor sobolstate, int64_t dimension, int64_t num_generated, int dtype);
+raw_tensor atg__sobol_engine_ff_(gc_tensor self, int64_t n, gc_tensor sobolstate, int64_t dimension, int64_t num_generated);
+raw_tensor atg__sobol_engine_initialize_state_(gc_tensor self, int64_t dimension);
+raw_tensor atg__sobol_engine_scramble_(gc_tensor self, gc_tensor ltm, int64_t dimension);
+raw_tensor atg__softmax(gc_tensor self, int64_t dim, int half_to_float);
+raw_tensor atg__softmax_backward_data(gc_tensor grad_output, gc_tensor output, int64_t dim, int input_dtype);
+raw_tensor atg__softmax_backward_data_out(gc_tensor grad_input, gc_tensor grad_output, gc_tensor output, int64_t dim, int input_dtype);
+raw_tensor atg__softmax_out(gc_tensor out, gc_tensor self, int64_t dim, int half_to_float);
+raw_tensor atg__sparse_addmm(gc_tensor self, gc_tensor mat1, gc_tensor mat2);
+raw_tensor atg__sparse_addmm_out(gc_tensor out, gc_tensor self, gc_tensor mat1, gc_tensor mat2);
+raw_tensor atg__sparse_broadcast_to(gc_tensor self, int64_t *size_data, int size_len);
+raw_tensor atg__sparse_broadcast_to_copy(gc_tensor self, int64_t *size_data, int size_len);
+raw_tensor atg__sparse_broadcast_to_copy_out(gc_tensor out, gc_tensor self, int64_t *size_data, int size_len);
+raw_tensor atg__sparse_bsc_tensor_unsafe(gc_tensor ccol_indices, gc_tensor row_indices, gc_tensor values, int64_t *size_data, int size_len, int options_kind, int options_device);
+raw_tensor atg__sparse_bsr_tensor_unsafe(gc_tensor crow_indices, gc_tensor col_indices, gc_tensor values, int64_t *size_data, int size_len, int options_kind, int options_device);
+raw_tensor atg__sparse_compressed_tensor_unsafe(gc_tensor compressed_indices, gc_tensor plain_indices, gc_tensor values, int64_t *size_data, int size_len, int options_kind, int options_device);
+raw_tensor atg__sparse_coo_tensor_unsafe(gc_tensor indices, gc_tensor values, int64_t *size_data, int size_len, int options_kind, int options_device, int is_coalesced);
+raw_tensor atg__sparse_coo_tensor_with_dims(int64_t sparse_dim, int64_t dense_dim, int64_t *size_data, int size_len, int options_kind, int options_device);
+raw_tensor atg__sparse_coo_tensor_with_dims_and_tensors(int64_t sparse_dim, int64_t dense_dim, int64_t *size_data, int size_len, gc_tensor indices, gc_tensor values, int options_kind, int options_device, int is_coalesced);
+raw_tensor atg__sparse_coo_tensor_with_dims_and_tensors_out(gc_tensor out, int64_t sparse_dim, int64_t dense_dim, int64_t *size_data, int size_len, gc_tensor indices, gc_tensor values, int is_coalesced);
+raw_tensor atg__sparse_coo_tensor_with_dims_out(gc_tensor out, int64_t sparse_dim, int64_t dense_dim, int64_t *size_data, int size_len);
+raw_tensor atg__sparse_csc_tensor_unsafe(gc_tensor ccol_indices, gc_tensor row_indices, gc_tensor values, int64_t *size_data, int size_len, int options_kind, int options_device);
+raw_tensor atg__sparse_csr_prod(gc_tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype);
+raw_tensor atg__sparse_csr_prod_dim_dtype_out(gc_tensor out, gc_tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype);
+raw_tensor atg__sparse_csr_sum(gc_tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype);
+raw_tensor atg__sparse_csr_sum_dim_dtype_out(gc_tensor out, gc_tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype);
+raw_tensor atg__sparse_csr_tensor_unsafe(gc_tensor crow_indices, gc_tensor col_indices, gc_tensor values, int64_t *size_data, int size_len, int options_kind, int options_device);
+raw_tensor atg__sparse_log_softmax(gc_tensor self, int64_t dim, int half_to_float);
+raw_tensor atg__sparse_log_softmax_backward_data(gc_tensor grad_output, gc_tensor output, int64_t dim, gc_tensor self);
+raw_tensor atg__sparse_log_softmax_backward_data_out(gc_tensor out, gc_tensor grad_output, gc_tensor output, int64_t dim, gc_tensor self);
+raw_tensor atg__sparse_log_softmax_int(gc_tensor self, int64_t dim, int dtype);
+raw_tensor atg__sparse_log_softmax_out(gc_tensor out, gc_tensor self, int64_t dim, int half_to_float);
+raw_tensor atg__sparse_mask_projection(gc_tensor self, gc_tensor mask, int accumulate_matches);
+raw_tensor atg__sparse_mask_projection_out(gc_tensor out, gc_tensor self, gc_tensor mask, int accumulate_matches);
+raw_tensor atg__sparse_mm(gc_tensor sparse, gc_tensor dense);
+raw_tensor atg__sparse_mm_reduce(gc_tensor sparse, gc_tensor dense, char * reduce);
+void atg__sparse_mm_reduce_impl(raw_tensor *, gc_tensor self, gc_tensor other, char * reduce);
+raw_tensor atg__sparse_semi_structured_linear(gc_tensor input, gc_tensor weight, gc_tensor meta, gc_tensor bias, char * activation);
+raw_tensor atg__sparse_softmax(gc_tensor self, int64_t dim, int half_to_float);
+raw_tensor atg__sparse_softmax_backward_data(gc_tensor grad_output, gc_tensor output, int64_t dim, gc_tensor self);
+raw_tensor atg__sparse_softmax_backward_data_out(gc_tensor out, gc_tensor grad_output, gc_tensor output, int64_t dim, gc_tensor self);
+raw_tensor atg__sparse_softmax_int(gc_tensor self, int64_t dim, int dtype);
+raw_tensor atg__sparse_softmax_out(gc_tensor out, gc_tensor self, int64_t dim, int half_to_float);
+raw_tensor atg__sparse_sparse_matmul(gc_tensor self, gc_tensor other);
+raw_tensor atg__sparse_sparse_matmul_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg__sparse_sum(gc_tensor self);
+raw_tensor atg__sparse_sum_backward(gc_tensor grad, gc_tensor self, int64_t *dim_data, int dim_len);
+raw_tensor atg__sparse_sum_backward_out(gc_tensor out, gc_tensor grad, gc_tensor self, int64_t *dim_data, int dim_len);
+raw_tensor atg__sparse_sum_dim(gc_tensor self, int64_t *dim_data, int dim_len);
+raw_tensor atg__sparse_sum_dim_dtype(gc_tensor self, int64_t *dim_data, int dim_len, int dtype);
+raw_tensor atg__sparse_sum_dim_out(gc_tensor out, gc_tensor self, int64_t *dim_data, int dim_len);
+raw_tensor atg__sparse_sum_dtype(gc_tensor self, int dtype);
+raw_tensor atg__spdiags(gc_tensor diagonals, gc_tensor offsets, int64_t *shape_data, int shape_len);
+raw_tensor atg__spdiags_out(gc_tensor out, gc_tensor diagonals, gc_tensor offsets, int64_t *shape_data, int shape_len);
+raw_tensor atg__stack(gc_tensor *tensors_data, int tensors_len, int64_t dim);
+raw_tensor atg__stack_out(gc_tensor out, gc_tensor *tensors_data, int tensors_len, int64_t dim);
+raw_tensor atg__standard_gamma(gc_tensor self);
+raw_tensor atg__standard_gamma_grad(gc_tensor self, gc_tensor output);
+raw_tensor atg__standard_gamma_grad_out(gc_tensor out, gc_tensor self, gc_tensor output);
+raw_tensor atg__standard_gamma_out(gc_tensor out, gc_tensor self);
+raw_tensor atg__test_ambiguous_defaults(gc_tensor dummy, int64_t a, int64_t b);
+raw_tensor atg__test_ambiguous_defaults_b(gc_tensor dummy, int64_t a, char * b);
+raw_tensor atg__test_autograd_multiple_dispatch(gc_tensor self);
+raw_tensor atg__test_autograd_multiple_dispatch_fullcoverage_out(gc_tensor out, gc_tensor self);
+raw_tensor atg__test_autograd_multiple_dispatch_ntonly(gc_tensor self, int b);
+raw_tensor atg__test_autograd_multiple_dispatch_view(gc_tensor self);
+raw_tensor atg__test_autograd_multiple_dispatch_view_copy(gc_tensor self);
+raw_tensor atg__test_autograd_multiple_dispatch_view_copy_out(gc_tensor out, gc_tensor self);
+raw_tensor atg__test_check_tensor(gc_tensor self);
+raw_tensor atg__test_functorch_fallback(gc_tensor self, gc_tensor other);
+raw_tensor atg__test_functorch_fallback_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg__test_optional_filled_intlist(gc_tensor values, int64_t *addends_data, int addends_len);
+raw_tensor atg__test_optional_filled_intlist_out(gc_tensor out, gc_tensor values, int64_t *addends_data, int addends_len);
+raw_tensor atg__test_optional_floatlist(gc_tensor values, double *addends_data, int addends_len);
+raw_tensor atg__test_optional_floatlist_out(gc_tensor out, gc_tensor values, double *addends_data, int addends_len);
+raw_tensor atg__test_optional_intlist(gc_tensor values, int64_t *addends_data, int addends_len);
+raw_tensor atg__test_optional_intlist_out(gc_tensor out, gc_tensor values, int64_t *addends_data, int addends_len);
+raw_tensor atg__test_serialization_subcmul(gc_tensor self, gc_tensor other);
+raw_tensor atg__test_string_default(gc_tensor dummy, char * a, char * b);
+raw_tensor atg__test_warn_in_autograd(gc_tensor self);
+raw_tensor atg__test_warn_in_autograd_out(gc_tensor out, gc_tensor self);
+void atg__thnn_differentiable_gru_cell_backward(raw_tensor *, gc_tensor grad_hy, gc_tensor input_gates, gc_tensor hidden_gates, gc_tensor hx, gc_tensor input_bias, gc_tensor hidden_bias);
+void atg__thnn_differentiable_lstm_cell_backward(raw_tensor *, gc_tensor grad_hy, gc_tensor grad_cy, gc_tensor input_gates, gc_tensor hidden_gates, gc_tensor input_bias, gc_tensor hidden_bias, gc_tensor cx, gc_tensor cy);
+void atg__thnn_fused_gru_cell(raw_tensor *, gc_tensor input_gates, gc_tensor hidden_gates, gc_tensor hx, gc_tensor input_bias, gc_tensor hidden_bias);
+void atg__thnn_fused_gru_cell_backward(raw_tensor *, gc_tensor grad_hy, gc_tensor workspace, int has_bias);
+void atg__thnn_fused_gru_cell_backward_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor out3, gc_tensor out4, gc_tensor grad_hy, gc_tensor workspace, int has_bias);
+void atg__thnn_fused_gru_cell_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor input_gates, gc_tensor hidden_gates, gc_tensor hx, gc_tensor input_bias, gc_tensor hidden_bias);
+void atg__thnn_fused_lstm_cell(raw_tensor *, gc_tensor input_gates, gc_tensor hidden_gates, gc_tensor cx, gc_tensor input_bias, gc_tensor hidden_bias);
+void atg__thnn_fused_lstm_cell_backward(raw_tensor *, gc_tensor grad_hy, gc_tensor grad_cy, gc_tensor cx, gc_tensor cy, gc_tensor workspace, int has_bias);
+void atg__thnn_fused_lstm_cell_backward_impl(raw_tensor *, gc_tensor grad_hy, gc_tensor grad_cy, gc_tensor cx, gc_tensor cy, gc_tensor workspace, int has_bias);
+void atg__thnn_fused_lstm_cell_backward_impl_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor grad_hy, gc_tensor grad_cy, gc_tensor cx, gc_tensor cy, gc_tensor workspace, int has_bias);
+void atg__thnn_fused_lstm_cell_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor input_gates, gc_tensor hidden_gates, gc_tensor cx, gc_tensor input_bias, gc_tensor hidden_bias);
+raw_tensor atg__to_copy(gc_tensor self, int options_kind, int options_device, int non_blocking);
+raw_tensor atg__to_copy_out(gc_tensor out, gc_tensor self, int non_blocking);
+raw_tensor *atg__to_cpu(gc_tensor *tensors_data, int tensors_len);
+raw_tensor atg__to_dense(gc_tensor self, int dtype, int masked_grad);
+raw_tensor atg__to_dense_out(gc_tensor out, gc_tensor self, int dtype, int masked_grad);
+raw_tensor atg__to_sparse_bsc(gc_tensor self, int64_t *blocksize_data, int blocksize_len, int64_t dense_dim_v, int dense_dim_null);
+raw_tensor atg__to_sparse_bsc_out(gc_tensor out, gc_tensor self, int64_t *blocksize_data, int blocksize_len, int64_t dense_dim_v, int dense_dim_null);
+raw_tensor atg__to_sparse_bsr(gc_tensor self, int64_t *blocksize_data, int blocksize_len, int64_t dense_dim_v, int dense_dim_null);
+raw_tensor atg__to_sparse_bsr_out(gc_tensor out, gc_tensor self, int64_t *blocksize_data, int blocksize_len, int64_t dense_dim_v, int dense_dim_null);
+raw_tensor atg__to_sparse_csc(gc_tensor self, int64_t dense_dim_v, int dense_dim_null);
+raw_tensor atg__to_sparse_csc_out(gc_tensor out, gc_tensor self, int64_t dense_dim_v, int dense_dim_null);
+raw_tensor atg__to_sparse_csr(gc_tensor self, int64_t dense_dim_v, int dense_dim_null);
+raw_tensor atg__to_sparse_csr_out(gc_tensor out, gc_tensor self, int64_t dense_dim_v, int dense_dim_null);
+void atg__to_sparse_semi_structured(raw_tensor *, gc_tensor dense);
+void atg__transform_bias_rescale_qkv(raw_tensor *, gc_tensor qkv, gc_tensor qkv_bias, int64_t num_heads);
+void atg__transform_bias_rescale_qkv_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor qkv, gc_tensor qkv_bias, int64_t num_heads);
+raw_tensor atg__transformer_encoder_layer_fwd(gc_tensor src, int64_t embed_dim, int64_t num_heads, gc_tensor qkv_weight, gc_tensor qkv_bias, gc_tensor proj_weight, gc_tensor proj_bias, int use_gelu, int norm_first, double eps, gc_tensor norm_weight_1, gc_tensor norm_bias_1, gc_tensor norm_weight_2, gc_tensor norm_bias_2, gc_tensor ffn_weight_1, gc_tensor ffn_bias_1, gc_tensor ffn_weight_2, gc_tensor ffn_bias_2, gc_tensor mask, int64_t mask_type_v, int mask_type_null);
+raw_tensor atg__transformer_encoder_layer_fwd_out(gc_tensor out, gc_tensor src, int64_t embed_dim, int64_t num_heads, gc_tensor qkv_weight, gc_tensor qkv_bias, gc_tensor proj_weight, gc_tensor proj_bias, int use_gelu, int norm_first, double eps, gc_tensor norm_weight_1, gc_tensor norm_bias_1, gc_tensor norm_weight_2, gc_tensor norm_bias_2, gc_tensor ffn_weight_1, gc_tensor ffn_bias_1, gc_tensor ffn_weight_2, gc_tensor ffn_bias_2, gc_tensor mask, int64_t mask_type_v, int mask_type_null);
+raw_tensor atg__trilinear(gc_tensor i1, gc_tensor i2, gc_tensor i3, int64_t *expand1_data, int expand1_len, int64_t *expand2_data, int expand2_len, int64_t *expand3_data, int expand3_len, int64_t *sumdim_data, int sumdim_len, int64_t unroll_dim);
+raw_tensor atg__trilinear_out(gc_tensor out, gc_tensor i1, gc_tensor i2, gc_tensor i3, int64_t *expand1_data, int expand1_len, int64_t *expand2_data, int expand2_len, int64_t *expand3_data, int expand3_len, int64_t *sumdim_data, int sumdim_len, int64_t unroll_dim);
+raw_tensor atg__triton_multi_head_attention(gc_tensor query, gc_tensor key, gc_tensor value, int64_t embed_dim, int64_t num_head, gc_tensor qkv_weight, gc_tensor qkv_bias, gc_tensor proj_weight, gc_tensor proj_bias, gc_tensor mask);
+raw_tensor atg__triton_multi_head_attention_out(gc_tensor out, gc_tensor query, gc_tensor key, gc_tensor value, int64_t embed_dim, int64_t num_head, gc_tensor qkv_weight, gc_tensor qkv_bias, gc_tensor proj_weight, gc_tensor proj_bias, gc_tensor mask);
+raw_tensor atg__triton_scaled_dot_attention(gc_tensor q, gc_tensor k, gc_tensor v, double dropout_p);
+raw_tensor atg__triton_scaled_dot_attention_out(gc_tensor out, gc_tensor q, gc_tensor k, gc_tensor v, double dropout_p);
+void atg__unique(raw_tensor *, gc_tensor self, int sorted, int return_inverse);
+void atg__unique2(raw_tensor *, gc_tensor self, int sorted, int return_inverse, int return_counts);
+void atg__unique2_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor self, int sorted, int return_inverse, int return_counts);
+void atg__unique_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor self, int sorted, int return_inverse);
+void atg__unpack_dual(raw_tensor *, gc_tensor dual, int64_t level);
+raw_tensor atg__unsafe_index(gc_tensor self, gc_tensor *indices_data, int indices_len);
+raw_tensor atg__unsafe_index_put(gc_tensor self, gc_tensor *indices_data, int indices_len, gc_tensor values, int accumulate);
+raw_tensor atg__unsafe_view(gc_tensor self, int64_t *size_data, int size_len);
+raw_tensor atg__unsafe_view_out(gc_tensor out, gc_tensor self, int64_t *size_data, int size_len);
+raw_tensor atg__upsample_bicubic2d_aa(gc_tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
+raw_tensor atg__upsample_bicubic2d_aa_backward(gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
+raw_tensor atg__upsample_bicubic2d_aa_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
+raw_tensor atg__upsample_bicubic2d_aa_out(gc_tensor out, gc_tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
+raw_tensor atg__upsample_bicubic2d_aa_vec(gc_tensor input, int64_t *output_size_data, int output_size_len, int align_corners, double *scale_factors_data, int scale_factors_len);
+raw_tensor atg__upsample_bilinear2d_aa(gc_tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
+raw_tensor atg__upsample_bilinear2d_aa_backward(gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
+raw_tensor atg__upsample_bilinear2d_aa_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
+raw_tensor atg__upsample_bilinear2d_aa_out(gc_tensor out, gc_tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
+raw_tensor atg__upsample_bilinear2d_aa_vec(gc_tensor input, int64_t *output_size_data, int output_size_len, int align_corners, double *scale_factors_data, int scale_factors_len);
+raw_tensor atg__upsample_nearest_exact1d(gc_tensor self, int64_t *output_size_data, int output_size_len, double scales_v, int scales_null);
+raw_tensor atg__upsample_nearest_exact1d_backward(gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_v, int scales_null);
+raw_tensor atg__upsample_nearest_exact1d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_v, int scales_null);
+raw_tensor atg__upsample_nearest_exact1d_out(gc_tensor out, gc_tensor self, int64_t *output_size_data, int output_size_len, double scales_v, int scales_null);
+raw_tensor atg__upsample_nearest_exact1d_vec(gc_tensor input, int64_t *output_size_data, int output_size_len, double *scale_factors_data, int scale_factors_len);
+raw_tensor atg__upsample_nearest_exact2d(gc_tensor self, int64_t *output_size_data, int output_size_len, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
+raw_tensor atg__upsample_nearest_exact2d_backward(gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
+raw_tensor atg__upsample_nearest_exact2d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
+raw_tensor atg__upsample_nearest_exact2d_out(gc_tensor out, gc_tensor self, int64_t *output_size_data, int output_size_len, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
+raw_tensor atg__upsample_nearest_exact2d_vec(gc_tensor input, int64_t *output_size_data, int output_size_len, double *scale_factors_data, int scale_factors_len);
+raw_tensor atg__upsample_nearest_exact3d(gc_tensor self, int64_t *output_size_data, int output_size_len, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
+raw_tensor atg__upsample_nearest_exact3d_backward(gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
+raw_tensor atg__upsample_nearest_exact3d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
+raw_tensor atg__upsample_nearest_exact3d_out(gc_tensor out, gc_tensor self, int64_t *output_size_data, int output_size_len, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
+raw_tensor atg__upsample_nearest_exact3d_vec(gc_tensor input, int64_t *output_size_data, int output_size_len, double *scale_factors_data, int scale_factors_len);
+int atg__use_cudnn_ctc_loss(gc_tensor log_probs, gc_tensor targets, int64_t *input_lengths_data, int input_lengths_len, int64_t *target_lengths_data, int target_lengths_len, int64_t blank);
+int atg__use_cudnn_ctc_loss_tensor(gc_tensor log_probs, gc_tensor targets, gc_tensor input_lengths, gc_tensor target_lengths, int64_t blank);
 int atg__use_cudnn_rnn_flatten_weight();
-void atg__validate_compressed_sparse_indices(int is_crow, tensor compressed_idx, tensor plain_idx, int64_t cdim, int64_t dim, int64_t nnz);
-void atg__validate_sparse_bsc_tensor_args(tensor ccol_indices, tensor row_indices, tensor values, int64_t *size_data, int size_len);
-void atg__validate_sparse_bsr_tensor_args(tensor crow_indices, tensor col_indices, tensor values, int64_t *size_data, int size_len);
-void atg__validate_sparse_csc_tensor_args(tensor ccol_indices, tensor row_indices, tensor values, int64_t *size_data, int size_len);
-void atg__values(tensor *, tensor self);
-void atg__values_copy(tensor *, tensor self);
-void atg__values_copy_out(tensor *, tensor out, tensor self);
-int64_t atg__version(tensor self);
-void atg__weight_norm(tensor *, tensor v, tensor g, int64_t dim);
-void atg__weight_norm_differentiable_backward(tensor *, tensor grad_w, tensor saved_v, tensor saved_g, tensor saved_norms, int64_t dim);
-void atg__weight_norm_interface(tensor *, tensor v, tensor g, int64_t dim);
-void atg__weight_norm_interface_backward(tensor *, tensor grad_w, tensor saved_v, tensor saved_g, tensor saved_norms, int64_t dim);
-void atg__weight_norm_interface_backward_out(tensor *, tensor out0, tensor out1, tensor grad_w, tensor saved_v, tensor saved_g, tensor saved_norms, int64_t dim);
-void atg__weight_norm_interface_out(tensor *, tensor out0, tensor out1, tensor v, tensor g, int64_t dim);
-void atg_abs(tensor *, tensor self);
-void atg_abs_(tensor *, tensor self);
-void atg_abs_out(tensor *, tensor out, tensor self);
-void atg_absolute(tensor *, tensor self);
-void atg_absolute_(tensor *, tensor self);
-void atg_absolute_out(tensor *, tensor out, tensor self);
-void atg_acos(tensor *, tensor self);
-void atg_acos_(tensor *, tensor self);
-void atg_acos_out(tensor *, tensor out, tensor self);
-void atg_acosh(tensor *, tensor self);
-void atg_acosh_(tensor *, tensor self);
-void atg_acosh_out(tensor *, tensor out, tensor self);
-void atg_adaptive_avg_pool1d(tensor *, tensor self, int64_t *output_size_data, int output_size_len);
-void atg_adaptive_avg_pool2d(tensor *, tensor self, int64_t *output_size_data, int output_size_len);
-void atg_adaptive_avg_pool2d_out(tensor *, tensor out, tensor self, int64_t *output_size_data, int output_size_len);
-void atg_adaptive_avg_pool3d(tensor *, tensor self, int64_t *output_size_data, int output_size_len);
-void atg_adaptive_avg_pool3d_backward(tensor *, tensor grad_input, tensor grad_output, tensor self);
-void atg_adaptive_avg_pool3d_out(tensor *, tensor out, tensor self, int64_t *output_size_data, int output_size_len);
-void atg_adaptive_max_pool1d(tensor *, tensor self, int64_t *output_size_data, int output_size_len);
-void atg_adaptive_max_pool2d(tensor *, tensor self, int64_t *output_size_data, int output_size_len);
-void atg_adaptive_max_pool2d_backward(tensor *, tensor grad_output, tensor self, tensor indices);
-void atg_adaptive_max_pool2d_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, tensor self, tensor indices);
-void atg_adaptive_max_pool2d_out(tensor *, tensor out, tensor indices, tensor self, int64_t *output_size_data, int output_size_len);
-void atg_adaptive_max_pool3d(tensor *, tensor self, int64_t *output_size_data, int output_size_len);
-void atg_adaptive_max_pool3d_backward(tensor *, tensor grad_output, tensor self, tensor indices);
-void atg_adaptive_max_pool3d_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, tensor self, tensor indices);
-void atg_adaptive_max_pool3d_out(tensor *, tensor out, tensor indices, tensor self, int64_t *output_size_data, int output_size_len);
-void atg_add(tensor *, tensor self, tensor other);
-void atg_add_(tensor *, tensor self, tensor other);
-void atg_add_out(tensor *, tensor out, tensor self, tensor other);
-void atg_add_scalar(tensor *, tensor self, scalar other);
-void atg_add_scalar_(tensor *, tensor self, scalar other);
-void atg_add_scalar_out(tensor *, tensor out, tensor self, scalar other);
-void atg_addbmm(tensor *, tensor self, tensor batch1, tensor batch2);
-void atg_addbmm_(tensor *, tensor self, tensor batch1, tensor batch2);
-void atg_addbmm_out(tensor *, tensor out, tensor self, tensor batch1, tensor batch2);
-void atg_addcdiv(tensor *, tensor self, tensor tensor1, tensor tensor2);
-void atg_addcdiv_(tensor *, tensor self, tensor tensor1, tensor tensor2);
-void atg_addcdiv_out(tensor *, tensor out, tensor self, tensor tensor1, tensor tensor2);
-void atg_addcmul(tensor *, tensor self, tensor tensor1, tensor tensor2);
-void atg_addcmul_(tensor *, tensor self, tensor tensor1, tensor tensor2);
-void atg_addcmul_out(tensor *, tensor out, tensor self, tensor tensor1, tensor tensor2);
-void atg_addmm(tensor *, tensor self, tensor mat1, tensor mat2);
-void atg_addmm_(tensor *, tensor self, tensor mat1, tensor mat2);
-void atg_addmm_out(tensor *, tensor out, tensor self, tensor mat1, tensor mat2);
-void atg_addmv(tensor *, tensor self, tensor mat, tensor vec);
-void atg_addmv_(tensor *, tensor self, tensor mat, tensor vec);
-void atg_addmv_out(tensor *, tensor out, tensor self, tensor mat, tensor vec);
-void atg_addr(tensor *, tensor self, tensor vec1, tensor vec2);
-void atg_addr_(tensor *, tensor self, tensor vec1, tensor vec2);
-void atg_addr_out(tensor *, tensor out, tensor self, tensor vec1, tensor vec2);
-void atg_adjoint(tensor *, tensor self);
-void atg_affine_grid_generator(tensor *, tensor theta, int64_t *size_data, int size_len, int align_corners);
-void atg_affine_grid_generator_backward(tensor *, tensor grad, int64_t *size_data, int size_len, int align_corners);
-void atg_affine_grid_generator_out(tensor *, tensor out, tensor theta, int64_t *size_data, int size_len, int align_corners);
-void atg_alias(tensor *, tensor self);
-void atg_alias_copy(tensor *, tensor self);
-void atg_alias_copy_out(tensor *, tensor out, tensor self);
-void atg_align_as(tensor *, tensor self, tensor other);
-tensor *atg_align_tensors(tensor *tensors_data, int tensors_len);
-void atg_all(tensor *, tensor self);
-void atg_all_all_out(tensor *, tensor out, tensor self);
-void atg_all_dim(tensor *, tensor self, int64_t dim, int keepdim);
-void atg_all_out(tensor *, tensor out, tensor self, int64_t dim, int keepdim);
-int atg_allclose(tensor self, tensor other, double rtol, double atol, int equal_nan);
-void atg_alpha_dropout(tensor *, tensor input, double p, int train);
-void atg_alpha_dropout_(tensor *, tensor self, double p, int train);
-void atg_amax(tensor *, tensor self, int64_t *dim_data, int dim_len, int keepdim);
-void atg_amax_out(tensor *, tensor out, tensor self, int64_t *dim_data, int dim_len, int keepdim);
-void atg_amin(tensor *, tensor self, int64_t *dim_data, int dim_len, int keepdim);
-void atg_amin_out(tensor *, tensor out, tensor self, int64_t *dim_data, int dim_len, int keepdim);
-void atg_aminmax(tensor *, tensor self, int64_t dim_v, int dim_null, int keepdim);
-void atg_aminmax_out(tensor *, tensor min, tensor max, tensor self, int64_t dim_v, int dim_null, int keepdim);
-void atg_angle(tensor *, tensor self);
-void atg_angle_out(tensor *, tensor out, tensor self);
-void atg_any(tensor *, tensor self);
-void atg_any_all_out(tensor *, tensor out, tensor self);
-void atg_any_dim(tensor *, tensor self, int64_t dim, int keepdim);
-void atg_any_out(tensor *, tensor out, tensor self, int64_t dim, int keepdim);
-void atg_arange(tensor *, scalar end, int options_kind, int options_device);
-void atg_arange_start(tensor *, scalar start, scalar end, int options_kind, int options_device);
-void atg_arange_start_step(tensor *, scalar start, scalar end, int options_kind, int options_device);
-void atg_arccos(tensor *, tensor self);
-void atg_arccos_(tensor *, tensor self);
-void atg_arccos_out(tensor *, tensor out, tensor self);
-void atg_arccosh(tensor *, tensor self);
-void atg_arccosh_(tensor *, tensor self);
-void atg_arccosh_out(tensor *, tensor out, tensor self);
-void atg_arcsin(tensor *, tensor self);
-void atg_arcsin_(tensor *, tensor self);
-void atg_arcsin_out(tensor *, tensor out, tensor self);
-void atg_arcsinh(tensor *, tensor self);
-void atg_arcsinh_(tensor *, tensor self);
-void atg_arcsinh_out(tensor *, tensor out, tensor self);
-void atg_arctan(tensor *, tensor self);
-void atg_arctan2(tensor *, tensor self, tensor other);
-void atg_arctan2_(tensor *, tensor self, tensor other);
-void atg_arctan2_out(tensor *, tensor out, tensor self, tensor other);
-void atg_arctan_(tensor *, tensor self);
-void atg_arctan_out(tensor *, tensor out, tensor self);
-void atg_arctanh(tensor *, tensor self);
-void atg_arctanh_(tensor *, tensor self);
-void atg_arctanh_out(tensor *, tensor out, tensor self);
-void atg_argmax(tensor *, tensor self, int64_t dim_v, int dim_null, int keepdim);
-void atg_argmax_out(tensor *, tensor out, tensor self, int64_t dim_v, int dim_null, int keepdim);
-void atg_argmin(tensor *, tensor self, int64_t dim_v, int dim_null, int keepdim);
-void atg_argmin_out(tensor *, tensor out, tensor self, int64_t dim_v, int dim_null, int keepdim);
-void atg_argsort(tensor *, tensor self, int64_t dim, int descending);
-void atg_argsort_stable(tensor *, tensor self, int stable, int64_t dim, int descending);
-void atg_argsort_stable_out(tensor *, tensor out, tensor self, int stable, int64_t dim, int descending);
-void atg_argwhere(tensor *, tensor self);
-void atg_as_strided(tensor *, tensor self, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len, int64_t storage_offset_v, int storage_offset_null);
-void atg_as_strided_(tensor *, tensor self, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len, int64_t storage_offset_v, int storage_offset_null);
-void atg_as_strided_copy(tensor *, tensor self, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len, int64_t storage_offset_v, int storage_offset_null);
-void atg_as_strided_copy_out(tensor *, tensor out, tensor self, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len, int64_t storage_offset_v, int storage_offset_null);
-void atg_as_strided_scatter(tensor *, tensor self, tensor src, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len, int64_t storage_offset_v, int storage_offset_null);
-void atg_as_strided_scatter_out(tensor *, tensor out, tensor self, tensor src, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len, int64_t storage_offset_v, int storage_offset_null);
-void atg_asin(tensor *, tensor self);
-void atg_asin_(tensor *, tensor self);
-void atg_asin_out(tensor *, tensor out, tensor self);
-void atg_asinh(tensor *, tensor self);
-void atg_asinh_(tensor *, tensor self);
-void atg_asinh_out(tensor *, tensor out, tensor self);
-void atg_atan(tensor *, tensor self);
-void atg_atan2(tensor *, tensor self, tensor other);
-void atg_atan2_(tensor *, tensor self, tensor other);
-void atg_atan2_out(tensor *, tensor out, tensor self, tensor other);
-void atg_atan_(tensor *, tensor self);
-void atg_atan_out(tensor *, tensor out, tensor self);
-void atg_atanh(tensor *, tensor self);
-void atg_atanh_(tensor *, tensor self);
-void atg_atanh_out(tensor *, tensor out, tensor self);
-void atg_atleast_1d(tensor *, tensor self);
-tensor *atg_atleast_1d_sequence(tensor *tensors_data, int tensors_len);
-void atg_atleast_2d(tensor *, tensor self);
-tensor *atg_atleast_2d_sequence(tensor *tensors_data, int tensors_len);
-void atg_atleast_3d(tensor *, tensor self);
-tensor *atg_atleast_3d_sequence(tensor *tensors_data, int tensors_len);
-void atg_avg_pool1d(tensor *, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int ceil_mode, int count_include_pad);
-void atg_avg_pool2d(tensor *, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int ceil_mode, int count_include_pad, int64_t divisor_override_v, int divisor_override_null);
-void atg_avg_pool2d_backward(tensor *, tensor grad_output, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int ceil_mode, int count_include_pad, int64_t divisor_override_v, int divisor_override_null);
-void atg_avg_pool2d_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int ceil_mode, int count_include_pad, int64_t divisor_override_v, int divisor_override_null);
-void atg_avg_pool2d_out(tensor *, tensor out, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int ceil_mode, int count_include_pad, int64_t divisor_override_v, int divisor_override_null);
-void atg_avg_pool3d(tensor *, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int ceil_mode, int count_include_pad, int64_t divisor_override_v, int divisor_override_null);
-void atg_avg_pool3d_backward(tensor *, tensor grad_output, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int ceil_mode, int count_include_pad, int64_t divisor_override_v, int divisor_override_null);
-void atg_avg_pool3d_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int ceil_mode, int count_include_pad, int64_t divisor_override_v, int divisor_override_null);
-void atg_avg_pool3d_out(tensor *, tensor out, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int ceil_mode, int count_include_pad, int64_t divisor_override_v, int divisor_override_null);
-void atg_baddbmm(tensor *, tensor self, tensor batch1, tensor batch2);
-void atg_baddbmm_(tensor *, tensor self, tensor batch1, tensor batch2);
-void atg_baddbmm_out(tensor *, tensor out, tensor self, tensor batch1, tensor batch2);
-void atg_bartlett_window(tensor *, int64_t window_length, int options_kind, int options_device);
-void atg_bartlett_window_out(tensor *, tensor out, int64_t window_length);
-void atg_bartlett_window_periodic(tensor *, int64_t window_length, int periodic, int options_kind, int options_device);
-void atg_bartlett_window_periodic_out(tensor *, tensor out, int64_t window_length, int periodic);
-void atg_batch_norm(tensor *, tensor input, tensor weight, tensor bias, tensor running_mean, tensor running_var, int training, double momentum, double eps, int cudnn_enabled);
-void atg_batch_norm_backward_elemt(tensor *, tensor grad_out, tensor input, tensor mean, tensor invstd, tensor weight, tensor sum_dy, tensor sum_dy_xmu, tensor count);
-void atg_batch_norm_backward_elemt_out(tensor *, tensor out, tensor grad_out, tensor input, tensor mean, tensor invstd, tensor weight, tensor sum_dy, tensor sum_dy_xmu, tensor count);
-void atg_batch_norm_backward_reduce(tensor *, tensor grad_out, tensor input, tensor mean, tensor invstd, tensor weight, int input_g, int weight_g, int bias_g);
-void atg_batch_norm_backward_reduce_out(tensor *, tensor out0, tensor out1, tensor out2, tensor out3, tensor grad_out, tensor input, tensor mean, tensor invstd, tensor weight, int input_g, int weight_g, int bias_g);
-void atg_batch_norm_elemt(tensor *, tensor input, tensor weight, tensor bias, tensor mean, tensor invstd, double eps);
-void atg_batch_norm_elemt_out(tensor *, tensor out, tensor input, tensor weight, tensor bias, tensor mean, tensor invstd, double eps);
-void atg_batch_norm_gather_stats(tensor *, tensor input, tensor mean, tensor invstd, tensor running_mean, tensor running_var, double momentum, double eps, int64_t count);
-void atg_batch_norm_gather_stats_out(tensor *, tensor out0, tensor out1, tensor input, tensor mean, tensor invstd, tensor running_mean, tensor running_var, double momentum, double eps, int64_t count);
-void atg_batch_norm_gather_stats_with_counts(tensor *, tensor input, tensor mean, tensor invstd, tensor running_mean, tensor running_var, double momentum, double eps, tensor counts);
-void atg_batch_norm_gather_stats_with_counts_out(tensor *, tensor out0, tensor out1, tensor input, tensor mean, tensor invstd, tensor running_mean, tensor running_var, double momentum, double eps, tensor counts);
-void atg_batch_norm_stats(tensor *, tensor input, double eps);
-void atg_batch_norm_stats_out(tensor *, tensor out0, tensor out1, tensor input, double eps);
-void atg_batch_norm_update_stats(tensor *, tensor input, tensor running_mean, tensor running_var, double momentum);
-void atg_batch_norm_update_stats_out(tensor *, tensor out0, tensor out1, tensor input, tensor running_mean, tensor running_var, double momentum);
-void atg_bernoulli(tensor *, tensor self);
-void atg_bernoulli_(tensor *, tensor self, tensor p);
-void atg_bernoulli_float_(tensor *, tensor self, double p);
-void atg_bernoulli_p(tensor *, tensor self, double p);
-void atg_bernoulli_tensor(tensor *, tensor self, tensor p);
-void atg_bilinear(tensor *, tensor input1, tensor input2, tensor weight, tensor bias);
-void atg_binary_cross_entropy(tensor *, tensor self, tensor target, tensor weight, int64_t reduction);
-void atg_binary_cross_entropy_backward(tensor *, tensor grad_output, tensor self, tensor target, tensor weight, int64_t reduction);
-void atg_binary_cross_entropy_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, tensor self, tensor target, tensor weight, int64_t reduction);
-void atg_binary_cross_entropy_out(tensor *, tensor out, tensor self, tensor target, tensor weight, int64_t reduction);
-void atg_binary_cross_entropy_with_logits(tensor *, tensor self, tensor target, tensor weight, tensor pos_weight, int64_t reduction);
-void atg_binary_cross_entropy_with_logits_out(tensor *, tensor out, tensor self, tensor target, tensor weight, tensor pos_weight, int64_t reduction);
-void atg_bincount(tensor *, tensor self, tensor weights, int64_t minlength);
-void atg_bincount_out(tensor *, tensor out, tensor self, tensor weights, int64_t minlength);
-void atg_binomial(tensor *, tensor count, tensor prob);
-void atg_binomial_out(tensor *, tensor out, tensor count, tensor prob);
-void atg_bitwise_and(tensor *, tensor self, scalar other);
-void atg_bitwise_and_(tensor *, tensor self, scalar other);
-void atg_bitwise_and_scalar_out(tensor *, tensor out, tensor self, scalar other);
-void atg_bitwise_and_scalar_tensor(tensor *, scalar self, tensor other);
-void atg_bitwise_and_scalar_tensor_out(tensor *, tensor out, scalar self, tensor other);
-void atg_bitwise_and_tensor(tensor *, tensor self, tensor other);
-void atg_bitwise_and_tensor_(tensor *, tensor self, tensor other);
-void atg_bitwise_and_tensor_out(tensor *, tensor out, tensor self, tensor other);
-void atg_bitwise_left_shift(tensor *, tensor self, tensor other);
-void atg_bitwise_left_shift_(tensor *, tensor self, tensor other);
-void atg_bitwise_left_shift_scalar_tensor(tensor *, scalar self, tensor other);
-void atg_bitwise_left_shift_scalar_tensor_out(tensor *, tensor out, scalar self, tensor other);
-void atg_bitwise_left_shift_tensor_out(tensor *, tensor out, tensor self, tensor other);
-void atg_bitwise_left_shift_tensor_scalar(tensor *, tensor self, scalar other);
-void atg_bitwise_left_shift_tensor_scalar_(tensor *, tensor self, scalar other);
-void atg_bitwise_left_shift_tensor_scalar_out(tensor *, tensor out, tensor self, scalar other);
-void atg_bitwise_not(tensor *, tensor self);
-void atg_bitwise_not_(tensor *, tensor self);
-void atg_bitwise_not_out(tensor *, tensor out, tensor self);
-void atg_bitwise_or(tensor *, tensor self, scalar other);
-void atg_bitwise_or_(tensor *, tensor self, scalar other);
-void atg_bitwise_or_scalar_out(tensor *, tensor out, tensor self, scalar other);
-void atg_bitwise_or_scalar_tensor(tensor *, scalar self, tensor other);
-void atg_bitwise_or_scalar_tensor_out(tensor *, tensor out, scalar self, tensor other);
-void atg_bitwise_or_tensor(tensor *, tensor self, tensor other);
-void atg_bitwise_or_tensor_(tensor *, tensor self, tensor other);
-void atg_bitwise_or_tensor_out(tensor *, tensor out, tensor self, tensor other);
-void atg_bitwise_right_shift(tensor *, tensor self, tensor other);
-void atg_bitwise_right_shift_(tensor *, tensor self, tensor other);
-void atg_bitwise_right_shift_scalar_tensor(tensor *, scalar self, tensor other);
-void atg_bitwise_right_shift_scalar_tensor_out(tensor *, tensor out, scalar self, tensor other);
-void atg_bitwise_right_shift_tensor_out(tensor *, tensor out, tensor self, tensor other);
-void atg_bitwise_right_shift_tensor_scalar(tensor *, tensor self, scalar other);
-void atg_bitwise_right_shift_tensor_scalar_(tensor *, tensor self, scalar other);
-void atg_bitwise_right_shift_tensor_scalar_out(tensor *, tensor out, tensor self, scalar other);
-void atg_bitwise_xor(tensor *, tensor self, scalar other);
-void atg_bitwise_xor_(tensor *, tensor self, scalar other);
-void atg_bitwise_xor_scalar_out(tensor *, tensor out, tensor self, scalar other);
-void atg_bitwise_xor_scalar_tensor(tensor *, scalar self, tensor other);
-void atg_bitwise_xor_scalar_tensor_out(tensor *, tensor out, scalar self, tensor other);
-void atg_bitwise_xor_tensor(tensor *, tensor self, tensor other);
-void atg_bitwise_xor_tensor_(tensor *, tensor self, tensor other);
-void atg_bitwise_xor_tensor_out(tensor *, tensor out, tensor self, tensor other);
-void atg_blackman_window(tensor *, int64_t window_length, int options_kind, int options_device);
-void atg_blackman_window_out(tensor *, tensor out, int64_t window_length);
-void atg_blackman_window_periodic(tensor *, int64_t window_length, int periodic, int options_kind, int options_device);
-void atg_blackman_window_periodic_out(tensor *, tensor out, int64_t window_length, int periodic);
-void atg_block_diag(tensor *, tensor *tensors_data, int tensors_len);
-void atg_block_diag_out(tensor *, tensor out, tensor *tensors_data, int tensors_len);
-void atg_bmm(tensor *, tensor self, tensor mat2);
-void atg_bmm_out(tensor *, tensor out, tensor self, tensor mat2);
-tensor *atg_broadcast_tensors(tensor *tensors_data, int tensors_len);
-void atg_broadcast_to(tensor *, tensor self, int64_t *size_data, int size_len);
-void atg_bucketize(tensor *, tensor self, tensor boundaries, int out_int32, int right);
-void atg_bucketize_scalar(tensor *, scalar self, tensor boundaries, int out_int32, int right);
-void atg_bucketize_scalar_out(tensor *, tensor out, scalar self, tensor boundaries, int out_int32, int right);
-void atg_bucketize_tensor_out(tensor *, tensor out, tensor self, tensor boundaries, int out_int32, int right);
+void atg__validate_compressed_sparse_indices(int is_crow, gc_tensor compressed_idx, gc_tensor plain_idx, int64_t cdim, int64_t dim, int64_t nnz);
+void atg__validate_sparse_bsc_tensor_args(gc_tensor ccol_indices, gc_tensor row_indices, gc_tensor values, int64_t *size_data, int size_len);
+void atg__validate_sparse_bsr_tensor_args(gc_tensor crow_indices, gc_tensor col_indices, gc_tensor values, int64_t *size_data, int size_len);
+void atg__validate_sparse_csc_tensor_args(gc_tensor ccol_indices, gc_tensor row_indices, gc_tensor values, int64_t *size_data, int size_len);
+raw_tensor atg__values(gc_tensor self);
+raw_tensor atg__values_copy(gc_tensor self);
+raw_tensor atg__values_copy_out(gc_tensor out, gc_tensor self);
+int64_t atg__version(gc_tensor self);
+raw_tensor atg__weight_norm(gc_tensor v, gc_tensor g, int64_t dim);
+void atg__weight_norm_differentiable_backward(raw_tensor *, gc_tensor grad_w, gc_tensor saved_v, gc_tensor saved_g, gc_tensor saved_norms, int64_t dim);
+void atg__weight_norm_interface(raw_tensor *, gc_tensor v, gc_tensor g, int64_t dim);
+void atg__weight_norm_interface_backward(raw_tensor *, gc_tensor grad_w, gc_tensor saved_v, gc_tensor saved_g, gc_tensor saved_norms, int64_t dim);
+void atg__weight_norm_interface_backward_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor grad_w, gc_tensor saved_v, gc_tensor saved_g, gc_tensor saved_norms, int64_t dim);
+void atg__weight_norm_interface_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor v, gc_tensor g, int64_t dim);
+raw_tensor atg_abs(gc_tensor self);
+raw_tensor atg_abs_(gc_tensor self);
+raw_tensor atg_abs_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_absolute(gc_tensor self);
+raw_tensor atg_absolute_(gc_tensor self);
+raw_tensor atg_absolute_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_acos(gc_tensor self);
+raw_tensor atg_acos_(gc_tensor self);
+raw_tensor atg_acos_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_acosh(gc_tensor self);
+raw_tensor atg_acosh_(gc_tensor self);
+raw_tensor atg_acosh_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_adaptive_avg_pool1d(gc_tensor self, int64_t *output_size_data, int output_size_len);
+raw_tensor atg_adaptive_avg_pool2d(gc_tensor self, int64_t *output_size_data, int output_size_len);
+raw_tensor atg_adaptive_avg_pool2d_out(gc_tensor out, gc_tensor self, int64_t *output_size_data, int output_size_len);
+raw_tensor atg_adaptive_avg_pool3d(gc_tensor self, int64_t *output_size_data, int output_size_len);
+raw_tensor atg_adaptive_avg_pool3d_backward(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self);
+raw_tensor atg_adaptive_avg_pool3d_out(gc_tensor out, gc_tensor self, int64_t *output_size_data, int output_size_len);
+void atg_adaptive_max_pool1d(raw_tensor *, gc_tensor self, int64_t *output_size_data, int output_size_len);
+void atg_adaptive_max_pool2d(raw_tensor *, gc_tensor self, int64_t *output_size_data, int output_size_len);
+raw_tensor atg_adaptive_max_pool2d_backward(gc_tensor grad_output, gc_tensor self, gc_tensor indices);
+raw_tensor atg_adaptive_max_pool2d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, gc_tensor indices);
+void atg_adaptive_max_pool2d_out(raw_tensor *, gc_tensor out, gc_tensor indices, gc_tensor self, int64_t *output_size_data, int output_size_len);
+void atg_adaptive_max_pool3d(raw_tensor *, gc_tensor self, int64_t *output_size_data, int output_size_len);
+raw_tensor atg_adaptive_max_pool3d_backward(gc_tensor grad_output, gc_tensor self, gc_tensor indices);
+raw_tensor atg_adaptive_max_pool3d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, gc_tensor indices);
+void atg_adaptive_max_pool3d_out(raw_tensor *, gc_tensor out, gc_tensor indices, gc_tensor self, int64_t *output_size_data, int output_size_len);
+raw_tensor atg_add(gc_tensor self, gc_tensor other);
+raw_tensor atg_add_(gc_tensor self, gc_tensor other);
+raw_tensor atg_add_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_add_scalar(gc_tensor self, scalar other);
+raw_tensor atg_add_scalar_(gc_tensor self, scalar other);
+raw_tensor atg_add_scalar_out(gc_tensor out, gc_tensor self, scalar other);
+raw_tensor atg_addbmm(gc_tensor self, gc_tensor batch1, gc_tensor batch2);
+raw_tensor atg_addbmm_(gc_tensor self, gc_tensor batch1, gc_tensor batch2);
+raw_tensor atg_addbmm_out(gc_tensor out, gc_tensor self, gc_tensor batch1, gc_tensor batch2);
+raw_tensor atg_addcdiv(gc_tensor self, gc_tensor tensor1, gc_tensor tensor2);
+raw_tensor atg_addcdiv_(gc_tensor self, gc_tensor tensor1, gc_tensor tensor2);
+raw_tensor atg_addcdiv_out(gc_tensor out, gc_tensor self, gc_tensor tensor1, gc_tensor tensor2);
+raw_tensor atg_addcmul(gc_tensor self, gc_tensor tensor1, gc_tensor tensor2);
+raw_tensor atg_addcmul_(gc_tensor self, gc_tensor tensor1, gc_tensor tensor2);
+raw_tensor atg_addcmul_out(gc_tensor out, gc_tensor self, gc_tensor tensor1, gc_tensor tensor2);
+raw_tensor atg_addmm(gc_tensor self, gc_tensor mat1, gc_tensor mat2);
+raw_tensor atg_addmm_(gc_tensor self, gc_tensor mat1, gc_tensor mat2);
+raw_tensor atg_addmm_out(gc_tensor out, gc_tensor self, gc_tensor mat1, gc_tensor mat2);
+raw_tensor atg_addmv(gc_tensor self, gc_tensor mat, gc_tensor vec);
+raw_tensor atg_addmv_(gc_tensor self, gc_tensor mat, gc_tensor vec);
+raw_tensor atg_addmv_out(gc_tensor out, gc_tensor self, gc_tensor mat, gc_tensor vec);
+raw_tensor atg_addr(gc_tensor self, gc_tensor vec1, gc_tensor vec2);
+raw_tensor atg_addr_(gc_tensor self, gc_tensor vec1, gc_tensor vec2);
+raw_tensor atg_addr_out(gc_tensor out, gc_tensor self, gc_tensor vec1, gc_tensor vec2);
+raw_tensor atg_adjoint(gc_tensor self);
+raw_tensor atg_affine_grid_generator(gc_tensor theta, int64_t *size_data, int size_len, int align_corners);
+raw_tensor atg_affine_grid_generator_backward(gc_tensor grad, int64_t *size_data, int size_len, int align_corners);
+raw_tensor atg_affine_grid_generator_out(gc_tensor out, gc_tensor theta, int64_t *size_data, int size_len, int align_corners);
+raw_tensor atg_alias(gc_tensor self);
+raw_tensor atg_alias_copy(gc_tensor self);
+raw_tensor atg_alias_copy_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_align_as(gc_tensor self, gc_tensor other);
+raw_tensor *atg_align_tensors(gc_tensor *tensors_data, int tensors_len);
+raw_tensor atg_all(gc_tensor self);
+raw_tensor atg_all_all_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_all_dim(gc_tensor self, int64_t dim, int keepdim);
+raw_tensor atg_all_out(gc_tensor out, gc_tensor self, int64_t dim, int keepdim);
+int atg_allclose(gc_tensor self, gc_tensor other, double rtol, double atol, int equal_nan);
+raw_tensor atg_alpha_dropout(gc_tensor input, double p, int train);
+raw_tensor atg_alpha_dropout_(gc_tensor self, double p, int train);
+raw_tensor atg_amax(gc_tensor self, int64_t *dim_data, int dim_len, int keepdim);
+raw_tensor atg_amax_out(gc_tensor out, gc_tensor self, int64_t *dim_data, int dim_len, int keepdim);
+raw_tensor atg_amin(gc_tensor self, int64_t *dim_data, int dim_len, int keepdim);
+raw_tensor atg_amin_out(gc_tensor out, gc_tensor self, int64_t *dim_data, int dim_len, int keepdim);
+void atg_aminmax(raw_tensor *, gc_tensor self, int64_t dim_v, int dim_null, int keepdim);
+void atg_aminmax_out(raw_tensor *, gc_tensor min, gc_tensor max, gc_tensor self, int64_t dim_v, int dim_null, int keepdim);
+raw_tensor atg_angle(gc_tensor self);
+raw_tensor atg_angle_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_any(gc_tensor self);
+raw_tensor atg_any_all_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_any_dim(gc_tensor self, int64_t dim, int keepdim);
+raw_tensor atg_any_out(gc_tensor out, gc_tensor self, int64_t dim, int keepdim);
+raw_tensor atg_arange(scalar end, int options_kind, int options_device);
+raw_tensor atg_arange_start(scalar start, scalar end, int options_kind, int options_device);
+raw_tensor atg_arange_start_step(scalar start, scalar end, int options_kind, int options_device);
+raw_tensor atg_arccos(gc_tensor self);
+raw_tensor atg_arccos_(gc_tensor self);
+raw_tensor atg_arccos_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_arccosh(gc_tensor self);
+raw_tensor atg_arccosh_(gc_tensor self);
+raw_tensor atg_arccosh_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_arcsin(gc_tensor self);
+raw_tensor atg_arcsin_(gc_tensor self);
+raw_tensor atg_arcsin_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_arcsinh(gc_tensor self);
+raw_tensor atg_arcsinh_(gc_tensor self);
+raw_tensor atg_arcsinh_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_arctan(gc_tensor self);
+raw_tensor atg_arctan2(gc_tensor self, gc_tensor other);
+raw_tensor atg_arctan2_(gc_tensor self, gc_tensor other);
+raw_tensor atg_arctan2_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_arctan_(gc_tensor self);
+raw_tensor atg_arctan_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_arctanh(gc_tensor self);
+raw_tensor atg_arctanh_(gc_tensor self);
+raw_tensor atg_arctanh_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_argmax(gc_tensor self, int64_t dim_v, int dim_null, int keepdim);
+raw_tensor atg_argmax_out(gc_tensor out, gc_tensor self, int64_t dim_v, int dim_null, int keepdim);
+raw_tensor atg_argmin(gc_tensor self, int64_t dim_v, int dim_null, int keepdim);
+raw_tensor atg_argmin_out(gc_tensor out, gc_tensor self, int64_t dim_v, int dim_null, int keepdim);
+raw_tensor atg_argsort(gc_tensor self, int64_t dim, int descending);
+raw_tensor atg_argsort_stable(gc_tensor self, int stable, int64_t dim, int descending);
+raw_tensor atg_argsort_stable_out(gc_tensor out, gc_tensor self, int stable, int64_t dim, int descending);
+raw_tensor atg_argwhere(gc_tensor self);
+raw_tensor atg_as_strided(gc_tensor self, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len, int64_t storage_offset_v, int storage_offset_null);
+raw_tensor atg_as_strided_(gc_tensor self, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len, int64_t storage_offset_v, int storage_offset_null);
+raw_tensor atg_as_strided_copy(gc_tensor self, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len, int64_t storage_offset_v, int storage_offset_null);
+raw_tensor atg_as_strided_copy_out(gc_tensor out, gc_tensor self, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len, int64_t storage_offset_v, int storage_offset_null);
+raw_tensor atg_as_strided_scatter(gc_tensor self, gc_tensor src, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len, int64_t storage_offset_v, int storage_offset_null);
+raw_tensor atg_as_strided_scatter_out(gc_tensor out, gc_tensor self, gc_tensor src, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len, int64_t storage_offset_v, int storage_offset_null);
+raw_tensor atg_asin(gc_tensor self);
+raw_tensor atg_asin_(gc_tensor self);
+raw_tensor atg_asin_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_asinh(gc_tensor self);
+raw_tensor atg_asinh_(gc_tensor self);
+raw_tensor atg_asinh_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_atan(gc_tensor self);
+raw_tensor atg_atan2(gc_tensor self, gc_tensor other);
+raw_tensor atg_atan2_(gc_tensor self, gc_tensor other);
+raw_tensor atg_atan2_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_atan_(gc_tensor self);
+raw_tensor atg_atan_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_atanh(gc_tensor self);
+raw_tensor atg_atanh_(gc_tensor self);
+raw_tensor atg_atanh_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_atleast_1d(gc_tensor self);
+raw_tensor *atg_atleast_1d_sequence(gc_tensor *tensors_data, int tensors_len);
+raw_tensor atg_atleast_2d(gc_tensor self);
+raw_tensor *atg_atleast_2d_sequence(gc_tensor *tensors_data, int tensors_len);
+raw_tensor atg_atleast_3d(gc_tensor self);
+raw_tensor *atg_atleast_3d_sequence(gc_tensor *tensors_data, int tensors_len);
+raw_tensor atg_avg_pool1d(gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int ceil_mode, int count_include_pad);
+raw_tensor atg_avg_pool2d(gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int ceil_mode, int count_include_pad, int64_t divisor_override_v, int divisor_override_null);
+raw_tensor atg_avg_pool2d_backward(gc_tensor grad_output, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int ceil_mode, int count_include_pad, int64_t divisor_override_v, int divisor_override_null);
+raw_tensor atg_avg_pool2d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int ceil_mode, int count_include_pad, int64_t divisor_override_v, int divisor_override_null);
+raw_tensor atg_avg_pool2d_out(gc_tensor out, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int ceil_mode, int count_include_pad, int64_t divisor_override_v, int divisor_override_null);
+raw_tensor atg_avg_pool3d(gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int ceil_mode, int count_include_pad, int64_t divisor_override_v, int divisor_override_null);
+raw_tensor atg_avg_pool3d_backward(gc_tensor grad_output, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int ceil_mode, int count_include_pad, int64_t divisor_override_v, int divisor_override_null);
+raw_tensor atg_avg_pool3d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int ceil_mode, int count_include_pad, int64_t divisor_override_v, int divisor_override_null);
+raw_tensor atg_avg_pool3d_out(gc_tensor out, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int ceil_mode, int count_include_pad, int64_t divisor_override_v, int divisor_override_null);
+raw_tensor atg_baddbmm(gc_tensor self, gc_tensor batch1, gc_tensor batch2);
+raw_tensor atg_baddbmm_(gc_tensor self, gc_tensor batch1, gc_tensor batch2);
+raw_tensor atg_baddbmm_out(gc_tensor out, gc_tensor self, gc_tensor batch1, gc_tensor batch2);
+raw_tensor atg_bartlett_window(int64_t window_length, int options_kind, int options_device);
+raw_tensor atg_bartlett_window_out(gc_tensor out, int64_t window_length);
+raw_tensor atg_bartlett_window_periodic(int64_t window_length, int periodic, int options_kind, int options_device);
+raw_tensor atg_bartlett_window_periodic_out(gc_tensor out, int64_t window_length, int periodic);
+raw_tensor atg_batch_norm(gc_tensor input, gc_tensor weight, gc_tensor bias, gc_tensor running_mean, gc_tensor running_var, int training, double momentum, double eps, int cudnn_enabled);
+raw_tensor atg_batch_norm_backward_elemt(gc_tensor grad_out, gc_tensor input, gc_tensor mean, gc_tensor invstd, gc_tensor weight, gc_tensor sum_dy, gc_tensor sum_dy_xmu, gc_tensor count);
+raw_tensor atg_batch_norm_backward_elemt_out(gc_tensor out, gc_tensor grad_out, gc_tensor input, gc_tensor mean, gc_tensor invstd, gc_tensor weight, gc_tensor sum_dy, gc_tensor sum_dy_xmu, gc_tensor count);
+void atg_batch_norm_backward_reduce(raw_tensor *, gc_tensor grad_out, gc_tensor input, gc_tensor mean, gc_tensor invstd, gc_tensor weight, int input_g, int weight_g, int bias_g);
+void atg_batch_norm_backward_reduce_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor out3, gc_tensor grad_out, gc_tensor input, gc_tensor mean, gc_tensor invstd, gc_tensor weight, int input_g, int weight_g, int bias_g);
+raw_tensor atg_batch_norm_elemt(gc_tensor input, gc_tensor weight, gc_tensor bias, gc_tensor mean, gc_tensor invstd, double eps);
+raw_tensor atg_batch_norm_elemt_out(gc_tensor out, gc_tensor input, gc_tensor weight, gc_tensor bias, gc_tensor mean, gc_tensor invstd, double eps);
+void atg_batch_norm_gather_stats(raw_tensor *, gc_tensor input, gc_tensor mean, gc_tensor invstd, gc_tensor running_mean, gc_tensor running_var, double momentum, double eps, int64_t count);
+void atg_batch_norm_gather_stats_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor input, gc_tensor mean, gc_tensor invstd, gc_tensor running_mean, gc_tensor running_var, double momentum, double eps, int64_t count);
+void atg_batch_norm_gather_stats_with_counts(raw_tensor *, gc_tensor input, gc_tensor mean, gc_tensor invstd, gc_tensor running_mean, gc_tensor running_var, double momentum, double eps, gc_tensor counts);
+void atg_batch_norm_gather_stats_with_counts_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor input, gc_tensor mean, gc_tensor invstd, gc_tensor running_mean, gc_tensor running_var, double momentum, double eps, gc_tensor counts);
+void atg_batch_norm_stats(raw_tensor *, gc_tensor input, double eps);
+void atg_batch_norm_stats_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor input, double eps);
+void atg_batch_norm_update_stats(raw_tensor *, gc_tensor input, gc_tensor running_mean, gc_tensor running_var, double momentum);
+void atg_batch_norm_update_stats_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor input, gc_tensor running_mean, gc_tensor running_var, double momentum);
+raw_tensor atg_bernoulli(gc_tensor self);
+raw_tensor atg_bernoulli_(gc_tensor self, gc_tensor p);
+raw_tensor atg_bernoulli_float_(gc_tensor self, double p);
+raw_tensor atg_bernoulli_p(gc_tensor self, double p);
+raw_tensor atg_bernoulli_tensor(gc_tensor self, gc_tensor p);
+raw_tensor atg_bilinear(gc_tensor input1, gc_tensor input2, gc_tensor weight, gc_tensor bias);
+raw_tensor atg_binary_cross_entropy(gc_tensor self, gc_tensor target, gc_tensor weight, int64_t reduction);
+raw_tensor atg_binary_cross_entropy_backward(gc_tensor grad_output, gc_tensor self, gc_tensor target, gc_tensor weight, int64_t reduction);
+raw_tensor atg_binary_cross_entropy_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, gc_tensor target, gc_tensor weight, int64_t reduction);
+raw_tensor atg_binary_cross_entropy_out(gc_tensor out, gc_tensor self, gc_tensor target, gc_tensor weight, int64_t reduction);
+raw_tensor atg_binary_cross_entropy_with_logits(gc_tensor self, gc_tensor target, gc_tensor weight, gc_tensor pos_weight, int64_t reduction);
+raw_tensor atg_binary_cross_entropy_with_logits_out(gc_tensor out, gc_tensor self, gc_tensor target, gc_tensor weight, gc_tensor pos_weight, int64_t reduction);
+raw_tensor atg_bincount(gc_tensor self, gc_tensor weights, int64_t minlength);
+raw_tensor atg_bincount_out(gc_tensor out, gc_tensor self, gc_tensor weights, int64_t minlength);
+raw_tensor atg_binomial(gc_tensor count, gc_tensor prob);
+raw_tensor atg_binomial_out(gc_tensor out, gc_tensor count, gc_tensor prob);
+raw_tensor atg_bitwise_and(gc_tensor self, scalar other);
+raw_tensor atg_bitwise_and_(gc_tensor self, scalar other);
+raw_tensor atg_bitwise_and_scalar_out(gc_tensor out, gc_tensor self, scalar other);
+raw_tensor atg_bitwise_and_scalar_tensor(scalar self, gc_tensor other);
+raw_tensor atg_bitwise_and_scalar_tensor_out(gc_tensor out, scalar self, gc_tensor other);
+raw_tensor atg_bitwise_and_tensor(gc_tensor self, gc_tensor other);
+raw_tensor atg_bitwise_and_tensor_(gc_tensor self, gc_tensor other);
+raw_tensor atg_bitwise_and_tensor_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_bitwise_left_shift(gc_tensor self, gc_tensor other);
+raw_tensor atg_bitwise_left_shift_(gc_tensor self, gc_tensor other);
+raw_tensor atg_bitwise_left_shift_scalar_tensor(scalar self, gc_tensor other);
+raw_tensor atg_bitwise_left_shift_scalar_tensor_out(gc_tensor out, scalar self, gc_tensor other);
+raw_tensor atg_bitwise_left_shift_tensor_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_bitwise_left_shift_tensor_scalar(gc_tensor self, scalar other);
+raw_tensor atg_bitwise_left_shift_tensor_scalar_(gc_tensor self, scalar other);
+raw_tensor atg_bitwise_left_shift_tensor_scalar_out(gc_tensor out, gc_tensor self, scalar other);
+raw_tensor atg_bitwise_not(gc_tensor self);
+raw_tensor atg_bitwise_not_(gc_tensor self);
+raw_tensor atg_bitwise_not_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_bitwise_or(gc_tensor self, scalar other);
+raw_tensor atg_bitwise_or_(gc_tensor self, scalar other);
+raw_tensor atg_bitwise_or_scalar_out(gc_tensor out, gc_tensor self, scalar other);
+raw_tensor atg_bitwise_or_scalar_tensor(scalar self, gc_tensor other);
+raw_tensor atg_bitwise_or_scalar_tensor_out(gc_tensor out, scalar self, gc_tensor other);
+raw_tensor atg_bitwise_or_tensor(gc_tensor self, gc_tensor other);
+raw_tensor atg_bitwise_or_tensor_(gc_tensor self, gc_tensor other);
+raw_tensor atg_bitwise_or_tensor_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_bitwise_right_shift(gc_tensor self, gc_tensor other);
+raw_tensor atg_bitwise_right_shift_(gc_tensor self, gc_tensor other);
+raw_tensor atg_bitwise_right_shift_scalar_tensor(scalar self, gc_tensor other);
+raw_tensor atg_bitwise_right_shift_scalar_tensor_out(gc_tensor out, scalar self, gc_tensor other);
+raw_tensor atg_bitwise_right_shift_tensor_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_bitwise_right_shift_tensor_scalar(gc_tensor self, scalar other);
+raw_tensor atg_bitwise_right_shift_tensor_scalar_(gc_tensor self, scalar other);
+raw_tensor atg_bitwise_right_shift_tensor_scalar_out(gc_tensor out, gc_tensor self, scalar other);
+raw_tensor atg_bitwise_xor(gc_tensor self, scalar other);
+raw_tensor atg_bitwise_xor_(gc_tensor self, scalar other);
+raw_tensor atg_bitwise_xor_scalar_out(gc_tensor out, gc_tensor self, scalar other);
+raw_tensor atg_bitwise_xor_scalar_tensor(scalar self, gc_tensor other);
+raw_tensor atg_bitwise_xor_scalar_tensor_out(gc_tensor out, scalar self, gc_tensor other);
+raw_tensor atg_bitwise_xor_tensor(gc_tensor self, gc_tensor other);
+raw_tensor atg_bitwise_xor_tensor_(gc_tensor self, gc_tensor other);
+raw_tensor atg_bitwise_xor_tensor_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_blackman_window(int64_t window_length, int options_kind, int options_device);
+raw_tensor atg_blackman_window_out(gc_tensor out, int64_t window_length);
+raw_tensor atg_blackman_window_periodic(int64_t window_length, int periodic, int options_kind, int options_device);
+raw_tensor atg_blackman_window_periodic_out(gc_tensor out, int64_t window_length, int periodic);
+raw_tensor atg_block_diag(gc_tensor *tensors_data, int tensors_len);
+raw_tensor atg_block_diag_out(gc_tensor out, gc_tensor *tensors_data, int tensors_len);
+raw_tensor atg_bmm(gc_tensor self, gc_tensor mat2);
+raw_tensor atg_bmm_out(gc_tensor out, gc_tensor self, gc_tensor mat2);
+raw_tensor *atg_broadcast_tensors(gc_tensor *tensors_data, int tensors_len);
+raw_tensor atg_broadcast_to(gc_tensor self, int64_t *size_data, int size_len);
+raw_tensor atg_bucketize(gc_tensor self, gc_tensor boundaries, int out_int32, int right);
+raw_tensor atg_bucketize_scalar(scalar self, gc_tensor boundaries, int out_int32, int right);
+raw_tensor atg_bucketize_scalar_out(gc_tensor out, scalar self, gc_tensor boundaries, int out_int32, int right);
+raw_tensor atg_bucketize_tensor_out(gc_tensor out, gc_tensor self, gc_tensor boundaries, int out_int32, int right);
 int atg_can_cast(int from, int to);
-void atg_cartesian_prod(tensor *, tensor *tensors_data, int tensors_len);
-void atg_cat(tensor *, tensor *tensors_data, int tensors_len, int64_t dim);
-void atg_cat_out(tensor *, tensor out, tensor *tensors_data, int tensors_len, int64_t dim);
-void atg_cauchy(tensor *, tensor self, double median, double sigma);
-void atg_cauchy_(tensor *, tensor self, double median, double sigma);
-void atg_cauchy_out(tensor *, tensor out, tensor self, double median, double sigma);
-void atg_ccol_indices(tensor *, tensor self);
-void atg_ccol_indices_copy(tensor *, tensor self);
-void atg_ccol_indices_copy_out(tensor *, tensor out, tensor self);
-void atg_cdist(tensor *, tensor x1, tensor x2, double p, int64_t compute_mode_v, int compute_mode_null);
-void atg_ceil(tensor *, tensor self);
-void atg_ceil_(tensor *, tensor self);
-void atg_ceil_out(tensor *, tensor out, tensor self);
-void atg_celu(tensor *, tensor self);
-void atg_celu_(tensor *, tensor self);
-void atg_celu_out(tensor *, tensor out, tensor self);
-void atg_chain_matmul(tensor *, tensor *matrices_data, int matrices_len);
-void atg_chain_matmul_out(tensor *, tensor out, tensor *matrices_data, int matrices_len);
-void atg_chalf(tensor *, tensor self);
-void atg_channel_shuffle(tensor *, tensor self, int64_t groups);
-void atg_channel_shuffle_out(tensor *, tensor out, tensor self, int64_t groups);
-void atg_cholesky(tensor *, tensor self, int upper);
-void atg_cholesky_inverse(tensor *, tensor self, int upper);
-void atg_cholesky_inverse_out(tensor *, tensor out, tensor self, int upper);
-void atg_cholesky_out(tensor *, tensor out, tensor self, int upper);
-void atg_cholesky_solve(tensor *, tensor self, tensor input2, int upper);
-void atg_cholesky_solve_out(tensor *, tensor out, tensor self, tensor input2, int upper);
-void atg_choose_qparams_optimized(tensor *, tensor input, int64_t numel, int64_t n_bins, double ratio, int64_t bit_width);
-tensor *atg_chunk(tensor self, int64_t chunks, int64_t dim);
-void atg_clamp(tensor *, tensor self, scalar min, scalar max);
-void atg_clamp_(tensor *, tensor self, scalar min, scalar max);
-void atg_clamp_max(tensor *, tensor self, scalar max);
-void atg_clamp_max_(tensor *, tensor self, scalar max);
-void atg_clamp_max_out(tensor *, tensor out, tensor self, scalar max);
-void atg_clamp_max_tensor(tensor *, tensor self, tensor max);
-void atg_clamp_max_tensor_(tensor *, tensor self, tensor max);
-void atg_clamp_max_tensor_out(tensor *, tensor out, tensor self, tensor max);
-void atg_clamp_min(tensor *, tensor self, scalar min);
-void atg_clamp_min_(tensor *, tensor self, scalar min);
-void atg_clamp_min_out(tensor *, tensor out, tensor self, scalar min);
-void atg_clamp_min_tensor(tensor *, tensor self, tensor min);
-void atg_clamp_min_tensor_(tensor *, tensor self, tensor min);
-void atg_clamp_min_tensor_out(tensor *, tensor out, tensor self, tensor min);
-void atg_clamp_out(tensor *, tensor out, tensor self, scalar min, scalar max);
-void atg_clamp_tensor(tensor *, tensor self, tensor min, tensor max);
-void atg_clamp_tensor_(tensor *, tensor self, tensor min, tensor max);
-void atg_clamp_tensor_out(tensor *, tensor out, tensor self, tensor min, tensor max);
-void atg_clip(tensor *, tensor self, scalar min, scalar max);
-void atg_clip_(tensor *, tensor self, scalar min, scalar max);
-void atg_clip_out(tensor *, tensor out, tensor self, scalar min, scalar max);
-void atg_clip_tensor(tensor *, tensor self, tensor min, tensor max);
-void atg_clip_tensor_(tensor *, tensor self, tensor min, tensor max);
-void atg_clip_tensor_out(tensor *, tensor out, tensor self, tensor min, tensor max);
-void atg_clone(tensor *, tensor self);
-void atg_clone_out(tensor *, tensor out, tensor self);
-void atg_coalesce(tensor *, tensor self);
-void atg_col2im(tensor *, tensor self, int64_t *output_size_data, int output_size_len, int64_t *kernel_size_data, int kernel_size_len, int64_t *dilation_data, int dilation_len, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len);
-void atg_col2im_out(tensor *, tensor out, tensor self, int64_t *output_size_data, int output_size_len, int64_t *kernel_size_data, int kernel_size_len, int64_t *dilation_data, int dilation_len, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len);
-void atg_col_indices(tensor *, tensor self);
-void atg_col_indices_copy(tensor *, tensor self);
-void atg_col_indices_copy_out(tensor *, tensor out, tensor self);
-void atg_column_stack(tensor *, tensor *tensors_data, int tensors_len);
-void atg_column_stack_out(tensor *, tensor out, tensor *tensors_data, int tensors_len);
-void atg_combinations(tensor *, tensor self, int64_t r, int with_replacement);
-void atg_complex(tensor *, tensor real, tensor imag);
-void atg_complex_out(tensor *, tensor out, tensor real, tensor imag);
-void atg_concat(tensor *, tensor *tensors_data, int tensors_len, int64_t dim);
-void atg_concat_out(tensor *, tensor out, tensor *tensors_data, int tensors_len, int64_t dim);
-void atg_concatenate(tensor *, tensor *tensors_data, int tensors_len, int64_t dim);
-void atg_concatenate_out(tensor *, tensor out, tensor *tensors_data, int tensors_len, int64_t dim);
-void atg_conj(tensor *, tensor self);
-void atg_conj_physical(tensor *, tensor self);
-void atg_conj_physical_(tensor *, tensor self);
-void atg_conj_physical_out(tensor *, tensor out, tensor self);
-void atg_constant_pad_nd(tensor *, tensor self, int64_t *pad_data, int pad_len);
-void atg_constant_pad_nd_out(tensor *, tensor out, tensor self, int64_t *pad_data, int pad_len);
-void atg_contiguous(tensor *, tensor self);
-void atg_conv1d(tensor *, tensor input, tensor weight, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int64_t groups);
-void atg_conv1d_padding(tensor *, tensor input, tensor weight, tensor bias, int64_t *stride_data, int stride_len, char * padding, int64_t *dilation_data, int dilation_len, int64_t groups);
-void atg_conv2d(tensor *, tensor input, tensor weight, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int64_t groups);
-void atg_conv2d_padding(tensor *, tensor input, tensor weight, tensor bias, int64_t *stride_data, int stride_len, char * padding, int64_t *dilation_data, int dilation_len, int64_t groups);
-void atg_conv3d(tensor *, tensor input, tensor weight, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int64_t groups);
-void atg_conv3d_padding(tensor *, tensor input, tensor weight, tensor bias, int64_t *stride_data, int stride_len, char * padding, int64_t *dilation_data, int dilation_len, int64_t groups);
-void atg_conv_depthwise3d(tensor *, tensor self, tensor weight, int64_t *kernel_size_data, int kernel_size_len, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len);
-void atg_conv_depthwise3d_out(tensor *, tensor out, tensor self, tensor weight, int64_t *kernel_size_data, int kernel_size_len, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len);
-void atg_conv_tbc(tensor *, tensor self, tensor weight, tensor bias, int64_t pad);
-void atg_conv_tbc_backward(tensor *, tensor self, tensor input, tensor weight, tensor bias, int64_t pad);
-void atg_conv_tbc_out(tensor *, tensor out, tensor self, tensor weight, tensor bias, int64_t pad);
-void atg_conv_transpose1d(tensor *, tensor input, tensor weight, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t groups, int64_t *dilation_data, int dilation_len);
-void atg_conv_transpose2d(tensor *, tensor input, tensor weight, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t groups, int64_t *dilation_data, int dilation_len);
-void atg_conv_transpose3d(tensor *, tensor input, tensor weight, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t groups, int64_t *dilation_data, int dilation_len);
-void atg_convolution(tensor *, tensor input, tensor weight, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int transposed, int64_t *output_padding_data, int output_padding_len, int64_t groups);
-void atg_convolution_out(tensor *, tensor out, tensor input, tensor weight, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int transposed, int64_t *output_padding_data, int output_padding_len, int64_t groups);
-void atg_convolution_overrideable(tensor *, tensor input, tensor weight, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int transposed, int64_t *output_padding_data, int output_padding_len, int64_t groups);
-void atg_convolution_overrideable_out(tensor *, tensor out, tensor input, tensor weight, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int transposed, int64_t *output_padding_data, int output_padding_len, int64_t groups);
-void atg_copy(tensor *, tensor self, tensor src, int non_blocking);
-void atg_copy_out(tensor *, tensor out, tensor self, tensor src, int non_blocking);
-void atg_copy_sparse_to_sparse(tensor *, tensor self, tensor src, int non_blocking);
-void atg_copy_sparse_to_sparse_(tensor *, tensor self, tensor src, int non_blocking);
-void atg_copy_sparse_to_sparse_out(tensor *, tensor out, tensor self, tensor src, int non_blocking);
-void atg_copysign(tensor *, tensor self, tensor other);
-void atg_copysign_(tensor *, tensor self, tensor other);
-void atg_copysign_out(tensor *, tensor out, tensor self, tensor other);
-void atg_copysign_scalar(tensor *, tensor self, scalar other);
-void atg_copysign_scalar_(tensor *, tensor self, scalar other);
-void atg_copysign_scalar_out(tensor *, tensor out, tensor self, scalar other);
-void atg_corrcoef(tensor *, tensor self);
-void atg_cos(tensor *, tensor self);
-void atg_cos_(tensor *, tensor self);
-void atg_cos_out(tensor *, tensor out, tensor self);
-void atg_cosh(tensor *, tensor self);
-void atg_cosh_(tensor *, tensor self);
-void atg_cosh_out(tensor *, tensor out, tensor self);
-void atg_cosine_embedding_loss(tensor *, tensor input1, tensor input2, tensor target, double margin, int64_t reduction);
-void atg_cosine_similarity(tensor *, tensor x1, tensor x2, int64_t dim, double eps);
-void atg_count_nonzero(tensor *, tensor out, tensor self, int64_t *dim_data, int dim_len);
-void atg_count_nonzero_out(tensor *, tensor out, tensor self, int64_t dim_v, int dim_null);
-void atg_cov(tensor *, tensor self, int64_t correction, tensor fweights, tensor aweights);
-void atg_cross(tensor *, tensor self, tensor other, int64_t dim_v, int dim_null);
-void atg_cross_entropy_loss(tensor *, tensor self, tensor target, tensor weight, int64_t reduction, int64_t ignore_index, double label_smoothing);
-void atg_cross_out(tensor *, tensor out, tensor self, tensor other, int64_t dim_v, int dim_null);
-void atg_crow_indices(tensor *, tensor self);
-void atg_crow_indices_copy(tensor *, tensor self);
-void atg_crow_indices_copy_out(tensor *, tensor out, tensor self);
-void atg_ctc_loss(tensor *, tensor log_probs, tensor targets, int64_t *input_lengths_data, int input_lengths_len, int64_t *target_lengths_data, int target_lengths_len, int64_t blank, int64_t reduction, int zero_infinity);
-void atg_ctc_loss_tensor(tensor *, tensor log_probs, tensor targets, tensor input_lengths, tensor target_lengths, int64_t blank, int64_t reduction, int zero_infinity);
-void atg_cudnn_affine_grid_generator(tensor *, tensor theta, int64_t n, int64_t C, int64_t H, int64_t W);
-void atg_cudnn_affine_grid_generator_backward(tensor *, tensor grad, int64_t n, int64_t C, int64_t H, int64_t W);
-void atg_cudnn_affine_grid_generator_backward_out(tensor *, tensor out, tensor grad, int64_t n, int64_t C, int64_t H, int64_t W);
-void atg_cudnn_affine_grid_generator_out(tensor *, tensor out, tensor theta, int64_t n, int64_t C, int64_t H, int64_t W);
-void atg_cudnn_batch_norm(tensor *, tensor input, tensor weight, tensor bias, tensor running_mean, tensor running_var, int training, double exponential_average_factor, double epsilon);
-void atg_cudnn_batch_norm_backward(tensor *, tensor input, tensor grad_output, tensor weight, tensor running_mean, tensor running_var, tensor save_mean, tensor save_var, double epsilon, tensor reserveSpace);
-void atg_cudnn_batch_norm_backward_out(tensor *, tensor out0, tensor out1, tensor out2, tensor input, tensor grad_output, tensor weight, tensor running_mean, tensor running_var, tensor save_mean, tensor save_var, double epsilon, tensor reserveSpace);
-void atg_cudnn_batch_norm_out(tensor *, tensor out0, tensor out1, tensor out2, tensor out3, tensor input, tensor weight, tensor bias, tensor running_mean, tensor running_var, int training, double exponential_average_factor, double epsilon);
-void atg_cudnn_convolution(tensor *, tensor self, tensor weight, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int benchmark, int deterministic, int allow_tf32);
-void atg_cudnn_convolution_add_relu(tensor *, tensor self, tensor weight, tensor z, scalar alpha, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int64_t groups);
-void atg_cudnn_convolution_add_relu_out(tensor *, tensor out, tensor self, tensor weight, tensor z, scalar alpha, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int64_t groups);
-void atg_cudnn_convolution_out(tensor *, tensor out, tensor self, tensor weight, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int benchmark, int deterministic, int allow_tf32);
-void atg_cudnn_convolution_relu(tensor *, tensor self, tensor weight, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int64_t groups);
-void atg_cudnn_convolution_relu_out(tensor *, tensor out, tensor self, tensor weight, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int64_t groups);
-void atg_cudnn_convolution_transpose(tensor *, tensor self, tensor weight, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int benchmark, int deterministic, int allow_tf32);
-void atg_cudnn_convolution_transpose_out(tensor *, tensor out, tensor self, tensor weight, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int benchmark, int deterministic, int allow_tf32);
-void atg_cudnn_grid_sampler(tensor *, tensor self, tensor grid);
-void atg_cudnn_grid_sampler_backward(tensor *, tensor self, tensor grid, tensor grad_output);
-void atg_cudnn_grid_sampler_backward_out(tensor *, tensor out0, tensor out1, tensor self, tensor grid, tensor grad_output);
-void atg_cudnn_grid_sampler_out(tensor *, tensor out, tensor self, tensor grid);
-int atg_cudnn_is_acceptable(tensor self);
-void atg_cummax(tensor *, tensor self, int64_t dim);
-void atg_cummax_out(tensor *, tensor values, tensor indices, tensor self, int64_t dim);
-void atg_cummaxmin_backward(tensor *, tensor grad, tensor input, tensor indices, int64_t dim);
-void atg_cummin(tensor *, tensor self, int64_t dim);
-void atg_cummin_out(tensor *, tensor values, tensor indices, tensor self, int64_t dim);
-void atg_cumprod(tensor *, tensor self, int64_t dim, int dtype);
-void atg_cumprod_(tensor *, tensor self, int64_t dim, int dtype);
-void atg_cumprod_backward(tensor *, tensor grad, tensor input, int64_t dim, tensor output);
-void atg_cumprod_out(tensor *, tensor out, tensor self, int64_t dim, int dtype);
-void atg_cumsum(tensor *, tensor self, int64_t dim, int dtype);
-void atg_cumsum_(tensor *, tensor self, int64_t dim, int dtype);
-void atg_cumsum_out(tensor *, tensor out, tensor self, int64_t dim, int dtype);
-void atg_cumulative_trapezoid(tensor *, tensor y, int64_t dim);
-void atg_cumulative_trapezoid_x(tensor *, tensor y, tensor x, int64_t dim);
-void atg_data(tensor *, tensor self);
-void atg_deg2rad(tensor *, tensor self);
-void atg_deg2rad_(tensor *, tensor self);
-void atg_deg2rad_out(tensor *, tensor out, tensor self);
-int64_t atg_dense_dim(tensor self);
-void atg_dequantize(tensor *, tensor self);
-void atg_dequantize_self_out(tensor *, tensor out, tensor self);
-tensor *atg_dequantize_tensors(tensor *tensors_data, int tensors_len);
-void atg_dequantize_tensors_out(tensor *out_data, int out_len, tensor *tensors_data, int tensors_len);
-void atg_det(tensor *, tensor self);
-void atg_detach(tensor *, tensor self);
-void atg_detach_(tensor *, tensor self);
-void atg_detach_copy(tensor *, tensor self);
-void atg_detach_copy_out(tensor *, tensor out, tensor self);
-void atg_diag(tensor *, tensor self, int64_t diagonal);
-void atg_diag_embed(tensor *, tensor self, int64_t offset, int64_t dim1, int64_t dim2);
-void atg_diag_embed_out(tensor *, tensor out, tensor self, int64_t offset, int64_t dim1, int64_t dim2);
-void atg_diag_out(tensor *, tensor out, tensor self, int64_t diagonal);
-void atg_diagflat(tensor *, tensor self, int64_t offset);
-void atg_diagonal(tensor *, tensor self, int64_t offset, int64_t dim1, int64_t dim2);
-void atg_diagonal_backward(tensor *, tensor grad_output, int64_t *input_sizes_data, int input_sizes_len, int64_t offset, int64_t dim1, int64_t dim2);
-void atg_diagonal_backward_out(tensor *, tensor out, tensor grad_output, int64_t *input_sizes_data, int input_sizes_len, int64_t offset, int64_t dim1, int64_t dim2);
-void atg_diagonal_copy(tensor *, tensor self, int64_t offset, int64_t dim1, int64_t dim2);
-void atg_diagonal_copy_out(tensor *, tensor out, tensor self, int64_t offset, int64_t dim1, int64_t dim2);
-void atg_diagonal_scatter(tensor *, tensor self, tensor src, int64_t offset, int64_t dim1, int64_t dim2);
-void atg_diagonal_scatter_out(tensor *, tensor out, tensor self, tensor src, int64_t offset, int64_t dim1, int64_t dim2);
-void atg_diff(tensor *, tensor self, int64_t n, int64_t dim, tensor prepend, tensor append);
-void atg_diff_out(tensor *, tensor out, tensor self, int64_t n, int64_t dim, tensor prepend, tensor append);
-void atg_digamma(tensor *, tensor self);
-void atg_digamma_(tensor *, tensor self);
-void atg_digamma_out(tensor *, tensor out, tensor self);
-void atg_dist(tensor *, tensor self, tensor other);
-void atg_dist_out(tensor *, tensor out, tensor self, tensor other);
-void atg_div(tensor *, tensor self, tensor other);
-void atg_div_(tensor *, tensor self, tensor other);
-void atg_div_out(tensor *, tensor out, tensor self, tensor other);
-void atg_div_out_mode(tensor *, tensor out, tensor self, tensor other, char * rounding_mode);
-void atg_div_scalar(tensor *, tensor self, scalar other);
-void atg_div_scalar_(tensor *, tensor self, scalar other);
-void atg_div_scalar_mode(tensor *, tensor self, scalar other, char * rounding_mode);
-void atg_div_scalar_mode_(tensor *, tensor self, scalar other, char * rounding_mode);
-void atg_div_scalar_mode_out(tensor *, tensor out, tensor self, scalar other, char * rounding_mode);
-void atg_div_scalar_out(tensor *, tensor out, tensor self, scalar other);
-void atg_div_tensor_mode(tensor *, tensor self, tensor other, char * rounding_mode);
-void atg_div_tensor_mode_(tensor *, tensor self, tensor other, char * rounding_mode);
-void atg_divide(tensor *, tensor self, tensor other);
-void atg_divide_(tensor *, tensor self, tensor other);
-void atg_divide_out(tensor *, tensor out, tensor self, tensor other);
-void atg_divide_out_mode(tensor *, tensor out, tensor self, tensor other, char * rounding_mode);
-void atg_divide_scalar(tensor *, tensor self, scalar other);
-void atg_divide_scalar_(tensor *, tensor self, scalar other);
-void atg_divide_scalar_mode(tensor *, tensor self, scalar other, char * rounding_mode);
-void atg_divide_scalar_mode_(tensor *, tensor self, scalar other, char * rounding_mode);
-void atg_divide_tensor_mode(tensor *, tensor self, tensor other, char * rounding_mode);
-void atg_divide_tensor_mode_(tensor *, tensor self, tensor other, char * rounding_mode);
-void atg_dot(tensor *, tensor self, tensor tensor);
-void atg_dot_out(tensor *, tensor out, tensor self, tensor tensor);
-void atg_dropout(tensor *, tensor input, double p, int train);
-void atg_dropout_(tensor *, tensor self, double p, int train);
-tensor *atg_dsplit(tensor self, int64_t sections);
-tensor *atg_dsplit_array(tensor self, int64_t *indices_data, int indices_len);
-void atg_dstack(tensor *, tensor *tensors_data, int tensors_len);
-void atg_dstack_out(tensor *, tensor out, tensor *tensors_data, int tensors_len);
-void atg_einsum(tensor *, char * equation, tensor *tensors_data, int tensors_len, int64_t *path_data, int path_len);
-void atg_elu(tensor *, tensor self);
-void atg_elu_(tensor *, tensor self);
-void atg_elu_backward(tensor *, tensor grad_output, scalar alpha, scalar scale, scalar input_scale, int is_result, tensor self_or_result);
-void atg_elu_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, scalar alpha, scalar scale, scalar input_scale, int is_result, tensor self_or_result);
-void atg_elu_out(tensor *, tensor out, tensor self);
-void atg_embedding(tensor *, tensor weight, tensor indices, int64_t padding_idx, int scale_grad_by_freq, int sparse);
-void atg_embedding_backward(tensor *, tensor grad, tensor indices, int64_t num_weights, int64_t padding_idx, int scale_grad_by_freq, int sparse);
-void atg_embedding_bag(tensor *, tensor weight, tensor indices, tensor offsets, int scale_grad_by_freq, int64_t mode, int sparse, tensor per_sample_weights, int include_last_offset);
-void atg_embedding_bag_padding_idx(tensor *, tensor weight, tensor indices, tensor offsets, int scale_grad_by_freq, int64_t mode, int sparse, tensor per_sample_weights, int include_last_offset, int64_t padding_idx_v, int padding_idx_null);
-void atg_embedding_dense_backward(tensor *, tensor grad_output, tensor indices, int64_t num_weights, int64_t padding_idx, int scale_grad_by_freq);
-void atg_embedding_dense_backward_out(tensor *, tensor out, tensor grad_output, tensor indices, int64_t num_weights, int64_t padding_idx, int scale_grad_by_freq);
-void atg_embedding_out(tensor *, tensor out, tensor weight, tensor indices, int64_t padding_idx, int scale_grad_by_freq, int sparse);
-void atg_embedding_renorm(tensor *, tensor self, tensor indices, double max_norm, double norm_type);
-void atg_embedding_renorm_(tensor *, tensor self, tensor indices, double max_norm, double norm_type);
-void atg_embedding_renorm_out(tensor *, tensor out, tensor self, tensor indices, double max_norm, double norm_type);
-void atg_embedding_sparse_backward(tensor *, tensor grad, tensor indices, int64_t num_weights, int64_t padding_idx, int scale_grad_by_freq);
-void atg_empty(tensor *, int64_t *size_data, int size_len, int options_kind, int options_device);
-void atg_empty_like(tensor *, tensor self);
-void atg_empty_like_out(tensor *, tensor out, tensor self);
-void atg_empty_out(tensor *, tensor out, int64_t *size_data, int size_len);
-void atg_empty_permuted(tensor *, int64_t *size_data, int size_len, int64_t *physical_layout_data, int physical_layout_len, int options_kind, int options_device);
-void atg_empty_permuted_out(tensor *, tensor out, int64_t *size_data, int size_len, int64_t *physical_layout_data, int physical_layout_len);
-void atg_empty_quantized(tensor *, int64_t *size_data, int size_len, tensor qtensor, int options_kind, int options_device);
-void atg_empty_quantized_out(tensor *, tensor out, int64_t *size_data, int size_len, tensor qtensor);
-void atg_empty_strided(tensor *, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len, int options_kind, int options_device);
-void atg_empty_strided_out(tensor *, tensor out, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len);
-void atg_eq(tensor *, tensor self, scalar other);
-void atg_eq_(tensor *, tensor self, scalar other);
-void atg_eq_scalar_out(tensor *, tensor out, tensor self, scalar other);
-void atg_eq_tensor(tensor *, tensor self, tensor other);
-void atg_eq_tensor_(tensor *, tensor self, tensor other);
-void atg_eq_tensor_out(tensor *, tensor out, tensor self, tensor other);
-int atg_equal(tensor self, tensor other);
-void atg_erf(tensor *, tensor self);
-void atg_erf_(tensor *, tensor self);
-void atg_erf_out(tensor *, tensor out, tensor self);
-void atg_erfc(tensor *, tensor self);
-void atg_erfc_(tensor *, tensor self);
-void atg_erfc_out(tensor *, tensor out, tensor self);
-void atg_erfinv(tensor *, tensor self);
-void atg_erfinv_(tensor *, tensor self);
-void atg_erfinv_out(tensor *, tensor out, tensor self);
-void atg_exp(tensor *, tensor self);
-void atg_exp2(tensor *, tensor self);
-void atg_exp2_(tensor *, tensor self);
-void atg_exp2_out(tensor *, tensor out, tensor self);
-void atg_exp_(tensor *, tensor self);
-void atg_exp_out(tensor *, tensor out, tensor self);
-void atg_expand(tensor *, tensor self, int64_t *size_data, int size_len, int implicit);
-void atg_expand_as(tensor *, tensor self, tensor other);
-void atg_expand_copy(tensor *, tensor self, int64_t *size_data, int size_len, int implicit);
-void atg_expand_copy_out(tensor *, tensor out, tensor self, int64_t *size_data, int size_len, int implicit);
-void atg_expm1(tensor *, tensor self);
-void atg_expm1_(tensor *, tensor self);
-void atg_expm1_out(tensor *, tensor out, tensor self);
-void atg_exponential(tensor *, tensor self, double lambd);
-void atg_exponential_(tensor *, tensor self, double lambd);
-void atg_exponential_out(tensor *, tensor out, tensor self, double lambd);
-void atg_eye(tensor *, int64_t n, int options_kind, int options_device);
-void atg_eye_m(tensor *, int64_t n, int64_t m, int options_kind, int options_device);
-void atg_eye_m_out(tensor *, tensor out, int64_t n, int64_t m);
-void atg_eye_out(tensor *, tensor out, int64_t n);
-void atg_fake_quantize_per_channel_affine(tensor *, tensor self, tensor scale, tensor zero_point, int64_t axis, int64_t quant_min, int64_t quant_max);
-void atg_fake_quantize_per_channel_affine_cachemask(tensor *, tensor self, tensor scale, tensor zero_point, int64_t axis, int64_t quant_min, int64_t quant_max);
-void atg_fake_quantize_per_channel_affine_cachemask_backward(tensor *, tensor grad, tensor mask);
-void atg_fake_quantize_per_channel_affine_cachemask_out(tensor *, tensor out0, tensor out1, tensor self, tensor scale, tensor zero_point, int64_t axis, int64_t quant_min, int64_t quant_max);
-void atg_fake_quantize_per_tensor_affine(tensor *, tensor self, double scale, int64_t zero_point, int64_t quant_min, int64_t quant_max);
-void atg_fake_quantize_per_tensor_affine_cachemask(tensor *, tensor self, double scale, int64_t zero_point, int64_t quant_min, int64_t quant_max);
-void atg_fake_quantize_per_tensor_affine_cachemask_backward(tensor *, tensor grad, tensor mask);
-void atg_fake_quantize_per_tensor_affine_cachemask_out(tensor *, tensor out0, tensor out1, tensor self, double scale, int64_t zero_point, int64_t quant_min, int64_t quant_max);
-void atg_fake_quantize_per_tensor_affine_tensor_qparams(tensor *, tensor self, tensor scale, tensor zero_point, int64_t quant_min, int64_t quant_max);
-void atg_fbgemm_linear_fp16_weight(tensor *, tensor input, tensor packed_weight, tensor bias);
-void atg_fbgemm_linear_fp16_weight_fp32_activation(tensor *, tensor input, tensor packed_weight, tensor bias);
-void atg_fbgemm_linear_int8_weight(tensor *, tensor input, tensor weight, tensor packed, tensor col_offsets, scalar weight_scale, scalar weight_zero_point, tensor bias);
-void atg_fbgemm_linear_int8_weight_fp32_activation(tensor *, tensor input, tensor weight, tensor packed, tensor col_offsets, scalar weight_scale, scalar weight_zero_point, tensor bias);
-void atg_fbgemm_pack_gemm_matrix_fp16(tensor *, tensor input);
-void atg_fbgemm_pack_quantized_matrix(tensor *, tensor input);
-void atg_fbgemm_pack_quantized_matrix_kn(tensor *, tensor input, int64_t K, int64_t n);
-void atg_feature_alpha_dropout(tensor *, tensor input, double p, int train);
-void atg_feature_alpha_dropout_(tensor *, tensor self, double p, int train);
-void atg_feature_dropout(tensor *, tensor input, double p, int train);
-void atg_feature_dropout_(tensor *, tensor self, double p, int train);
-void atg_fft_fft(tensor *, tensor self, int64_t n_v, int n_null, int64_t dim, char * norm);
-void atg_fft_fft2(tensor *, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
-void atg_fft_fft2_out(tensor *, tensor out, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
-void atg_fft_fft_out(tensor *, tensor out, tensor self, int64_t n_v, int n_null, int64_t dim, char * norm);
-void atg_fft_fftfreq(tensor *, int64_t n, double d, int options_kind, int options_device);
-void atg_fft_fftfreq_out(tensor *, tensor out, int64_t n, double d);
-void atg_fft_fftn(tensor *, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
-void atg_fft_fftn_out(tensor *, tensor out, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
-void atg_fft_fftshift(tensor *, tensor self, int64_t *dim_data, int dim_len);
-void atg_fft_hfft(tensor *, tensor self, int64_t n_v, int n_null, int64_t dim, char * norm);
-void atg_fft_hfft2(tensor *, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
-void atg_fft_hfft2_out(tensor *, tensor out, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
-void atg_fft_hfft_out(tensor *, tensor out, tensor self, int64_t n_v, int n_null, int64_t dim, char * norm);
-void atg_fft_hfftn(tensor *, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
-void atg_fft_hfftn_out(tensor *, tensor out, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
-void atg_fft_ifft(tensor *, tensor self, int64_t n_v, int n_null, int64_t dim, char * norm);
-void atg_fft_ifft2(tensor *, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
-void atg_fft_ifft2_out(tensor *, tensor out, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
-void atg_fft_ifft_out(tensor *, tensor out, tensor self, int64_t n_v, int n_null, int64_t dim, char * norm);
-void atg_fft_ifftn(tensor *, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
-void atg_fft_ifftn_out(tensor *, tensor out, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
-void atg_fft_ifftshift(tensor *, tensor self, int64_t *dim_data, int dim_len);
-void atg_fft_ihfft(tensor *, tensor self, int64_t n_v, int n_null, int64_t dim, char * norm);
-void atg_fft_ihfft2(tensor *, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
-void atg_fft_ihfft2_out(tensor *, tensor out, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
-void atg_fft_ihfft_out(tensor *, tensor out, tensor self, int64_t n_v, int n_null, int64_t dim, char * norm);
-void atg_fft_ihfftn(tensor *, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
-void atg_fft_ihfftn_out(tensor *, tensor out, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
-void atg_fft_irfft(tensor *, tensor self, int64_t n_v, int n_null, int64_t dim, char * norm);
-void atg_fft_irfft2(tensor *, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
-void atg_fft_irfft2_out(tensor *, tensor out, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
-void atg_fft_irfft_out(tensor *, tensor out, tensor self, int64_t n_v, int n_null, int64_t dim, char * norm);
-void atg_fft_irfftn(tensor *, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
-void atg_fft_irfftn_out(tensor *, tensor out, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
-void atg_fft_rfft(tensor *, tensor self, int64_t n_v, int n_null, int64_t dim, char * norm);
-void atg_fft_rfft2(tensor *, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
-void atg_fft_rfft2_out(tensor *, tensor out, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
-void atg_fft_rfft_out(tensor *, tensor out, tensor self, int64_t n_v, int n_null, int64_t dim, char * norm);
-void atg_fft_rfftfreq(tensor *, int64_t n, double d, int options_kind, int options_device);
-void atg_fft_rfftfreq_out(tensor *, tensor out, int64_t n, double d);
-void atg_fft_rfftn(tensor *, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
-void atg_fft_rfftn_out(tensor *, tensor out, tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
-void atg_fill(tensor *, tensor self, scalar value);
-void atg_fill_(tensor *, tensor self, scalar value);
-void atg_fill_diagonal_(tensor *, tensor self, scalar fill_value, int wrap);
-void atg_fill_scalar_out(tensor *, tensor out, tensor self, scalar value);
-void atg_fill_tensor(tensor *, tensor self, tensor value);
-void atg_fill_tensor_(tensor *, tensor self, tensor value);
-void atg_fill_tensor_out(tensor *, tensor out, tensor self, tensor value);
-void atg_fix(tensor *, tensor self);
-void atg_fix_(tensor *, tensor self);
-void atg_fix_out(tensor *, tensor out, tensor self);
-void atg_flatten(tensor *, tensor self, int64_t start_dim, int64_t end_dim);
-void atg_flatten_dense_tensors(tensor *, tensor *tensors_data, int tensors_len);
-void atg_flip(tensor *, tensor self, int64_t *dims_data, int dims_len);
-void atg_flip_out(tensor *, tensor out, tensor self, int64_t *dims_data, int dims_len);
-void atg_fliplr(tensor *, tensor self);
-void atg_flipud(tensor *, tensor self);
-void atg_float_power(tensor *, tensor self, tensor exponent);
-void atg_float_power_(tensor *, tensor self, scalar exponent);
-void atg_float_power_scalar(tensor *, scalar self, tensor exponent);
-void atg_float_power_scalar_out(tensor *, tensor out, scalar self, tensor exponent);
-void atg_float_power_tensor_(tensor *, tensor self, tensor exponent);
-void atg_float_power_tensor_scalar(tensor *, tensor self, scalar exponent);
-void atg_float_power_tensor_scalar_out(tensor *, tensor out, tensor self, scalar exponent);
-void atg_float_power_tensor_tensor_out(tensor *, tensor out, tensor self, tensor exponent);
-void atg_floor(tensor *, tensor self);
-void atg_floor_(tensor *, tensor self);
-void atg_floor_divide(tensor *, tensor self, tensor other);
-void atg_floor_divide_(tensor *, tensor self, tensor other);
-void atg_floor_divide_out(tensor *, tensor out, tensor self, tensor other);
-void atg_floor_divide_scalar(tensor *, tensor self, scalar other);
-void atg_floor_divide_scalar_(tensor *, tensor self, scalar other);
-void atg_floor_out(tensor *, tensor out, tensor self);
-void atg_fmax(tensor *, tensor self, tensor other);
-void atg_fmax_out(tensor *, tensor out, tensor self, tensor other);
-void atg_fmin(tensor *, tensor self, tensor other);
-void atg_fmin_out(tensor *, tensor out, tensor self, tensor other);
-void atg_fmod(tensor *, tensor self, scalar other);
-void atg_fmod_(tensor *, tensor self, scalar other);
-void atg_fmod_scalar_out(tensor *, tensor out, tensor self, scalar other);
-void atg_fmod_tensor(tensor *, tensor self, tensor other);
-void atg_fmod_tensor_(tensor *, tensor self, tensor other);
-void atg_fmod_tensor_out(tensor *, tensor out, tensor self, tensor other);
-void atg_frac(tensor *, tensor self);
-void atg_frac_(tensor *, tensor self);
-void atg_frac_out(tensor *, tensor out, tensor self);
-void atg_fractional_max_pool2d(tensor *, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *output_size_data, int output_size_len, tensor random_samples);
-void atg_fractional_max_pool2d_backward(tensor *, tensor grad_output, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *output_size_data, int output_size_len, tensor indices);
-void atg_fractional_max_pool2d_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *output_size_data, int output_size_len, tensor indices);
-void atg_fractional_max_pool2d_output(tensor *, tensor output, tensor indices, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *output_size_data, int output_size_len, tensor random_samples);
-void atg_fractional_max_pool3d(tensor *, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *output_size_data, int output_size_len, tensor random_samples);
-void atg_fractional_max_pool3d_backward(tensor *, tensor grad_output, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *output_size_data, int output_size_len, tensor indices);
-void atg_fractional_max_pool3d_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *output_size_data, int output_size_len, tensor indices);
-void atg_fractional_max_pool3d_output(tensor *, tensor output, tensor indices, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *output_size_data, int output_size_len, tensor random_samples);
-void atg_frexp(tensor *, tensor self);
-void atg_frexp_tensor_out(tensor *, tensor mantissa, tensor exponent, tensor self);
-void atg_frobenius_norm(tensor *, tensor self, int64_t *dim_data, int dim_len, int keepdim);
-void atg_frobenius_norm_out(tensor *, tensor out, tensor self, int64_t *dim_data, int dim_len, int keepdim);
-void atg_from_file(tensor *, char * filename, int shared, int64_t size_v, int size_null, int options_kind, int options_device);
-void atg_from_file_out(tensor *, tensor out, char * filename, int shared, int64_t size_v, int size_null);
-void atg_full(tensor *, int64_t *size_data, int size_len, scalar fill_value, int options_kind, int options_device);
-void atg_full_like(tensor *, tensor self, scalar fill_value);
-void atg_full_like_out(tensor *, tensor out, tensor self, scalar fill_value);
-void atg_full_out(tensor *, tensor out, int64_t *size_data, int size_len, scalar fill_value);
-void atg_fused_moving_avg_obs_fake_quant(tensor *, tensor self, tensor observer_on, tensor fake_quant_on, tensor running_min, tensor running_max, tensor scale, tensor zero_point, double averaging_const, int64_t quant_min, int64_t quant_max, int64_t ch_axis, int per_row_fake_quant, int symmetric_quant);
-void atg_gather(tensor *, tensor self, int64_t dim, tensor index, int sparse_grad);
-void atg_gather_backward(tensor *, tensor grad, tensor self, int64_t dim, tensor index, int sparse_grad);
-void atg_gather_out(tensor *, tensor out, tensor self, int64_t dim, tensor index, int sparse_grad);
-void atg_gcd(tensor *, tensor self, tensor other);
-void atg_gcd_(tensor *, tensor self, tensor other);
-void atg_gcd_out(tensor *, tensor out, tensor self, tensor other);
-void atg_ge(tensor *, tensor self, scalar other);
-void atg_ge_(tensor *, tensor self, scalar other);
-void atg_ge_scalar_out(tensor *, tensor out, tensor self, scalar other);
-void atg_ge_tensor(tensor *, tensor self, tensor other);
-void atg_ge_tensor_(tensor *, tensor self, tensor other);
-void atg_ge_tensor_out(tensor *, tensor out, tensor self, tensor other);
-void atg_gelu(tensor *, tensor self, char * approximate);
-void atg_gelu_(tensor *, tensor self, char * approximate);
-void atg_gelu_backward(tensor *, tensor grad_output, tensor self, char * approximate);
-void atg_gelu_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, tensor self, char * approximate);
-void atg_gelu_out(tensor *, tensor out, tensor self, char * approximate);
-void atg_geometric(tensor *, tensor self, double p);
-void atg_geometric_(tensor *, tensor self, double p);
-void atg_geometric_out(tensor *, tensor out, tensor self, double p);
-void atg_geqrf(tensor *, tensor self);
-void atg_geqrf_a(tensor *, tensor a, tensor tau, tensor self);
-void atg_ger(tensor *, tensor self, tensor vec2);
-void atg_ger_out(tensor *, tensor out, tensor self, tensor vec2);
-void atg_glu(tensor *, tensor self, int64_t dim);
-void atg_glu_backward(tensor *, tensor grad_output, tensor self, int64_t dim);
-void atg_glu_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, tensor self, int64_t dim);
-void atg_glu_backward_jvp(tensor *, tensor grad_x, tensor grad_glu, tensor x, tensor dgrad_glu, tensor dx, int64_t dim);
-void atg_glu_backward_jvp_out(tensor *, tensor out, tensor grad_x, tensor grad_glu, tensor x, tensor dgrad_glu, tensor dx, int64_t dim);
-void atg_glu_jvp(tensor *, tensor glu, tensor x, tensor dx, int64_t dim);
-void atg_glu_jvp_out(tensor *, tensor out, tensor glu, tensor x, tensor dx, int64_t dim);
-void atg_glu_out(tensor *, tensor out, tensor self, int64_t dim);
-void atg_grad(tensor *, tensor self);
-void atg_greater(tensor *, tensor self, scalar other);
-void atg_greater_(tensor *, tensor self, scalar other);
-void atg_greater_equal(tensor *, tensor self, scalar other);
-void atg_greater_equal_(tensor *, tensor self, scalar other);
-void atg_greater_equal_scalar_out(tensor *, tensor out, tensor self, scalar other);
-void atg_greater_equal_tensor(tensor *, tensor self, tensor other);
-void atg_greater_equal_tensor_(tensor *, tensor self, tensor other);
-void atg_greater_equal_tensor_out(tensor *, tensor out, tensor self, tensor other);
-void atg_greater_scalar_out(tensor *, tensor out, tensor self, scalar other);
-void atg_greater_tensor(tensor *, tensor self, tensor other);
-void atg_greater_tensor_(tensor *, tensor self, tensor other);
-void atg_greater_tensor_out(tensor *, tensor out, tensor self, tensor other);
-void atg_grid_sampler(tensor *, tensor input, tensor grid, int64_t interpolation_mode, int64_t padding_mode, int align_corners);
-void atg_grid_sampler_2d(tensor *, tensor input, tensor grid, int64_t interpolation_mode, int64_t padding_mode, int align_corners);
-void atg_grid_sampler_2d_out(tensor *, tensor out, tensor input, tensor grid, int64_t interpolation_mode, int64_t padding_mode, int align_corners);
-void atg_grid_sampler_3d(tensor *, tensor input, tensor grid, int64_t interpolation_mode, int64_t padding_mode, int align_corners);
-void atg_grid_sampler_3d_out(tensor *, tensor out, tensor input, tensor grid, int64_t interpolation_mode, int64_t padding_mode, int align_corners);
-void atg_group_norm(tensor *, tensor input, int64_t num_groups, tensor weight, tensor bias, double eps, int cudnn_enabled);
-void atg_gru(tensor *, tensor input, tensor hx, tensor *params_data, int params_len, int has_biases, int64_t num_layers, double dropout, int train, int bidirectional, int batch_first);
-void atg_gru_cell(tensor *, tensor input, tensor hx, tensor w_ih, tensor w_hh, tensor b_ih, tensor b_hh);
-void atg_gru_data(tensor *, tensor data, tensor batch_sizes, tensor hx, tensor *params_data, int params_len, int has_biases, int64_t num_layers, double dropout, int train, int bidirectional);
-void atg_gt(tensor *, tensor self, scalar other);
-void atg_gt_(tensor *, tensor self, scalar other);
-void atg_gt_scalar_out(tensor *, tensor out, tensor self, scalar other);
-void atg_gt_tensor(tensor *, tensor self, tensor other);
-void atg_gt_tensor_(tensor *, tensor self, tensor other);
-void atg_gt_tensor_out(tensor *, tensor out, tensor self, tensor other);
-void atg_hamming_window(tensor *, int64_t window_length, int options_kind, int options_device);
-void atg_hamming_window_out(tensor *, tensor out, int64_t window_length);
-void atg_hamming_window_periodic(tensor *, int64_t window_length, int periodic, int options_kind, int options_device);
-void atg_hamming_window_periodic_alpha(tensor *, int64_t window_length, int periodic, double alpha, int options_kind, int options_device);
-void atg_hamming_window_periodic_alpha_beta(tensor *, int64_t window_length, int periodic, double alpha, double beta, int options_kind, int options_device);
-void atg_hamming_window_periodic_alpha_beta_out(tensor *, tensor out, int64_t window_length, int periodic, double alpha, double beta);
-void atg_hamming_window_periodic_alpha_out(tensor *, tensor out, int64_t window_length, int periodic, double alpha);
-void atg_hamming_window_periodic_out(tensor *, tensor out, int64_t window_length, int periodic);
-void atg_hann_window(tensor *, int64_t window_length, int options_kind, int options_device);
-void atg_hann_window_out(tensor *, tensor out, int64_t window_length);
-void atg_hann_window_periodic(tensor *, int64_t window_length, int periodic, int options_kind, int options_device);
-void atg_hann_window_periodic_out(tensor *, tensor out, int64_t window_length, int periodic);
-void atg_hardshrink(tensor *, tensor self);
-void atg_hardshrink_backward(tensor *, tensor grad_out, tensor self, scalar lambd);
-void atg_hardshrink_backward_grad_input(tensor *, tensor grad_input, tensor grad_out, tensor self, scalar lambd);
-void atg_hardshrink_out(tensor *, tensor out, tensor self);
-void atg_hardsigmoid(tensor *, tensor self);
-void atg_hardsigmoid_(tensor *, tensor self);
-void atg_hardsigmoid_backward(tensor *, tensor grad_output, tensor self);
-void atg_hardsigmoid_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, tensor self);
-void atg_hardsigmoid_out(tensor *, tensor out, tensor self);
-void atg_hardswish(tensor *, tensor self);
-void atg_hardswish_(tensor *, tensor self);
-void atg_hardswish_backward(tensor *, tensor grad_output, tensor self);
-void atg_hardswish_backward_out(tensor *, tensor out, tensor grad_output, tensor self);
-void atg_hardswish_out(tensor *, tensor out, tensor self);
-void atg_hardtanh(tensor *, tensor self);
-void atg_hardtanh_(tensor *, tensor self);
-void atg_hardtanh_backward(tensor *, tensor grad_output, tensor self, scalar min_val, scalar max_val);
-void atg_hardtanh_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, tensor self, scalar min_val, scalar max_val);
-void atg_hardtanh_out(tensor *, tensor out, tensor self);
-void atg_heaviside(tensor *, tensor self, tensor values);
-void atg_heaviside_(tensor *, tensor self, tensor values);
-void atg_heaviside_out(tensor *, tensor out, tensor self, tensor values);
-void atg_hinge_embedding_loss(tensor *, tensor self, tensor target, double margin, int64_t reduction);
-void atg_histc(tensor *, tensor self, int64_t bins);
-void atg_histc_out(tensor *, tensor out, tensor self, int64_t bins);
-tensor *atg_hsplit(tensor self, int64_t sections);
-tensor *atg_hsplit_array(tensor self, int64_t *indices_data, int indices_len);
-void atg_hspmm(tensor *, tensor mat1, tensor mat2);
-void atg_hspmm_out(tensor *, tensor out, tensor mat1, tensor mat2);
-void atg_hstack(tensor *, tensor *tensors_data, int tensors_len);
-void atg_hstack_out(tensor *, tensor out, tensor *tensors_data, int tensors_len);
-void atg_huber_loss(tensor *, tensor self, tensor target, int64_t reduction, double delta);
-void atg_huber_loss_backward(tensor *, tensor grad_output, tensor self, tensor target, int64_t reduction, double delta);
-void atg_huber_loss_backward_out(tensor *, tensor grad_input, tensor grad_output, tensor self, tensor target, int64_t reduction, double delta);
-void atg_huber_loss_out(tensor *, tensor out, tensor self, tensor target, int64_t reduction, double delta);
-void atg_hypot(tensor *, tensor self, tensor other);
-void atg_hypot_(tensor *, tensor self, tensor other);
-void atg_hypot_out(tensor *, tensor out, tensor self, tensor other);
-void atg_i0(tensor *, tensor self);
-void atg_i0_(tensor *, tensor self);
-void atg_i0_out(tensor *, tensor out, tensor self);
-void atg_igamma(tensor *, tensor self, tensor other);
-void atg_igamma_(tensor *, tensor self, tensor other);
-void atg_igamma_out(tensor *, tensor out, tensor self, tensor other);
-void atg_igammac(tensor *, tensor self, tensor other);
-void atg_igammac_(tensor *, tensor self, tensor other);
-void atg_igammac_out(tensor *, tensor out, tensor self, tensor other);
-void atg_im2col(tensor *, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *dilation_data, int dilation_len, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len);
-void atg_im2col_out(tensor *, tensor out, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *dilation_data, int dilation_len, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len);
-void atg_imag(tensor *, tensor self);
-void atg_index(tensor *, tensor self, tensor *indices_data, int indices_len);
-void atg_index_add(tensor *, tensor self, int64_t dim, tensor index, tensor source);
-void atg_index_add_(tensor *, tensor self, int64_t dim, tensor index, tensor source);
-void atg_index_add_out(tensor *, tensor out, tensor self, int64_t dim, tensor index, tensor source);
-void atg_index_copy(tensor *, tensor self, int64_t dim, tensor index, tensor source);
-void atg_index_copy_(tensor *, tensor self, int64_t dim, tensor index, tensor source);
-void atg_index_copy_out(tensor *, tensor out, tensor self, int64_t dim, tensor index, tensor source);
-void atg_index_fill(tensor *, tensor self, int64_t dim, tensor index, scalar value);
-void atg_index_fill_(tensor *, tensor self, int64_t dim, tensor index, scalar value);
-void atg_index_fill_int_scalar_out(tensor *, tensor out, tensor self, int64_t dim, tensor index, scalar value);
-void atg_index_fill_int_tensor(tensor *, tensor self, int64_t dim, tensor index, tensor value);
-void atg_index_fill_int_tensor_(tensor *, tensor self, int64_t dim, tensor index, tensor value);
-void atg_index_fill_int_tensor_out(tensor *, tensor out, tensor self, int64_t dim, tensor index, tensor value);
-void atg_index_put(tensor *, tensor self, tensor *indices_data, int indices_len, tensor values, int accumulate);
-void atg_index_put_(tensor *, tensor self, tensor *indices_data, int indices_len, tensor values, int accumulate);
-void atg_index_put_out(tensor *, tensor out, tensor self, tensor *indices_data, int indices_len, tensor values, int accumulate);
-void atg_index_reduce(tensor *, tensor self, int64_t dim, tensor index, tensor source, char * reduce, int include_self);
-void atg_index_reduce_(tensor *, tensor self, int64_t dim, tensor index, tensor source, char * reduce, int include_self);
-void atg_index_reduce_out(tensor *, tensor out, tensor self, int64_t dim, tensor index, tensor source, char * reduce, int include_self);
-void atg_index_select(tensor *, tensor self, int64_t dim, tensor index);
-void atg_index_select_backward(tensor *, tensor grad, int64_t *self_sizes_data, int self_sizes_len, int64_t dim, tensor index);
-void atg_index_select_out(tensor *, tensor out, tensor self, int64_t dim, tensor index);
-void atg_index_tensor_out(tensor *, tensor out, tensor self, tensor *indices_data, int indices_len);
-void atg_indices(tensor *, tensor self);
-void atg_indices_copy(tensor *, tensor self);
-void atg_indices_copy_out(tensor *, tensor out, tensor self);
-void atg_infinitely_differentiable_gelu_backward(tensor *, tensor grad, tensor self);
-void atg_inner(tensor *, tensor self, tensor other);
-void atg_inner_out(tensor *, tensor out, tensor self, tensor other);
-void atg_instance_norm(tensor *, tensor input, tensor weight, tensor bias, tensor running_mean, tensor running_var, int use_input_stats, double momentum, double eps, int cudnn_enabled);
-void atg_int_repr(tensor *, tensor self);
-void atg_int_repr_out(tensor *, tensor out, tensor self);
-void atg_inverse(tensor *, tensor self);
-void atg_inverse_out(tensor *, tensor out, tensor self);
-int atg_is_coalesced(tensor self);
-int atg_is_complex(tensor self);
-int atg_is_conj(tensor self);
-int atg_is_distributed(tensor self);
-int atg_is_floating_point(tensor self);
-int atg_is_inference(tensor self);
-int atg_is_leaf(tensor self);
-int atg_is_neg(tensor self);
-int atg_is_nonzero(tensor self);
-int atg_is_pinned(tensor self, int device);
-int atg_is_same_size(tensor self, tensor other);
-int atg_is_set_to(tensor self, tensor tensor);
-int atg_is_signed(tensor self);
+raw_tensor atg_cartesian_prod(gc_tensor *tensors_data, int tensors_len);
+raw_tensor atg_cat(gc_tensor *tensors_data, int tensors_len, int64_t dim);
+raw_tensor atg_cat_out(gc_tensor out, gc_tensor *tensors_data, int tensors_len, int64_t dim);
+raw_tensor atg_cauchy(gc_tensor self, double median, double sigma);
+raw_tensor atg_cauchy_(gc_tensor self, double median, double sigma);
+raw_tensor atg_cauchy_out(gc_tensor out, gc_tensor self, double median, double sigma);
+raw_tensor atg_ccol_indices(gc_tensor self);
+raw_tensor atg_ccol_indices_copy(gc_tensor self);
+raw_tensor atg_ccol_indices_copy_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_cdist(gc_tensor x1, gc_tensor x2, double p, int64_t compute_mode_v, int compute_mode_null);
+raw_tensor atg_ceil(gc_tensor self);
+raw_tensor atg_ceil_(gc_tensor self);
+raw_tensor atg_ceil_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_celu(gc_tensor self);
+raw_tensor atg_celu_(gc_tensor self);
+raw_tensor atg_celu_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_chain_matmul(gc_tensor *matrices_data, int matrices_len);
+raw_tensor atg_chain_matmul_out(gc_tensor out, gc_tensor *matrices_data, int matrices_len);
+raw_tensor atg_chalf(gc_tensor self);
+raw_tensor atg_channel_shuffle(gc_tensor self, int64_t groups);
+raw_tensor atg_channel_shuffle_out(gc_tensor out, gc_tensor self, int64_t groups);
+raw_tensor atg_cholesky(gc_tensor self, int upper);
+raw_tensor atg_cholesky_inverse(gc_tensor self, int upper);
+raw_tensor atg_cholesky_inverse_out(gc_tensor out, gc_tensor self, int upper);
+raw_tensor atg_cholesky_out(gc_tensor out, gc_tensor self, int upper);
+raw_tensor atg_cholesky_solve(gc_tensor self, gc_tensor input2, int upper);
+raw_tensor atg_cholesky_solve_out(gc_tensor out, gc_tensor self, gc_tensor input2, int upper);
+void atg_choose_qparams_optimized(raw_tensor *, gc_tensor input, int64_t numel, int64_t n_bins, double ratio, int64_t bit_width);
+raw_tensor *atg_chunk(gc_tensor self, int64_t chunks, int64_t dim);
+raw_tensor atg_clamp(gc_tensor self, scalar min, scalar max);
+raw_tensor atg_clamp_(gc_tensor self, scalar min, scalar max);
+raw_tensor atg_clamp_max(gc_tensor self, scalar max);
+raw_tensor atg_clamp_max_(gc_tensor self, scalar max);
+raw_tensor atg_clamp_max_out(gc_tensor out, gc_tensor self, scalar max);
+raw_tensor atg_clamp_max_tensor(gc_tensor self, gc_tensor max);
+raw_tensor atg_clamp_max_tensor_(gc_tensor self, gc_tensor max);
+raw_tensor atg_clamp_max_tensor_out(gc_tensor out, gc_tensor self, gc_tensor max);
+raw_tensor atg_clamp_min(gc_tensor self, scalar min);
+raw_tensor atg_clamp_min_(gc_tensor self, scalar min);
+raw_tensor atg_clamp_min_out(gc_tensor out, gc_tensor self, scalar min);
+raw_tensor atg_clamp_min_tensor(gc_tensor self, gc_tensor min);
+raw_tensor atg_clamp_min_tensor_(gc_tensor self, gc_tensor min);
+raw_tensor atg_clamp_min_tensor_out(gc_tensor out, gc_tensor self, gc_tensor min);
+raw_tensor atg_clamp_out(gc_tensor out, gc_tensor self, scalar min, scalar max);
+raw_tensor atg_clamp_tensor(gc_tensor self, gc_tensor min, gc_tensor max);
+raw_tensor atg_clamp_tensor_(gc_tensor self, gc_tensor min, gc_tensor max);
+raw_tensor atg_clamp_tensor_out(gc_tensor out, gc_tensor self, gc_tensor min, gc_tensor max);
+raw_tensor atg_clip(gc_tensor self, scalar min, scalar max);
+raw_tensor atg_clip_(gc_tensor self, scalar min, scalar max);
+raw_tensor atg_clip_out(gc_tensor out, gc_tensor self, scalar min, scalar max);
+raw_tensor atg_clip_tensor(gc_tensor self, gc_tensor min, gc_tensor max);
+raw_tensor atg_clip_tensor_(gc_tensor self, gc_tensor min, gc_tensor max);
+raw_tensor atg_clip_tensor_out(gc_tensor out, gc_tensor self, gc_tensor min, gc_tensor max);
+raw_tensor atg_clone(gc_tensor self);
+raw_tensor atg_clone_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_coalesce(gc_tensor self);
+raw_tensor atg_col2im(gc_tensor self, int64_t *output_size_data, int output_size_len, int64_t *kernel_size_data, int kernel_size_len, int64_t *dilation_data, int dilation_len, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len);
+raw_tensor atg_col2im_out(gc_tensor out, gc_tensor self, int64_t *output_size_data, int output_size_len, int64_t *kernel_size_data, int kernel_size_len, int64_t *dilation_data, int dilation_len, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len);
+raw_tensor atg_col_indices(gc_tensor self);
+raw_tensor atg_col_indices_copy(gc_tensor self);
+raw_tensor atg_col_indices_copy_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_column_stack(gc_tensor *tensors_data, int tensors_len);
+raw_tensor atg_column_stack_out(gc_tensor out, gc_tensor *tensors_data, int tensors_len);
+raw_tensor atg_combinations(gc_tensor self, int64_t r, int with_replacement);
+raw_tensor atg_complex(gc_tensor real, gc_tensor imag);
+raw_tensor atg_complex_out(gc_tensor out, gc_tensor real, gc_tensor imag);
+raw_tensor atg_concat(gc_tensor *tensors_data, int tensors_len, int64_t dim);
+raw_tensor atg_concat_out(gc_tensor out, gc_tensor *tensors_data, int tensors_len, int64_t dim);
+raw_tensor atg_concatenate(gc_tensor *tensors_data, int tensors_len, int64_t dim);
+raw_tensor atg_concatenate_out(gc_tensor out, gc_tensor *tensors_data, int tensors_len, int64_t dim);
+raw_tensor atg_conj(gc_tensor self);
+raw_tensor atg_conj_physical(gc_tensor self);
+raw_tensor atg_conj_physical_(gc_tensor self);
+raw_tensor atg_conj_physical_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_constant_pad_nd(gc_tensor self, int64_t *pad_data, int pad_len);
+raw_tensor atg_constant_pad_nd_out(gc_tensor out, gc_tensor self, int64_t *pad_data, int pad_len);
+raw_tensor atg_contiguous(gc_tensor self);
+raw_tensor atg_conv1d(gc_tensor input, gc_tensor weight, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int64_t groups);
+raw_tensor atg_conv1d_padding(gc_tensor input, gc_tensor weight, gc_tensor bias, int64_t *stride_data, int stride_len, char * padding, int64_t *dilation_data, int dilation_len, int64_t groups);
+raw_tensor atg_conv2d(gc_tensor input, gc_tensor weight, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int64_t groups);
+raw_tensor atg_conv2d_padding(gc_tensor input, gc_tensor weight, gc_tensor bias, int64_t *stride_data, int stride_len, char * padding, int64_t *dilation_data, int dilation_len, int64_t groups);
+raw_tensor atg_conv3d(gc_tensor input, gc_tensor weight, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int64_t groups);
+raw_tensor atg_conv3d_padding(gc_tensor input, gc_tensor weight, gc_tensor bias, int64_t *stride_data, int stride_len, char * padding, int64_t *dilation_data, int dilation_len, int64_t groups);
+raw_tensor atg_conv_depthwise3d(gc_tensor self, gc_tensor weight, int64_t *kernel_size_data, int kernel_size_len, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len);
+raw_tensor atg_conv_depthwise3d_out(gc_tensor out, gc_tensor self, gc_tensor weight, int64_t *kernel_size_data, int kernel_size_len, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len);
+raw_tensor atg_conv_tbc(gc_tensor self, gc_tensor weight, gc_tensor bias, int64_t pad);
+void atg_conv_tbc_backward(raw_tensor *, gc_tensor self, gc_tensor input, gc_tensor weight, gc_tensor bias, int64_t pad);
+raw_tensor atg_conv_tbc_out(gc_tensor out, gc_tensor self, gc_tensor weight, gc_tensor bias, int64_t pad);
+raw_tensor atg_conv_transpose1d(gc_tensor input, gc_tensor weight, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t groups, int64_t *dilation_data, int dilation_len);
+raw_tensor atg_conv_transpose2d(gc_tensor input, gc_tensor weight, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t groups, int64_t *dilation_data, int dilation_len);
+raw_tensor atg_conv_transpose3d(gc_tensor input, gc_tensor weight, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t groups, int64_t *dilation_data, int dilation_len);
+raw_tensor atg_convolution(gc_tensor input, gc_tensor weight, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int transposed, int64_t *output_padding_data, int output_padding_len, int64_t groups);
+raw_tensor atg_convolution_out(gc_tensor out, gc_tensor input, gc_tensor weight, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int transposed, int64_t *output_padding_data, int output_padding_len, int64_t groups);
+raw_tensor atg_convolution_overrideable(gc_tensor input, gc_tensor weight, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int transposed, int64_t *output_padding_data, int output_padding_len, int64_t groups);
+raw_tensor atg_convolution_overrideable_out(gc_tensor out, gc_tensor input, gc_tensor weight, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int transposed, int64_t *output_padding_data, int output_padding_len, int64_t groups);
+raw_tensor atg_copy(gc_tensor self, gc_tensor src, int non_blocking);
+raw_tensor atg_copy_out(gc_tensor out, gc_tensor self, gc_tensor src, int non_blocking);
+raw_tensor atg_copy_sparse_to_sparse(gc_tensor self, gc_tensor src, int non_blocking);
+raw_tensor atg_copy_sparse_to_sparse_(gc_tensor self, gc_tensor src, int non_blocking);
+raw_tensor atg_copy_sparse_to_sparse_out(gc_tensor out, gc_tensor self, gc_tensor src, int non_blocking);
+raw_tensor atg_copysign(gc_tensor self, gc_tensor other);
+raw_tensor atg_copysign_(gc_tensor self, gc_tensor other);
+raw_tensor atg_copysign_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_copysign_scalar(gc_tensor self, scalar other);
+raw_tensor atg_copysign_scalar_(gc_tensor self, scalar other);
+raw_tensor atg_copysign_scalar_out(gc_tensor out, gc_tensor self, scalar other);
+raw_tensor atg_corrcoef(gc_tensor self);
+raw_tensor atg_cos(gc_tensor self);
+raw_tensor atg_cos_(gc_tensor self);
+raw_tensor atg_cos_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_cosh(gc_tensor self);
+raw_tensor atg_cosh_(gc_tensor self);
+raw_tensor atg_cosh_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_cosine_embedding_loss(gc_tensor input1, gc_tensor input2, gc_tensor target, double margin, int64_t reduction);
+raw_tensor atg_cosine_similarity(gc_tensor x1, gc_tensor x2, int64_t dim, double eps);
+raw_tensor atg_count_nonzero(gc_tensor out, gc_tensor self, int64_t *dim_data, int dim_len);
+raw_tensor atg_count_nonzero_out(gc_tensor out, gc_tensor self, int64_t dim_v, int dim_null);
+raw_tensor atg_cov(gc_tensor self, int64_t correction, gc_tensor fweights, gc_tensor aweights);
+raw_tensor atg_cross(gc_tensor self, gc_tensor other, int64_t dim_v, int dim_null);
+raw_tensor atg_cross_entropy_loss(gc_tensor self, gc_tensor target, gc_tensor weight, int64_t reduction, int64_t ignore_index, double label_smoothing);
+raw_tensor atg_cross_out(gc_tensor out, gc_tensor self, gc_tensor other, int64_t dim_v, int dim_null);
+raw_tensor atg_crow_indices(gc_tensor self);
+raw_tensor atg_crow_indices_copy(gc_tensor self);
+raw_tensor atg_crow_indices_copy_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_ctc_loss(gc_tensor log_probs, gc_tensor targets, int64_t *input_lengths_data, int input_lengths_len, int64_t *target_lengths_data, int target_lengths_len, int64_t blank, int64_t reduction, int zero_infinity);
+raw_tensor atg_ctc_loss_tensor(gc_tensor log_probs, gc_tensor targets, gc_tensor input_lengths, gc_tensor target_lengths, int64_t blank, int64_t reduction, int zero_infinity);
+raw_tensor atg_cudnn_affine_grid_generator(gc_tensor theta, int64_t n, int64_t C, int64_t H, int64_t W);
+raw_tensor atg_cudnn_affine_grid_generator_backward(gc_tensor grad, int64_t n, int64_t C, int64_t H, int64_t W);
+raw_tensor atg_cudnn_affine_grid_generator_backward_out(gc_tensor out, gc_tensor grad, int64_t n, int64_t C, int64_t H, int64_t W);
+raw_tensor atg_cudnn_affine_grid_generator_out(gc_tensor out, gc_tensor theta, int64_t n, int64_t C, int64_t H, int64_t W);
+void atg_cudnn_batch_norm(raw_tensor *, gc_tensor input, gc_tensor weight, gc_tensor bias, gc_tensor running_mean, gc_tensor running_var, int training, double exponential_average_factor, double epsilon);
+void atg_cudnn_batch_norm_backward(raw_tensor *, gc_tensor input, gc_tensor grad_output, gc_tensor weight, gc_tensor running_mean, gc_tensor running_var, gc_tensor save_mean, gc_tensor save_var, double epsilon, gc_tensor reserveSpace);
+void atg_cudnn_batch_norm_backward_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor input, gc_tensor grad_output, gc_tensor weight, gc_tensor running_mean, gc_tensor running_var, gc_tensor save_mean, gc_tensor save_var, double epsilon, gc_tensor reserveSpace);
+void atg_cudnn_batch_norm_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor out3, gc_tensor input, gc_tensor weight, gc_tensor bias, gc_tensor running_mean, gc_tensor running_var, int training, double exponential_average_factor, double epsilon);
+raw_tensor atg_cudnn_convolution(gc_tensor self, gc_tensor weight, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int benchmark, int deterministic, int allow_tf32);
+raw_tensor atg_cudnn_convolution_add_relu(gc_tensor self, gc_tensor weight, gc_tensor z, scalar alpha, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int64_t groups);
+raw_tensor atg_cudnn_convolution_add_relu_out(gc_tensor out, gc_tensor self, gc_tensor weight, gc_tensor z, scalar alpha, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int64_t groups);
+raw_tensor atg_cudnn_convolution_out(gc_tensor out, gc_tensor self, gc_tensor weight, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int benchmark, int deterministic, int allow_tf32);
+raw_tensor atg_cudnn_convolution_relu(gc_tensor self, gc_tensor weight, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int64_t groups);
+raw_tensor atg_cudnn_convolution_relu_out(gc_tensor out, gc_tensor self, gc_tensor weight, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int64_t groups);
+raw_tensor atg_cudnn_convolution_transpose(gc_tensor self, gc_tensor weight, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int benchmark, int deterministic, int allow_tf32);
+raw_tensor atg_cudnn_convolution_transpose_out(gc_tensor out, gc_tensor self, gc_tensor weight, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int benchmark, int deterministic, int allow_tf32);
+raw_tensor atg_cudnn_grid_sampler(gc_tensor self, gc_tensor grid);
+void atg_cudnn_grid_sampler_backward(raw_tensor *, gc_tensor self, gc_tensor grid, gc_tensor grad_output);
+void atg_cudnn_grid_sampler_backward_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor self, gc_tensor grid, gc_tensor grad_output);
+raw_tensor atg_cudnn_grid_sampler_out(gc_tensor out, gc_tensor self, gc_tensor grid);
+int atg_cudnn_is_acceptable(gc_tensor self);
+void atg_cummax(raw_tensor *, gc_tensor self, int64_t dim);
+void atg_cummax_out(raw_tensor *, gc_tensor values, gc_tensor indices, gc_tensor self, int64_t dim);
+raw_tensor atg_cummaxmin_backward(gc_tensor grad, gc_tensor input, gc_tensor indices, int64_t dim);
+void atg_cummin(raw_tensor *, gc_tensor self, int64_t dim);
+void atg_cummin_out(raw_tensor *, gc_tensor values, gc_tensor indices, gc_tensor self, int64_t dim);
+raw_tensor atg_cumprod(gc_tensor self, int64_t dim, int dtype);
+raw_tensor atg_cumprod_(gc_tensor self, int64_t dim, int dtype);
+raw_tensor atg_cumprod_backward(gc_tensor grad, gc_tensor input, int64_t dim, gc_tensor output);
+raw_tensor atg_cumprod_out(gc_tensor out, gc_tensor self, int64_t dim, int dtype);
+raw_tensor atg_cumsum(gc_tensor self, int64_t dim, int dtype);
+raw_tensor atg_cumsum_(gc_tensor self, int64_t dim, int dtype);
+raw_tensor atg_cumsum_out(gc_tensor out, gc_tensor self, int64_t dim, int dtype);
+raw_tensor atg_cumulative_trapezoid(gc_tensor y, int64_t dim);
+raw_tensor atg_cumulative_trapezoid_x(gc_tensor y, gc_tensor x, int64_t dim);
+raw_tensor atg_data(gc_tensor self);
+raw_tensor atg_deg2rad(gc_tensor self);
+raw_tensor atg_deg2rad_(gc_tensor self);
+raw_tensor atg_deg2rad_out(gc_tensor out, gc_tensor self);
+int64_t atg_dense_dim(gc_tensor self);
+raw_tensor atg_dequantize(gc_tensor self);
+raw_tensor atg_dequantize_self_out(gc_tensor out, gc_tensor self);
+raw_tensor *atg_dequantize_tensors(gc_tensor *tensors_data, int tensors_len);
+void atg_dequantize_tensors_out(gc_tensor *out_data, int out_len, gc_tensor *tensors_data, int tensors_len);
+raw_tensor atg_det(gc_tensor self);
+raw_tensor atg_detach(gc_tensor self);
+raw_tensor atg_detach_(gc_tensor self);
+raw_tensor atg_detach_copy(gc_tensor self);
+raw_tensor atg_detach_copy_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_diag(gc_tensor self, int64_t diagonal);
+raw_tensor atg_diag_embed(gc_tensor self, int64_t offset, int64_t dim1, int64_t dim2);
+raw_tensor atg_diag_embed_out(gc_tensor out, gc_tensor self, int64_t offset, int64_t dim1, int64_t dim2);
+raw_tensor atg_diag_out(gc_tensor out, gc_tensor self, int64_t diagonal);
+raw_tensor atg_diagflat(gc_tensor self, int64_t offset);
+raw_tensor atg_diagonal(gc_tensor self, int64_t offset, int64_t dim1, int64_t dim2);
+raw_tensor atg_diagonal_backward(gc_tensor grad_output, int64_t *input_sizes_data, int input_sizes_len, int64_t offset, int64_t dim1, int64_t dim2);
+raw_tensor atg_diagonal_backward_out(gc_tensor out, gc_tensor grad_output, int64_t *input_sizes_data, int input_sizes_len, int64_t offset, int64_t dim1, int64_t dim2);
+raw_tensor atg_diagonal_copy(gc_tensor self, int64_t offset, int64_t dim1, int64_t dim2);
+raw_tensor atg_diagonal_copy_out(gc_tensor out, gc_tensor self, int64_t offset, int64_t dim1, int64_t dim2);
+raw_tensor atg_diagonal_scatter(gc_tensor self, gc_tensor src, int64_t offset, int64_t dim1, int64_t dim2);
+raw_tensor atg_diagonal_scatter_out(gc_tensor out, gc_tensor self, gc_tensor src, int64_t offset, int64_t dim1, int64_t dim2);
+raw_tensor atg_diff(gc_tensor self, int64_t n, int64_t dim, gc_tensor prepend, gc_tensor append);
+raw_tensor atg_diff_out(gc_tensor out, gc_tensor self, int64_t n, int64_t dim, gc_tensor prepend, gc_tensor append);
+raw_tensor atg_digamma(gc_tensor self);
+raw_tensor atg_digamma_(gc_tensor self);
+raw_tensor atg_digamma_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_dist(gc_tensor self, gc_tensor other);
+raw_tensor atg_dist_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_div(gc_tensor self, gc_tensor other);
+raw_tensor atg_div_(gc_tensor self, gc_tensor other);
+raw_tensor atg_div_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_div_out_mode(gc_tensor out, gc_tensor self, gc_tensor other, char * rounding_mode);
+raw_tensor atg_div_scalar(gc_tensor self, scalar other);
+raw_tensor atg_div_scalar_(gc_tensor self, scalar other);
+raw_tensor atg_div_scalar_mode(gc_tensor self, scalar other, char * rounding_mode);
+raw_tensor atg_div_scalar_mode_(gc_tensor self, scalar other, char * rounding_mode);
+raw_tensor atg_div_scalar_mode_out(gc_tensor out, gc_tensor self, scalar other, char * rounding_mode);
+raw_tensor atg_div_scalar_out(gc_tensor out, gc_tensor self, scalar other);
+raw_tensor atg_div_tensor_mode(gc_tensor self, gc_tensor other, char * rounding_mode);
+raw_tensor atg_div_tensor_mode_(gc_tensor self, gc_tensor other, char * rounding_mode);
+raw_tensor atg_divide(gc_tensor self, gc_tensor other);
+raw_tensor atg_divide_(gc_tensor self, gc_tensor other);
+raw_tensor atg_divide_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_divide_out_mode(gc_tensor out, gc_tensor self, gc_tensor other, char * rounding_mode);
+raw_tensor atg_divide_scalar(gc_tensor self, scalar other);
+raw_tensor atg_divide_scalar_(gc_tensor self, scalar other);
+raw_tensor atg_divide_scalar_mode(gc_tensor self, scalar other, char * rounding_mode);
+raw_tensor atg_divide_scalar_mode_(gc_tensor self, scalar other, char * rounding_mode);
+raw_tensor atg_divide_tensor_mode(gc_tensor self, gc_tensor other, char * rounding_mode);
+raw_tensor atg_divide_tensor_mode_(gc_tensor self, gc_tensor other, char * rounding_mode);
+raw_tensor atg_dot(gc_tensor self, gc_tensor tensor);
+raw_tensor atg_dot_out(gc_tensor out, gc_tensor self, gc_tensor tensor);
+raw_tensor atg_dropout(gc_tensor input, double p, int train);
+raw_tensor atg_dropout_(gc_tensor self, double p, int train);
+raw_tensor *atg_dsplit(gc_tensor self, int64_t sections);
+raw_tensor *atg_dsplit_array(gc_tensor self, int64_t *indices_data, int indices_len);
+raw_tensor atg_dstack(gc_tensor *tensors_data, int tensors_len);
+raw_tensor atg_dstack_out(gc_tensor out, gc_tensor *tensors_data, int tensors_len);
+raw_tensor atg_einsum(char * equation, gc_tensor *tensors_data, int tensors_len, int64_t *path_data, int path_len);
+raw_tensor atg_elu(gc_tensor self);
+raw_tensor atg_elu_(gc_tensor self);
+raw_tensor atg_elu_backward(gc_tensor grad_output, scalar alpha, scalar scale, scalar input_scale, int is_result, gc_tensor self_or_result);
+raw_tensor atg_elu_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, scalar alpha, scalar scale, scalar input_scale, int is_result, gc_tensor self_or_result);
+raw_tensor atg_elu_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_embedding(gc_tensor weight, gc_tensor indices, int64_t padding_idx, int scale_grad_by_freq, int sparse);
+raw_tensor atg_embedding_backward(gc_tensor grad, gc_tensor indices, int64_t num_weights, int64_t padding_idx, int scale_grad_by_freq, int sparse);
+void atg_embedding_bag(raw_tensor *, gc_tensor weight, gc_tensor indices, gc_tensor offsets, int scale_grad_by_freq, int64_t mode, int sparse, gc_tensor per_sample_weights, int include_last_offset);
+void atg_embedding_bag_padding_idx(raw_tensor *, gc_tensor weight, gc_tensor indices, gc_tensor offsets, int scale_grad_by_freq, int64_t mode, int sparse, gc_tensor per_sample_weights, int include_last_offset, int64_t padding_idx_v, int padding_idx_null);
+raw_tensor atg_embedding_dense_backward(gc_tensor grad_output, gc_tensor indices, int64_t num_weights, int64_t padding_idx, int scale_grad_by_freq);
+raw_tensor atg_embedding_dense_backward_out(gc_tensor out, gc_tensor grad_output, gc_tensor indices, int64_t num_weights, int64_t padding_idx, int scale_grad_by_freq);
+raw_tensor atg_embedding_out(gc_tensor out, gc_tensor weight, gc_tensor indices, int64_t padding_idx, int scale_grad_by_freq, int sparse);
+raw_tensor atg_embedding_renorm(gc_tensor self, gc_tensor indices, double max_norm, double norm_type);
+raw_tensor atg_embedding_renorm_(gc_tensor self, gc_tensor indices, double max_norm, double norm_type);
+raw_tensor atg_embedding_renorm_out(gc_tensor out, gc_tensor self, gc_tensor indices, double max_norm, double norm_type);
+raw_tensor atg_embedding_sparse_backward(gc_tensor grad, gc_tensor indices, int64_t num_weights, int64_t padding_idx, int scale_grad_by_freq);
+raw_tensor atg_empty(int64_t *size_data, int size_len, int options_kind, int options_device);
+raw_tensor atg_empty_like(gc_tensor self);
+raw_tensor atg_empty_like_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_empty_out(gc_tensor out, int64_t *size_data, int size_len);
+raw_tensor atg_empty_permuted(int64_t *size_data, int size_len, int64_t *physical_layout_data, int physical_layout_len, int options_kind, int options_device);
+raw_tensor atg_empty_permuted_out(gc_tensor out, int64_t *size_data, int size_len, int64_t *physical_layout_data, int physical_layout_len);
+raw_tensor atg_empty_quantized(int64_t *size_data, int size_len, gc_tensor qtensor, int options_kind, int options_device);
+raw_tensor atg_empty_quantized_out(gc_tensor out, int64_t *size_data, int size_len, gc_tensor qtensor);
+raw_tensor atg_empty_strided(int64_t *size_data, int size_len, int64_t *stride_data, int stride_len, int options_kind, int options_device);
+raw_tensor atg_empty_strided_out(gc_tensor out, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len);
+raw_tensor atg_eq(gc_tensor self, scalar other);
+raw_tensor atg_eq_(gc_tensor self, scalar other);
+raw_tensor atg_eq_scalar_out(gc_tensor out, gc_tensor self, scalar other);
+raw_tensor atg_eq_tensor(gc_tensor self, gc_tensor other);
+raw_tensor atg_eq_tensor_(gc_tensor self, gc_tensor other);
+raw_tensor atg_eq_tensor_out(gc_tensor out, gc_tensor self, gc_tensor other);
+int atg_equal(gc_tensor self, gc_tensor other);
+raw_tensor atg_erf(gc_tensor self);
+raw_tensor atg_erf_(gc_tensor self);
+raw_tensor atg_erf_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_erfc(gc_tensor self);
+raw_tensor atg_erfc_(gc_tensor self);
+raw_tensor atg_erfc_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_erfinv(gc_tensor self);
+raw_tensor atg_erfinv_(gc_tensor self);
+raw_tensor atg_erfinv_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_exp(gc_tensor self);
+raw_tensor atg_exp2(gc_tensor self);
+raw_tensor atg_exp2_(gc_tensor self);
+raw_tensor atg_exp2_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_exp_(gc_tensor self);
+raw_tensor atg_exp_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_expand(gc_tensor self, int64_t *size_data, int size_len, int implicit);
+raw_tensor atg_expand_as(gc_tensor self, gc_tensor other);
+raw_tensor atg_expand_copy(gc_tensor self, int64_t *size_data, int size_len, int implicit);
+raw_tensor atg_expand_copy_out(gc_tensor out, gc_tensor self, int64_t *size_data, int size_len, int implicit);
+raw_tensor atg_expm1(gc_tensor self);
+raw_tensor atg_expm1_(gc_tensor self);
+raw_tensor atg_expm1_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_exponential(gc_tensor self, double lambd);
+raw_tensor atg_exponential_(gc_tensor self, double lambd);
+raw_tensor atg_exponential_out(gc_tensor out, gc_tensor self, double lambd);
+raw_tensor atg_eye(int64_t n, int options_kind, int options_device);
+raw_tensor atg_eye_m(int64_t n, int64_t m, int options_kind, int options_device);
+raw_tensor atg_eye_m_out(gc_tensor out, int64_t n, int64_t m);
+raw_tensor atg_eye_out(gc_tensor out, int64_t n);
+raw_tensor atg_fake_quantize_per_channel_affine(gc_tensor self, gc_tensor scale, gc_tensor zero_point, int64_t axis, int64_t quant_min, int64_t quant_max);
+void atg_fake_quantize_per_channel_affine_cachemask(raw_tensor *, gc_tensor self, gc_tensor scale, gc_tensor zero_point, int64_t axis, int64_t quant_min, int64_t quant_max);
+raw_tensor atg_fake_quantize_per_channel_affine_cachemask_backward(gc_tensor grad, gc_tensor mask);
+void atg_fake_quantize_per_channel_affine_cachemask_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor self, gc_tensor scale, gc_tensor zero_point, int64_t axis, int64_t quant_min, int64_t quant_max);
+raw_tensor atg_fake_quantize_per_tensor_affine(gc_tensor self, double scale, int64_t zero_point, int64_t quant_min, int64_t quant_max);
+void atg_fake_quantize_per_tensor_affine_cachemask(raw_tensor *, gc_tensor self, double scale, int64_t zero_point, int64_t quant_min, int64_t quant_max);
+raw_tensor atg_fake_quantize_per_tensor_affine_cachemask_backward(gc_tensor grad, gc_tensor mask);
+void atg_fake_quantize_per_tensor_affine_cachemask_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor self, double scale, int64_t zero_point, int64_t quant_min, int64_t quant_max);
+raw_tensor atg_fake_quantize_per_tensor_affine_tensor_qparams(gc_tensor self, gc_tensor scale, gc_tensor zero_point, int64_t quant_min, int64_t quant_max);
+raw_tensor atg_fbgemm_linear_fp16_weight(gc_tensor input, gc_tensor packed_weight, gc_tensor bias);
+raw_tensor atg_fbgemm_linear_fp16_weight_fp32_activation(gc_tensor input, gc_tensor packed_weight, gc_tensor bias);
+raw_tensor atg_fbgemm_linear_int8_weight(gc_tensor input, gc_tensor weight, gc_tensor packed, gc_tensor col_offsets, scalar weight_scale, scalar weight_zero_point, gc_tensor bias);
+raw_tensor atg_fbgemm_linear_int8_weight_fp32_activation(gc_tensor input, gc_tensor weight, gc_tensor packed, gc_tensor col_offsets, scalar weight_scale, scalar weight_zero_point, gc_tensor bias);
+raw_tensor atg_fbgemm_pack_gemm_matrix_fp16(gc_tensor input);
+raw_tensor atg_fbgemm_pack_quantized_matrix(gc_tensor input);
+raw_tensor atg_fbgemm_pack_quantized_matrix_kn(gc_tensor input, int64_t K, int64_t n);
+raw_tensor atg_feature_alpha_dropout(gc_tensor input, double p, int train);
+raw_tensor atg_feature_alpha_dropout_(gc_tensor self, double p, int train);
+raw_tensor atg_feature_dropout(gc_tensor input, double p, int train);
+raw_tensor atg_feature_dropout_(gc_tensor self, double p, int train);
+raw_tensor atg_fft_fft(gc_tensor self, int64_t n_v, int n_null, int64_t dim, char * norm);
+raw_tensor atg_fft_fft2(gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
+raw_tensor atg_fft_fft2_out(gc_tensor out, gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
+raw_tensor atg_fft_fft_out(gc_tensor out, gc_tensor self, int64_t n_v, int n_null, int64_t dim, char * norm);
+raw_tensor atg_fft_fftfreq(int64_t n, double d, int options_kind, int options_device);
+raw_tensor atg_fft_fftfreq_out(gc_tensor out, int64_t n, double d);
+raw_tensor atg_fft_fftn(gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
+raw_tensor atg_fft_fftn_out(gc_tensor out, gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
+raw_tensor atg_fft_fftshift(gc_tensor self, int64_t *dim_data, int dim_len);
+raw_tensor atg_fft_hfft(gc_tensor self, int64_t n_v, int n_null, int64_t dim, char * norm);
+raw_tensor atg_fft_hfft2(gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
+raw_tensor atg_fft_hfft2_out(gc_tensor out, gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
+raw_tensor atg_fft_hfft_out(gc_tensor out, gc_tensor self, int64_t n_v, int n_null, int64_t dim, char * norm);
+raw_tensor atg_fft_hfftn(gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
+raw_tensor atg_fft_hfftn_out(gc_tensor out, gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
+raw_tensor atg_fft_ifft(gc_tensor self, int64_t n_v, int n_null, int64_t dim, char * norm);
+raw_tensor atg_fft_ifft2(gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
+raw_tensor atg_fft_ifft2_out(gc_tensor out, gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
+raw_tensor atg_fft_ifft_out(gc_tensor out, gc_tensor self, int64_t n_v, int n_null, int64_t dim, char * norm);
+raw_tensor atg_fft_ifftn(gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
+raw_tensor atg_fft_ifftn_out(gc_tensor out, gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
+raw_tensor atg_fft_ifftshift(gc_tensor self, int64_t *dim_data, int dim_len);
+raw_tensor atg_fft_ihfft(gc_tensor self, int64_t n_v, int n_null, int64_t dim, char * norm);
+raw_tensor atg_fft_ihfft2(gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
+raw_tensor atg_fft_ihfft2_out(gc_tensor out, gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
+raw_tensor atg_fft_ihfft_out(gc_tensor out, gc_tensor self, int64_t n_v, int n_null, int64_t dim, char * norm);
+raw_tensor atg_fft_ihfftn(gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
+raw_tensor atg_fft_ihfftn_out(gc_tensor out, gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
+raw_tensor atg_fft_irfft(gc_tensor self, int64_t n_v, int n_null, int64_t dim, char * norm);
+raw_tensor atg_fft_irfft2(gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
+raw_tensor atg_fft_irfft2_out(gc_tensor out, gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
+raw_tensor atg_fft_irfft_out(gc_tensor out, gc_tensor self, int64_t n_v, int n_null, int64_t dim, char * norm);
+raw_tensor atg_fft_irfftn(gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
+raw_tensor atg_fft_irfftn_out(gc_tensor out, gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
+raw_tensor atg_fft_rfft(gc_tensor self, int64_t n_v, int n_null, int64_t dim, char * norm);
+raw_tensor atg_fft_rfft2(gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
+raw_tensor atg_fft_rfft2_out(gc_tensor out, gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
+raw_tensor atg_fft_rfft_out(gc_tensor out, gc_tensor self, int64_t n_v, int n_null, int64_t dim, char * norm);
+raw_tensor atg_fft_rfftfreq(int64_t n, double d, int options_kind, int options_device);
+raw_tensor atg_fft_rfftfreq_out(gc_tensor out, int64_t n, double d);
+raw_tensor atg_fft_rfftn(gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
+raw_tensor atg_fft_rfftn_out(gc_tensor out, gc_tensor self, int64_t *s_data, int s_len, int64_t *dim_data, int dim_len, char * norm);
+raw_tensor atg_fill(gc_tensor self, scalar value);
+raw_tensor atg_fill_(gc_tensor self, scalar value);
+raw_tensor atg_fill_diagonal_(gc_tensor self, scalar fill_value, int wrap);
+raw_tensor atg_fill_scalar_out(gc_tensor out, gc_tensor self, scalar value);
+raw_tensor atg_fill_tensor(gc_tensor self, gc_tensor value);
+raw_tensor atg_fill_tensor_(gc_tensor self, gc_tensor value);
+raw_tensor atg_fill_tensor_out(gc_tensor out, gc_tensor self, gc_tensor value);
+raw_tensor atg_fix(gc_tensor self);
+raw_tensor atg_fix_(gc_tensor self);
+raw_tensor atg_fix_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_flatten(gc_tensor self, int64_t start_dim, int64_t end_dim);
+raw_tensor atg_flatten_dense_tensors(gc_tensor *tensors_data, int tensors_len);
+raw_tensor atg_flip(gc_tensor self, int64_t *dims_data, int dims_len);
+raw_tensor atg_flip_out(gc_tensor out, gc_tensor self, int64_t *dims_data, int dims_len);
+raw_tensor atg_fliplr(gc_tensor self);
+raw_tensor atg_flipud(gc_tensor self);
+raw_tensor atg_float_power(gc_tensor self, gc_tensor exponent);
+raw_tensor atg_float_power_(gc_tensor self, scalar exponent);
+raw_tensor atg_float_power_scalar(scalar self, gc_tensor exponent);
+raw_tensor atg_float_power_scalar_out(gc_tensor out, scalar self, gc_tensor exponent);
+raw_tensor atg_float_power_tensor_(gc_tensor self, gc_tensor exponent);
+raw_tensor atg_float_power_tensor_scalar(gc_tensor self, scalar exponent);
+raw_tensor atg_float_power_tensor_scalar_out(gc_tensor out, gc_tensor self, scalar exponent);
+raw_tensor atg_float_power_tensor_tensor_out(gc_tensor out, gc_tensor self, gc_tensor exponent);
+raw_tensor atg_floor(gc_tensor self);
+raw_tensor atg_floor_(gc_tensor self);
+raw_tensor atg_floor_divide(gc_tensor self, gc_tensor other);
+raw_tensor atg_floor_divide_(gc_tensor self, gc_tensor other);
+raw_tensor atg_floor_divide_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_floor_divide_scalar(gc_tensor self, scalar other);
+raw_tensor atg_floor_divide_scalar_(gc_tensor self, scalar other);
+raw_tensor atg_floor_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_fmax(gc_tensor self, gc_tensor other);
+raw_tensor atg_fmax_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_fmin(gc_tensor self, gc_tensor other);
+raw_tensor atg_fmin_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_fmod(gc_tensor self, scalar other);
+raw_tensor atg_fmod_(gc_tensor self, scalar other);
+raw_tensor atg_fmod_scalar_out(gc_tensor out, gc_tensor self, scalar other);
+raw_tensor atg_fmod_tensor(gc_tensor self, gc_tensor other);
+raw_tensor atg_fmod_tensor_(gc_tensor self, gc_tensor other);
+raw_tensor atg_fmod_tensor_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_frac(gc_tensor self);
+raw_tensor atg_frac_(gc_tensor self);
+raw_tensor atg_frac_out(gc_tensor out, gc_tensor self);
+void atg_fractional_max_pool2d(raw_tensor *, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *output_size_data, int output_size_len, gc_tensor random_samples);
+raw_tensor atg_fractional_max_pool2d_backward(gc_tensor grad_output, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *output_size_data, int output_size_len, gc_tensor indices);
+raw_tensor atg_fractional_max_pool2d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *output_size_data, int output_size_len, gc_tensor indices);
+void atg_fractional_max_pool2d_output(raw_tensor *, gc_tensor output, gc_tensor indices, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *output_size_data, int output_size_len, gc_tensor random_samples);
+void atg_fractional_max_pool3d(raw_tensor *, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *output_size_data, int output_size_len, gc_tensor random_samples);
+raw_tensor atg_fractional_max_pool3d_backward(gc_tensor grad_output, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *output_size_data, int output_size_len, gc_tensor indices);
+raw_tensor atg_fractional_max_pool3d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *output_size_data, int output_size_len, gc_tensor indices);
+void atg_fractional_max_pool3d_output(raw_tensor *, gc_tensor output, gc_tensor indices, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *output_size_data, int output_size_len, gc_tensor random_samples);
+void atg_frexp(raw_tensor *, gc_tensor self);
+void atg_frexp_tensor_out(raw_tensor *, gc_tensor mantissa, gc_tensor exponent, gc_tensor self);
+raw_tensor atg_frobenius_norm(gc_tensor self, int64_t *dim_data, int dim_len, int keepdim);
+raw_tensor atg_frobenius_norm_out(gc_tensor out, gc_tensor self, int64_t *dim_data, int dim_len, int keepdim);
+raw_tensor atg_from_file(char * filename, int shared, int64_t size_v, int size_null, int options_kind, int options_device);
+raw_tensor atg_from_file_out(gc_tensor out, char * filename, int shared, int64_t size_v, int size_null);
+raw_tensor atg_full(int64_t *size_data, int size_len, scalar fill_value, int options_kind, int options_device);
+raw_tensor atg_full_like(gc_tensor self, scalar fill_value);
+raw_tensor atg_full_like_out(gc_tensor out, gc_tensor self, scalar fill_value);
+raw_tensor atg_full_out(gc_tensor out, int64_t *size_data, int size_len, scalar fill_value);
+raw_tensor atg_fused_moving_avg_obs_fake_quant(gc_tensor self, gc_tensor observer_on, gc_tensor fake_quant_on, gc_tensor running_min, gc_tensor running_max, gc_tensor scale, gc_tensor zero_point, double averaging_const, int64_t quant_min, int64_t quant_max, int64_t ch_axis, int per_row_fake_quant, int symmetric_quant);
+raw_tensor atg_gather(gc_tensor self, int64_t dim, gc_tensor index, int sparse_grad);
+raw_tensor atg_gather_backward(gc_tensor grad, gc_tensor self, int64_t dim, gc_tensor index, int sparse_grad);
+raw_tensor atg_gather_out(gc_tensor out, gc_tensor self, int64_t dim, gc_tensor index, int sparse_grad);
+raw_tensor atg_gcd(gc_tensor self, gc_tensor other);
+raw_tensor atg_gcd_(gc_tensor self, gc_tensor other);
+raw_tensor atg_gcd_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_ge(gc_tensor self, scalar other);
+raw_tensor atg_ge_(gc_tensor self, scalar other);
+raw_tensor atg_ge_scalar_out(gc_tensor out, gc_tensor self, scalar other);
+raw_tensor atg_ge_tensor(gc_tensor self, gc_tensor other);
+raw_tensor atg_ge_tensor_(gc_tensor self, gc_tensor other);
+raw_tensor atg_ge_tensor_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_gelu(gc_tensor self, char * approximate);
+raw_tensor atg_gelu_(gc_tensor self, char * approximate);
+raw_tensor atg_gelu_backward(gc_tensor grad_output, gc_tensor self, char * approximate);
+raw_tensor atg_gelu_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, char * approximate);
+raw_tensor atg_gelu_out(gc_tensor out, gc_tensor self, char * approximate);
+raw_tensor atg_geometric(gc_tensor self, double p);
+raw_tensor atg_geometric_(gc_tensor self, double p);
+raw_tensor atg_geometric_out(gc_tensor out, gc_tensor self, double p);
+void atg_geqrf(raw_tensor *, gc_tensor self);
+void atg_geqrf_a(raw_tensor *, gc_tensor a, gc_tensor tau, gc_tensor self);
+raw_tensor atg_ger(gc_tensor self, gc_tensor vec2);
+raw_tensor atg_ger_out(gc_tensor out, gc_tensor self, gc_tensor vec2);
+raw_tensor atg_glu(gc_tensor self, int64_t dim);
+raw_tensor atg_glu_backward(gc_tensor grad_output, gc_tensor self, int64_t dim);
+raw_tensor atg_glu_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, int64_t dim);
+raw_tensor atg_glu_backward_jvp(gc_tensor grad_x, gc_tensor grad_glu, gc_tensor x, gc_tensor dgrad_glu, gc_tensor dx, int64_t dim);
+raw_tensor atg_glu_backward_jvp_out(gc_tensor out, gc_tensor grad_x, gc_tensor grad_glu, gc_tensor x, gc_tensor dgrad_glu, gc_tensor dx, int64_t dim);
+raw_tensor atg_glu_jvp(gc_tensor glu, gc_tensor x, gc_tensor dx, int64_t dim);
+raw_tensor atg_glu_jvp_out(gc_tensor out, gc_tensor glu, gc_tensor x, gc_tensor dx, int64_t dim);
+raw_tensor atg_glu_out(gc_tensor out, gc_tensor self, int64_t dim);
+raw_tensor atg_grad(gc_tensor self);
+raw_tensor atg_greater(gc_tensor self, scalar other);
+raw_tensor atg_greater_(gc_tensor self, scalar other);
+raw_tensor atg_greater_equal(gc_tensor self, scalar other);
+raw_tensor atg_greater_equal_(gc_tensor self, scalar other);
+raw_tensor atg_greater_equal_scalar_out(gc_tensor out, gc_tensor self, scalar other);
+raw_tensor atg_greater_equal_tensor(gc_tensor self, gc_tensor other);
+raw_tensor atg_greater_equal_tensor_(gc_tensor self, gc_tensor other);
+raw_tensor atg_greater_equal_tensor_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_greater_scalar_out(gc_tensor out, gc_tensor self, scalar other);
+raw_tensor atg_greater_tensor(gc_tensor self, gc_tensor other);
+raw_tensor atg_greater_tensor_(gc_tensor self, gc_tensor other);
+raw_tensor atg_greater_tensor_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_grid_sampler(gc_tensor input, gc_tensor grid, int64_t interpolation_mode, int64_t padding_mode, int align_corners);
+raw_tensor atg_grid_sampler_2d(gc_tensor input, gc_tensor grid, int64_t interpolation_mode, int64_t padding_mode, int align_corners);
+raw_tensor atg_grid_sampler_2d_out(gc_tensor out, gc_tensor input, gc_tensor grid, int64_t interpolation_mode, int64_t padding_mode, int align_corners);
+raw_tensor atg_grid_sampler_3d(gc_tensor input, gc_tensor grid, int64_t interpolation_mode, int64_t padding_mode, int align_corners);
+raw_tensor atg_grid_sampler_3d_out(gc_tensor out, gc_tensor input, gc_tensor grid, int64_t interpolation_mode, int64_t padding_mode, int align_corners);
+raw_tensor atg_group_norm(gc_tensor input, int64_t num_groups, gc_tensor weight, gc_tensor bias, double eps, int cudnn_enabled);
+void atg_gru(raw_tensor *, gc_tensor input, gc_tensor hx, gc_tensor *params_data, int params_len, int has_biases, int64_t num_layers, double dropout, int train, int bidirectional, int batch_first);
+raw_tensor atg_gru_cell(gc_tensor input, gc_tensor hx, gc_tensor w_ih, gc_tensor w_hh, gc_tensor b_ih, gc_tensor b_hh);
+void atg_gru_data(raw_tensor *, gc_tensor data, gc_tensor batch_sizes, gc_tensor hx, gc_tensor *params_data, int params_len, int has_biases, int64_t num_layers, double dropout, int train, int bidirectional);
+raw_tensor atg_gt(gc_tensor self, scalar other);
+raw_tensor atg_gt_(gc_tensor self, scalar other);
+raw_tensor atg_gt_scalar_out(gc_tensor out, gc_tensor self, scalar other);
+raw_tensor atg_gt_tensor(gc_tensor self, gc_tensor other);
+raw_tensor atg_gt_tensor_(gc_tensor self, gc_tensor other);
+raw_tensor atg_gt_tensor_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_hamming_window(int64_t window_length, int options_kind, int options_device);
+raw_tensor atg_hamming_window_out(gc_tensor out, int64_t window_length);
+raw_tensor atg_hamming_window_periodic(int64_t window_length, int periodic, int options_kind, int options_device);
+raw_tensor atg_hamming_window_periodic_alpha(int64_t window_length, int periodic, double alpha, int options_kind, int options_device);
+raw_tensor atg_hamming_window_periodic_alpha_beta(int64_t window_length, int periodic, double alpha, double beta, int options_kind, int options_device);
+raw_tensor atg_hamming_window_periodic_alpha_beta_out(gc_tensor out, int64_t window_length, int periodic, double alpha, double beta);
+raw_tensor atg_hamming_window_periodic_alpha_out(gc_tensor out, int64_t window_length, int periodic, double alpha);
+raw_tensor atg_hamming_window_periodic_out(gc_tensor out, int64_t window_length, int periodic);
+raw_tensor atg_hann_window(int64_t window_length, int options_kind, int options_device);
+raw_tensor atg_hann_window_out(gc_tensor out, int64_t window_length);
+raw_tensor atg_hann_window_periodic(int64_t window_length, int periodic, int options_kind, int options_device);
+raw_tensor atg_hann_window_periodic_out(gc_tensor out, int64_t window_length, int periodic);
+raw_tensor atg_hardshrink(gc_tensor self);
+raw_tensor atg_hardshrink_backward(gc_tensor grad_out, gc_tensor self, scalar lambd);
+raw_tensor atg_hardshrink_backward_grad_input(gc_tensor grad_input, gc_tensor grad_out, gc_tensor self, scalar lambd);
+raw_tensor atg_hardshrink_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_hardsigmoid(gc_tensor self);
+raw_tensor atg_hardsigmoid_(gc_tensor self);
+raw_tensor atg_hardsigmoid_backward(gc_tensor grad_output, gc_tensor self);
+raw_tensor atg_hardsigmoid_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self);
+raw_tensor atg_hardsigmoid_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_hardswish(gc_tensor self);
+raw_tensor atg_hardswish_(gc_tensor self);
+raw_tensor atg_hardswish_backward(gc_tensor grad_output, gc_tensor self);
+raw_tensor atg_hardswish_backward_out(gc_tensor out, gc_tensor grad_output, gc_tensor self);
+raw_tensor atg_hardswish_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_hardtanh(gc_tensor self);
+raw_tensor atg_hardtanh_(gc_tensor self);
+raw_tensor atg_hardtanh_backward(gc_tensor grad_output, gc_tensor self, scalar min_val, scalar max_val);
+raw_tensor atg_hardtanh_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, scalar min_val, scalar max_val);
+raw_tensor atg_hardtanh_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_heaviside(gc_tensor self, gc_tensor values);
+raw_tensor atg_heaviside_(gc_tensor self, gc_tensor values);
+raw_tensor atg_heaviside_out(gc_tensor out, gc_tensor self, gc_tensor values);
+raw_tensor atg_hinge_embedding_loss(gc_tensor self, gc_tensor target, double margin, int64_t reduction);
+raw_tensor atg_histc(gc_tensor self, int64_t bins);
+raw_tensor atg_histc_out(gc_tensor out, gc_tensor self, int64_t bins);
+raw_tensor *atg_hsplit(gc_tensor self, int64_t sections);
+raw_tensor *atg_hsplit_array(gc_tensor self, int64_t *indices_data, int indices_len);
+raw_tensor atg_hspmm(gc_tensor mat1, gc_tensor mat2);
+raw_tensor atg_hspmm_out(gc_tensor out, gc_tensor mat1, gc_tensor mat2);
+raw_tensor atg_hstack(gc_tensor *tensors_data, int tensors_len);
+raw_tensor atg_hstack_out(gc_tensor out, gc_tensor *tensors_data, int tensors_len);
+raw_tensor atg_huber_loss(gc_tensor self, gc_tensor target, int64_t reduction, double delta);
+raw_tensor atg_huber_loss_backward(gc_tensor grad_output, gc_tensor self, gc_tensor target, int64_t reduction, double delta);
+raw_tensor atg_huber_loss_backward_out(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, gc_tensor target, int64_t reduction, double delta);
+raw_tensor atg_huber_loss_out(gc_tensor out, gc_tensor self, gc_tensor target, int64_t reduction, double delta);
+raw_tensor atg_hypot(gc_tensor self, gc_tensor other);
+raw_tensor atg_hypot_(gc_tensor self, gc_tensor other);
+raw_tensor atg_hypot_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_i0(gc_tensor self);
+raw_tensor atg_i0_(gc_tensor self);
+raw_tensor atg_i0_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_igamma(gc_tensor self, gc_tensor other);
+raw_tensor atg_igamma_(gc_tensor self, gc_tensor other);
+raw_tensor atg_igamma_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_igammac(gc_tensor self, gc_tensor other);
+raw_tensor atg_igammac_(gc_tensor self, gc_tensor other);
+raw_tensor atg_igammac_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_im2col(gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *dilation_data, int dilation_len, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len);
+raw_tensor atg_im2col_out(gc_tensor out, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *dilation_data, int dilation_len, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len);
+raw_tensor atg_imag(gc_tensor self);
+raw_tensor atg_index(gc_tensor self, gc_tensor *indices_data, int indices_len);
+raw_tensor atg_index_add(gc_tensor self, int64_t dim, gc_tensor index, gc_tensor source);
+raw_tensor atg_index_add_(gc_tensor self, int64_t dim, gc_tensor index, gc_tensor source);
+raw_tensor atg_index_add_out(gc_tensor out, gc_tensor self, int64_t dim, gc_tensor index, gc_tensor source);
+raw_tensor atg_index_copy(gc_tensor self, int64_t dim, gc_tensor index, gc_tensor source);
+raw_tensor atg_index_copy_(gc_tensor self, int64_t dim, gc_tensor index, gc_tensor source);
+raw_tensor atg_index_copy_out(gc_tensor out, gc_tensor self, int64_t dim, gc_tensor index, gc_tensor source);
+raw_tensor atg_index_fill(gc_tensor self, int64_t dim, gc_tensor index, scalar value);
+raw_tensor atg_index_fill_(gc_tensor self, int64_t dim, gc_tensor index, scalar value);
+raw_tensor atg_index_fill_int_scalar_out(gc_tensor out, gc_tensor self, int64_t dim, gc_tensor index, scalar value);
+raw_tensor atg_index_fill_int_tensor(gc_tensor self, int64_t dim, gc_tensor index, gc_tensor value);
+raw_tensor atg_index_fill_int_tensor_(gc_tensor self, int64_t dim, gc_tensor index, gc_tensor value);
+raw_tensor atg_index_fill_int_tensor_out(gc_tensor out, gc_tensor self, int64_t dim, gc_tensor index, gc_tensor value);
+raw_tensor atg_index_put(gc_tensor self, gc_tensor *indices_data, int indices_len, gc_tensor values, int accumulate);
+raw_tensor atg_index_put_(gc_tensor self, gc_tensor *indices_data, int indices_len, gc_tensor values, int accumulate);
+raw_tensor atg_index_put_out(gc_tensor out, gc_tensor self, gc_tensor *indices_data, int indices_len, gc_tensor values, int accumulate);
+raw_tensor atg_index_reduce(gc_tensor self, int64_t dim, gc_tensor index, gc_tensor source, char * reduce, int include_self);
+raw_tensor atg_index_reduce_(gc_tensor self, int64_t dim, gc_tensor index, gc_tensor source, char * reduce, int include_self);
+raw_tensor atg_index_reduce_out(gc_tensor out, gc_tensor self, int64_t dim, gc_tensor index, gc_tensor source, char * reduce, int include_self);
+raw_tensor atg_index_select(gc_tensor self, int64_t dim, gc_tensor index);
+raw_tensor atg_index_select_backward(gc_tensor grad, int64_t *self_sizes_data, int self_sizes_len, int64_t dim, gc_tensor index);
+raw_tensor atg_index_select_out(gc_tensor out, gc_tensor self, int64_t dim, gc_tensor index);
+raw_tensor atg_index_tensor_out(gc_tensor out, gc_tensor self, gc_tensor *indices_data, int indices_len);
+raw_tensor atg_indices(gc_tensor self);
+raw_tensor atg_indices_copy(gc_tensor self);
+raw_tensor atg_indices_copy_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_infinitely_differentiable_gelu_backward(gc_tensor grad, gc_tensor self);
+raw_tensor atg_inner(gc_tensor self, gc_tensor other);
+raw_tensor atg_inner_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_instance_norm(gc_tensor input, gc_tensor weight, gc_tensor bias, gc_tensor running_mean, gc_tensor running_var, int use_input_stats, double momentum, double eps, int cudnn_enabled);
+raw_tensor atg_int_repr(gc_tensor self);
+raw_tensor atg_int_repr_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_inverse(gc_tensor self);
+raw_tensor atg_inverse_out(gc_tensor out, gc_tensor self);
+int atg_is_coalesced(gc_tensor self);
+int atg_is_complex(gc_tensor self);
+int atg_is_conj(gc_tensor self);
+int atg_is_distributed(gc_tensor self);
+int atg_is_floating_point(gc_tensor self);
+int atg_is_inference(gc_tensor self);
+int atg_is_leaf(gc_tensor self);
+int atg_is_neg(gc_tensor self);
+int atg_is_nonzero(gc_tensor self);
+int atg_is_pinned(gc_tensor self, int device);
+int atg_is_same_size(gc_tensor self, gc_tensor other);
+int atg_is_set_to(gc_tensor self, gc_tensor tensor);
+int atg_is_signed(gc_tensor self);
 int atg_is_vulkan_available();
-void atg_isclose(tensor *, tensor self, tensor other, double rtol, double atol, int equal_nan);
-void atg_isfinite(tensor *, tensor self);
-void atg_isin(tensor *, tensor elements, tensor test_elements, int assume_unique, int invert);
-void atg_isin_scalar_tensor(tensor *, scalar element, tensor test_elements, int assume_unique, int invert);
-void atg_isin_scalar_tensor_out(tensor *, tensor out, scalar element, tensor test_elements, int assume_unique, int invert);
-void atg_isin_tensor_scalar(tensor *, tensor elements, scalar test_element, int assume_unique, int invert);
-void atg_isin_tensor_scalar_out(tensor *, tensor out, tensor elements, scalar test_element, int assume_unique, int invert);
-void atg_isin_tensor_tensor_out(tensor *, tensor out, tensor elements, tensor test_elements, int assume_unique, int invert);
-void atg_isinf(tensor *, tensor self);
-void atg_isinf_out(tensor *, tensor out, tensor self);
-void atg_isnan(tensor *, tensor self);
-void atg_isnan_out(tensor *, tensor out, tensor self);
-void atg_isneginf(tensor *, tensor self);
-void atg_isneginf_out(tensor *, tensor out, tensor self);
-void atg_isposinf(tensor *, tensor self);
-void atg_isposinf_out(tensor *, tensor out, tensor self);
-void atg_isreal(tensor *, tensor self);
-void atg_istft(tensor *, tensor self, int64_t n_fft, int64_t hop_length_v, int hop_length_null, int64_t win_length_v, int win_length_null, tensor window, int center, int normalized, int onesided, int64_t length_v, int length_null, int return_complex);
-void atg_kaiser_window(tensor *, int64_t window_length, int options_kind, int options_device);
-void atg_kaiser_window_beta(tensor *, int64_t window_length, int periodic, double beta, int options_kind, int options_device);
-void atg_kaiser_window_beta_out(tensor *, tensor out, int64_t window_length, int periodic, double beta);
-void atg_kaiser_window_out(tensor *, tensor out, int64_t window_length);
-void atg_kaiser_window_periodic(tensor *, int64_t window_length, int periodic, int options_kind, int options_device);
-void atg_kaiser_window_periodic_out(tensor *, tensor out, int64_t window_length, int periodic);
-void atg_kl_div(tensor *, tensor self, tensor target, int64_t reduction, int log_target);
-void atg_kron(tensor *, tensor self, tensor other);
-void atg_kron_out(tensor *, tensor out, tensor self, tensor other);
-void atg_kthvalue(tensor *, tensor self, int64_t k, int64_t dim, int keepdim);
-void atg_kthvalue_values(tensor *, tensor values, tensor indices, tensor self, int64_t k, int64_t dim, int keepdim);
-void atg_l1_loss(tensor *, tensor self, tensor target, int64_t reduction);
-void atg_layer_norm(tensor *, tensor input, int64_t *normalized_shape_data, int normalized_shape_len, tensor weight, tensor bias, double eps, int cudnn_enable);
-void atg_lcm(tensor *, tensor self, tensor other);
-void atg_lcm_(tensor *, tensor self, tensor other);
-void atg_lcm_out(tensor *, tensor out, tensor self, tensor other);
-void atg_ldexp(tensor *, tensor self, tensor other);
-void atg_ldexp_(tensor *, tensor self, tensor other);
-void atg_ldexp_out(tensor *, tensor out, tensor self, tensor other);
-void atg_le(tensor *, tensor self, scalar other);
-void atg_le_(tensor *, tensor self, scalar other);
-void atg_le_scalar_out(tensor *, tensor out, tensor self, scalar other);
-void atg_le_tensor(tensor *, tensor self, tensor other);
-void atg_le_tensor_(tensor *, tensor self, tensor other);
-void atg_le_tensor_out(tensor *, tensor out, tensor self, tensor other);
-void atg_leaky_relu(tensor *, tensor self);
-void atg_leaky_relu_(tensor *, tensor self);
-void atg_leaky_relu_backward(tensor *, tensor grad_output, tensor self, scalar negative_slope, int self_is_result);
-void atg_leaky_relu_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, tensor self, scalar negative_slope, int self_is_result);
-void atg_leaky_relu_out(tensor *, tensor out, tensor self);
-void atg_lerp(tensor *, tensor self, tensor end, scalar weight);
-void atg_lerp_(tensor *, tensor self, tensor end, scalar weight);
-void atg_lerp_scalar_out(tensor *, tensor out, tensor self, tensor end, scalar weight);
-void atg_lerp_tensor(tensor *, tensor self, tensor end, tensor weight);
-void atg_lerp_tensor_(tensor *, tensor self, tensor end, tensor weight);
-void atg_lerp_tensor_out(tensor *, tensor out, tensor self, tensor end, tensor weight);
-void atg_less(tensor *, tensor self, scalar other);
-void atg_less_(tensor *, tensor self, scalar other);
-void atg_less_equal(tensor *, tensor self, scalar other);
-void atg_less_equal_(tensor *, tensor self, scalar other);
-void atg_less_equal_scalar_out(tensor *, tensor out, tensor self, scalar other);
-void atg_less_equal_tensor(tensor *, tensor self, tensor other);
-void atg_less_equal_tensor_(tensor *, tensor self, tensor other);
-void atg_less_equal_tensor_out(tensor *, tensor out, tensor self, tensor other);
-void atg_less_scalar_out(tensor *, tensor out, tensor self, scalar other);
-void atg_less_tensor(tensor *, tensor self, tensor other);
-void atg_less_tensor_(tensor *, tensor self, tensor other);
-void atg_less_tensor_out(tensor *, tensor out, tensor self, tensor other);
-void atg_lgamma(tensor *, tensor self);
-void atg_lgamma_(tensor *, tensor self);
-void atg_lgamma_out(tensor *, tensor out, tensor self);
-void atg_lift(tensor *, tensor self);
-void atg_lift_fresh(tensor *, tensor self);
-void atg_lift_fresh_copy(tensor *, tensor self);
-void atg_lift_fresh_copy_out(tensor *, tensor out, tensor self);
-void atg_lift_out(tensor *, tensor out, tensor self);
-void atg_linalg_cholesky(tensor *, tensor self, int upper);
-void atg_linalg_cholesky_ex(tensor *, tensor self, int upper, int check_errors);
-void atg_linalg_cholesky_ex_l(tensor *, tensor L, tensor info, tensor self, int upper, int check_errors);
-void atg_linalg_cholesky_out(tensor *, tensor out, tensor self, int upper);
-void atg_linalg_cond(tensor *, tensor self, scalar p);
-void atg_linalg_cond_out(tensor *, tensor out, tensor self, scalar p);
-void atg_linalg_cond_p_str(tensor *, tensor self, char * p);
-void atg_linalg_cond_p_str_out(tensor *, tensor out, tensor self, char * p);
-void atg_linalg_cross(tensor *, tensor self, tensor other, int64_t dim);
-void atg_linalg_cross_out(tensor *, tensor out, tensor self, tensor other, int64_t dim);
-void atg_linalg_det(tensor *, tensor A);
-void atg_linalg_det_out(tensor *, tensor out, tensor A);
-void atg_linalg_diagonal(tensor *, tensor A, int64_t offset, int64_t dim1, int64_t dim2);
-void atg_linalg_eig(tensor *, tensor self);
-void atg_linalg_eig_out(tensor *, tensor eigenvalues, tensor eigenvectors, tensor self);
-void atg_linalg_eigh(tensor *, tensor self, char * UPLO);
-void atg_linalg_eigh_eigvals(tensor *, tensor eigvals, tensor eigvecs, tensor self, char * UPLO);
-void atg_linalg_eigvals(tensor *, tensor self);
-void atg_linalg_eigvals_out(tensor *, tensor out, tensor self);
-void atg_linalg_eigvalsh(tensor *, tensor self, char * UPLO);
-void atg_linalg_eigvalsh_out(tensor *, tensor out, tensor self, char * UPLO);
-void atg_linalg_householder_product(tensor *, tensor input, tensor tau);
-void atg_linalg_householder_product_out(tensor *, tensor out, tensor input, tensor tau);
-void atg_linalg_inv(tensor *, tensor A);
-void atg_linalg_inv_ex(tensor *, tensor A, int check_errors);
-void atg_linalg_inv_ex_inverse(tensor *, tensor inverse, tensor info, tensor A, int check_errors);
-void atg_linalg_inv_out(tensor *, tensor out, tensor A);
-void atg_linalg_ldl_factor(tensor *, tensor self, int hermitian);
-void atg_linalg_ldl_factor_ex(tensor *, tensor self, int hermitian, int check_errors);
-void atg_linalg_ldl_factor_ex_out(tensor *, tensor LD, tensor pivots, tensor info, tensor self, int hermitian, int check_errors);
-void atg_linalg_ldl_factor_out(tensor *, tensor LD, tensor pivots, tensor self, int hermitian);
-void atg_linalg_ldl_solve(tensor *, tensor LD, tensor pivots, tensor B, int hermitian);
-void atg_linalg_ldl_solve_out(tensor *, tensor out, tensor LD, tensor pivots, tensor B, int hermitian);
-void atg_linalg_lstsq(tensor *, tensor self, tensor b, double rcond_v, int rcond_null, char * driver);
-void atg_linalg_lstsq_out(tensor *, tensor solution, tensor residuals, tensor rank, tensor singular_values, tensor self, tensor b, double rcond_v, int rcond_null, char * driver);
-void atg_linalg_lu(tensor *, tensor A, int pivot);
-void atg_linalg_lu_factor(tensor *, tensor A, int pivot);
-void atg_linalg_lu_factor_ex(tensor *, tensor A, int pivot, int check_errors);
-void atg_linalg_lu_factor_ex_out(tensor *, tensor LU, tensor pivots, tensor info, tensor A, int pivot, int check_errors);
-void atg_linalg_lu_factor_out(tensor *, tensor LU, tensor pivots, tensor A, int pivot);
-void atg_linalg_lu_out(tensor *, tensor P, tensor L, tensor U, tensor A, int pivot);
-void atg_linalg_lu_solve(tensor *, tensor LU, tensor pivots, tensor B, int left, int adjoint);
-void atg_linalg_lu_solve_out(tensor *, tensor out, tensor LU, tensor pivots, tensor B, int left, int adjoint);
-void atg_linalg_matmul(tensor *, tensor self, tensor other);
-void atg_linalg_matmul_out(tensor *, tensor out, tensor self, tensor other);
-void atg_linalg_matrix_exp(tensor *, tensor self);
-void atg_linalg_matrix_exp_out(tensor *, tensor out, tensor self);
-void atg_linalg_matrix_power(tensor *, tensor self, int64_t n);
-void atg_linalg_matrix_power_out(tensor *, tensor out, tensor self, int64_t n);
-void atg_linalg_matrix_rank(tensor *, tensor self, double tol, int hermitian);
-void atg_linalg_matrix_rank_atol_rtol_float(tensor *, tensor self, double atol_v, int atol_null, double rtol_v, int rtol_null, int hermitian);
-void atg_linalg_matrix_rank_atol_rtol_float_out(tensor *, tensor out, tensor self, double atol_v, int atol_null, double rtol_v, int rtol_null, int hermitian);
-void atg_linalg_matrix_rank_atol_rtol_tensor(tensor *, tensor input, tensor atol, tensor rtol, int hermitian);
-void atg_linalg_matrix_rank_atol_rtol_tensor_out(tensor *, tensor out, tensor input, tensor atol, tensor rtol, int hermitian);
-void atg_linalg_matrix_rank_out(tensor *, tensor out, tensor self, double tol, int hermitian);
-void atg_linalg_matrix_rank_out_tol_tensor(tensor *, tensor out, tensor input, tensor tol, int hermitian);
-void atg_linalg_matrix_rank_tol_tensor(tensor *, tensor input, tensor tol, int hermitian);
-void atg_linalg_multi_dot(tensor *, tensor *tensors_data, int tensors_len);
-void atg_linalg_multi_dot_out(tensor *, tensor out, tensor *tensors_data, int tensors_len);
-void atg_linalg_pinv(tensor *, tensor self, double rcond, int hermitian);
-void atg_linalg_pinv_atol_rtol_float(tensor *, tensor self, double atol_v, int atol_null, double rtol_v, int rtol_null, int hermitian);
-void atg_linalg_pinv_atol_rtol_float_out(tensor *, tensor out, tensor self, double atol_v, int atol_null, double rtol_v, int rtol_null, int hermitian);
-void atg_linalg_pinv_atol_rtol_tensor(tensor *, tensor self, tensor atol, tensor rtol, int hermitian);
-void atg_linalg_pinv_atol_rtol_tensor_out(tensor *, tensor out, tensor self, tensor atol, tensor rtol, int hermitian);
-void atg_linalg_pinv_out(tensor *, tensor out, tensor self, double rcond, int hermitian);
-void atg_linalg_pinv_out_rcond_tensor(tensor *, tensor out, tensor self, tensor rcond, int hermitian);
-void atg_linalg_pinv_rcond_tensor(tensor *, tensor self, tensor rcond, int hermitian);
-void atg_linalg_qr(tensor *, tensor A, char * mode);
-void atg_linalg_qr_out(tensor *, tensor Q, tensor R, tensor A, char * mode);
-void atg_linalg_slogdet(tensor *, tensor A);
-void atg_linalg_slogdet_out(tensor *, tensor sign, tensor logabsdet, tensor A);
-void atg_linalg_solve(tensor *, tensor A, tensor B, int left);
-void atg_linalg_solve_ex(tensor *, tensor A, tensor B, int left, int check_errors);
-void atg_linalg_solve_ex_out(tensor *, tensor result, tensor info, tensor A, tensor B, int left, int check_errors);
-void atg_linalg_solve_out(tensor *, tensor out, tensor A, tensor B, int left);
-void atg_linalg_solve_triangular(tensor *, tensor self, tensor B, int upper, int left, int unitriangular);
-void atg_linalg_solve_triangular_out(tensor *, tensor out, tensor self, tensor B, int upper, int left, int unitriangular);
-void atg_linalg_svd(tensor *, tensor A, int full_matrices, char * driver);
-void atg_linalg_svd_u(tensor *, tensor U, tensor S, tensor Vh, tensor A, int full_matrices, char * driver);
-void atg_linalg_svdvals(tensor *, tensor A, char * driver);
-void atg_linalg_svdvals_out(tensor *, tensor out, tensor A, char * driver);
-void atg_linalg_tensorinv(tensor *, tensor self, int64_t ind);
-void atg_linalg_tensorinv_out(tensor *, tensor out, tensor self, int64_t ind);
-void atg_linalg_tensorsolve(tensor *, tensor self, tensor other, int64_t *dims_data, int dims_len);
-void atg_linalg_tensorsolve_out(tensor *, tensor out, tensor self, tensor other, int64_t *dims_data, int dims_len);
-void atg_linalg_vander(tensor *, tensor x, int64_t n_v, int n_null);
-void atg_linalg_vecdot(tensor *, tensor x, tensor y, int64_t dim);
-void atg_linalg_vecdot_out(tensor *, tensor out, tensor x, tensor y, int64_t dim);
-void atg_linear(tensor *, tensor input, tensor weight, tensor bias);
-void atg_linear_out(tensor *, tensor out, tensor input, tensor weight, tensor bias);
-void atg_linspace(tensor *, scalar start, scalar end, int64_t steps, int options_kind, int options_device);
-void atg_linspace_out(tensor *, tensor out, scalar start, scalar end, int64_t steps);
-void atg_log(tensor *, tensor self);
-void atg_log10(tensor *, tensor self);
-void atg_log10_(tensor *, tensor self);
-void atg_log10_out(tensor *, tensor out, tensor self);
-void atg_log1p(tensor *, tensor self);
-void atg_log1p_(tensor *, tensor self);
-void atg_log1p_out(tensor *, tensor out, tensor self);
-void atg_log2(tensor *, tensor self);
-void atg_log2_(tensor *, tensor self);
-void atg_log2_out(tensor *, tensor out, tensor self);
-void atg_log_(tensor *, tensor self);
-void atg_log_normal(tensor *, tensor self, double mean, double std);
-void atg_log_normal_(tensor *, tensor self, double mean, double std);
-void atg_log_normal_out(tensor *, tensor out, tensor self, double mean, double std);
-void atg_log_out(tensor *, tensor out, tensor self);
-void atg_log_sigmoid(tensor *, tensor self);
-void atg_log_sigmoid_backward(tensor *, tensor grad_output, tensor self, tensor buffer);
-void atg_log_sigmoid_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, tensor self, tensor buffer);
-void atg_log_sigmoid_out(tensor *, tensor out, tensor self);
-void atg_log_softmax(tensor *, tensor self, int64_t dim, int dtype);
-void atg_log_softmax_int_out(tensor *, tensor out, tensor self, int64_t dim, int dtype);
-void atg_logaddexp(tensor *, tensor self, tensor other);
-void atg_logaddexp2(tensor *, tensor self, tensor other);
-void atg_logaddexp2_out(tensor *, tensor out, tensor self, tensor other);
-void atg_logaddexp_out(tensor *, tensor out, tensor self, tensor other);
-void atg_logcumsumexp(tensor *, tensor self, int64_t dim);
-void atg_logcumsumexp_out(tensor *, tensor out, tensor self, int64_t dim);
-void atg_logdet(tensor *, tensor self);
-void atg_logical_and(tensor *, tensor self, tensor other);
-void atg_logical_and_(tensor *, tensor self, tensor other);
-void atg_logical_and_out(tensor *, tensor out, tensor self, tensor other);
-void atg_logical_not(tensor *, tensor self);
-void atg_logical_not_(tensor *, tensor self);
-void atg_logical_not_out(tensor *, tensor out, tensor self);
-void atg_logical_or(tensor *, tensor self, tensor other);
-void atg_logical_or_(tensor *, tensor self, tensor other);
-void atg_logical_or_out(tensor *, tensor out, tensor self, tensor other);
-void atg_logical_xor(tensor *, tensor self, tensor other);
-void atg_logical_xor_(tensor *, tensor self, tensor other);
-void atg_logical_xor_out(tensor *, tensor out, tensor self, tensor other);
-void atg_logit(tensor *, tensor self, double eps_v, int eps_null);
-void atg_logit_(tensor *, tensor self, double eps_v, int eps_null);
-void atg_logit_backward(tensor *, tensor grad_output, tensor self, double eps_v, int eps_null);
-void atg_logit_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, tensor self, double eps_v, int eps_null);
-void atg_logit_out(tensor *, tensor out, tensor self, double eps_v, int eps_null);
-void atg_logspace(tensor *, scalar start, scalar end, int64_t steps, double base, int options_kind, int options_device);
-void atg_logspace_out(tensor *, tensor out, scalar start, scalar end, int64_t steps, double base);
-void atg_logsumexp(tensor *, tensor self, int64_t *dim_data, int dim_len, int keepdim);
-void atg_logsumexp_out(tensor *, tensor out, tensor self, int64_t *dim_data, int dim_len, int keepdim);
-void atg_lstm(tensor *, tensor input, tensor *hx_data, int hx_len, tensor *params_data, int params_len, int has_biases, int64_t num_layers, double dropout, int train, int bidirectional, int batch_first);
-void atg_lstm_cell(tensor *, tensor input, tensor *hx_data, int hx_len, tensor w_ih, tensor w_hh, tensor b_ih, tensor b_hh);
-void atg_lstm_data(tensor *, tensor data, tensor batch_sizes, tensor *hx_data, int hx_len, tensor *params_data, int params_len, int has_biases, int64_t num_layers, double dropout, int train, int bidirectional);
-void atg_lstm_mps_backward(tensor out0, tensor *out1_data, int out1_len, tensor *out2_data, int out2_len, tensor grad_y, tensor grad_hy, tensor grad_cy, tensor z_state, tensor cell_state_fwd, tensor input, tensor layersOutputs, tensor *hx_data, int hx_len, tensor *params_data, int params_len, int has_biases, int64_t num_layers, double dropout, int train, int bidirectional, int batch_first);
-void atg_lt(tensor *, tensor self, scalar other);
-void atg_lt_(tensor *, tensor self, scalar other);
-void atg_lt_scalar_out(tensor *, tensor out, tensor self, scalar other);
-void atg_lt_tensor(tensor *, tensor self, tensor other);
-void atg_lt_tensor_(tensor *, tensor self, tensor other);
-void atg_lt_tensor_out(tensor *, tensor out, tensor self, tensor other);
-void atg_lu_solve(tensor *, tensor self, tensor LU_data, tensor LU_pivots);
-void atg_lu_solve_out(tensor *, tensor out, tensor self, tensor LU_data, tensor LU_pivots);
-void atg_lu_unpack(tensor *, tensor LU_data, tensor LU_pivots, int unpack_data, int unpack_pivots);
-void atg_lu_unpack_out(tensor *, tensor P, tensor L, tensor U, tensor LU_data, tensor LU_pivots, int unpack_data, int unpack_pivots);
-void atg_margin_ranking_loss(tensor *, tensor input1, tensor input2, tensor target, double margin, int64_t reduction);
-void atg_masked_fill(tensor *, tensor self, tensor mask, scalar value);
-void atg_masked_fill_(tensor *, tensor self, tensor mask, scalar value);
-void atg_masked_fill_scalar_out(tensor *, tensor out, tensor self, tensor mask, scalar value);
-void atg_masked_fill_tensor(tensor *, tensor self, tensor mask, tensor value);
-void atg_masked_fill_tensor_(tensor *, tensor self, tensor mask, tensor value);
-void atg_masked_fill_tensor_out(tensor *, tensor out, tensor self, tensor mask, tensor value);
-void atg_masked_scatter(tensor *, tensor self, tensor mask, tensor source);
-void atg_masked_scatter_(tensor *, tensor self, tensor mask, tensor source);
-void atg_masked_scatter_out(tensor *, tensor out, tensor self, tensor mask, tensor source);
-void atg_masked_select(tensor *, tensor self, tensor mask);
-void atg_masked_select_backward(tensor *, tensor grad, tensor input, tensor mask);
-void atg_masked_select_out(tensor *, tensor out, tensor self, tensor mask);
-void atg_matmul(tensor *, tensor self, tensor other);
-void atg_matmul_out(tensor *, tensor out, tensor self, tensor other);
-void atg_matrix_exp(tensor *, tensor self);
-void atg_matrix_exp_backward(tensor *, tensor self, tensor grad);
-void atg_matrix_h(tensor *, tensor self);
-void atg_matrix_power(tensor *, tensor self, int64_t n);
-void atg_matrix_power_out(tensor *, tensor out, tensor self, int64_t n);
-void atg_max(tensor *, tensor self);
-void atg_max_dim(tensor *, tensor self, int64_t dim, int keepdim);
-void atg_max_dim_max(tensor *, tensor max, tensor max_values, tensor self, int64_t dim, int keepdim);
-void atg_max_other(tensor *, tensor self, tensor other);
-void atg_max_out(tensor *, tensor out, tensor self, tensor other);
-void atg_max_pool1d(tensor *, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
-void atg_max_pool1d_with_indices(tensor *, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
-void atg_max_pool2d(tensor *, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
-void atg_max_pool2d_backward(tensor *, tensor grad_output, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
-void atg_max_pool2d_backward_out(tensor *, tensor out, tensor grad_output, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
-void atg_max_pool2d_with_indices(tensor *, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
-void atg_max_pool2d_with_indices_backward(tensor *, tensor grad_output, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode, tensor indices);
-void atg_max_pool2d_with_indices_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode, tensor indices);
-void atg_max_pool2d_with_indices_out(tensor *, tensor out, tensor indices, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
-void atg_max_pool3d(tensor *, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
-void atg_max_pool3d_with_indices(tensor *, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
-void atg_max_pool3d_with_indices_backward(tensor *, tensor grad_output, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode, tensor indices);
-void atg_max_pool3d_with_indices_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode, tensor indices);
-void atg_max_pool3d_with_indices_out(tensor *, tensor out, tensor indices, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
-void atg_max_unary_out(tensor *, tensor out, tensor self);
-void atg_max_unpool2d(tensor *, tensor self, tensor indices, int64_t *output_size_data, int output_size_len);
-void atg_max_unpool2d_out(tensor *, tensor out, tensor self, tensor indices, int64_t *output_size_data, int output_size_len);
-void atg_max_unpool3d(tensor *, tensor self, tensor indices, int64_t *output_size_data, int output_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len);
-void atg_max_unpool3d_out(tensor *, tensor out, tensor self, tensor indices, int64_t *output_size_data, int output_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len);
-void atg_maximum(tensor *, tensor self, tensor other);
-void atg_maximum_out(tensor *, tensor out, tensor self, tensor other);
-void atg_mean(tensor *, tensor self, int dtype);
-void atg_mean_dim(tensor *, tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype);
-void atg_mean_out(tensor *, tensor out, tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype);
-void atg_median(tensor *, tensor self);
-void atg_median_dim(tensor *, tensor self, int64_t dim, int keepdim);
-void atg_median_dim_values(tensor *, tensor values, tensor indices, tensor self, int64_t dim, int keepdim);
-void atg_median_out(tensor *, tensor out, tensor self);
-tensor *atg_meshgrid(tensor *tensors_data, int tensors_len);
-tensor *atg_meshgrid_indexing(tensor *tensors_data, int tensors_len, char * indexing);
-void atg_mh(tensor *, tensor self);
-void atg_min(tensor *, tensor self);
-void atg_min_dim(tensor *, tensor self, int64_t dim, int keepdim);
-void atg_min_dim_min(tensor *, tensor min, tensor min_indices, tensor self, int64_t dim, int keepdim);
-void atg_min_other(tensor *, tensor self, tensor other);
-void atg_min_out(tensor *, tensor out, tensor self, tensor other);
-void atg_min_unary_out(tensor *, tensor out, tensor self);
-void atg_minimum(tensor *, tensor self, tensor other);
-void atg_minimum_out(tensor *, tensor out, tensor self, tensor other);
-void atg_miopen_batch_norm(tensor *, tensor input, tensor weight, tensor bias, tensor running_mean, tensor running_var, int training, double exponential_average_factor, double epsilon);
-void atg_miopen_batch_norm_backward(tensor *, tensor input, tensor grad_output, tensor weight, tensor running_mean, tensor running_var, tensor save_mean, tensor save_var, double epsilon);
-void atg_miopen_batch_norm_backward_out(tensor *, tensor out0, tensor out1, tensor out2, tensor input, tensor grad_output, tensor weight, tensor running_mean, tensor running_var, tensor save_mean, tensor save_var, double epsilon);
-void atg_miopen_batch_norm_out(tensor *, tensor out0, tensor out1, tensor out2, tensor input, tensor weight, tensor bias, tensor running_mean, tensor running_var, int training, double exponential_average_factor, double epsilon);
-void atg_miopen_convolution(tensor *, tensor self, tensor weight, tensor bias, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int benchmark, int deterministic);
-void atg_miopen_convolution_add_relu(tensor *, tensor self, tensor weight, tensor z, scalar alpha, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int64_t groups);
-void atg_miopen_convolution_out(tensor *, tensor out, tensor self, tensor weight, tensor bias, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int benchmark, int deterministic);
-void atg_miopen_convolution_relu(tensor *, tensor self, tensor weight, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int64_t groups);
-void atg_miopen_convolution_transpose(tensor *, tensor self, tensor weight, tensor bias, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int benchmark, int deterministic);
-void atg_miopen_convolution_transpose_out(tensor *, tensor out, tensor self, tensor weight, tensor bias, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int benchmark, int deterministic);
-void atg_miopen_depthwise_convolution(tensor *, tensor self, tensor weight, tensor bias, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int benchmark, int deterministic);
-void atg_miopen_depthwise_convolution_out(tensor *, tensor out, tensor self, tensor weight, tensor bias, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int benchmark, int deterministic);
-void atg_miopen_rnn(tensor *, tensor input, tensor *weight_data, int weight_len, int64_t weight_stride0, tensor hx, tensor cx, int64_t mode, int64_t hidden_size, int64_t num_layers, int batch_first, double dropout, int train, int bidirectional, int64_t *batch_sizes_data, int batch_sizes_len, tensor dropout_state);
-void atg_miopen_rnn_out(tensor *, tensor out0, tensor out1, tensor out2, tensor out3, tensor out4, tensor input, tensor *weight_data, int weight_len, int64_t weight_stride0, tensor hx, tensor cx, int64_t mode, int64_t hidden_size, int64_t num_layers, int batch_first, double dropout, int train, int bidirectional, int64_t *batch_sizes_data, int batch_sizes_len, tensor dropout_state);
-void atg_mish(tensor *, tensor self);
-void atg_mish_(tensor *, tensor self);
-void atg_mish_backward(tensor *, tensor grad_output, tensor self);
-void atg_mish_out(tensor *, tensor out, tensor self);
-void atg_mkldnn_adaptive_avg_pool2d(tensor *, tensor self, int64_t *output_size_data, int output_size_len);
-void atg_mkldnn_adaptive_avg_pool2d_backward(tensor *, tensor grad_output, tensor self);
-void atg_mkldnn_adaptive_avg_pool2d_backward_out(tensor *, tensor out, tensor grad_output, tensor self);
-void atg_mkldnn_adaptive_avg_pool2d_out(tensor *, tensor out, tensor self, int64_t *output_size_data, int output_size_len);
-void atg_mkldnn_convolution(tensor *, tensor self, tensor weight, tensor bias, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups);
-void atg_mkldnn_convolution_out(tensor *, tensor out, tensor self, tensor weight, tensor bias, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups);
-void atg_mkldnn_linear(tensor *, tensor self, tensor weight, tensor bias);
-void atg_mkldnn_linear_backward_input(tensor *, int64_t *input_size_data, int input_size_len, tensor grad_output, tensor weight);
-void atg_mkldnn_linear_backward_input_out(tensor *, tensor out, int64_t *input_size_data, int input_size_len, tensor grad_output, tensor weight);
-void atg_mkldnn_linear_backward_weights(tensor *, tensor grad_output, tensor input, tensor weight, int bias_defined);
-void atg_mkldnn_linear_backward_weights_out(tensor *, tensor out0, tensor out1, tensor grad_output, tensor input, tensor weight, int bias_defined);
-void atg_mkldnn_linear_out(tensor *, tensor out, tensor self, tensor weight, tensor bias);
-void atg_mkldnn_max_pool2d(tensor *, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
-void atg_mkldnn_max_pool2d_backward(tensor *, tensor grad_output, tensor output, tensor input, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
-void atg_mkldnn_max_pool2d_backward_out(tensor *, tensor out, tensor grad_output, tensor output, tensor input, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
-void atg_mkldnn_max_pool2d_out(tensor *, tensor out, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
-void atg_mkldnn_max_pool3d(tensor *, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
-void atg_mkldnn_max_pool3d_backward(tensor *, tensor grad_output, tensor output, tensor input, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
-void atg_mkldnn_max_pool3d_backward_out(tensor *, tensor out, tensor grad_output, tensor output, tensor input, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
-void atg_mkldnn_max_pool3d_out(tensor *, tensor out, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
-void atg_mkldnn_reorder_conv2d_weight(tensor *, tensor self, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int64_t *input_size_data, int input_size_len);
-void atg_mkldnn_reorder_conv2d_weight_out(tensor *, tensor out, tensor self, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int64_t *input_size_data, int input_size_len);
-void atg_mkldnn_reorder_conv3d_weight(tensor *, tensor self, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups);
-void atg_mkldnn_reorder_conv3d_weight_out(tensor *, tensor out, tensor self, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups);
-void atg_mkldnn_rnn_layer(tensor *, tensor input, tensor weight0, tensor weight1, tensor weight2, tensor weight3, tensor hx_, tensor cx_, int reverse, int64_t *batch_sizes_data, int batch_sizes_len, int64_t mode, int64_t hidden_size, int64_t num_layers, int has_biases, int bidirectional, int batch_first, int train);
-void atg_mkldnn_rnn_layer_backward(tensor *, tensor input, tensor weight1, tensor weight2, tensor weight3, tensor weight4, tensor hx_, tensor cx_tmp, tensor output, tensor hy_, tensor cy_, tensor grad_output, tensor grad_hy, tensor grad_cy, int reverse, int64_t mode, int64_t hidden_size, int64_t num_layers, int has_biases, int train, int bidirectional, int64_t *batch_sizes_data, int batch_sizes_len, int batch_first, tensor workspace);
-void atg_mkldnn_rnn_layer_backward_out(tensor *, tensor out0, tensor out1, tensor out2, tensor out3, tensor out4, tensor out5, tensor out6, tensor input, tensor weight1, tensor weight2, tensor weight3, tensor weight4, tensor hx_, tensor cx_tmp, tensor output, tensor hy_, tensor cy_, tensor grad_output, tensor grad_hy, tensor grad_cy, int reverse, int64_t mode, int64_t hidden_size, int64_t num_layers, int has_biases, int train, int bidirectional, int64_t *batch_sizes_data, int batch_sizes_len, int batch_first, tensor workspace);
-void atg_mkldnn_rnn_layer_out(tensor *, tensor out0, tensor out1, tensor out2, tensor out3, tensor input, tensor weight0, tensor weight1, tensor weight2, tensor weight3, tensor hx_, tensor cx_, int reverse, int64_t *batch_sizes_data, int batch_sizes_len, int64_t mode, int64_t hidden_size, int64_t num_layers, int has_biases, int bidirectional, int batch_first, int train);
-void atg_mm(tensor *, tensor self, tensor mat2);
-void atg_mm_out(tensor *, tensor out, tensor self, tensor mat2);
-void atg_mode(tensor *, tensor self, int64_t dim, int keepdim);
-void atg_mode_values(tensor *, tensor values, tensor indices, tensor self, int64_t dim, int keepdim);
-void atg_moveaxis(tensor *, tensor self, int64_t *source_data, int source_len, int64_t *destination_data, int destination_len);
-void atg_moveaxis_int(tensor *, tensor self, int64_t source, int64_t destination);
-void atg_movedim(tensor *, tensor self, int64_t *source_data, int source_len, int64_t *destination_data, int destination_len);
-void atg_movedim_int(tensor *, tensor self, int64_t source, int64_t destination);
-void atg_mse_loss(tensor *, tensor self, tensor target, int64_t reduction);
-void atg_mse_loss_backward(tensor *, tensor grad_output, tensor self, tensor target, int64_t reduction);
-void atg_mse_loss_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, tensor self, tensor target, int64_t reduction);
-void atg_mse_loss_out(tensor *, tensor out, tensor self, tensor target, int64_t reduction);
-void atg_msort(tensor *, tensor self);
-void atg_msort_out(tensor *, tensor out, tensor self);
-void atg_mt(tensor *, tensor self);
-void atg_mul(tensor *, tensor self, tensor other);
-void atg_mul_(tensor *, tensor self, tensor other);
-void atg_mul_out(tensor *, tensor out, tensor self, tensor other);
-void atg_mul_scalar(tensor *, tensor self, scalar other);
-void atg_mul_scalar_(tensor *, tensor self, scalar other);
-void atg_mul_scalar_out(tensor *, tensor out, tensor self, scalar other);
-void atg_multi_margin_loss_backward(tensor *, tensor grad_output, tensor self, tensor target, scalar p, scalar margin, tensor weight, int64_t reduction);
-void atg_multi_margin_loss_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, tensor self, tensor target, scalar p, scalar margin, tensor weight, int64_t reduction);
-void atg_multilabel_margin_loss(tensor *, tensor self, tensor target, int64_t reduction);
-void atg_multilabel_margin_loss_backward(tensor *, tensor grad_output, tensor self, tensor target, int64_t reduction, tensor is_target);
-void atg_multilabel_margin_loss_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, tensor self, tensor target, int64_t reduction, tensor is_target);
-void atg_multilabel_margin_loss_out(tensor *, tensor out, tensor self, tensor target, int64_t reduction);
-void atg_multinomial(tensor *, tensor self, int64_t num_samples, int replacement);
-void atg_multinomial_out(tensor *, tensor out, tensor self, int64_t num_samples, int replacement);
-void atg_multiply(tensor *, tensor self, tensor other);
-void atg_multiply_(tensor *, tensor self, tensor other);
-void atg_multiply_out(tensor *, tensor out, tensor self, tensor other);
-void atg_multiply_scalar(tensor *, tensor self, scalar other);
-void atg_multiply_scalar_(tensor *, tensor self, scalar other);
-void atg_mv(tensor *, tensor self, tensor vec);
-void atg_mv_out(tensor *, tensor out, tensor self, tensor vec);
-void atg_mvlgamma(tensor *, tensor self, int64_t p);
-void atg_mvlgamma_(tensor *, tensor self, int64_t p);
-void atg_mvlgamma_out(tensor *, tensor out, tensor self, int64_t p);
-void atg_nan_to_num(tensor *, tensor self, double nan_v, int nan_null, double posinf_v, int posinf_null, double neginf_v, int neginf_null);
-void atg_nan_to_num_(tensor *, tensor self, double nan_v, int nan_null, double posinf_v, int posinf_null, double neginf_v, int neginf_null);
-void atg_nan_to_num_out(tensor *, tensor out, tensor self, double nan_v, int nan_null, double posinf_v, int posinf_null, double neginf_v, int neginf_null);
-void atg_nanmean(tensor *, tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype);
-void atg_nanmean_out(tensor *, tensor out, tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype);
-void atg_nanmedian(tensor *, tensor self);
-void atg_nanmedian_dim(tensor *, tensor self, int64_t dim, int keepdim);
-void atg_nanmedian_dim_values(tensor *, tensor values, tensor indices, tensor self, int64_t dim, int keepdim);
-void atg_nanmedian_out(tensor *, tensor out, tensor self);
-void atg_nanquantile(tensor *, tensor self, tensor q, int64_t dim_v, int dim_null, int keepdim, char * interpolation);
-void atg_nanquantile_out(tensor *, tensor out, tensor self, tensor q, int64_t dim_v, int dim_null, int keepdim, char * interpolation);
-void atg_nanquantile_scalar(tensor *, tensor self, double q, int64_t dim_v, int dim_null, int keepdim, char * interpolation);
-void atg_nanquantile_scalar_out(tensor *, tensor out, tensor self, double q, int64_t dim_v, int dim_null, int keepdim, char * interpolation);
-void atg_nansum(tensor *, tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype);
-void atg_nansum_out(tensor *, tensor out, tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype);
-void atg_narrow(tensor *, tensor self, int64_t dim, int64_t start, int64_t length);
-void atg_narrow_copy(tensor *, tensor self, int64_t dim, int64_t start, int64_t length);
-void atg_narrow_copy_out(tensor *, tensor out, tensor self, int64_t dim, int64_t start, int64_t length);
-void atg_narrow_tensor(tensor *, tensor self, int64_t dim, tensor start, int64_t length);
-void atg_native_batch_norm(tensor *, tensor input, tensor weight, tensor bias, tensor running_mean, tensor running_var, int training, double momentum, double eps);
-void atg_native_batch_norm_out(tensor *, tensor out, tensor save_mean, tensor save_invstd, tensor input, tensor weight, tensor bias, tensor running_mean, tensor running_var, int training, double momentum, double eps);
-void atg_native_channel_shuffle(tensor *, tensor self, int64_t groups);
-void atg_native_dropout(tensor *, tensor input, double p, int train);
-void atg_native_dropout_backward(tensor *, tensor grad_output, tensor mask, double scale);
-void atg_native_dropout_backward_out(tensor *, tensor out, tensor grad_output, tensor mask, double scale);
-void atg_native_dropout_out(tensor *, tensor out0, tensor out1, tensor input, double p, int train);
-void atg_native_group_norm(tensor *, tensor input, tensor weight, tensor bias, int64_t n, int64_t C, int64_t HxW, int64_t group, double eps);
-void atg_native_group_norm_out(tensor *, tensor out0, tensor out1, tensor out2, tensor input, tensor weight, tensor bias, int64_t n, int64_t C, int64_t HxW, int64_t group, double eps);
-void atg_native_layer_norm(tensor *, tensor input, int64_t *normalized_shape_data, int normalized_shape_len, tensor weight, tensor bias, double eps);
-void atg_native_layer_norm_out(tensor *, tensor out0, tensor out1, tensor out2, tensor input, int64_t *normalized_shape_data, int normalized_shape_len, tensor weight, tensor bias, double eps);
-void atg_native_norm(tensor *, tensor self);
-void atg_native_norm_out(tensor *, tensor out, tensor self);
-void atg_native_norm_scalaropt_dim_dtype(tensor *, tensor self, scalar p, int64_t *dim_data, int dim_len, int keepdim, int dtype);
-void atg_native_norm_scalaropt_dim_dtype_out(tensor *, tensor out, tensor self, scalar p, int64_t *dim_data, int dim_len, int keepdim, int dtype);
-void atg_ne(tensor *, tensor self, scalar other);
-void atg_ne_(tensor *, tensor self, scalar other);
-void atg_ne_scalar_out(tensor *, tensor out, tensor self, scalar other);
-void atg_ne_tensor(tensor *, tensor self, tensor other);
-void atg_ne_tensor_(tensor *, tensor self, tensor other);
-void atg_ne_tensor_out(tensor *, tensor out, tensor self, tensor other);
-void atg_neg(tensor *, tensor self);
-void atg_neg_(tensor *, tensor self);
-void atg_neg_out(tensor *, tensor out, tensor self);
-void atg_negative(tensor *, tensor self);
-void atg_negative_(tensor *, tensor self);
-void atg_negative_out(tensor *, tensor out, tensor self);
-void atg_nested_to_padded_tensor(tensor *, tensor self, double padding, int64_t *output_size_data, int output_size_len);
-void atg_new_empty(tensor *, tensor self, int64_t *size_data, int size_len, int options_kind, int options_device);
-void atg_new_empty_out(tensor *, tensor out, tensor self, int64_t *size_data, int size_len);
-void atg_new_empty_strided(tensor *, tensor self, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len, int options_kind, int options_device);
-void atg_new_empty_strided_out(tensor *, tensor out, tensor self, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len);
-void atg_new_full(tensor *, tensor self, int64_t *size_data, int size_len, scalar fill_value, int options_kind, int options_device);
-void atg_new_full_out(tensor *, tensor out, tensor self, int64_t *size_data, int size_len, scalar fill_value);
-void atg_new_ones(tensor *, tensor self, int64_t *size_data, int size_len, int options_kind, int options_device);
-void atg_new_ones_out(tensor *, tensor out, tensor self, int64_t *size_data, int size_len);
-void atg_new_zeros(tensor *, tensor self, int64_t *size_data, int size_len, int options_kind, int options_device);
-void atg_new_zeros_out(tensor *, tensor out, tensor self, int64_t *size_data, int size_len);
-void atg_nextafter(tensor *, tensor self, tensor other);
-void atg_nextafter_(tensor *, tensor self, tensor other);
-void atg_nextafter_out(tensor *, tensor out, tensor self, tensor other);
-void atg_nll_loss(tensor *, tensor self, tensor target, tensor weight, int64_t reduction, int64_t ignore_index);
-void atg_nll_loss2d(tensor *, tensor self, tensor target, tensor weight, int64_t reduction, int64_t ignore_index);
-void atg_nll_loss2d_backward(tensor *, tensor grad_output, tensor self, tensor target, tensor weight, int64_t reduction, int64_t ignore_index, tensor total_weight);
-void atg_nll_loss2d_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, tensor self, tensor target, tensor weight, int64_t reduction, int64_t ignore_index, tensor total_weight);
-void atg_nll_loss2d_out(tensor *, tensor out, tensor self, tensor target, tensor weight, int64_t reduction, int64_t ignore_index);
-void atg_nll_loss_backward(tensor *, tensor grad_output, tensor self, tensor target, tensor weight, int64_t reduction, int64_t ignore_index, tensor total_weight);
-void atg_nll_loss_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, tensor self, tensor target, tensor weight, int64_t reduction, int64_t ignore_index, tensor total_weight);
-void atg_nll_loss_nd(tensor *, tensor self, tensor target, tensor weight, int64_t reduction, int64_t ignore_index);
-void atg_nll_loss_out(tensor *, tensor out, tensor self, tensor target, tensor weight, int64_t reduction, int64_t ignore_index);
-void atg_nonzero(tensor *, tensor self);
-tensor *atg_nonzero_numpy(tensor self);
-void atg_nonzero_out(tensor *, tensor out, tensor self);
-void atg_nonzero_static(tensor *, tensor self, int64_t size, int64_t fill_value);
-void atg_nonzero_static_out(tensor *, tensor out, tensor self, int64_t size, int64_t fill_value);
-void atg_norm(tensor *, tensor self);
-void atg_norm_dtype_out(tensor *, tensor out, tensor self, scalar p, int64_t *dim_data, int dim_len, int keepdim, int dtype);
-void atg_norm_except_dim(tensor *, tensor v, int64_t pow, int64_t dim);
-void atg_norm_out(tensor *, tensor out, tensor self, scalar p, int64_t *dim_data, int dim_len, int keepdim);
-void atg_norm_scalar_out(tensor *, tensor out, tensor self);
-void atg_norm_scalaropt_dim(tensor *, tensor self, scalar p, int64_t *dim_data, int dim_len, int keepdim);
-void atg_norm_scalaropt_dim_dtype(tensor *, tensor self, scalar p, int64_t *dim_data, int dim_len, int keepdim, int dtype);
-void atg_norm_scalaropt_dtype(tensor *, tensor self, scalar p, int dtype);
-void atg_norm_scalaropt_dtype_out(tensor *, tensor out, tensor self, scalar p, int dtype);
-void atg_normal_(tensor *, tensor self, double mean, double std);
-void atg_normal_functional(tensor *, tensor self, double mean, double std);
-void atg_not_equal(tensor *, tensor self, scalar other);
-void atg_not_equal_(tensor *, tensor self, scalar other);
-void atg_not_equal_scalar_out(tensor *, tensor out, tensor self, scalar other);
-void atg_not_equal_tensor(tensor *, tensor self, tensor other);
-void atg_not_equal_tensor_(tensor *, tensor self, tensor other);
-void atg_not_equal_tensor_out(tensor *, tensor out, tensor self, tensor other);
-void atg_nuclear_norm(tensor *, tensor self, int keepdim);
-void atg_nuclear_norm_dim(tensor *, tensor self, int64_t *dim_data, int dim_len, int keepdim);
-void atg_nuclear_norm_dim_out(tensor *, tensor out, tensor self, int64_t *dim_data, int dim_len, int keepdim);
-void atg_nuclear_norm_out(tensor *, tensor out, tensor self, int keepdim);
-void atg_numpy_t(tensor *, tensor self);
-void atg_one_hot(tensor *, tensor self, int64_t num_classes);
-void atg_ones(tensor *, int64_t *size_data, int size_len, int options_kind, int options_device);
-void atg_ones_like(tensor *, tensor self);
-void atg_ones_like_out(tensor *, tensor out, tensor self);
-void atg_ones_out(tensor *, tensor out, int64_t *size_data, int size_len);
-void atg_orgqr(tensor *, tensor self, tensor input2);
-void atg_orgqr_out(tensor *, tensor out, tensor self, tensor input2);
-void atg_ormqr(tensor *, tensor self, tensor input2, tensor input3, int left, int transpose);
-void atg_ormqr_out(tensor *, tensor out, tensor self, tensor input2, tensor input3, int left, int transpose);
-void atg_outer(tensor *, tensor self, tensor vec2);
-void atg_outer_out(tensor *, tensor out, tensor self, tensor vec2);
-int64_t atg_output_nr(tensor self);
-void atg_pad(tensor *, tensor self, int64_t *pad_data, int pad_len, char * mode, double value_v, int value_null);
-void atg_pad_sequence(tensor *, tensor *sequences_data, int sequences_len, int batch_first, double padding_value);
-void atg_pairwise_distance(tensor *, tensor x1, tensor x2, double p, double eps, int keepdim);
-void atg_pdist(tensor *, tensor self, double p);
-void atg_permute(tensor *, tensor self, int64_t *dims_data, int dims_len);
-void atg_permute_copy(tensor *, tensor self, int64_t *dims_data, int dims_len);
-void atg_permute_copy_out(tensor *, tensor out, tensor self, int64_t *dims_data, int dims_len);
-void atg_pin_memory(tensor *, tensor self, int device);
-void atg_pinverse(tensor *, tensor self, double rcond);
-void atg_pixel_shuffle(tensor *, tensor self, int64_t upscale_factor);
-void atg_pixel_shuffle_out(tensor *, tensor out, tensor self, int64_t upscale_factor);
-void atg_pixel_unshuffle(tensor *, tensor self, int64_t downscale_factor);
-void atg_pixel_unshuffle_out(tensor *, tensor out, tensor self, int64_t downscale_factor);
-void atg_poisson(tensor *, tensor self);
-void atg_poisson_nll_loss(tensor *, tensor input, tensor target, int log_input, int full, double eps, int64_t reduction);
-void atg_poisson_out(tensor *, tensor out, tensor self);
-void atg_polar(tensor *, tensor abs, tensor angle);
-void atg_polar_out(tensor *, tensor out, tensor abs, tensor angle);
-void atg_polygamma(tensor *, int64_t n, tensor self);
-void atg_polygamma_(tensor *, tensor self, int64_t n);
-void atg_polygamma_out(tensor *, tensor out, int64_t n, tensor self);
-void atg_positive(tensor *, tensor self);
-void atg_pow(tensor *, tensor self, tensor exponent);
-void atg_pow_(tensor *, tensor self, scalar exponent);
-void atg_pow_scalar(tensor *, scalar self, tensor exponent);
-void atg_pow_scalar_out(tensor *, tensor out, scalar self, tensor exponent);
-void atg_pow_tensor_(tensor *, tensor self, tensor exponent);
-void atg_pow_tensor_scalar(tensor *, tensor self, scalar exponent);
-void atg_pow_tensor_scalar_out(tensor *, tensor out, tensor self, scalar exponent);
-void atg_pow_tensor_tensor_out(tensor *, tensor out, tensor self, tensor exponent);
-void atg_prelu(tensor *, tensor self, tensor weight);
-void atg_prod(tensor *, tensor self, int dtype);
-void atg_prod_dim_int(tensor *, tensor self, int64_t dim, int keepdim, int dtype);
-void atg_prod_int_out(tensor *, tensor out, tensor self, int64_t dim, int keepdim, int dtype);
-void atg_prod_out(tensor *, tensor out, tensor self, int dtype);
-void atg_put(tensor *, tensor self, tensor index, tensor source, int accumulate);
-void atg_put_(tensor *, tensor self, tensor index, tensor source, int accumulate);
-void atg_put_out(tensor *, tensor out, tensor self, tensor index, tensor source, int accumulate);
-int64_t atg_q_per_channel_axis(tensor self);
-void atg_q_per_channel_scales(tensor *, tensor self);
-void atg_q_per_channel_scales_out(tensor *, tensor out, tensor self);
-void atg_q_per_channel_zero_points(tensor *, tensor self);
-void atg_q_per_channel_zero_points_out(tensor *, tensor out, tensor self);
-double atg_q_scale(tensor self);
-int64_t atg_q_zero_point(tensor self);
-void atg_qr(tensor *, tensor self, int some);
-void atg_qr_q(tensor *, tensor Q, tensor R, tensor self, int some);
-void atg_quantile(tensor *, tensor self, tensor q, int64_t dim_v, int dim_null, int keepdim, char * interpolation);
-void atg_quantile_out(tensor *, tensor out, tensor self, tensor q, int64_t dim_v, int dim_null, int keepdim, char * interpolation);
-void atg_quantile_scalar(tensor *, tensor self, double q, int64_t dim_v, int dim_null, int keepdim, char * interpolation);
-void atg_quantile_scalar_out(tensor *, tensor out, tensor self, double q, int64_t dim_v, int dim_null, int keepdim, char * interpolation);
-void atg_quantize_per_channel(tensor *, tensor self, tensor scales, tensor zero_points, int64_t axis, int dtype);
-void atg_quantize_per_channel_out(tensor *, tensor out, tensor self, tensor scales, tensor zero_points, int64_t axis, int dtype);
-void atg_quantize_per_tensor(tensor *, tensor self, double scale, int64_t zero_point, int dtype);
-void atg_quantize_per_tensor_dynamic(tensor *, tensor self, int dtype, int reduce_range);
-void atg_quantize_per_tensor_dynamic_out(tensor *, tensor out, tensor self, int dtype, int reduce_range);
-void atg_quantize_per_tensor_out(tensor *, tensor out, tensor self, double scale, int64_t zero_point, int dtype);
-void atg_quantize_per_tensor_tensor_qparams(tensor *, tensor self, tensor scale, tensor zero_point, int dtype);
-void atg_quantize_per_tensor_tensor_qparams_out(tensor *, tensor out, tensor self, tensor scale, tensor zero_point, int dtype);
-tensor *atg_quantize_per_tensor_tensors(tensor *tensors_data, int tensors_len, tensor scales, tensor zero_points, int dtype);
-void atg_quantize_per_tensor_tensors_out(tensor *out_data, int out_len, tensor *tensors_data, int tensors_len, tensor scales, tensor zero_points, int dtype);
-void atg_quantized_batch_norm(tensor *, tensor input, tensor weight, tensor bias, tensor mean, tensor var, double eps, double output_scale, int64_t output_zero_point);
-void atg_quantized_batch_norm_out(tensor *, tensor out, tensor input, tensor weight, tensor bias, tensor mean, tensor var, double eps, double output_scale, int64_t output_zero_point);
-void atg_quantized_gru_cell(tensor *, tensor input, tensor hx, tensor w_ih, tensor w_hh, tensor b_ih, tensor b_hh, tensor packed_ih, tensor packed_hh, tensor col_offsets_ih, tensor col_offsets_hh, scalar scale_ih, scalar scale_hh, scalar zero_point_ih, scalar zero_point_hh);
-void atg_quantized_lstm_cell(tensor *, tensor input, tensor *hx_data, int hx_len, tensor w_ih, tensor w_hh, tensor b_ih, tensor b_hh, tensor packed_ih, tensor packed_hh, tensor col_offsets_ih, tensor col_offsets_hh, scalar scale_ih, scalar scale_hh, scalar zero_point_ih, scalar zero_point_hh);
-void atg_quantized_max_pool1d(tensor *, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
-void atg_quantized_max_pool1d_out(tensor *, tensor out, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
-void atg_quantized_max_pool2d(tensor *, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
-void atg_quantized_max_pool2d_out(tensor *, tensor out, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
-void atg_quantized_max_pool3d(tensor *, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
-void atg_quantized_max_pool3d_out(tensor *, tensor out, tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
-void atg_quantized_rnn_relu_cell(tensor *, tensor input, tensor hx, tensor w_ih, tensor w_hh, tensor b_ih, tensor b_hh, tensor packed_ih, tensor packed_hh, tensor col_offsets_ih, tensor col_offsets_hh, scalar scale_ih, scalar scale_hh, scalar zero_point_ih, scalar zero_point_hh);
-void atg_quantized_rnn_tanh_cell(tensor *, tensor input, tensor hx, tensor w_ih, tensor w_hh, tensor b_ih, tensor b_hh, tensor packed_ih, tensor packed_hh, tensor col_offsets_ih, tensor col_offsets_hh, scalar scale_ih, scalar scale_hh, scalar zero_point_ih, scalar zero_point_hh);
-void atg_rad2deg(tensor *, tensor self);
-void atg_rad2deg_(tensor *, tensor self);
-void atg_rad2deg_out(tensor *, tensor out, tensor self);
-void atg_rand(tensor *, int64_t *size_data, int size_len, int options_kind, int options_device);
-void atg_rand_like(tensor *, tensor self);
-void atg_rand_like_out(tensor *, tensor out, tensor self);
-void atg_rand_out(tensor *, tensor out, int64_t *size_data, int size_len);
-void atg_randint(tensor *, int64_t high, int64_t *size_data, int size_len, int options_kind, int options_device);
-void atg_randint_like(tensor *, tensor self, int64_t high);
-void atg_randint_like_low_dtype(tensor *, tensor self, int64_t low, int64_t high);
-void atg_randint_like_low_dtype_out(tensor *, tensor out, tensor self, int64_t low, int64_t high);
-void atg_randint_like_out(tensor *, tensor out, tensor self, int64_t high);
-void atg_randint_low(tensor *, int64_t low, int64_t high, int64_t *size_data, int size_len, int options_kind, int options_device);
-void atg_randint_low_out(tensor *, tensor out, int64_t low, int64_t high, int64_t *size_data, int size_len);
-void atg_randint_out(tensor *, tensor out, int64_t high, int64_t *size_data, int size_len);
-void atg_randn(tensor *, int64_t *size_data, int size_len, int options_kind, int options_device);
-void atg_randn_like(tensor *, tensor self);
-void atg_randn_like_out(tensor *, tensor out, tensor self);
-void atg_randn_out(tensor *, tensor out, int64_t *size_data, int size_len);
-void atg_random(tensor *, tensor self);
-void atg_random_(tensor *, tensor self);
-void atg_random_from(tensor *, tensor self, int64_t from, int64_t to_v, int to_null);
-void atg_random_from_(tensor *, tensor self, int64_t from, int64_t to_v, int to_null);
-void atg_random_from_out(tensor *, tensor out, tensor self, int64_t from, int64_t to_v, int to_null);
-void atg_random_out(tensor *, tensor out, tensor self);
-void atg_random_to(tensor *, tensor self, int64_t to);
-void atg_random_to_(tensor *, tensor self, int64_t to);
-void atg_random_to_out(tensor *, tensor out, tensor self, int64_t to);
-void atg_randperm(tensor *, int64_t n, int options_kind, int options_device);
-void atg_randperm_out(tensor *, tensor out, int64_t n);
-void atg_range(tensor *, scalar start, scalar end, int options_kind, int options_device);
-void atg_range_out(tensor *, tensor out, scalar start, scalar end);
-void atg_range_out_(tensor *, tensor out, scalar start, scalar end);
-void atg_range_step(tensor *, scalar start, scalar end, int options_kind, int options_device);
-void atg_ravel(tensor *, tensor self);
-void atg_real(tensor *, tensor self);
-void atg_reciprocal(tensor *, tensor self);
-void atg_reciprocal_(tensor *, tensor self);
-void atg_reciprocal_out(tensor *, tensor out, tensor self);
-void atg_reflection_pad1d(tensor *, tensor self, int64_t *padding_data, int padding_len);
-void atg_reflection_pad1d_backward(tensor *, tensor grad_output, tensor self, int64_t *padding_data, int padding_len);
-void atg_reflection_pad1d_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, tensor self, int64_t *padding_data, int padding_len);
-void atg_reflection_pad1d_out(tensor *, tensor out, tensor self, int64_t *padding_data, int padding_len);
-void atg_reflection_pad2d(tensor *, tensor self, int64_t *padding_data, int padding_len);
-void atg_reflection_pad2d_backward(tensor *, tensor grad_output, tensor self, int64_t *padding_data, int padding_len);
-void atg_reflection_pad2d_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, tensor self, int64_t *padding_data, int padding_len);
-void atg_reflection_pad2d_out(tensor *, tensor out, tensor self, int64_t *padding_data, int padding_len);
-void atg_reflection_pad3d(tensor *, tensor self, int64_t *padding_data, int padding_len);
-void atg_reflection_pad3d_backward(tensor *, tensor grad_output, tensor self, int64_t *padding_data, int padding_len);
-void atg_reflection_pad3d_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, tensor self, int64_t *padding_data, int padding_len);
-void atg_reflection_pad3d_out(tensor *, tensor out, tensor self, int64_t *padding_data, int padding_len);
-void atg_relu(tensor *, tensor self);
-void atg_relu6(tensor *, tensor self);
-void atg_relu6_(tensor *, tensor self);
-void atg_relu_(tensor *, tensor self);
-void atg_relu_out(tensor *, tensor out, tensor self);
-void atg_remainder(tensor *, tensor self, scalar other);
-void atg_remainder_(tensor *, tensor self, scalar other);
-void atg_remainder_scalar_out(tensor *, tensor out, tensor self, scalar other);
-void atg_remainder_scalar_tensor(tensor *, scalar self, tensor other);
-void atg_remainder_scalar_tensor_out(tensor *, tensor out, scalar self, tensor other);
-void atg_remainder_tensor(tensor *, tensor self, tensor other);
-void atg_remainder_tensor_(tensor *, tensor self, tensor other);
-void atg_remainder_tensor_out(tensor *, tensor out, tensor self, tensor other);
-void atg_renorm(tensor *, tensor self, scalar p, int64_t dim, scalar maxnorm);
-void atg_renorm_(tensor *, tensor self, scalar p, int64_t dim, scalar maxnorm);
-void atg_renorm_out(tensor *, tensor out, tensor self, scalar p, int64_t dim, scalar maxnorm);
-void atg_repeat(tensor *, tensor self, int64_t *repeats_data, int repeats_len);
-void atg_repeat_interleave(tensor *, tensor repeats, int64_t output_size_v, int output_size_null);
-void atg_repeat_interleave_self_int(tensor *, tensor self, int64_t repeats, int64_t dim_v, int dim_null, int64_t output_size_v, int output_size_null);
-void atg_repeat_interleave_self_tensor(tensor *, tensor self, tensor repeats, int64_t dim_v, int dim_null, int64_t output_size_v, int output_size_null);
-void atg_repeat_interleave_tensor_out(tensor *, tensor out, tensor repeats, int64_t output_size_v, int output_size_null);
-void atg_repeat_out(tensor *, tensor out, tensor self, int64_t *repeats_data, int repeats_len);
-void atg_replication_pad1d(tensor *, tensor self, int64_t *padding_data, int padding_len);
-void atg_replication_pad1d_backward(tensor *, tensor grad_output, tensor self, int64_t *padding_data, int padding_len);
-void atg_replication_pad1d_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, tensor self, int64_t *padding_data, int padding_len);
-void atg_replication_pad1d_out(tensor *, tensor out, tensor self, int64_t *padding_data, int padding_len);
-void atg_replication_pad2d(tensor *, tensor self, int64_t *padding_data, int padding_len);
-void atg_replication_pad2d_backward(tensor *, tensor grad_output, tensor self, int64_t *padding_data, int padding_len);
-void atg_replication_pad2d_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, tensor self, int64_t *padding_data, int padding_len);
-void atg_replication_pad2d_out(tensor *, tensor out, tensor self, int64_t *padding_data, int padding_len);
-void atg_replication_pad3d(tensor *, tensor self, int64_t *padding_data, int padding_len);
-void atg_replication_pad3d_backward(tensor *, tensor grad_output, tensor self, int64_t *padding_data, int padding_len);
-void atg_replication_pad3d_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, tensor self, int64_t *padding_data, int padding_len);
-void atg_replication_pad3d_out(tensor *, tensor out, tensor self, int64_t *padding_data, int padding_len);
-void atg_requires_grad_(tensor *, tensor self, int requires_grad);
-void atg_reshape(tensor *, tensor self, int64_t *shape_data, int shape_len);
-void atg_reshape_as(tensor *, tensor self, tensor other);
-void atg_resize(tensor *, tensor self, int64_t *size_data, int size_len);
-void atg_resize_(tensor *, tensor self, int64_t *size_data, int size_len);
-void atg_resize_as(tensor *, tensor self, tensor the_template);
-void atg_resize_as_(tensor *, tensor self, tensor the_template);
-void atg_resize_as_out(tensor *, tensor out, tensor self, tensor the_template);
-void atg_resize_as_sparse(tensor *, tensor self, tensor the_template);
-void atg_resize_as_sparse_(tensor *, tensor self, tensor the_template);
-void atg_resize_as_sparse_out(tensor *, tensor out, tensor self, tensor the_template);
-void atg_resize_out(tensor *, tensor out, tensor self, int64_t *size_data, int size_len);
-void atg_resolve_conj(tensor *, tensor self);
-void atg_resolve_neg(tensor *, tensor self);
-int atg_retains_grad(tensor self);
-void atg_rnn_relu(tensor *, tensor input, tensor hx, tensor *params_data, int params_len, int has_biases, int64_t num_layers, double dropout, int train, int bidirectional, int batch_first);
-void atg_rnn_relu_cell(tensor *, tensor input, tensor hx, tensor w_ih, tensor w_hh, tensor b_ih, tensor b_hh);
-void atg_rnn_relu_data(tensor *, tensor data, tensor batch_sizes, tensor hx, tensor *params_data, int params_len, int has_biases, int64_t num_layers, double dropout, int train, int bidirectional);
-void atg_rnn_tanh(tensor *, tensor input, tensor hx, tensor *params_data, int params_len, int has_biases, int64_t num_layers, double dropout, int train, int bidirectional, int batch_first);
-void atg_rnn_tanh_cell(tensor *, tensor input, tensor hx, tensor w_ih, tensor w_hh, tensor b_ih, tensor b_hh);
-void atg_rnn_tanh_data(tensor *, tensor data, tensor batch_sizes, tensor hx, tensor *params_data, int params_len, int has_biases, int64_t num_layers, double dropout, int train, int bidirectional);
-void atg_roll(tensor *, tensor self, int64_t *shifts_data, int shifts_len, int64_t *dims_data, int dims_len);
-void atg_roll_out(tensor *, tensor out, tensor self, int64_t *shifts_data, int shifts_len, int64_t *dims_data, int dims_len);
-void atg_rot90(tensor *, tensor self, int64_t k, int64_t *dims_data, int dims_len);
-void atg_rot90_out(tensor *, tensor out, tensor self, int64_t k, int64_t *dims_data, int dims_len);
-void atg_round(tensor *, tensor self);
-void atg_round_(tensor *, tensor self);
-void atg_round_decimals(tensor *, tensor self, int64_t decimals);
-void atg_round_decimals_(tensor *, tensor self, int64_t decimals);
-void atg_round_decimals_out(tensor *, tensor out, tensor self, int64_t decimals);
-void atg_round_out(tensor *, tensor out, tensor self);
-void atg_row_indices(tensor *, tensor self);
-void atg_row_indices_copy(tensor *, tensor self);
-void atg_row_indices_copy_out(tensor *, tensor out, tensor self);
-void atg_row_stack(tensor *, tensor *tensors_data, int tensors_len);
-void atg_row_stack_out(tensor *, tensor out, tensor *tensors_data, int tensors_len);
-void atg_rrelu(tensor *, tensor self, int training);
-void atg_rrelu_(tensor *, tensor self, int training);
-void atg_rrelu_with_noise(tensor *, tensor self, tensor noise, int training);
-void atg_rrelu_with_noise_(tensor *, tensor self, tensor noise, int training);
-void atg_rrelu_with_noise_backward(tensor *, tensor grad_output, tensor self, tensor noise, scalar lower, scalar upper, int training, int self_is_result);
-void atg_rrelu_with_noise_backward_out(tensor *, tensor out, tensor grad_output, tensor self, tensor noise, scalar lower, scalar upper, int training, int self_is_result);
-void atg_rrelu_with_noise_out(tensor *, tensor out, tensor self, tensor noise, int training);
-void atg_rsqrt(tensor *, tensor self);
-void atg_rsqrt_(tensor *, tensor self);
-void atg_rsqrt_out(tensor *, tensor out, tensor self);
-void atg_rsub(tensor *, tensor self, tensor other);
-void atg_rsub_scalar(tensor *, tensor self, scalar other);
-void atg_rsub_scalar_out(tensor *, tensor out, tensor self, scalar other);
-void atg_rsub_tensor_out(tensor *, tensor out, tensor self, tensor other);
-void atg_scalar_tensor(tensor *, scalar s, int options_kind, int options_device);
-void atg_scalar_tensor_out(tensor *, tensor out, scalar s);
-void atg_scaled_dot_product_attention(tensor *, tensor query, tensor key, tensor value, tensor attn_mask, double dropout_p, int is_causal, double scale_v, int scale_null);
-void atg_scatter(tensor *, tensor self, int64_t dim, tensor index, tensor src);
-void atg_scatter_(tensor *, tensor self, int64_t dim, tensor index, tensor src);
-void atg_scatter_add(tensor *, tensor self, int64_t dim, tensor index, tensor src);
-void atg_scatter_add_(tensor *, tensor self, int64_t dim, tensor index, tensor src);
-void atg_scatter_add_out(tensor *, tensor out, tensor self, int64_t dim, tensor index, tensor src);
-void atg_scatter_reduce(tensor *, tensor self, int64_t dim, tensor index, tensor src, char * reduce);
-void atg_scatter_reduce_(tensor *, tensor self, int64_t dim, tensor index, tensor src, char * reduce);
-void atg_scatter_reduce_out(tensor *, tensor out, tensor self, int64_t dim, tensor index, tensor src, char * reduce);
-void atg_scatter_src_out(tensor *, tensor out, tensor self, int64_t dim, tensor index, tensor src);
-void atg_scatter_value(tensor *, tensor self, int64_t dim, tensor index, scalar value);
-void atg_scatter_value_(tensor *, tensor self, int64_t dim, tensor index, scalar value);
-void atg_scatter_value_out(tensor *, tensor out, tensor self, int64_t dim, tensor index, scalar value);
-void atg_scatter_value_reduce(tensor *, tensor self, int64_t dim, tensor index, scalar value, char * reduce);
-void atg_scatter_value_reduce_(tensor *, tensor self, int64_t dim, tensor index, scalar value, char * reduce);
-void atg_scatter_value_reduce_out(tensor *, tensor out, tensor self, int64_t dim, tensor index, scalar value, char * reduce);
-void atg_searchsorted(tensor *, tensor sorted_sequence, tensor self, int out_int32, int right, char * side, tensor sorter);
-void atg_searchsorted_scalar(tensor *, tensor sorted_sequence, scalar self, int out_int32, int right, char * side, tensor sorter);
-void atg_searchsorted_scalar_out(tensor *, tensor out, tensor sorted_sequence, scalar self, int out_int32, int right, char * side, tensor sorter);
-void atg_searchsorted_tensor_out(tensor *, tensor out, tensor sorted_sequence, tensor self, int out_int32, int right, char * side, tensor sorter);
-void atg_segment_reduce(tensor *, tensor data, char * reduce, tensor lengths, tensor indices, tensor offsets, int64_t axis, int unsafe, scalar initial);
-void atg_segment_reduce_out(tensor *, tensor out, tensor data, char * reduce, tensor lengths, tensor indices, tensor offsets, int64_t axis, int unsafe, scalar initial);
-void atg_select(tensor *, tensor self, int64_t dim, int64_t index);
-void atg_select_backward(tensor *, tensor grad_output, int64_t *input_sizes_data, int input_sizes_len, int64_t dim, int64_t index);
-void atg_select_backward_out(tensor *, tensor out, tensor grad_output, int64_t *input_sizes_data, int input_sizes_len, int64_t dim, int64_t index);
-void atg_select_copy(tensor *, tensor self, int64_t dim, int64_t index);
-void atg_select_copy_int_out(tensor *, tensor out, tensor self, int64_t dim, int64_t index);
-void atg_select_scatter(tensor *, tensor self, tensor src, int64_t dim, int64_t index);
-void atg_select_scatter_out(tensor *, tensor out, tensor self, tensor src, int64_t dim, int64_t index);
-void atg_selu(tensor *, tensor self);
-void atg_selu_(tensor *, tensor self);
-void atg_set(tensor *, tensor self);
-void atg_set_(tensor *, tensor self);
-void atg_set_out(tensor *, tensor out, tensor self);
-void atg_set_requires_grad(tensor *, tensor self, int r);
-void atg_set_source_tensor(tensor *, tensor self, tensor source);
-void atg_set_source_tensor_(tensor *, tensor self, tensor source);
-void atg_set_source_tensor_out(tensor *, tensor out, tensor self, tensor source);
-void atg_set_source_tensor_storage_offset_(tensor *, tensor self, tensor source, int64_t storage_offset, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len);
-void atg_sgn(tensor *, tensor self);
-void atg_sgn_(tensor *, tensor self);
-void atg_sgn_out(tensor *, tensor out, tensor self);
-void atg_sigmoid(tensor *, tensor self);
-void atg_sigmoid_(tensor *, tensor self);
-void atg_sigmoid_backward(tensor *, tensor grad_output, tensor output);
-void atg_sigmoid_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, tensor output);
-void atg_sigmoid_out(tensor *, tensor out, tensor self);
-void atg_sign(tensor *, tensor self);
-void atg_sign_(tensor *, tensor self);
-void atg_sign_out(tensor *, tensor out, tensor self);
-void atg_signbit(tensor *, tensor self);
-void atg_signbit_out(tensor *, tensor out, tensor self);
-void atg_silu(tensor *, tensor self);
-void atg_silu_(tensor *, tensor self);
-void atg_silu_backward(tensor *, tensor grad_output, tensor self);
-void atg_silu_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, tensor self);
-void atg_silu_out(tensor *, tensor out, tensor self);
-void atg_sin(tensor *, tensor self);
-void atg_sin_(tensor *, tensor self);
-void atg_sin_out(tensor *, tensor out, tensor self);
-void atg_sinc(tensor *, tensor self);
-void atg_sinc_(tensor *, tensor self);
-void atg_sinc_out(tensor *, tensor out, tensor self);
-void atg_sinh(tensor *, tensor self);
-void atg_sinh_(tensor *, tensor self);
-void atg_sinh_out(tensor *, tensor out, tensor self);
-int64_t atg_size(tensor self, int64_t dim);
-void atg_slice(tensor *, tensor self, int64_t dim, int64_t start_v, int start_null, int64_t end_v, int end_null, int64_t step);
-void atg_slice_backward(tensor *, tensor grad_output, int64_t *input_sizes_data, int input_sizes_len, int64_t dim, int64_t start, int64_t end, int64_t step);
-void atg_slice_backward_out(tensor *, tensor out, tensor grad_output, int64_t *input_sizes_data, int input_sizes_len, int64_t dim, int64_t start, int64_t end, int64_t step);
-void atg_slice_copy(tensor *, tensor self, int64_t dim, int64_t start_v, int start_null, int64_t end_v, int end_null, int64_t step);
-void atg_slice_copy_tensor_out(tensor *, tensor out, tensor self, int64_t dim, int64_t start_v, int start_null, int64_t end_v, int end_null, int64_t step);
-void atg_slice_scatter(tensor *, tensor self, tensor src, int64_t dim, int64_t start_v, int start_null, int64_t end_v, int end_null, int64_t step);
-void atg_slice_scatter_out(tensor *, tensor out, tensor self, tensor src, int64_t dim, int64_t start_v, int start_null, int64_t end_v, int end_null, int64_t step);
-void atg_slogdet(tensor *, tensor self);
-void atg_slogdet_out(tensor *, tensor sign, tensor logabsdet, tensor self);
-void atg_slow_conv3d(tensor *, tensor self, tensor weight, int64_t *kernel_size_data, int kernel_size_len, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len);
-void atg_slow_conv3d_out(tensor *, tensor out, tensor self, tensor weight, int64_t *kernel_size_data, int kernel_size_len, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len);
-void atg_slow_conv_dilated2d(tensor *, tensor self, tensor weight, int64_t *kernel_size_data, int kernel_size_len, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len);
-void atg_slow_conv_dilated2d_out(tensor *, tensor out, tensor self, tensor weight, int64_t *kernel_size_data, int kernel_size_len, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len);
-void atg_slow_conv_dilated3d(tensor *, tensor self, tensor weight, int64_t *kernel_size_data, int kernel_size_len, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len);
-void atg_slow_conv_dilated3d_out(tensor *, tensor out, tensor self, tensor weight, int64_t *kernel_size_data, int kernel_size_len, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len);
-void atg_slow_conv_transpose2d(tensor *, tensor self, tensor weight, int64_t *kernel_size_data, int kernel_size_len, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t *dilation_data, int dilation_len);
-void atg_slow_conv_transpose2d_out(tensor *, tensor out, tensor self, tensor weight, int64_t *kernel_size_data, int kernel_size_len, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t *dilation_data, int dilation_len);
-void atg_slow_conv_transpose3d(tensor *, tensor self, tensor weight, int64_t *kernel_size_data, int kernel_size_len, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t *dilation_data, int dilation_len);
-void atg_slow_conv_transpose3d_out(tensor *, tensor out, tensor self, tensor weight, int64_t *kernel_size_data, int kernel_size_len, tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t *dilation_data, int dilation_len);
-void atg_smm(tensor *, tensor self, tensor mat2);
-void atg_smooth_l1_loss(tensor *, tensor self, tensor target, int64_t reduction, double beta);
-void atg_smooth_l1_loss_backward(tensor *, tensor grad_output, tensor self, tensor target, int64_t reduction, double beta);
-void atg_smooth_l1_loss_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, tensor self, tensor target, int64_t reduction, double beta);
-void atg_smooth_l1_loss_out(tensor *, tensor out, tensor self, tensor target, int64_t reduction, double beta);
-void atg_soft_margin_loss(tensor *, tensor self, tensor target, int64_t reduction);
-void atg_soft_margin_loss_backward(tensor *, tensor grad_output, tensor self, tensor target, int64_t reduction);
-void atg_soft_margin_loss_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, tensor self, tensor target, int64_t reduction);
-void atg_soft_margin_loss_out(tensor *, tensor out, tensor self, tensor target, int64_t reduction);
-void atg_softmax(tensor *, tensor self, int64_t dim, int dtype);
-void atg_softmax_int_out(tensor *, tensor out, tensor self, int64_t dim, int dtype);
-void atg_softplus(tensor *, tensor self);
-void atg_softplus_backward(tensor *, tensor grad_output, tensor self, scalar beta, scalar threshold);
-void atg_softplus_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, tensor self, scalar beta, scalar threshold);
-void atg_softplus_out(tensor *, tensor out, tensor self);
-void atg_softshrink(tensor *, tensor self);
-void atg_softshrink_backward(tensor *, tensor grad_output, tensor self, scalar lambd);
-void atg_softshrink_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, tensor self, scalar lambd);
-void atg_softshrink_out(tensor *, tensor out, tensor self);
-void atg_sort(tensor *, tensor self, int64_t dim, int descending);
-void atg_sort_stable(tensor *, tensor self, int stable, int64_t dim, int descending);
-void atg_sort_values(tensor *, tensor values, tensor indices, tensor self, int64_t dim, int descending);
-void atg_sort_values_stable(tensor *, tensor values, tensor indices, tensor self, int stable, int64_t dim, int descending);
-void atg_sparse_bsc_tensor(tensor *, tensor ccol_indices, tensor row_indices, tensor values, int options_kind, int options_device);
-void atg_sparse_bsc_tensor_ccol_row_value_size(tensor *, tensor ccol_indices, tensor row_indices, tensor values, int64_t *size_data, int size_len, int options_kind, int options_device);
-void atg_sparse_bsr_tensor(tensor *, tensor crow_indices, tensor col_indices, tensor values, int options_kind, int options_device);
-void atg_sparse_bsr_tensor_crow_col_value_size(tensor *, tensor crow_indices, tensor col_indices, tensor values, int64_t *size_data, int size_len, int options_kind, int options_device);
-void atg_sparse_compressed_tensor(tensor *, tensor compressed_indices, tensor plain_indices, tensor values, int options_kind, int options_device);
-void atg_sparse_compressed_tensor_comp_plain_value_size(tensor *, tensor compressed_indices, tensor plain_indices, tensor values, int64_t *size_data, int size_len, int options_kind, int options_device);
-void atg_sparse_coo_tensor(tensor *, int64_t *size_data, int size_len, int options_kind, int options_device);
-void atg_sparse_coo_tensor_indices(tensor *, tensor indices, tensor values, int options_kind, int options_device, int is_coalesced);
-void atg_sparse_coo_tensor_indices_size(tensor *, tensor indices, tensor values, int64_t *size_data, int size_len, int options_kind, int options_device, int is_coalesced);
-void atg_sparse_coo_tensor_size_out(tensor *, tensor out, int64_t *size_data, int size_len);
-void atg_sparse_csc_tensor(tensor *, tensor ccol_indices, tensor row_indices, tensor values, int options_kind, int options_device);
-void atg_sparse_csc_tensor_ccol_row_value_size(tensor *, tensor ccol_indices, tensor row_indices, tensor values, int64_t *size_data, int size_len, int options_kind, int options_device);
-void atg_sparse_csr_tensor(tensor *, tensor crow_indices, tensor col_indices, tensor values, int options_kind, int options_device);
-void atg_sparse_csr_tensor_crow_col_value_size(tensor *, tensor crow_indices, tensor col_indices, tensor values, int64_t *size_data, int size_len, int options_kind, int options_device);
-int64_t atg_sparse_dim(tensor self);
-void atg_sparse_mask(tensor *, tensor self, tensor mask);
-void atg_sparse_mask_out(tensor *, tensor out, tensor self, tensor mask);
-void atg_sparse_resize(tensor *, tensor self, int64_t *size_data, int size_len, int64_t sparse_dim, int64_t dense_dim);
-void atg_sparse_resize_(tensor *, tensor self, int64_t *size_data, int size_len, int64_t sparse_dim, int64_t dense_dim);
-void atg_sparse_resize_and_clear(tensor *, tensor self, int64_t *size_data, int size_len, int64_t sparse_dim, int64_t dense_dim);
-void atg_sparse_resize_and_clear_(tensor *, tensor self, int64_t *size_data, int size_len, int64_t sparse_dim, int64_t dense_dim);
-void atg_sparse_resize_and_clear_out(tensor *, tensor out, tensor self, int64_t *size_data, int size_len, int64_t sparse_dim, int64_t dense_dim);
-void atg_sparse_resize_out(tensor *, tensor out, tensor self, int64_t *size_data, int size_len, int64_t sparse_dim, int64_t dense_dim);
-void atg_sparse_sampled_addmm(tensor *, tensor self, tensor mat1, tensor mat2);
-void atg_sparse_sampled_addmm_out(tensor *, tensor out, tensor self, tensor mat1, tensor mat2);
-void atg_special_airy_ai(tensor *, tensor x);
-void atg_special_airy_ai_out(tensor *, tensor out, tensor x);
-void atg_special_bessel_j0(tensor *, tensor self);
-void atg_special_bessel_j0_out(tensor *, tensor out, tensor self);
-void atg_special_bessel_j1(tensor *, tensor self);
-void atg_special_bessel_j1_out(tensor *, tensor out, tensor self);
-void atg_special_bessel_y0(tensor *, tensor self);
-void atg_special_bessel_y0_out(tensor *, tensor out, tensor self);
-void atg_special_bessel_y1(tensor *, tensor self);
-void atg_special_bessel_y1_out(tensor *, tensor out, tensor self);
-void atg_special_chebyshev_polynomial_t(tensor *, tensor x, tensor n);
-void atg_special_chebyshev_polynomial_t_n_scalar(tensor *, tensor x, scalar n);
-void atg_special_chebyshev_polynomial_t_n_scalar_out(tensor *, tensor out, tensor x, scalar n);
-void atg_special_chebyshev_polynomial_t_out(tensor *, tensor out, tensor x, tensor n);
-void atg_special_chebyshev_polynomial_t_x_scalar(tensor *, scalar x, tensor n);
-void atg_special_chebyshev_polynomial_t_x_scalar_out(tensor *, tensor out, scalar x, tensor n);
-void atg_special_chebyshev_polynomial_u(tensor *, tensor x, tensor n);
-void atg_special_chebyshev_polynomial_u_n_scalar(tensor *, tensor x, scalar n);
-void atg_special_chebyshev_polynomial_u_n_scalar_out(tensor *, tensor out, tensor x, scalar n);
-void atg_special_chebyshev_polynomial_u_out(tensor *, tensor out, tensor x, tensor n);
-void atg_special_chebyshev_polynomial_u_x_scalar(tensor *, scalar x, tensor n);
-void atg_special_chebyshev_polynomial_u_x_scalar_out(tensor *, tensor out, scalar x, tensor n);
-void atg_special_chebyshev_polynomial_v(tensor *, tensor x, tensor n);
-void atg_special_chebyshev_polynomial_v_n_scalar(tensor *, tensor x, scalar n);
-void atg_special_chebyshev_polynomial_v_n_scalar_out(tensor *, tensor out, tensor x, scalar n);
-void atg_special_chebyshev_polynomial_v_out(tensor *, tensor out, tensor x, tensor n);
-void atg_special_chebyshev_polynomial_v_x_scalar(tensor *, scalar x, tensor n);
-void atg_special_chebyshev_polynomial_v_x_scalar_out(tensor *, tensor out, scalar x, tensor n);
-void atg_special_chebyshev_polynomial_w(tensor *, tensor x, tensor n);
-void atg_special_chebyshev_polynomial_w_n_scalar(tensor *, tensor x, scalar n);
-void atg_special_chebyshev_polynomial_w_n_scalar_out(tensor *, tensor out, tensor x, scalar n);
-void atg_special_chebyshev_polynomial_w_out(tensor *, tensor out, tensor x, tensor n);
-void atg_special_chebyshev_polynomial_w_x_scalar(tensor *, scalar x, tensor n);
-void atg_special_chebyshev_polynomial_w_x_scalar_out(tensor *, tensor out, scalar x, tensor n);
-void atg_special_digamma(tensor *, tensor self);
-void atg_special_digamma_out(tensor *, tensor out, tensor self);
-void atg_special_entr(tensor *, tensor self);
-void atg_special_entr_out(tensor *, tensor out, tensor self);
-void atg_special_erf(tensor *, tensor self);
-void atg_special_erf_out(tensor *, tensor out, tensor self);
-void atg_special_erfc(tensor *, tensor self);
-void atg_special_erfc_out(tensor *, tensor out, tensor self);
-void atg_special_erfcx(tensor *, tensor self);
-void atg_special_erfcx_out(tensor *, tensor out, tensor self);
-void atg_special_erfinv(tensor *, tensor self);
-void atg_special_erfinv_out(tensor *, tensor out, tensor self);
-void atg_special_exp2(tensor *, tensor self);
-void atg_special_exp2_out(tensor *, tensor out, tensor self);
-void atg_special_expit(tensor *, tensor self);
-void atg_special_expit_out(tensor *, tensor out, tensor self);
-void atg_special_expm1(tensor *, tensor self);
-void atg_special_expm1_out(tensor *, tensor out, tensor self);
-void atg_special_gammainc(tensor *, tensor self, tensor other);
-void atg_special_gammainc_out(tensor *, tensor out, tensor self, tensor other);
-void atg_special_gammaincc(tensor *, tensor self, tensor other);
-void atg_special_gammaincc_out(tensor *, tensor out, tensor self, tensor other);
-void atg_special_gammaln(tensor *, tensor self);
-void atg_special_gammaln_out(tensor *, tensor out, tensor self);
-void atg_special_hermite_polynomial_h(tensor *, tensor x, tensor n);
-void atg_special_hermite_polynomial_h_n_scalar(tensor *, tensor x, scalar n);
-void atg_special_hermite_polynomial_h_n_scalar_out(tensor *, tensor out, tensor x, scalar n);
-void atg_special_hermite_polynomial_h_out(tensor *, tensor out, tensor x, tensor n);
-void atg_special_hermite_polynomial_h_x_scalar(tensor *, scalar x, tensor n);
-void atg_special_hermite_polynomial_h_x_scalar_out(tensor *, tensor out, scalar x, tensor n);
-void atg_special_hermite_polynomial_he(tensor *, tensor x, tensor n);
-void atg_special_hermite_polynomial_he_n_scalar(tensor *, tensor x, scalar n);
-void atg_special_hermite_polynomial_he_n_scalar_out(tensor *, tensor out, tensor x, scalar n);
-void atg_special_hermite_polynomial_he_out(tensor *, tensor out, tensor x, tensor n);
-void atg_special_hermite_polynomial_he_x_scalar(tensor *, scalar x, tensor n);
-void atg_special_hermite_polynomial_he_x_scalar_out(tensor *, tensor out, scalar x, tensor n);
-void atg_special_i0(tensor *, tensor self);
-void atg_special_i0_out(tensor *, tensor out, tensor self);
-void atg_special_i0e(tensor *, tensor self);
-void atg_special_i0e_out(tensor *, tensor out, tensor self);
-void atg_special_i1(tensor *, tensor self);
-void atg_special_i1_out(tensor *, tensor out, tensor self);
-void atg_special_i1e(tensor *, tensor self);
-void atg_special_i1e_out(tensor *, tensor out, tensor self);
-void atg_special_laguerre_polynomial_l(tensor *, tensor x, tensor n);
-void atg_special_laguerre_polynomial_l_n_scalar(tensor *, tensor x, scalar n);
-void atg_special_laguerre_polynomial_l_n_scalar_out(tensor *, tensor out, tensor x, scalar n);
-void atg_special_laguerre_polynomial_l_out(tensor *, tensor out, tensor x, tensor n);
-void atg_special_laguerre_polynomial_l_x_scalar(tensor *, scalar x, tensor n);
-void atg_special_laguerre_polynomial_l_x_scalar_out(tensor *, tensor out, scalar x, tensor n);
-void atg_special_legendre_polynomial_p(tensor *, tensor x, tensor n);
-void atg_special_legendre_polynomial_p_n_scalar(tensor *, tensor x, scalar n);
-void atg_special_legendre_polynomial_p_n_scalar_out(tensor *, tensor out, tensor x, scalar n);
-void atg_special_legendre_polynomial_p_out(tensor *, tensor out, tensor x, tensor n);
-void atg_special_legendre_polynomial_p_x_scalar(tensor *, scalar x, tensor n);
-void atg_special_legendre_polynomial_p_x_scalar_out(tensor *, tensor out, scalar x, tensor n);
-void atg_special_log1p(tensor *, tensor self);
-void atg_special_log1p_out(tensor *, tensor out, tensor self);
-void atg_special_log_ndtr(tensor *, tensor self);
-void atg_special_log_ndtr_out(tensor *, tensor out, tensor self);
-void atg_special_log_softmax(tensor *, tensor self, int64_t dim, int dtype);
-void atg_special_logit(tensor *, tensor self, double eps_v, int eps_null);
-void atg_special_logit_out(tensor *, tensor out, tensor self, double eps_v, int eps_null);
-void atg_special_logsumexp(tensor *, tensor self, int64_t *dim_data, int dim_len, int keepdim);
-void atg_special_logsumexp_out(tensor *, tensor out, tensor self, int64_t *dim_data, int dim_len, int keepdim);
-void atg_special_modified_bessel_i0(tensor *, tensor self);
-void atg_special_modified_bessel_i0_out(tensor *, tensor out, tensor self);
-void atg_special_modified_bessel_i1(tensor *, tensor self);
-void atg_special_modified_bessel_i1_out(tensor *, tensor out, tensor self);
-void atg_special_modified_bessel_k0(tensor *, tensor self);
-void atg_special_modified_bessel_k0_out(tensor *, tensor out, tensor self);
-void atg_special_modified_bessel_k1(tensor *, tensor self);
-void atg_special_modified_bessel_k1_out(tensor *, tensor out, tensor self);
-void atg_special_multigammaln(tensor *, tensor self, int64_t p);
-void atg_special_multigammaln_out(tensor *, tensor out, tensor self, int64_t p);
-void atg_special_ndtr(tensor *, tensor self);
-void atg_special_ndtr_out(tensor *, tensor out, tensor self);
-void atg_special_ndtri(tensor *, tensor self);
-void atg_special_ndtri_out(tensor *, tensor out, tensor self);
-void atg_special_polygamma(tensor *, int64_t n, tensor self);
-void atg_special_polygamma_out(tensor *, tensor out, int64_t n, tensor self);
-void atg_special_psi(tensor *, tensor self);
-void atg_special_psi_out(tensor *, tensor out, tensor self);
-void atg_special_round(tensor *, tensor self, int64_t decimals);
-void atg_special_round_out(tensor *, tensor out, tensor self, int64_t decimals);
-void atg_special_scaled_modified_bessel_k0(tensor *, tensor x);
-void atg_special_scaled_modified_bessel_k0_out(tensor *, tensor out, tensor x);
-void atg_special_scaled_modified_bessel_k1(tensor *, tensor x);
-void atg_special_scaled_modified_bessel_k1_out(tensor *, tensor out, tensor x);
-void atg_special_shifted_chebyshev_polynomial_t(tensor *, tensor x, tensor n);
-void atg_special_shifted_chebyshev_polynomial_t_n_scalar(tensor *, tensor x, scalar n);
-void atg_special_shifted_chebyshev_polynomial_t_n_scalar_out(tensor *, tensor out, tensor x, scalar n);
-void atg_special_shifted_chebyshev_polynomial_t_out(tensor *, tensor out, tensor x, tensor n);
-void atg_special_shifted_chebyshev_polynomial_t_x_scalar(tensor *, scalar x, tensor n);
-void atg_special_shifted_chebyshev_polynomial_t_x_scalar_out(tensor *, tensor out, scalar x, tensor n);
-void atg_special_shifted_chebyshev_polynomial_u(tensor *, tensor x, tensor n);
-void atg_special_shifted_chebyshev_polynomial_u_n_scalar(tensor *, tensor x, scalar n);
-void atg_special_shifted_chebyshev_polynomial_u_n_scalar_out(tensor *, tensor out, tensor x, scalar n);
-void atg_special_shifted_chebyshev_polynomial_u_out(tensor *, tensor out, tensor x, tensor n);
-void atg_special_shifted_chebyshev_polynomial_u_x_scalar(tensor *, scalar x, tensor n);
-void atg_special_shifted_chebyshev_polynomial_u_x_scalar_out(tensor *, tensor out, scalar x, tensor n);
-void atg_special_shifted_chebyshev_polynomial_v(tensor *, tensor x, tensor n);
-void atg_special_shifted_chebyshev_polynomial_v_n_scalar(tensor *, tensor x, scalar n);
-void atg_special_shifted_chebyshev_polynomial_v_n_scalar_out(tensor *, tensor out, tensor x, scalar n);
-void atg_special_shifted_chebyshev_polynomial_v_out(tensor *, tensor out, tensor x, tensor n);
-void atg_special_shifted_chebyshev_polynomial_v_x_scalar(tensor *, scalar x, tensor n);
-void atg_special_shifted_chebyshev_polynomial_v_x_scalar_out(tensor *, tensor out, scalar x, tensor n);
-void atg_special_shifted_chebyshev_polynomial_w(tensor *, tensor x, tensor n);
-void atg_special_shifted_chebyshev_polynomial_w_n_scalar(tensor *, tensor x, scalar n);
-void atg_special_shifted_chebyshev_polynomial_w_n_scalar_out(tensor *, tensor out, tensor x, scalar n);
-void atg_special_shifted_chebyshev_polynomial_w_out(tensor *, tensor out, tensor x, tensor n);
-void atg_special_shifted_chebyshev_polynomial_w_x_scalar(tensor *, scalar x, tensor n);
-void atg_special_shifted_chebyshev_polynomial_w_x_scalar_out(tensor *, tensor out, scalar x, tensor n);
-void atg_special_sinc(tensor *, tensor self);
-void atg_special_sinc_out(tensor *, tensor out, tensor self);
-void atg_special_softmax(tensor *, tensor self, int64_t dim, int dtype);
-void atg_special_spherical_bessel_j0(tensor *, tensor x);
-void atg_special_spherical_bessel_j0_out(tensor *, tensor out, tensor x);
-void atg_special_xlog1py(tensor *, tensor self, tensor other);
-void atg_special_xlog1py_other_scalar(tensor *, tensor self, scalar other);
-void atg_special_xlog1py_other_scalar_out(tensor *, tensor out, tensor self, scalar other);
-void atg_special_xlog1py_out(tensor *, tensor out, tensor self, tensor other);
-void atg_special_xlog1py_self_scalar(tensor *, scalar self, tensor other);
-void atg_special_xlog1py_self_scalar_out(tensor *, tensor out, scalar self, tensor other);
-void atg_special_xlogy(tensor *, tensor self, tensor other);
-void atg_special_xlogy_other_scalar(tensor *, tensor self, scalar other);
-void atg_special_xlogy_other_scalar_out(tensor *, tensor out, tensor self, scalar other);
-void atg_special_xlogy_out(tensor *, tensor out, tensor self, tensor other);
-void atg_special_xlogy_self_scalar(tensor *, scalar self, tensor other);
-void atg_special_xlogy_self_scalar_out(tensor *, tensor out, scalar self, tensor other);
-void atg_special_zeta(tensor *, tensor self, tensor other);
-void atg_special_zeta_other_scalar(tensor *, tensor self, scalar other);
-void atg_special_zeta_other_scalar_out(tensor *, tensor out, tensor self, scalar other);
-void atg_special_zeta_out(tensor *, tensor out, tensor self, tensor other);
-void atg_special_zeta_self_scalar(tensor *, scalar self, tensor other);
-void atg_special_zeta_self_scalar_out(tensor *, tensor out, scalar self, tensor other);
-tensor *atg_split(tensor self, int64_t split_size, int64_t dim);
-tensor *atg_split_copy(tensor self, int64_t split_size, int64_t dim);
-void atg_split_copy_tensor_out(tensor *out_data, int out_len, tensor self, int64_t split_size, int64_t dim);
-tensor *atg_split_sizes(tensor self, int64_t *split_size_data, int split_size_len, int64_t dim);
-tensor *atg_split_with_sizes(tensor self, int64_t *split_sizes_data, int split_sizes_len, int64_t dim);
-tensor *atg_split_with_sizes_copy(tensor self, int64_t *split_sizes_data, int split_sizes_len, int64_t dim);
-void atg_split_with_sizes_copy_out(tensor *out_data, int out_len, tensor self, int64_t *split_sizes_data, int split_sizes_len, int64_t dim);
-void atg_sqrt(tensor *, tensor self);
-void atg_sqrt_(tensor *, tensor self);
-void atg_sqrt_out(tensor *, tensor out, tensor self);
-void atg_square(tensor *, tensor self);
-void atg_square_(tensor *, tensor self);
-void atg_square_out(tensor *, tensor out, tensor self);
-void atg_squeeze(tensor *, tensor self);
-void atg_squeeze_(tensor *, tensor self);
-void atg_squeeze_copy(tensor *, tensor self);
-void atg_squeeze_copy_dim(tensor *, tensor self, int64_t dim);
-void atg_squeeze_copy_dim_out(tensor *, tensor out, tensor self, int64_t dim);
-void atg_squeeze_copy_dims(tensor *, tensor self, int64_t *dim_data, int dim_len);
-void atg_squeeze_copy_dims_out(tensor *, tensor out, tensor self, int64_t *dim_data, int dim_len);
-void atg_squeeze_copy_out(tensor *, tensor out, tensor self);
-void atg_squeeze_dim(tensor *, tensor self, int64_t dim);
-void atg_squeeze_dim_(tensor *, tensor self, int64_t dim);
-void atg_squeeze_dims(tensor *, tensor self, int64_t *dim_data, int dim_len);
-void atg_squeeze_dims_(tensor *, tensor self, int64_t *dim_data, int dim_len);
-void atg_sspaddmm(tensor *, tensor self, tensor mat1, tensor mat2);
-void atg_sspaddmm_out(tensor *, tensor out, tensor self, tensor mat1, tensor mat2);
-void atg_stack(tensor *, tensor *tensors_data, int tensors_len, int64_t dim);
-void atg_stack_out(tensor *, tensor out, tensor *tensors_data, int tensors_len, int64_t dim);
-void atg_std(tensor *, tensor self, int unbiased);
-void atg_std_correction(tensor *, tensor self, int64_t *dim_data, int dim_len, scalar correction, int keepdim);
-void atg_std_correction_out(tensor *, tensor out, tensor self, int64_t *dim_data, int dim_len, scalar correction, int keepdim);
-void atg_std_dim(tensor *, tensor self, int64_t *dim_data, int dim_len, int unbiased, int keepdim);
-void atg_std_mean(tensor *, tensor self, int unbiased);
-void atg_std_mean_correction(tensor *, tensor self, int64_t *dim_data, int dim_len, scalar correction, int keepdim);
-void atg_std_mean_correction_out(tensor *, tensor out0, tensor out1, tensor self, int64_t *dim_data, int dim_len, scalar correction, int keepdim);
-void atg_std_mean_dim(tensor *, tensor self, int64_t *dim_data, int dim_len, int unbiased, int keepdim);
-void atg_std_out(tensor *, tensor out, tensor self, int64_t *dim_data, int dim_len, int unbiased, int keepdim);
-void atg_stft(tensor *, tensor self, int64_t n_fft, int64_t hop_length_v, int hop_length_null, int64_t win_length_v, int win_length_null, tensor window, int normalized, int onesided, int return_complex);
-void atg_stft_center(tensor *, tensor self, int64_t n_fft, int64_t hop_length_v, int hop_length_null, int64_t win_length_v, int win_length_null, tensor window, int center, char * pad_mode, int normalized, int onesided, int return_complex);
-int64_t atg_stride(tensor self, int64_t dim);
-void atg_sub(tensor *, tensor self, tensor other);
-void atg_sub_(tensor *, tensor self, tensor other);
-void atg_sub_out(tensor *, tensor out, tensor self, tensor other);
-void atg_sub_scalar(tensor *, tensor self, scalar other);
-void atg_sub_scalar_(tensor *, tensor self, scalar other);
-void atg_sub_scalar_out(tensor *, tensor out, tensor self, scalar other);
-void atg_subtract(tensor *, tensor self, tensor other);
-void atg_subtract_(tensor *, tensor self, tensor other);
-void atg_subtract_out(tensor *, tensor out, tensor self, tensor other);
-void atg_subtract_scalar(tensor *, tensor self, scalar other);
-void atg_subtract_scalar_(tensor *, tensor self, scalar other);
-void atg_sum(tensor *, tensor self, int dtype);
-void atg_sum_dim_intlist(tensor *, tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype);
-void atg_sum_intlist_out(tensor *, tensor out, tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype);
-void atg_sum_out(tensor *, tensor out, tensor self, int dtype);
-void atg_sum_to_size(tensor *, tensor self, int64_t *size_data, int size_len);
-void atg_svd(tensor *, tensor self, int some, int compute_uv);
-void atg_svd_u(tensor *, tensor U, tensor S, tensor V, tensor self, int some, int compute_uv);
-void atg_swapaxes(tensor *, tensor self, int64_t axis0, int64_t axis1);
-void atg_swapaxes_(tensor *, tensor self, int64_t axis0, int64_t axis1);
-void atg_swapdims(tensor *, tensor self, int64_t dim0, int64_t dim1);
-void atg_swapdims_(tensor *, tensor self, int64_t dim0, int64_t dim1);
-void atg_t(tensor *, tensor self);
-void atg_t_(tensor *, tensor self);
-void atg_t_copy(tensor *, tensor self);
-void atg_t_copy_out(tensor *, tensor out, tensor self);
-void atg_take(tensor *, tensor self, tensor index);
-void atg_take_along_dim(tensor *, tensor self, tensor indices, int64_t dim_v, int dim_null);
-void atg_take_along_dim_out(tensor *, tensor out, tensor self, tensor indices, int64_t dim_v, int dim_null);
-void atg_take_out(tensor *, tensor out, tensor self, tensor index);
-void atg_tan(tensor *, tensor self);
-void atg_tan_(tensor *, tensor self);
-void atg_tan_out(tensor *, tensor out, tensor self);
-void atg_tanh(tensor *, tensor self);
-void atg_tanh_(tensor *, tensor self);
-void atg_tanh_backward(tensor *, tensor grad_output, tensor output);
-void atg_tanh_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, tensor output);
-void atg_tanh_out(tensor *, tensor out, tensor self);
-tensor *atg_tensor_split(tensor self, int64_t sections, int64_t dim);
-tensor *atg_tensor_split_indices(tensor self, int64_t *indices_data, int indices_len, int64_t dim);
-tensor *atg_tensor_split_tensor_indices_or_sections(tensor self, tensor tensor_indices_or_sections, int64_t dim);
-void atg_tensordot(tensor *, tensor self, tensor other, int64_t *dims_self_data, int dims_self_len, int64_t *dims_other_data, int dims_other_len);
-void atg_tensordot_out(tensor *, tensor out, tensor self, tensor other, int64_t *dims_self_data, int dims_self_len, int64_t *dims_other_data, int dims_other_len);
-void atg_threshold(tensor *, tensor self, scalar threshold, scalar value);
-void atg_threshold_(tensor *, tensor self, scalar threshold, scalar value);
-void atg_threshold_backward(tensor *, tensor grad_output, tensor self, scalar threshold);
-void atg_threshold_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, tensor self, scalar threshold);
-void atg_threshold_out(tensor *, tensor out, tensor self, scalar threshold, scalar value);
-void atg_tile(tensor *, tensor self, int64_t *dims_data, int dims_len);
-void atg_to(tensor *, tensor self, int device);
-void atg_to_dense(tensor *, tensor self, int dtype, int masked_grad);
-void atg_to_dense_backward(tensor *, tensor grad, tensor input, int masked_grad);
-void atg_to_device(tensor *, tensor self, int device, int dtype, int non_blocking, int copy);
-void atg_to_dtype(tensor *, tensor self, int dtype, int non_blocking, int copy);
-void atg_to_dtype_layout(tensor *, tensor self, int options_kind, int options_device, int non_blocking, int copy);
-void atg_to_mkldnn(tensor *, tensor self, int dtype);
-void atg_to_mkldnn_backward(tensor *, tensor grad, tensor input);
-void atg_to_mkldnn_out(tensor *, tensor out, tensor self, int dtype);
-void atg_to_other(tensor *, tensor self, tensor other, int non_blocking, int copy);
-void atg_to_padded_tensor(tensor *, tensor self, double padding, int64_t *output_size_data, int output_size_len);
-void atg_to_padded_tensor_out(tensor *, tensor out, tensor self, double padding, int64_t *output_size_data, int output_size_len);
-void atg_topk(tensor *, tensor self, int64_t k, int64_t dim, int largest, int sorted);
-void atg_topk_values(tensor *, tensor values, tensor indices, tensor self, int64_t k, int64_t dim, int largest, int sorted);
-void atg_totype(tensor *, tensor self, int scalar_type);
-void atg_trace(tensor *, tensor self);
-void atg_trace_backward(tensor *, tensor grad, int64_t *sizes_data, int sizes_len);
-void atg_trace_out(tensor *, tensor out, tensor self);
-void atg_transpose(tensor *, tensor self, int64_t dim0, int64_t dim1);
-void atg_transpose_(tensor *, tensor self, int64_t dim0, int64_t dim1);
-void atg_transpose_copy(tensor *, tensor self, int64_t dim0, int64_t dim1);
-void atg_transpose_copy_int_out(tensor *, tensor out, tensor self, int64_t dim0, int64_t dim1);
-void atg_trapezoid(tensor *, tensor y, int64_t dim);
-void atg_trapezoid_x(tensor *, tensor y, tensor x, int64_t dim);
-void atg_trapz(tensor *, tensor y, tensor x, int64_t dim);
-void atg_trapz_dx(tensor *, tensor y, double dx, int64_t dim);
-void atg_triangular_solve(tensor *, tensor self, tensor A, int upper, int transpose, int unitriangular);
-void atg_triangular_solve_x(tensor *, tensor X, tensor M, tensor self, tensor A, int upper, int transpose, int unitriangular);
-void atg_tril(tensor *, tensor self, int64_t diagonal);
-void atg_tril_(tensor *, tensor self, int64_t diagonal);
-void atg_tril_indices(tensor *, int64_t row, int64_t col, int64_t offset, int options_kind, int options_device);
-void atg_tril_indices_out(tensor *, tensor out, int64_t row, int64_t col, int64_t offset);
-void atg_tril_out(tensor *, tensor out, tensor self, int64_t diagonal);
-void atg_triplet_margin_loss(tensor *, tensor anchor, tensor positive, tensor negative, double margin, double p, double eps, int swap, int64_t reduction);
-void atg_triu(tensor *, tensor self, int64_t diagonal);
-void atg_triu_(tensor *, tensor self, int64_t diagonal);
-void atg_triu_indices(tensor *, int64_t row, int64_t col, int64_t offset, int options_kind, int options_device);
-void atg_triu_indices_out(tensor *, tensor out, int64_t row, int64_t col, int64_t offset);
-void atg_triu_out(tensor *, tensor out, tensor self, int64_t diagonal);
-void atg_true_divide(tensor *, tensor self, tensor other);
-void atg_true_divide_(tensor *, tensor self, tensor other);
-void atg_true_divide_out(tensor *, tensor out, tensor self, tensor other);
-void atg_true_divide_scalar(tensor *, tensor self, scalar other);
-void atg_true_divide_scalar_(tensor *, tensor self, scalar other);
-void atg_trunc(tensor *, tensor self);
-void atg_trunc_(tensor *, tensor self);
-void atg_trunc_out(tensor *, tensor out, tensor self);
-void atg_type_as(tensor *, tensor self, tensor other);
-tensor *atg_unbind(tensor self, int64_t dim);
-tensor *atg_unbind_copy(tensor self, int64_t dim);
-void atg_unbind_copy_int_out(tensor *out_data, int out_len, tensor self, int64_t dim);
-void atg_unflatten(tensor *, tensor self, int64_t dim, int64_t *sizes_data, int sizes_len);
-tensor *atg_unflatten_dense_tensors(tensor flat, tensor *tensors_data, int tensors_len);
-void atg_unfold(tensor *, tensor self, int64_t dimension, int64_t size, int64_t step);
-void atg_unfold_backward(tensor *, tensor grad_in, int64_t *input_sizes_data, int input_sizes_len, int64_t dim, int64_t size, int64_t step);
-void atg_unfold_backward_out(tensor *, tensor out, tensor grad_in, int64_t *input_sizes_data, int input_sizes_len, int64_t dim, int64_t size, int64_t step);
-void atg_unfold_copy(tensor *, tensor self, int64_t dimension, int64_t size, int64_t step);
-void atg_unfold_copy_out(tensor *, tensor out, tensor self, int64_t dimension, int64_t size, int64_t step);
-void atg_uniform(tensor *, tensor self, double from, double to);
-void atg_uniform_(tensor *, tensor self, double from, double to);
-void atg_uniform_out(tensor *, tensor out, tensor self, double from, double to);
-void atg_unique_consecutive(tensor *, tensor self, int return_inverse, int return_counts, int64_t dim_v, int dim_null);
-void atg_unique_consecutive_out(tensor *, tensor out0, tensor out1, tensor out2, tensor self, int return_inverse, int return_counts, int64_t dim_v, int dim_null);
-void atg_unique_dim(tensor *, tensor self, int64_t dim, int sorted, int return_inverse, int return_counts);
-void atg_unique_dim_consecutive(tensor *, tensor self, int64_t dim, int return_inverse, int return_counts);
-void atg_unique_dim_consecutive_out(tensor *, tensor out0, tensor out1, tensor out2, tensor self, int64_t dim, int return_inverse, int return_counts);
-void atg_unique_dim_out(tensor *, tensor out0, tensor out1, tensor out2, tensor self, int64_t dim, int sorted, int return_inverse, int return_counts);
-tensor *atg_unsafe_chunk(tensor self, int64_t chunks, int64_t dim);
-tensor *atg_unsafe_split(tensor self, int64_t split_size, int64_t dim);
-void atg_unsafe_split_tensor_out(tensor *out_data, int out_len, tensor self, int64_t split_size, int64_t dim);
-tensor *atg_unsafe_split_with_sizes(tensor self, int64_t *split_sizes_data, int split_sizes_len, int64_t dim);
-void atg_unsafe_split_with_sizes_out(tensor *out_data, int out_len, tensor self, int64_t *split_sizes_data, int split_sizes_len, int64_t dim);
-void atg_unsqueeze(tensor *, tensor self, int64_t dim);
-void atg_unsqueeze_(tensor *, tensor self, int64_t dim);
-void atg_unsqueeze_copy(tensor *, tensor self, int64_t dim);
-void atg_unsqueeze_copy_out(tensor *, tensor out, tensor self, int64_t dim);
-void atg_upsample_bicubic2d(tensor *, tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
-void atg_upsample_bicubic2d_backward(tensor *, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
-void atg_upsample_bicubic2d_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
-void atg_upsample_bicubic2d_out(tensor *, tensor out, tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
-void atg_upsample_bicubic2d_vec(tensor *, tensor input, int64_t *output_size_data, int output_size_len, int align_corners, double *scale_factors_data, int scale_factors_len);
-void atg_upsample_bilinear2d(tensor *, tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
-void atg_upsample_bilinear2d_backward(tensor *, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
-void atg_upsample_bilinear2d_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
-void atg_upsample_bilinear2d_out(tensor *, tensor out, tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
-void atg_upsample_bilinear2d_vec(tensor *, tensor input, int64_t *output_size_data, int output_size_len, int align_corners, double *scale_factors_data, int scale_factors_len);
-void atg_upsample_linear1d(tensor *, tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_v, int scales_null);
-void atg_upsample_linear1d_backward(tensor *, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_v, int scales_null);
-void atg_upsample_linear1d_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_v, int scales_null);
-void atg_upsample_linear1d_out(tensor *, tensor out, tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_v, int scales_null);
-void atg_upsample_linear1d_vec(tensor *, tensor input, int64_t *output_size_data, int output_size_len, int align_corners, double *scale_factors_data, int scale_factors_len);
-void atg_upsample_nearest1d(tensor *, tensor self, int64_t *output_size_data, int output_size_len, double scales_v, int scales_null);
-void atg_upsample_nearest1d_backward(tensor *, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_v, int scales_null);
-void atg_upsample_nearest1d_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_v, int scales_null);
-void atg_upsample_nearest1d_out(tensor *, tensor out, tensor self, int64_t *output_size_data, int output_size_len, double scales_v, int scales_null);
-void atg_upsample_nearest1d_vec(tensor *, tensor input, int64_t *output_size_data, int output_size_len, double *scale_factors_data, int scale_factors_len);
-void atg_upsample_nearest2d(tensor *, tensor self, int64_t *output_size_data, int output_size_len, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
-void atg_upsample_nearest2d_backward(tensor *, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
-void atg_upsample_nearest2d_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
-void atg_upsample_nearest2d_out(tensor *, tensor out, tensor self, int64_t *output_size_data, int output_size_len, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
-void atg_upsample_nearest2d_vec(tensor *, tensor input, int64_t *output_size_data, int output_size_len, double *scale_factors_data, int scale_factors_len);
-void atg_upsample_nearest3d(tensor *, tensor self, int64_t *output_size_data, int output_size_len, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
-void atg_upsample_nearest3d_backward(tensor *, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
-void atg_upsample_nearest3d_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
-void atg_upsample_nearest3d_out(tensor *, tensor out, tensor self, int64_t *output_size_data, int output_size_len, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
-void atg_upsample_nearest3d_vec(tensor *, tensor input, int64_t *output_size_data, int output_size_len, double *scale_factors_data, int scale_factors_len);
-void atg_upsample_trilinear3d(tensor *, tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
-void atg_upsample_trilinear3d_backward(tensor *, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
-void atg_upsample_trilinear3d_backward_grad_input(tensor *, tensor grad_input, tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
-void atg_upsample_trilinear3d_out(tensor *, tensor out, tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
-void atg_upsample_trilinear3d_vec(tensor *, tensor input, int64_t *output_size_data, int output_size_len, int align_corners, double *scale_factors_data, int scale_factors_len);
-void atg_value_selecting_reduction_backward(tensor *, tensor grad, int64_t dim, tensor indices, int64_t *sizes_data, int sizes_len, int keepdim);
-void atg_values(tensor *, tensor self);
-void atg_values_copy(tensor *, tensor self);
-void atg_values_copy_out(tensor *, tensor out, tensor self);
-void atg_vander(tensor *, tensor x, int64_t n_v, int n_null, int increasing);
-void atg_var(tensor *, tensor self, int unbiased);
-void atg_var_correction(tensor *, tensor self, int64_t *dim_data, int dim_len, scalar correction, int keepdim);
-void atg_var_correction_out(tensor *, tensor out, tensor self, int64_t *dim_data, int dim_len, scalar correction, int keepdim);
-void atg_var_dim(tensor *, tensor self, int64_t *dim_data, int dim_len, int unbiased, int keepdim);
-void atg_var_mean(tensor *, tensor self, int unbiased);
-void atg_var_mean_correction(tensor *, tensor self, int64_t *dim_data, int dim_len, scalar correction, int keepdim);
-void atg_var_mean_correction_out(tensor *, tensor out0, tensor out1, tensor self, int64_t *dim_data, int dim_len, scalar correction, int keepdim);
-void atg_var_mean_dim(tensor *, tensor self, int64_t *dim_data, int dim_len, int unbiased, int keepdim);
-void atg_var_out(tensor *, tensor out, tensor self, int64_t *dim_data, int dim_len, int unbiased, int keepdim);
-void atg_vdot(tensor *, tensor self, tensor other);
-void atg_vdot_out(tensor *, tensor out, tensor self, tensor other);
-void atg_view(tensor *, tensor self, int64_t *size_data, int size_len);
-void atg_view_as(tensor *, tensor self, tensor other);
-void atg_view_as_complex(tensor *, tensor self);
-void atg_view_as_complex_copy(tensor *, tensor self);
-void atg_view_as_complex_copy_out(tensor *, tensor out, tensor self);
-void atg_view_as_real(tensor *, tensor self);
-void atg_view_as_real_copy(tensor *, tensor self);
-void atg_view_as_real_copy_out(tensor *, tensor out, tensor self);
-void atg_view_copy(tensor *, tensor self, int64_t *size_data, int size_len);
-void atg_view_copy_dtype(tensor *, tensor self, int dtype);
-void atg_view_copy_dtype_out(tensor *, tensor out, tensor self, int dtype);
-void atg_view_copy_out(tensor *, tensor out, tensor self, int64_t *size_data, int size_len);
-void atg_view_dtype(tensor *, tensor self, int dtype);
-tensor *atg_vsplit(tensor self, int64_t sections);
-tensor *atg_vsplit_array(tensor self, int64_t *indices_data, int indices_len);
-void atg_vstack(tensor *, tensor *tensors_data, int tensors_len);
-void atg_vstack_out(tensor *, tensor out, tensor *tensors_data, int tensors_len);
-tensor *atg_where(tensor condition);
-void atg_where_scalar(tensor *, tensor condition, scalar self, scalar other);
-void atg_where_scalarother(tensor *, tensor condition, tensor self, scalar other);
-void atg_where_scalarself(tensor *, tensor condition, scalar self, tensor other);
-void atg_where_self(tensor *, tensor condition, tensor self, tensor other);
-void atg_where_self_out(tensor *, tensor out, tensor condition, tensor self, tensor other);
-void atg_xlogy(tensor *, tensor self, tensor other);
-void atg_xlogy_(tensor *, tensor self, tensor other);
-void atg_xlogy_outscalar_other(tensor *, tensor out, tensor self, scalar other);
-void atg_xlogy_outscalar_self(tensor *, tensor out, scalar self, tensor other);
-void atg_xlogy_outtensor(tensor *, tensor out, tensor self, tensor other);
-void atg_xlogy_scalar_other(tensor *, tensor self, scalar other);
-void atg_xlogy_scalar_other_(tensor *, tensor self, scalar other);
-void atg_xlogy_scalar_self(tensor *, scalar self, tensor other);
-void atg_zero(tensor *, tensor self);
-void atg_zero_(tensor *, tensor self);
-void atg_zero_out(tensor *, tensor out, tensor self);
-void atg_zeros(tensor *, int64_t *size_data, int size_len, int options_kind, int options_device);
-void atg_zeros_like(tensor *, tensor self);
-void atg_zeros_like_out(tensor *, tensor out, tensor self);
-void atg_zeros_out(tensor *, tensor out, int64_t *size_data, int size_len);
+raw_tensor atg_isclose(gc_tensor self, gc_tensor other, double rtol, double atol, int equal_nan);
+raw_tensor atg_isfinite(gc_tensor self);
+raw_tensor atg_isin(gc_tensor elements, gc_tensor test_elements, int assume_unique, int invert);
+raw_tensor atg_isin_scalar_tensor(scalar element, gc_tensor test_elements, int assume_unique, int invert);
+raw_tensor atg_isin_scalar_tensor_out(gc_tensor out, scalar element, gc_tensor test_elements, int assume_unique, int invert);
+raw_tensor atg_isin_tensor_scalar(gc_tensor elements, scalar test_element, int assume_unique, int invert);
+raw_tensor atg_isin_tensor_scalar_out(gc_tensor out, gc_tensor elements, scalar test_element, int assume_unique, int invert);
+raw_tensor atg_isin_tensor_tensor_out(gc_tensor out, gc_tensor elements, gc_tensor test_elements, int assume_unique, int invert);
+raw_tensor atg_isinf(gc_tensor self);
+raw_tensor atg_isinf_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_isnan(gc_tensor self);
+raw_tensor atg_isnan_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_isneginf(gc_tensor self);
+raw_tensor atg_isneginf_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_isposinf(gc_tensor self);
+raw_tensor atg_isposinf_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_isreal(gc_tensor self);
+raw_tensor atg_istft(gc_tensor self, int64_t n_fft, int64_t hop_length_v, int hop_length_null, int64_t win_length_v, int win_length_null, gc_tensor window, int center, int normalized, int onesided, int64_t length_v, int length_null, int return_complex);
+raw_tensor atg_kaiser_window(int64_t window_length, int options_kind, int options_device);
+raw_tensor atg_kaiser_window_beta(int64_t window_length, int periodic, double beta, int options_kind, int options_device);
+raw_tensor atg_kaiser_window_beta_out(gc_tensor out, int64_t window_length, int periodic, double beta);
+raw_tensor atg_kaiser_window_out(gc_tensor out, int64_t window_length);
+raw_tensor atg_kaiser_window_periodic(int64_t window_length, int periodic, int options_kind, int options_device);
+raw_tensor atg_kaiser_window_periodic_out(gc_tensor out, int64_t window_length, int periodic);
+raw_tensor atg_kl_div(gc_tensor self, gc_tensor target, int64_t reduction, int log_target);
+raw_tensor atg_kron(gc_tensor self, gc_tensor other);
+raw_tensor atg_kron_out(gc_tensor out, gc_tensor self, gc_tensor other);
+void atg_kthvalue(raw_tensor *, gc_tensor self, int64_t k, int64_t dim, int keepdim);
+void atg_kthvalue_values(raw_tensor *, gc_tensor values, gc_tensor indices, gc_tensor self, int64_t k, int64_t dim, int keepdim);
+raw_tensor atg_l1_loss(gc_tensor self, gc_tensor target, int64_t reduction);
+raw_tensor atg_layer_norm(gc_tensor input, int64_t *normalized_shape_data, int normalized_shape_len, gc_tensor weight, gc_tensor bias, double eps, int cudnn_enable);
+raw_tensor atg_lcm(gc_tensor self, gc_tensor other);
+raw_tensor atg_lcm_(gc_tensor self, gc_tensor other);
+raw_tensor atg_lcm_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_ldexp(gc_tensor self, gc_tensor other);
+raw_tensor atg_ldexp_(gc_tensor self, gc_tensor other);
+raw_tensor atg_ldexp_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_le(gc_tensor self, scalar other);
+raw_tensor atg_le_(gc_tensor self, scalar other);
+raw_tensor atg_le_scalar_out(gc_tensor out, gc_tensor self, scalar other);
+raw_tensor atg_le_tensor(gc_tensor self, gc_tensor other);
+raw_tensor atg_le_tensor_(gc_tensor self, gc_tensor other);
+raw_tensor atg_le_tensor_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_leaky_relu(gc_tensor self);
+raw_tensor atg_leaky_relu_(gc_tensor self);
+raw_tensor atg_leaky_relu_backward(gc_tensor grad_output, gc_tensor self, scalar negative_slope, int self_is_result);
+raw_tensor atg_leaky_relu_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, scalar negative_slope, int self_is_result);
+raw_tensor atg_leaky_relu_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_lerp(gc_tensor self, gc_tensor end, scalar weight);
+raw_tensor atg_lerp_(gc_tensor self, gc_tensor end, scalar weight);
+raw_tensor atg_lerp_scalar_out(gc_tensor out, gc_tensor self, gc_tensor end, scalar weight);
+raw_tensor atg_lerp_tensor(gc_tensor self, gc_tensor end, gc_tensor weight);
+raw_tensor atg_lerp_tensor_(gc_tensor self, gc_tensor end, gc_tensor weight);
+raw_tensor atg_lerp_tensor_out(gc_tensor out, gc_tensor self, gc_tensor end, gc_tensor weight);
+raw_tensor atg_less(gc_tensor self, scalar other);
+raw_tensor atg_less_(gc_tensor self, scalar other);
+raw_tensor atg_less_equal(gc_tensor self, scalar other);
+raw_tensor atg_less_equal_(gc_tensor self, scalar other);
+raw_tensor atg_less_equal_scalar_out(gc_tensor out, gc_tensor self, scalar other);
+raw_tensor atg_less_equal_tensor(gc_tensor self, gc_tensor other);
+raw_tensor atg_less_equal_tensor_(gc_tensor self, gc_tensor other);
+raw_tensor atg_less_equal_tensor_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_less_scalar_out(gc_tensor out, gc_tensor self, scalar other);
+raw_tensor atg_less_tensor(gc_tensor self, gc_tensor other);
+raw_tensor atg_less_tensor_(gc_tensor self, gc_tensor other);
+raw_tensor atg_less_tensor_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_lgamma(gc_tensor self);
+raw_tensor atg_lgamma_(gc_tensor self);
+raw_tensor atg_lgamma_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_lift(gc_tensor self);
+raw_tensor atg_lift_fresh(gc_tensor self);
+raw_tensor atg_lift_fresh_copy(gc_tensor self);
+raw_tensor atg_lift_fresh_copy_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_lift_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_linalg_cholesky(gc_tensor self, int upper);
+void atg_linalg_cholesky_ex(raw_tensor *, gc_tensor self, int upper, int check_errors);
+void atg_linalg_cholesky_ex_l(raw_tensor *, gc_tensor L, gc_tensor info, gc_tensor self, int upper, int check_errors);
+raw_tensor atg_linalg_cholesky_out(gc_tensor out, gc_tensor self, int upper);
+raw_tensor atg_linalg_cond(gc_tensor self, scalar p);
+raw_tensor atg_linalg_cond_out(gc_tensor out, gc_tensor self, scalar p);
+raw_tensor atg_linalg_cond_p_str(gc_tensor self, char * p);
+raw_tensor atg_linalg_cond_p_str_out(gc_tensor out, gc_tensor self, char * p);
+raw_tensor atg_linalg_cross(gc_tensor self, gc_tensor other, int64_t dim);
+raw_tensor atg_linalg_cross_out(gc_tensor out, gc_tensor self, gc_tensor other, int64_t dim);
+raw_tensor atg_linalg_det(gc_tensor A);
+raw_tensor atg_linalg_det_out(gc_tensor out, gc_tensor A);
+raw_tensor atg_linalg_diagonal(gc_tensor A, int64_t offset, int64_t dim1, int64_t dim2);
+void atg_linalg_eig(raw_tensor *, gc_tensor self);
+void atg_linalg_eig_out(raw_tensor *, gc_tensor eigenvalues, gc_tensor eigenvectors, gc_tensor self);
+void atg_linalg_eigh(raw_tensor *, gc_tensor self, char * UPLO);
+void atg_linalg_eigh_eigvals(raw_tensor *, gc_tensor eigvals, gc_tensor eigvecs, gc_tensor self, char * UPLO);
+raw_tensor atg_linalg_eigvals(gc_tensor self);
+raw_tensor atg_linalg_eigvals_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_linalg_eigvalsh(gc_tensor self, char * UPLO);
+raw_tensor atg_linalg_eigvalsh_out(gc_tensor out, gc_tensor self, char * UPLO);
+raw_tensor atg_linalg_householder_product(gc_tensor input, gc_tensor tau);
+raw_tensor atg_linalg_householder_product_out(gc_tensor out, gc_tensor input, gc_tensor tau);
+raw_tensor atg_linalg_inv(gc_tensor A);
+void atg_linalg_inv_ex(raw_tensor *, gc_tensor A, int check_errors);
+void atg_linalg_inv_ex_inverse(raw_tensor *, gc_tensor inverse, gc_tensor info, gc_tensor A, int check_errors);
+raw_tensor atg_linalg_inv_out(gc_tensor out, gc_tensor A);
+void atg_linalg_ldl_factor(raw_tensor *, gc_tensor self, int hermitian);
+void atg_linalg_ldl_factor_ex(raw_tensor *, gc_tensor self, int hermitian, int check_errors);
+void atg_linalg_ldl_factor_ex_out(raw_tensor *, gc_tensor LD, gc_tensor pivots, gc_tensor info, gc_tensor self, int hermitian, int check_errors);
+void atg_linalg_ldl_factor_out(raw_tensor *, gc_tensor LD, gc_tensor pivots, gc_tensor self, int hermitian);
+raw_tensor atg_linalg_ldl_solve(gc_tensor LD, gc_tensor pivots, gc_tensor B, int hermitian);
+raw_tensor atg_linalg_ldl_solve_out(gc_tensor out, gc_tensor LD, gc_tensor pivots, gc_tensor B, int hermitian);
+void atg_linalg_lstsq(raw_tensor *, gc_tensor self, gc_tensor b, double rcond_v, int rcond_null, char * driver);
+void atg_linalg_lstsq_out(raw_tensor *, gc_tensor solution, gc_tensor residuals, gc_tensor rank, gc_tensor singular_values, gc_tensor self, gc_tensor b, double rcond_v, int rcond_null, char * driver);
+void atg_linalg_lu(raw_tensor *, gc_tensor A, int pivot);
+void atg_linalg_lu_factor(raw_tensor *, gc_tensor A, int pivot);
+void atg_linalg_lu_factor_ex(raw_tensor *, gc_tensor A, int pivot, int check_errors);
+void atg_linalg_lu_factor_ex_out(raw_tensor *, gc_tensor LU, gc_tensor pivots, gc_tensor info, gc_tensor A, int pivot, int check_errors);
+void atg_linalg_lu_factor_out(raw_tensor *, gc_tensor LU, gc_tensor pivots, gc_tensor A, int pivot);
+void atg_linalg_lu_out(raw_tensor *, gc_tensor P, gc_tensor L, gc_tensor U, gc_tensor A, int pivot);
+raw_tensor atg_linalg_lu_solve(gc_tensor LU, gc_tensor pivots, gc_tensor B, int left, int adjoint);
+raw_tensor atg_linalg_lu_solve_out(gc_tensor out, gc_tensor LU, gc_tensor pivots, gc_tensor B, int left, int adjoint);
+raw_tensor atg_linalg_matmul(gc_tensor self, gc_tensor other);
+raw_tensor atg_linalg_matmul_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_linalg_matrix_exp(gc_tensor self);
+raw_tensor atg_linalg_matrix_exp_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_linalg_matrix_power(gc_tensor self, int64_t n);
+raw_tensor atg_linalg_matrix_power_out(gc_tensor out, gc_tensor self, int64_t n);
+raw_tensor atg_linalg_matrix_rank(gc_tensor self, double tol, int hermitian);
+raw_tensor atg_linalg_matrix_rank_atol_rtol_float(gc_tensor self, double atol_v, int atol_null, double rtol_v, int rtol_null, int hermitian);
+raw_tensor atg_linalg_matrix_rank_atol_rtol_float_out(gc_tensor out, gc_tensor self, double atol_v, int atol_null, double rtol_v, int rtol_null, int hermitian);
+raw_tensor atg_linalg_matrix_rank_atol_rtol_tensor(gc_tensor input, gc_tensor atol, gc_tensor rtol, int hermitian);
+raw_tensor atg_linalg_matrix_rank_atol_rtol_tensor_out(gc_tensor out, gc_tensor input, gc_tensor atol, gc_tensor rtol, int hermitian);
+raw_tensor atg_linalg_matrix_rank_out(gc_tensor out, gc_tensor self, double tol, int hermitian);
+raw_tensor atg_linalg_matrix_rank_out_tol_tensor(gc_tensor out, gc_tensor input, gc_tensor tol, int hermitian);
+raw_tensor atg_linalg_matrix_rank_tol_tensor(gc_tensor input, gc_tensor tol, int hermitian);
+raw_tensor atg_linalg_multi_dot(gc_tensor *tensors_data, int tensors_len);
+raw_tensor atg_linalg_multi_dot_out(gc_tensor out, gc_tensor *tensors_data, int tensors_len);
+raw_tensor atg_linalg_pinv(gc_tensor self, double rcond, int hermitian);
+raw_tensor atg_linalg_pinv_atol_rtol_float(gc_tensor self, double atol_v, int atol_null, double rtol_v, int rtol_null, int hermitian);
+raw_tensor atg_linalg_pinv_atol_rtol_float_out(gc_tensor out, gc_tensor self, double atol_v, int atol_null, double rtol_v, int rtol_null, int hermitian);
+raw_tensor atg_linalg_pinv_atol_rtol_tensor(gc_tensor self, gc_tensor atol, gc_tensor rtol, int hermitian);
+raw_tensor atg_linalg_pinv_atol_rtol_tensor_out(gc_tensor out, gc_tensor self, gc_tensor atol, gc_tensor rtol, int hermitian);
+raw_tensor atg_linalg_pinv_out(gc_tensor out, gc_tensor self, double rcond, int hermitian);
+raw_tensor atg_linalg_pinv_out_rcond_tensor(gc_tensor out, gc_tensor self, gc_tensor rcond, int hermitian);
+raw_tensor atg_linalg_pinv_rcond_tensor(gc_tensor self, gc_tensor rcond, int hermitian);
+void atg_linalg_qr(raw_tensor *, gc_tensor A, char * mode);
+void atg_linalg_qr_out(raw_tensor *, gc_tensor Q, gc_tensor R, gc_tensor A, char * mode);
+void atg_linalg_slogdet(raw_tensor *, gc_tensor A);
+void atg_linalg_slogdet_out(raw_tensor *, gc_tensor sign, gc_tensor logabsdet, gc_tensor A);
+raw_tensor atg_linalg_solve(gc_tensor A, gc_tensor B, int left);
+void atg_linalg_solve_ex(raw_tensor *, gc_tensor A, gc_tensor B, int left, int check_errors);
+void atg_linalg_solve_ex_out(raw_tensor *, gc_tensor result, gc_tensor info, gc_tensor A, gc_tensor B, int left, int check_errors);
+raw_tensor atg_linalg_solve_out(gc_tensor out, gc_tensor A, gc_tensor B, int left);
+raw_tensor atg_linalg_solve_triangular(gc_tensor self, gc_tensor B, int upper, int left, int unitriangular);
+raw_tensor atg_linalg_solve_triangular_out(gc_tensor out, gc_tensor self, gc_tensor B, int upper, int left, int unitriangular);
+void atg_linalg_svd(raw_tensor *, gc_tensor A, int full_matrices, char * driver);
+void atg_linalg_svd_u(raw_tensor *, gc_tensor U, gc_tensor S, gc_tensor Vh, gc_tensor A, int full_matrices, char * driver);
+raw_tensor atg_linalg_svdvals(gc_tensor A, char * driver);
+raw_tensor atg_linalg_svdvals_out(gc_tensor out, gc_tensor A, char * driver);
+raw_tensor atg_linalg_tensorinv(gc_tensor self, int64_t ind);
+raw_tensor atg_linalg_tensorinv_out(gc_tensor out, gc_tensor self, int64_t ind);
+raw_tensor atg_linalg_tensorsolve(gc_tensor self, gc_tensor other, int64_t *dims_data, int dims_len);
+raw_tensor atg_linalg_tensorsolve_out(gc_tensor out, gc_tensor self, gc_tensor other, int64_t *dims_data, int dims_len);
+raw_tensor atg_linalg_vander(gc_tensor x, int64_t n_v, int n_null);
+raw_tensor atg_linalg_vecdot(gc_tensor x, gc_tensor y, int64_t dim);
+raw_tensor atg_linalg_vecdot_out(gc_tensor out, gc_tensor x, gc_tensor y, int64_t dim);
+raw_tensor atg_linear(gc_tensor input, gc_tensor weight, gc_tensor bias);
+raw_tensor atg_linear_out(gc_tensor out, gc_tensor input, gc_tensor weight, gc_tensor bias);
+raw_tensor atg_linspace(scalar start, scalar end, int64_t steps, int options_kind, int options_device);
+raw_tensor atg_linspace_out(gc_tensor out, scalar start, scalar end, int64_t steps);
+raw_tensor atg_log(gc_tensor self);
+raw_tensor atg_log10(gc_tensor self);
+raw_tensor atg_log10_(gc_tensor self);
+raw_tensor atg_log10_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_log1p(gc_tensor self);
+raw_tensor atg_log1p_(gc_tensor self);
+raw_tensor atg_log1p_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_log2(gc_tensor self);
+raw_tensor atg_log2_(gc_tensor self);
+raw_tensor atg_log2_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_log_(gc_tensor self);
+raw_tensor atg_log_normal(gc_tensor self, double mean, double std);
+raw_tensor atg_log_normal_(gc_tensor self, double mean, double std);
+raw_tensor atg_log_normal_out(gc_tensor out, gc_tensor self, double mean, double std);
+raw_tensor atg_log_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_log_sigmoid(gc_tensor self);
+raw_tensor atg_log_sigmoid_backward(gc_tensor grad_output, gc_tensor self, gc_tensor buffer);
+raw_tensor atg_log_sigmoid_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, gc_tensor buffer);
+raw_tensor atg_log_sigmoid_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_log_softmax(gc_tensor self, int64_t dim, int dtype);
+raw_tensor atg_log_softmax_int_out(gc_tensor out, gc_tensor self, int64_t dim, int dtype);
+raw_tensor atg_logaddexp(gc_tensor self, gc_tensor other);
+raw_tensor atg_logaddexp2(gc_tensor self, gc_tensor other);
+raw_tensor atg_logaddexp2_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_logaddexp_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_logcumsumexp(gc_tensor self, int64_t dim);
+raw_tensor atg_logcumsumexp_out(gc_tensor out, gc_tensor self, int64_t dim);
+raw_tensor atg_logdet(gc_tensor self);
+raw_tensor atg_logical_and(gc_tensor self, gc_tensor other);
+raw_tensor atg_logical_and_(gc_tensor self, gc_tensor other);
+raw_tensor atg_logical_and_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_logical_not(gc_tensor self);
+raw_tensor atg_logical_not_(gc_tensor self);
+raw_tensor atg_logical_not_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_logical_or(gc_tensor self, gc_tensor other);
+raw_tensor atg_logical_or_(gc_tensor self, gc_tensor other);
+raw_tensor atg_logical_or_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_logical_xor(gc_tensor self, gc_tensor other);
+raw_tensor atg_logical_xor_(gc_tensor self, gc_tensor other);
+raw_tensor atg_logical_xor_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_logit(gc_tensor self, double eps_v, int eps_null);
+raw_tensor atg_logit_(gc_tensor self, double eps_v, int eps_null);
+raw_tensor atg_logit_backward(gc_tensor grad_output, gc_tensor self, double eps_v, int eps_null);
+raw_tensor atg_logit_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, double eps_v, int eps_null);
+raw_tensor atg_logit_out(gc_tensor out, gc_tensor self, double eps_v, int eps_null);
+raw_tensor atg_logspace(scalar start, scalar end, int64_t steps, double base, int options_kind, int options_device);
+raw_tensor atg_logspace_out(gc_tensor out, scalar start, scalar end, int64_t steps, double base);
+raw_tensor atg_logsumexp(gc_tensor self, int64_t *dim_data, int dim_len, int keepdim);
+raw_tensor atg_logsumexp_out(gc_tensor out, gc_tensor self, int64_t *dim_data, int dim_len, int keepdim);
+void atg_lstm(raw_tensor *, gc_tensor input, gc_tensor *hx_data, int hx_len, gc_tensor *params_data, int params_len, int has_biases, int64_t num_layers, double dropout, int train, int bidirectional, int batch_first);
+void atg_lstm_cell(raw_tensor *, gc_tensor input, gc_tensor *hx_data, int hx_len, gc_tensor w_ih, gc_tensor w_hh, gc_tensor b_ih, gc_tensor b_hh);
+void atg_lstm_data(raw_tensor *, gc_tensor data, gc_tensor batch_sizes, gc_tensor *hx_data, int hx_len, gc_tensor *params_data, int params_len, int has_biases, int64_t num_layers, double dropout, int train, int bidirectional);
+void atg_lstm_mps_backward(gc_tensor out0, gc_tensor *out1_data, int out1_len, gc_tensor *out2_data, int out2_len, gc_tensor grad_y, gc_tensor grad_hy, gc_tensor grad_cy, gc_tensor z_state, gc_tensor cell_state_fwd, gc_tensor input, gc_tensor layersOutputs, gc_tensor *hx_data, int hx_len, gc_tensor *params_data, int params_len, int has_biases, int64_t num_layers, double dropout, int train, int bidirectional, int batch_first);
+raw_tensor atg_lt(gc_tensor self, scalar other);
+raw_tensor atg_lt_(gc_tensor self, scalar other);
+raw_tensor atg_lt_scalar_out(gc_tensor out, gc_tensor self, scalar other);
+raw_tensor atg_lt_tensor(gc_tensor self, gc_tensor other);
+raw_tensor atg_lt_tensor_(gc_tensor self, gc_tensor other);
+raw_tensor atg_lt_tensor_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_lu_solve(gc_tensor self, gc_tensor LU_data, gc_tensor LU_pivots);
+raw_tensor atg_lu_solve_out(gc_tensor out, gc_tensor self, gc_tensor LU_data, gc_tensor LU_pivots);
+void atg_lu_unpack(raw_tensor *, gc_tensor LU_data, gc_tensor LU_pivots, int unpack_data, int unpack_pivots);
+void atg_lu_unpack_out(raw_tensor *, gc_tensor P, gc_tensor L, gc_tensor U, gc_tensor LU_data, gc_tensor LU_pivots, int unpack_data, int unpack_pivots);
+raw_tensor atg_margin_ranking_loss(gc_tensor input1, gc_tensor input2, gc_tensor target, double margin, int64_t reduction);
+raw_tensor atg_masked_fill(gc_tensor self, gc_tensor mask, scalar value);
+raw_tensor atg_masked_fill_(gc_tensor self, gc_tensor mask, scalar value);
+raw_tensor atg_masked_fill_scalar_out(gc_tensor out, gc_tensor self, gc_tensor mask, scalar value);
+raw_tensor atg_masked_fill_tensor(gc_tensor self, gc_tensor mask, gc_tensor value);
+raw_tensor atg_masked_fill_tensor_(gc_tensor self, gc_tensor mask, gc_tensor value);
+raw_tensor atg_masked_fill_tensor_out(gc_tensor out, gc_tensor self, gc_tensor mask, gc_tensor value);
+raw_tensor atg_masked_scatter(gc_tensor self, gc_tensor mask, gc_tensor source);
+raw_tensor atg_masked_scatter_(gc_tensor self, gc_tensor mask, gc_tensor source);
+raw_tensor atg_masked_scatter_out(gc_tensor out, gc_tensor self, gc_tensor mask, gc_tensor source);
+raw_tensor atg_masked_select(gc_tensor self, gc_tensor mask);
+raw_tensor atg_masked_select_backward(gc_tensor grad, gc_tensor input, gc_tensor mask);
+raw_tensor atg_masked_select_out(gc_tensor out, gc_tensor self, gc_tensor mask);
+raw_tensor atg_matmul(gc_tensor self, gc_tensor other);
+raw_tensor atg_matmul_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_matrix_exp(gc_tensor self);
+raw_tensor atg_matrix_exp_backward(gc_tensor self, gc_tensor grad);
+raw_tensor atg_matrix_h(gc_tensor self);
+raw_tensor atg_matrix_power(gc_tensor self, int64_t n);
+raw_tensor atg_matrix_power_out(gc_tensor out, gc_tensor self, int64_t n);
+raw_tensor atg_max(gc_tensor self);
+void atg_max_dim(raw_tensor *, gc_tensor self, int64_t dim, int keepdim);
+void atg_max_dim_max(raw_tensor *, gc_tensor max, gc_tensor max_values, gc_tensor self, int64_t dim, int keepdim);
+raw_tensor atg_max_other(gc_tensor self, gc_tensor other);
+raw_tensor atg_max_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_max_pool1d(gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
+void atg_max_pool1d_with_indices(raw_tensor *, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
+raw_tensor atg_max_pool2d(gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
+raw_tensor atg_max_pool2d_backward(gc_tensor grad_output, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
+raw_tensor atg_max_pool2d_backward_out(gc_tensor out, gc_tensor grad_output, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
+void atg_max_pool2d_with_indices(raw_tensor *, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
+raw_tensor atg_max_pool2d_with_indices_backward(gc_tensor grad_output, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode, gc_tensor indices);
+raw_tensor atg_max_pool2d_with_indices_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode, gc_tensor indices);
+void atg_max_pool2d_with_indices_out(raw_tensor *, gc_tensor out, gc_tensor indices, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
+raw_tensor atg_max_pool3d(gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
+void atg_max_pool3d_with_indices(raw_tensor *, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
+raw_tensor atg_max_pool3d_with_indices_backward(gc_tensor grad_output, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode, gc_tensor indices);
+raw_tensor atg_max_pool3d_with_indices_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode, gc_tensor indices);
+void atg_max_pool3d_with_indices_out(raw_tensor *, gc_tensor out, gc_tensor indices, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
+raw_tensor atg_max_unary_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_max_unpool2d(gc_tensor self, gc_tensor indices, int64_t *output_size_data, int output_size_len);
+raw_tensor atg_max_unpool2d_out(gc_tensor out, gc_tensor self, gc_tensor indices, int64_t *output_size_data, int output_size_len);
+raw_tensor atg_max_unpool3d(gc_tensor self, gc_tensor indices, int64_t *output_size_data, int output_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len);
+raw_tensor atg_max_unpool3d_out(gc_tensor out, gc_tensor self, gc_tensor indices, int64_t *output_size_data, int output_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len);
+raw_tensor atg_maximum(gc_tensor self, gc_tensor other);
+raw_tensor atg_maximum_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_mean(gc_tensor self, int dtype);
+raw_tensor atg_mean_dim(gc_tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype);
+raw_tensor atg_mean_out(gc_tensor out, gc_tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype);
+raw_tensor atg_median(gc_tensor self);
+void atg_median_dim(raw_tensor *, gc_tensor self, int64_t dim, int keepdim);
+void atg_median_dim_values(raw_tensor *, gc_tensor values, gc_tensor indices, gc_tensor self, int64_t dim, int keepdim);
+raw_tensor atg_median_out(gc_tensor out, gc_tensor self);
+raw_tensor *atg_meshgrid(gc_tensor *tensors_data, int tensors_len);
+raw_tensor *atg_meshgrid_indexing(gc_tensor *tensors_data, int tensors_len, char * indexing);
+raw_tensor atg_mh(gc_tensor self);
+raw_tensor atg_min(gc_tensor self);
+void atg_min_dim(raw_tensor *, gc_tensor self, int64_t dim, int keepdim);
+void atg_min_dim_min(raw_tensor *, gc_tensor min, gc_tensor min_indices, gc_tensor self, int64_t dim, int keepdim);
+raw_tensor atg_min_other(gc_tensor self, gc_tensor other);
+raw_tensor atg_min_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_min_unary_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_minimum(gc_tensor self, gc_tensor other);
+raw_tensor atg_minimum_out(gc_tensor out, gc_tensor self, gc_tensor other);
+void atg_miopen_batch_norm(raw_tensor *, gc_tensor input, gc_tensor weight, gc_tensor bias, gc_tensor running_mean, gc_tensor running_var, int training, double exponential_average_factor, double epsilon);
+void atg_miopen_batch_norm_backward(raw_tensor *, gc_tensor input, gc_tensor grad_output, gc_tensor weight, gc_tensor running_mean, gc_tensor running_var, gc_tensor save_mean, gc_tensor save_var, double epsilon);
+void atg_miopen_batch_norm_backward_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor input, gc_tensor grad_output, gc_tensor weight, gc_tensor running_mean, gc_tensor running_var, gc_tensor save_mean, gc_tensor save_var, double epsilon);
+void atg_miopen_batch_norm_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor input, gc_tensor weight, gc_tensor bias, gc_tensor running_mean, gc_tensor running_var, int training, double exponential_average_factor, double epsilon);
+raw_tensor atg_miopen_convolution(gc_tensor self, gc_tensor weight, gc_tensor bias, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int benchmark, int deterministic);
+raw_tensor atg_miopen_convolution_add_relu(gc_tensor self, gc_tensor weight, gc_tensor z, scalar alpha, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int64_t groups);
+raw_tensor atg_miopen_convolution_out(gc_tensor out, gc_tensor self, gc_tensor weight, gc_tensor bias, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int benchmark, int deterministic);
+raw_tensor atg_miopen_convolution_relu(gc_tensor self, gc_tensor weight, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int64_t groups);
+raw_tensor atg_miopen_convolution_transpose(gc_tensor self, gc_tensor weight, gc_tensor bias, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int benchmark, int deterministic);
+raw_tensor atg_miopen_convolution_transpose_out(gc_tensor out, gc_tensor self, gc_tensor weight, gc_tensor bias, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int benchmark, int deterministic);
+raw_tensor atg_miopen_depthwise_convolution(gc_tensor self, gc_tensor weight, gc_tensor bias, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int benchmark, int deterministic);
+raw_tensor atg_miopen_depthwise_convolution_out(gc_tensor out, gc_tensor self, gc_tensor weight, gc_tensor bias, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int benchmark, int deterministic);
+void atg_miopen_rnn(raw_tensor *, gc_tensor input, gc_tensor *weight_data, int weight_len, int64_t weight_stride0, gc_tensor hx, gc_tensor cx, int64_t mode, int64_t hidden_size, int64_t num_layers, int batch_first, double dropout, int train, int bidirectional, int64_t *batch_sizes_data, int batch_sizes_len, gc_tensor dropout_state);
+void atg_miopen_rnn_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor out3, gc_tensor out4, gc_tensor input, gc_tensor *weight_data, int weight_len, int64_t weight_stride0, gc_tensor hx, gc_tensor cx, int64_t mode, int64_t hidden_size, int64_t num_layers, int batch_first, double dropout, int train, int bidirectional, int64_t *batch_sizes_data, int batch_sizes_len, gc_tensor dropout_state);
+raw_tensor atg_mish(gc_tensor self);
+raw_tensor atg_mish_(gc_tensor self);
+raw_tensor atg_mish_backward(gc_tensor grad_output, gc_tensor self);
+raw_tensor atg_mish_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_mkldnn_adaptive_avg_pool2d(gc_tensor self, int64_t *output_size_data, int output_size_len);
+raw_tensor atg_mkldnn_adaptive_avg_pool2d_backward(gc_tensor grad_output, gc_tensor self);
+raw_tensor atg_mkldnn_adaptive_avg_pool2d_backward_out(gc_tensor out, gc_tensor grad_output, gc_tensor self);
+raw_tensor atg_mkldnn_adaptive_avg_pool2d_out(gc_tensor out, gc_tensor self, int64_t *output_size_data, int output_size_len);
+raw_tensor atg_mkldnn_convolution(gc_tensor self, gc_tensor weight, gc_tensor bias, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups);
+raw_tensor atg_mkldnn_convolution_out(gc_tensor out, gc_tensor self, gc_tensor weight, gc_tensor bias, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups);
+raw_tensor atg_mkldnn_linear(gc_tensor self, gc_tensor weight, gc_tensor bias);
+raw_tensor atg_mkldnn_linear_backward_input(int64_t *input_size_data, int input_size_len, gc_tensor grad_output, gc_tensor weight);
+raw_tensor atg_mkldnn_linear_backward_input_out(gc_tensor out, int64_t *input_size_data, int input_size_len, gc_tensor grad_output, gc_tensor weight);
+void atg_mkldnn_linear_backward_weights(raw_tensor *, gc_tensor grad_output, gc_tensor input, gc_tensor weight, int bias_defined);
+void atg_mkldnn_linear_backward_weights_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor grad_output, gc_tensor input, gc_tensor weight, int bias_defined);
+raw_tensor atg_mkldnn_linear_out(gc_tensor out, gc_tensor self, gc_tensor weight, gc_tensor bias);
+raw_tensor atg_mkldnn_max_pool2d(gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
+raw_tensor atg_mkldnn_max_pool2d_backward(gc_tensor grad_output, gc_tensor output, gc_tensor input, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
+raw_tensor atg_mkldnn_max_pool2d_backward_out(gc_tensor out, gc_tensor grad_output, gc_tensor output, gc_tensor input, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
+raw_tensor atg_mkldnn_max_pool2d_out(gc_tensor out, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
+raw_tensor atg_mkldnn_max_pool3d(gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
+raw_tensor atg_mkldnn_max_pool3d_backward(gc_tensor grad_output, gc_tensor output, gc_tensor input, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
+raw_tensor atg_mkldnn_max_pool3d_backward_out(gc_tensor out, gc_tensor grad_output, gc_tensor output, gc_tensor input, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
+raw_tensor atg_mkldnn_max_pool3d_out(gc_tensor out, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
+raw_tensor atg_mkldnn_reorder_conv2d_weight(gc_tensor self, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int64_t *input_size_data, int input_size_len);
+raw_tensor atg_mkldnn_reorder_conv2d_weight_out(gc_tensor out, gc_tensor self, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups, int64_t *input_size_data, int input_size_len);
+raw_tensor atg_mkldnn_reorder_conv3d_weight(gc_tensor self, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups);
+raw_tensor atg_mkldnn_reorder_conv3d_weight_out(gc_tensor out, gc_tensor self, int64_t *padding_data, int padding_len, int64_t *stride_data, int stride_len, int64_t *dilation_data, int dilation_len, int64_t groups);
+void atg_mkldnn_rnn_layer(raw_tensor *, gc_tensor input, gc_tensor weight0, gc_tensor weight1, gc_tensor weight2, gc_tensor weight3, gc_tensor hx_, gc_tensor cx_, int reverse, int64_t *batch_sizes_data, int batch_sizes_len, int64_t mode, int64_t hidden_size, int64_t num_layers, int has_biases, int bidirectional, int batch_first, int train);
+void atg_mkldnn_rnn_layer_backward(raw_tensor *, gc_tensor input, gc_tensor weight1, gc_tensor weight2, gc_tensor weight3, gc_tensor weight4, gc_tensor hx_, gc_tensor cx_tmp, gc_tensor output, gc_tensor hy_, gc_tensor cy_, gc_tensor grad_output, gc_tensor grad_hy, gc_tensor grad_cy, int reverse, int64_t mode, int64_t hidden_size, int64_t num_layers, int has_biases, int train, int bidirectional, int64_t *batch_sizes_data, int batch_sizes_len, int batch_first, gc_tensor workspace);
+void atg_mkldnn_rnn_layer_backward_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor out3, gc_tensor out4, gc_tensor out5, gc_tensor out6, gc_tensor input, gc_tensor weight1, gc_tensor weight2, gc_tensor weight3, gc_tensor weight4, gc_tensor hx_, gc_tensor cx_tmp, gc_tensor output, gc_tensor hy_, gc_tensor cy_, gc_tensor grad_output, gc_tensor grad_hy, gc_tensor grad_cy, int reverse, int64_t mode, int64_t hidden_size, int64_t num_layers, int has_biases, int train, int bidirectional, int64_t *batch_sizes_data, int batch_sizes_len, int batch_first, gc_tensor workspace);
+void atg_mkldnn_rnn_layer_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor out3, gc_tensor input, gc_tensor weight0, gc_tensor weight1, gc_tensor weight2, gc_tensor weight3, gc_tensor hx_, gc_tensor cx_, int reverse, int64_t *batch_sizes_data, int batch_sizes_len, int64_t mode, int64_t hidden_size, int64_t num_layers, int has_biases, int bidirectional, int batch_first, int train);
+raw_tensor atg_mm(gc_tensor self, gc_tensor mat2);
+raw_tensor atg_mm_out(gc_tensor out, gc_tensor self, gc_tensor mat2);
+void atg_mode(raw_tensor *, gc_tensor self, int64_t dim, int keepdim);
+void atg_mode_values(raw_tensor *, gc_tensor values, gc_tensor indices, gc_tensor self, int64_t dim, int keepdim);
+raw_tensor atg_moveaxis(gc_tensor self, int64_t *source_data, int source_len, int64_t *destination_data, int destination_len);
+raw_tensor atg_moveaxis_int(gc_tensor self, int64_t source, int64_t destination);
+raw_tensor atg_movedim(gc_tensor self, int64_t *source_data, int source_len, int64_t *destination_data, int destination_len);
+raw_tensor atg_movedim_int(gc_tensor self, int64_t source, int64_t destination);
+raw_tensor atg_mse_loss(gc_tensor self, gc_tensor target, int64_t reduction);
+raw_tensor atg_mse_loss_backward(gc_tensor grad_output, gc_tensor self, gc_tensor target, int64_t reduction);
+raw_tensor atg_mse_loss_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, gc_tensor target, int64_t reduction);
+raw_tensor atg_mse_loss_out(gc_tensor out, gc_tensor self, gc_tensor target, int64_t reduction);
+raw_tensor atg_msort(gc_tensor self);
+raw_tensor atg_msort_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_mt(gc_tensor self);
+raw_tensor atg_mul(gc_tensor self, gc_tensor other);
+raw_tensor atg_mul_(gc_tensor self, gc_tensor other);
+raw_tensor atg_mul_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_mul_scalar(gc_tensor self, scalar other);
+raw_tensor atg_mul_scalar_(gc_tensor self, scalar other);
+raw_tensor atg_mul_scalar_out(gc_tensor out, gc_tensor self, scalar other);
+raw_tensor atg_multi_margin_loss_backward(gc_tensor grad_output, gc_tensor self, gc_tensor target, scalar p, scalar margin, gc_tensor weight, int64_t reduction);
+raw_tensor atg_multi_margin_loss_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, gc_tensor target, scalar p, scalar margin, gc_tensor weight, int64_t reduction);
+raw_tensor atg_multilabel_margin_loss(gc_tensor self, gc_tensor target, int64_t reduction);
+raw_tensor atg_multilabel_margin_loss_backward(gc_tensor grad_output, gc_tensor self, gc_tensor target, int64_t reduction, gc_tensor is_target);
+raw_tensor atg_multilabel_margin_loss_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, gc_tensor target, int64_t reduction, gc_tensor is_target);
+raw_tensor atg_multilabel_margin_loss_out(gc_tensor out, gc_tensor self, gc_tensor target, int64_t reduction);
+raw_tensor atg_multinomial(gc_tensor self, int64_t num_samples, int replacement);
+raw_tensor atg_multinomial_out(gc_tensor out, gc_tensor self, int64_t num_samples, int replacement);
+raw_tensor atg_multiply(gc_tensor self, gc_tensor other);
+raw_tensor atg_multiply_(gc_tensor self, gc_tensor other);
+raw_tensor atg_multiply_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_multiply_scalar(gc_tensor self, scalar other);
+raw_tensor atg_multiply_scalar_(gc_tensor self, scalar other);
+raw_tensor atg_mv(gc_tensor self, gc_tensor vec);
+raw_tensor atg_mv_out(gc_tensor out, gc_tensor self, gc_tensor vec);
+raw_tensor atg_mvlgamma(gc_tensor self, int64_t p);
+raw_tensor atg_mvlgamma_(gc_tensor self, int64_t p);
+raw_tensor atg_mvlgamma_out(gc_tensor out, gc_tensor self, int64_t p);
+raw_tensor atg_nan_to_num(gc_tensor self, double nan_v, int nan_null, double posinf_v, int posinf_null, double neginf_v, int neginf_null);
+raw_tensor atg_nan_to_num_(gc_tensor self, double nan_v, int nan_null, double posinf_v, int posinf_null, double neginf_v, int neginf_null);
+raw_tensor atg_nan_to_num_out(gc_tensor out, gc_tensor self, double nan_v, int nan_null, double posinf_v, int posinf_null, double neginf_v, int neginf_null);
+raw_tensor atg_nanmean(gc_tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype);
+raw_tensor atg_nanmean_out(gc_tensor out, gc_tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype);
+raw_tensor atg_nanmedian(gc_tensor self);
+void atg_nanmedian_dim(raw_tensor *, gc_tensor self, int64_t dim, int keepdim);
+void atg_nanmedian_dim_values(raw_tensor *, gc_tensor values, gc_tensor indices, gc_tensor self, int64_t dim, int keepdim);
+raw_tensor atg_nanmedian_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_nanquantile(gc_tensor self, gc_tensor q, int64_t dim_v, int dim_null, int keepdim, char * interpolation);
+raw_tensor atg_nanquantile_out(gc_tensor out, gc_tensor self, gc_tensor q, int64_t dim_v, int dim_null, int keepdim, char * interpolation);
+raw_tensor atg_nanquantile_scalar(gc_tensor self, double q, int64_t dim_v, int dim_null, int keepdim, char * interpolation);
+raw_tensor atg_nanquantile_scalar_out(gc_tensor out, gc_tensor self, double q, int64_t dim_v, int dim_null, int keepdim, char * interpolation);
+raw_tensor atg_nansum(gc_tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype);
+raw_tensor atg_nansum_out(gc_tensor out, gc_tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype);
+raw_tensor atg_narrow(gc_tensor self, int64_t dim, int64_t start, int64_t length);
+raw_tensor atg_narrow_copy(gc_tensor self, int64_t dim, int64_t start, int64_t length);
+raw_tensor atg_narrow_copy_out(gc_tensor out, gc_tensor self, int64_t dim, int64_t start, int64_t length);
+raw_tensor atg_narrow_tensor(gc_tensor self, int64_t dim, gc_tensor start, int64_t length);
+void atg_native_batch_norm(raw_tensor *, gc_tensor input, gc_tensor weight, gc_tensor bias, gc_tensor running_mean, gc_tensor running_var, int training, double momentum, double eps);
+void atg_native_batch_norm_out(raw_tensor *, gc_tensor out, gc_tensor save_mean, gc_tensor save_invstd, gc_tensor input, gc_tensor weight, gc_tensor bias, gc_tensor running_mean, gc_tensor running_var, int training, double momentum, double eps);
+raw_tensor atg_native_channel_shuffle(gc_tensor self, int64_t groups);
+void atg_native_dropout(raw_tensor *, gc_tensor input, double p, int train);
+raw_tensor atg_native_dropout_backward(gc_tensor grad_output, gc_tensor mask, double scale);
+raw_tensor atg_native_dropout_backward_out(gc_tensor out, gc_tensor grad_output, gc_tensor mask, double scale);
+void atg_native_dropout_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor input, double p, int train);
+void atg_native_group_norm(raw_tensor *, gc_tensor input, gc_tensor weight, gc_tensor bias, int64_t n, int64_t C, int64_t HxW, int64_t group, double eps);
+void atg_native_group_norm_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor input, gc_tensor weight, gc_tensor bias, int64_t n, int64_t C, int64_t HxW, int64_t group, double eps);
+void atg_native_layer_norm(raw_tensor *, gc_tensor input, int64_t *normalized_shape_data, int normalized_shape_len, gc_tensor weight, gc_tensor bias, double eps);
+void atg_native_layer_norm_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor input, int64_t *normalized_shape_data, int normalized_shape_len, gc_tensor weight, gc_tensor bias, double eps);
+raw_tensor atg_native_norm(gc_tensor self);
+raw_tensor atg_native_norm_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_native_norm_scalaropt_dim_dtype(gc_tensor self, scalar p, int64_t *dim_data, int dim_len, int keepdim, int dtype);
+raw_tensor atg_native_norm_scalaropt_dim_dtype_out(gc_tensor out, gc_tensor self, scalar p, int64_t *dim_data, int dim_len, int keepdim, int dtype);
+raw_tensor atg_ne(gc_tensor self, scalar other);
+raw_tensor atg_ne_(gc_tensor self, scalar other);
+raw_tensor atg_ne_scalar_out(gc_tensor out, gc_tensor self, scalar other);
+raw_tensor atg_ne_tensor(gc_tensor self, gc_tensor other);
+raw_tensor atg_ne_tensor_(gc_tensor self, gc_tensor other);
+raw_tensor atg_ne_tensor_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_neg(gc_tensor self);
+raw_tensor atg_neg_(gc_tensor self);
+raw_tensor atg_neg_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_negative(gc_tensor self);
+raw_tensor atg_negative_(gc_tensor self);
+raw_tensor atg_negative_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_nested_to_padded_tensor(gc_tensor self, double padding, int64_t *output_size_data, int output_size_len);
+raw_tensor atg_new_empty(gc_tensor self, int64_t *size_data, int size_len, int options_kind, int options_device);
+raw_tensor atg_new_empty_out(gc_tensor out, gc_tensor self, int64_t *size_data, int size_len);
+raw_tensor atg_new_empty_strided(gc_tensor self, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len, int options_kind, int options_device);
+raw_tensor atg_new_empty_strided_out(gc_tensor out, gc_tensor self, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len);
+raw_tensor atg_new_full(gc_tensor self, int64_t *size_data, int size_len, scalar fill_value, int options_kind, int options_device);
+raw_tensor atg_new_full_out(gc_tensor out, gc_tensor self, int64_t *size_data, int size_len, scalar fill_value);
+raw_tensor atg_new_ones(gc_tensor self, int64_t *size_data, int size_len, int options_kind, int options_device);
+raw_tensor atg_new_ones_out(gc_tensor out, gc_tensor self, int64_t *size_data, int size_len);
+raw_tensor atg_new_zeros(gc_tensor self, int64_t *size_data, int size_len, int options_kind, int options_device);
+raw_tensor atg_new_zeros_out(gc_tensor out, gc_tensor self, int64_t *size_data, int size_len);
+raw_tensor atg_nextafter(gc_tensor self, gc_tensor other);
+raw_tensor atg_nextafter_(gc_tensor self, gc_tensor other);
+raw_tensor atg_nextafter_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_nll_loss(gc_tensor self, gc_tensor target, gc_tensor weight, int64_t reduction, int64_t ignore_index);
+raw_tensor atg_nll_loss2d(gc_tensor self, gc_tensor target, gc_tensor weight, int64_t reduction, int64_t ignore_index);
+raw_tensor atg_nll_loss2d_backward(gc_tensor grad_output, gc_tensor self, gc_tensor target, gc_tensor weight, int64_t reduction, int64_t ignore_index, gc_tensor total_weight);
+raw_tensor atg_nll_loss2d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, gc_tensor target, gc_tensor weight, int64_t reduction, int64_t ignore_index, gc_tensor total_weight);
+raw_tensor atg_nll_loss2d_out(gc_tensor out, gc_tensor self, gc_tensor target, gc_tensor weight, int64_t reduction, int64_t ignore_index);
+raw_tensor atg_nll_loss_backward(gc_tensor grad_output, gc_tensor self, gc_tensor target, gc_tensor weight, int64_t reduction, int64_t ignore_index, gc_tensor total_weight);
+raw_tensor atg_nll_loss_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, gc_tensor target, gc_tensor weight, int64_t reduction, int64_t ignore_index, gc_tensor total_weight);
+raw_tensor atg_nll_loss_nd(gc_tensor self, gc_tensor target, gc_tensor weight, int64_t reduction, int64_t ignore_index);
+raw_tensor atg_nll_loss_out(gc_tensor out, gc_tensor self, gc_tensor target, gc_tensor weight, int64_t reduction, int64_t ignore_index);
+raw_tensor atg_nonzero(gc_tensor self);
+raw_tensor *atg_nonzero_numpy(gc_tensor self);
+raw_tensor atg_nonzero_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_nonzero_static(gc_tensor self, int64_t size, int64_t fill_value);
+raw_tensor atg_nonzero_static_out(gc_tensor out, gc_tensor self, int64_t size, int64_t fill_value);
+raw_tensor atg_norm(gc_tensor self);
+raw_tensor atg_norm_dtype_out(gc_tensor out, gc_tensor self, scalar p, int64_t *dim_data, int dim_len, int keepdim, int dtype);
+raw_tensor atg_norm_except_dim(gc_tensor v, int64_t pow, int64_t dim);
+raw_tensor atg_norm_out(gc_tensor out, gc_tensor self, scalar p, int64_t *dim_data, int dim_len, int keepdim);
+raw_tensor atg_norm_scalar_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_norm_scalaropt_dim(gc_tensor self, scalar p, int64_t *dim_data, int dim_len, int keepdim);
+raw_tensor atg_norm_scalaropt_dim_dtype(gc_tensor self, scalar p, int64_t *dim_data, int dim_len, int keepdim, int dtype);
+raw_tensor atg_norm_scalaropt_dtype(gc_tensor self, scalar p, int dtype);
+raw_tensor atg_norm_scalaropt_dtype_out(gc_tensor out, gc_tensor self, scalar p, int dtype);
+raw_tensor atg_normal_(gc_tensor self, double mean, double std);
+raw_tensor atg_normal_functional(gc_tensor self, double mean, double std);
+raw_tensor atg_not_equal(gc_tensor self, scalar other);
+raw_tensor atg_not_equal_(gc_tensor self, scalar other);
+raw_tensor atg_not_equal_scalar_out(gc_tensor out, gc_tensor self, scalar other);
+raw_tensor atg_not_equal_tensor(gc_tensor self, gc_tensor other);
+raw_tensor atg_not_equal_tensor_(gc_tensor self, gc_tensor other);
+raw_tensor atg_not_equal_tensor_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_nuclear_norm(gc_tensor self, int keepdim);
+raw_tensor atg_nuclear_norm_dim(gc_tensor self, int64_t *dim_data, int dim_len, int keepdim);
+raw_tensor atg_nuclear_norm_dim_out(gc_tensor out, gc_tensor self, int64_t *dim_data, int dim_len, int keepdim);
+raw_tensor atg_nuclear_norm_out(gc_tensor out, gc_tensor self, int keepdim);
+raw_tensor atg_numpy_t(gc_tensor self);
+raw_tensor atg_one_hot(gc_tensor self, int64_t num_classes);
+raw_tensor atg_ones(int64_t *size_data, int size_len, int options_kind, int options_device);
+raw_tensor atg_ones_like(gc_tensor self);
+raw_tensor atg_ones_like_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_ones_out(gc_tensor out, int64_t *size_data, int size_len);
+raw_tensor atg_orgqr(gc_tensor self, gc_tensor input2);
+raw_tensor atg_orgqr_out(gc_tensor out, gc_tensor self, gc_tensor input2);
+raw_tensor atg_ormqr(gc_tensor self, gc_tensor input2, gc_tensor input3, int left, int transpose);
+raw_tensor atg_ormqr_out(gc_tensor out, gc_tensor self, gc_tensor input2, gc_tensor input3, int left, int transpose);
+raw_tensor atg_outer(gc_tensor self, gc_tensor vec2);
+raw_tensor atg_outer_out(gc_tensor out, gc_tensor self, gc_tensor vec2);
+int64_t atg_output_nr(gc_tensor self);
+raw_tensor atg_pad(gc_tensor self, int64_t *pad_data, int pad_len, char * mode, double value_v, int value_null);
+raw_tensor atg_pad_sequence(gc_tensor *sequences_data, int sequences_len, int batch_first, double padding_value);
+raw_tensor atg_pairwise_distance(gc_tensor x1, gc_tensor x2, double p, double eps, int keepdim);
+raw_tensor atg_pdist(gc_tensor self, double p);
+raw_tensor atg_permute(gc_tensor self, int64_t *dims_data, int dims_len);
+raw_tensor atg_permute_copy(gc_tensor self, int64_t *dims_data, int dims_len);
+raw_tensor atg_permute_copy_out(gc_tensor out, gc_tensor self, int64_t *dims_data, int dims_len);
+raw_tensor atg_pin_memory(gc_tensor self, int device);
+raw_tensor atg_pinverse(gc_tensor self, double rcond);
+raw_tensor atg_pixel_shuffle(gc_tensor self, int64_t upscale_factor);
+raw_tensor atg_pixel_shuffle_out(gc_tensor out, gc_tensor self, int64_t upscale_factor);
+raw_tensor atg_pixel_unshuffle(gc_tensor self, int64_t downscale_factor);
+raw_tensor atg_pixel_unshuffle_out(gc_tensor out, gc_tensor self, int64_t downscale_factor);
+raw_tensor atg_poisson(gc_tensor self);
+raw_tensor atg_poisson_nll_loss(gc_tensor input, gc_tensor target, int log_input, int full, double eps, int64_t reduction);
+raw_tensor atg_poisson_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_polar(gc_tensor abs, gc_tensor angle);
+raw_tensor atg_polar_out(gc_tensor out, gc_tensor abs, gc_tensor angle);
+raw_tensor atg_polygamma(int64_t n, gc_tensor self);
+raw_tensor atg_polygamma_(gc_tensor self, int64_t n);
+raw_tensor atg_polygamma_out(gc_tensor out, int64_t n, gc_tensor self);
+raw_tensor atg_positive(gc_tensor self);
+raw_tensor atg_pow(gc_tensor self, gc_tensor exponent);
+raw_tensor atg_pow_(gc_tensor self, scalar exponent);
+raw_tensor atg_pow_scalar(scalar self, gc_tensor exponent);
+raw_tensor atg_pow_scalar_out(gc_tensor out, scalar self, gc_tensor exponent);
+raw_tensor atg_pow_tensor_(gc_tensor self, gc_tensor exponent);
+raw_tensor atg_pow_tensor_scalar(gc_tensor self, scalar exponent);
+raw_tensor atg_pow_tensor_scalar_out(gc_tensor out, gc_tensor self, scalar exponent);
+raw_tensor atg_pow_tensor_tensor_out(gc_tensor out, gc_tensor self, gc_tensor exponent);
+raw_tensor atg_prelu(gc_tensor self, gc_tensor weight);
+raw_tensor atg_prod(gc_tensor self, int dtype);
+raw_tensor atg_prod_dim_int(gc_tensor self, int64_t dim, int keepdim, int dtype);
+raw_tensor atg_prod_int_out(gc_tensor out, gc_tensor self, int64_t dim, int keepdim, int dtype);
+raw_tensor atg_prod_out(gc_tensor out, gc_tensor self, int dtype);
+raw_tensor atg_put(gc_tensor self, gc_tensor index, gc_tensor source, int accumulate);
+raw_tensor atg_put_(gc_tensor self, gc_tensor index, gc_tensor source, int accumulate);
+raw_tensor atg_put_out(gc_tensor out, gc_tensor self, gc_tensor index, gc_tensor source, int accumulate);
+int64_t atg_q_per_channel_axis(gc_tensor self);
+raw_tensor atg_q_per_channel_scales(gc_tensor self);
+raw_tensor atg_q_per_channel_scales_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_q_per_channel_zero_points(gc_tensor self);
+raw_tensor atg_q_per_channel_zero_points_out(gc_tensor out, gc_tensor self);
+double atg_q_scale(gc_tensor self);
+int64_t atg_q_zero_point(gc_tensor self);
+void atg_qr(raw_tensor *, gc_tensor self, int some);
+void atg_qr_q(raw_tensor *, gc_tensor Q, gc_tensor R, gc_tensor self, int some);
+raw_tensor atg_quantile(gc_tensor self, gc_tensor q, int64_t dim_v, int dim_null, int keepdim, char * interpolation);
+raw_tensor atg_quantile_out(gc_tensor out, gc_tensor self, gc_tensor q, int64_t dim_v, int dim_null, int keepdim, char * interpolation);
+raw_tensor atg_quantile_scalar(gc_tensor self, double q, int64_t dim_v, int dim_null, int keepdim, char * interpolation);
+raw_tensor atg_quantile_scalar_out(gc_tensor out, gc_tensor self, double q, int64_t dim_v, int dim_null, int keepdim, char * interpolation);
+raw_tensor atg_quantize_per_channel(gc_tensor self, gc_tensor scales, gc_tensor zero_points, int64_t axis, int dtype);
+raw_tensor atg_quantize_per_channel_out(gc_tensor out, gc_tensor self, gc_tensor scales, gc_tensor zero_points, int64_t axis, int dtype);
+raw_tensor atg_quantize_per_tensor(gc_tensor self, double scale, int64_t zero_point, int dtype);
+raw_tensor atg_quantize_per_tensor_dynamic(gc_tensor self, int dtype, int reduce_range);
+raw_tensor atg_quantize_per_tensor_dynamic_out(gc_tensor out, gc_tensor self, int dtype, int reduce_range);
+raw_tensor atg_quantize_per_tensor_out(gc_tensor out, gc_tensor self, double scale, int64_t zero_point, int dtype);
+raw_tensor atg_quantize_per_tensor_tensor_qparams(gc_tensor self, gc_tensor scale, gc_tensor zero_point, int dtype);
+raw_tensor atg_quantize_per_tensor_tensor_qparams_out(gc_tensor out, gc_tensor self, gc_tensor scale, gc_tensor zero_point, int dtype);
+raw_tensor *atg_quantize_per_tensor_tensors(gc_tensor *tensors_data, int tensors_len, gc_tensor scales, gc_tensor zero_points, int dtype);
+void atg_quantize_per_tensor_tensors_out(gc_tensor *out_data, int out_len, gc_tensor *tensors_data, int tensors_len, gc_tensor scales, gc_tensor zero_points, int dtype);
+raw_tensor atg_quantized_batch_norm(gc_tensor input, gc_tensor weight, gc_tensor bias, gc_tensor mean, gc_tensor var, double eps, double output_scale, int64_t output_zero_point);
+raw_tensor atg_quantized_batch_norm_out(gc_tensor out, gc_tensor input, gc_tensor weight, gc_tensor bias, gc_tensor mean, gc_tensor var, double eps, double output_scale, int64_t output_zero_point);
+raw_tensor atg_quantized_gru_cell(gc_tensor input, gc_tensor hx, gc_tensor w_ih, gc_tensor w_hh, gc_tensor b_ih, gc_tensor b_hh, gc_tensor packed_ih, gc_tensor packed_hh, gc_tensor col_offsets_ih, gc_tensor col_offsets_hh, scalar scale_ih, scalar scale_hh, scalar zero_point_ih, scalar zero_point_hh);
+void atg_quantized_lstm_cell(raw_tensor *, gc_tensor input, gc_tensor *hx_data, int hx_len, gc_tensor w_ih, gc_tensor w_hh, gc_tensor b_ih, gc_tensor b_hh, gc_tensor packed_ih, gc_tensor packed_hh, gc_tensor col_offsets_ih, gc_tensor col_offsets_hh, scalar scale_ih, scalar scale_hh, scalar zero_point_ih, scalar zero_point_hh);
+raw_tensor atg_quantized_max_pool1d(gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
+raw_tensor atg_quantized_max_pool1d_out(gc_tensor out, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
+raw_tensor atg_quantized_max_pool2d(gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
+raw_tensor atg_quantized_max_pool2d_out(gc_tensor out, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
+raw_tensor atg_quantized_max_pool3d(gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
+raw_tensor atg_quantized_max_pool3d_out(gc_tensor out, gc_tensor self, int64_t *kernel_size_data, int kernel_size_len, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len, int ceil_mode);
+raw_tensor atg_quantized_rnn_relu_cell(gc_tensor input, gc_tensor hx, gc_tensor w_ih, gc_tensor w_hh, gc_tensor b_ih, gc_tensor b_hh, gc_tensor packed_ih, gc_tensor packed_hh, gc_tensor col_offsets_ih, gc_tensor col_offsets_hh, scalar scale_ih, scalar scale_hh, scalar zero_point_ih, scalar zero_point_hh);
+raw_tensor atg_quantized_rnn_tanh_cell(gc_tensor input, gc_tensor hx, gc_tensor w_ih, gc_tensor w_hh, gc_tensor b_ih, gc_tensor b_hh, gc_tensor packed_ih, gc_tensor packed_hh, gc_tensor col_offsets_ih, gc_tensor col_offsets_hh, scalar scale_ih, scalar scale_hh, scalar zero_point_ih, scalar zero_point_hh);
+raw_tensor atg_rad2deg(gc_tensor self);
+raw_tensor atg_rad2deg_(gc_tensor self);
+raw_tensor atg_rad2deg_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_rand(int64_t *size_data, int size_len, int options_kind, int options_device);
+raw_tensor atg_rand_like(gc_tensor self);
+raw_tensor atg_rand_like_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_rand_out(gc_tensor out, int64_t *size_data, int size_len);
+raw_tensor atg_randint(int64_t high, int64_t *size_data, int size_len, int options_kind, int options_device);
+raw_tensor atg_randint_like(gc_tensor self, int64_t high);
+raw_tensor atg_randint_like_low_dtype(gc_tensor self, int64_t low, int64_t high);
+raw_tensor atg_randint_like_low_dtype_out(gc_tensor out, gc_tensor self, int64_t low, int64_t high);
+raw_tensor atg_randint_like_out(gc_tensor out, gc_tensor self, int64_t high);
+raw_tensor atg_randint_low(int64_t low, int64_t high, int64_t *size_data, int size_len, int options_kind, int options_device);
+raw_tensor atg_randint_low_out(gc_tensor out, int64_t low, int64_t high, int64_t *size_data, int size_len);
+raw_tensor atg_randint_out(gc_tensor out, int64_t high, int64_t *size_data, int size_len);
+raw_tensor atg_randn(int64_t *size_data, int size_len, int options_kind, int options_device);
+raw_tensor atg_randn_like(gc_tensor self);
+raw_tensor atg_randn_like_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_randn_out(gc_tensor out, int64_t *size_data, int size_len);
+raw_tensor atg_random(gc_tensor self);
+raw_tensor atg_random_(gc_tensor self);
+raw_tensor atg_random_from(gc_tensor self, int64_t from, int64_t to_v, int to_null);
+raw_tensor atg_random_from_(gc_tensor self, int64_t from, int64_t to_v, int to_null);
+raw_tensor atg_random_from_out(gc_tensor out, gc_tensor self, int64_t from, int64_t to_v, int to_null);
+raw_tensor atg_random_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_random_to(gc_tensor self, int64_t to);
+raw_tensor atg_random_to_(gc_tensor self, int64_t to);
+raw_tensor atg_random_to_out(gc_tensor out, gc_tensor self, int64_t to);
+raw_tensor atg_randperm(int64_t n, int options_kind, int options_device);
+raw_tensor atg_randperm_out(gc_tensor out, int64_t n);
+raw_tensor atg_range(scalar start, scalar end, int options_kind, int options_device);
+raw_tensor atg_range_out(gc_tensor out, scalar start, scalar end);
+raw_tensor atg_range_out_(gc_tensor out, scalar start, scalar end);
+raw_tensor atg_range_step(scalar start, scalar end, int options_kind, int options_device);
+raw_tensor atg_ravel(gc_tensor self);
+raw_tensor atg_real(gc_tensor self);
+raw_tensor atg_reciprocal(gc_tensor self);
+raw_tensor atg_reciprocal_(gc_tensor self);
+raw_tensor atg_reciprocal_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_reflection_pad1d(gc_tensor self, int64_t *padding_data, int padding_len);
+raw_tensor atg_reflection_pad1d_backward(gc_tensor grad_output, gc_tensor self, int64_t *padding_data, int padding_len);
+raw_tensor atg_reflection_pad1d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, int64_t *padding_data, int padding_len);
+raw_tensor atg_reflection_pad1d_out(gc_tensor out, gc_tensor self, int64_t *padding_data, int padding_len);
+raw_tensor atg_reflection_pad2d(gc_tensor self, int64_t *padding_data, int padding_len);
+raw_tensor atg_reflection_pad2d_backward(gc_tensor grad_output, gc_tensor self, int64_t *padding_data, int padding_len);
+raw_tensor atg_reflection_pad2d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, int64_t *padding_data, int padding_len);
+raw_tensor atg_reflection_pad2d_out(gc_tensor out, gc_tensor self, int64_t *padding_data, int padding_len);
+raw_tensor atg_reflection_pad3d(gc_tensor self, int64_t *padding_data, int padding_len);
+raw_tensor atg_reflection_pad3d_backward(gc_tensor grad_output, gc_tensor self, int64_t *padding_data, int padding_len);
+raw_tensor atg_reflection_pad3d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, int64_t *padding_data, int padding_len);
+raw_tensor atg_reflection_pad3d_out(gc_tensor out, gc_tensor self, int64_t *padding_data, int padding_len);
+raw_tensor atg_relu(gc_tensor self);
+raw_tensor atg_relu6(gc_tensor self);
+raw_tensor atg_relu6_(gc_tensor self);
+raw_tensor atg_relu_(gc_tensor self);
+raw_tensor atg_relu_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_remainder(gc_tensor self, scalar other);
+raw_tensor atg_remainder_(gc_tensor self, scalar other);
+raw_tensor atg_remainder_scalar_out(gc_tensor out, gc_tensor self, scalar other);
+raw_tensor atg_remainder_scalar_tensor(scalar self, gc_tensor other);
+raw_tensor atg_remainder_scalar_tensor_out(gc_tensor out, scalar self, gc_tensor other);
+raw_tensor atg_remainder_tensor(gc_tensor self, gc_tensor other);
+raw_tensor atg_remainder_tensor_(gc_tensor self, gc_tensor other);
+raw_tensor atg_remainder_tensor_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_renorm(gc_tensor self, scalar p, int64_t dim, scalar maxnorm);
+raw_tensor atg_renorm_(gc_tensor self, scalar p, int64_t dim, scalar maxnorm);
+raw_tensor atg_renorm_out(gc_tensor out, gc_tensor self, scalar p, int64_t dim, scalar maxnorm);
+raw_tensor atg_repeat(gc_tensor self, int64_t *repeats_data, int repeats_len);
+raw_tensor atg_repeat_interleave(gc_tensor repeats, int64_t output_size_v, int output_size_null);
+raw_tensor atg_repeat_interleave_self_int(gc_tensor self, int64_t repeats, int64_t dim_v, int dim_null, int64_t output_size_v, int output_size_null);
+raw_tensor atg_repeat_interleave_self_tensor(gc_tensor self, gc_tensor repeats, int64_t dim_v, int dim_null, int64_t output_size_v, int output_size_null);
+raw_tensor atg_repeat_interleave_tensor_out(gc_tensor out, gc_tensor repeats, int64_t output_size_v, int output_size_null);
+raw_tensor atg_repeat_out(gc_tensor out, gc_tensor self, int64_t *repeats_data, int repeats_len);
+raw_tensor atg_replication_pad1d(gc_tensor self, int64_t *padding_data, int padding_len);
+raw_tensor atg_replication_pad1d_backward(gc_tensor grad_output, gc_tensor self, int64_t *padding_data, int padding_len);
+raw_tensor atg_replication_pad1d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, int64_t *padding_data, int padding_len);
+raw_tensor atg_replication_pad1d_out(gc_tensor out, gc_tensor self, int64_t *padding_data, int padding_len);
+raw_tensor atg_replication_pad2d(gc_tensor self, int64_t *padding_data, int padding_len);
+raw_tensor atg_replication_pad2d_backward(gc_tensor grad_output, gc_tensor self, int64_t *padding_data, int padding_len);
+raw_tensor atg_replication_pad2d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, int64_t *padding_data, int padding_len);
+raw_tensor atg_replication_pad2d_out(gc_tensor out, gc_tensor self, int64_t *padding_data, int padding_len);
+raw_tensor atg_replication_pad3d(gc_tensor self, int64_t *padding_data, int padding_len);
+raw_tensor atg_replication_pad3d_backward(gc_tensor grad_output, gc_tensor self, int64_t *padding_data, int padding_len);
+raw_tensor atg_replication_pad3d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, int64_t *padding_data, int padding_len);
+raw_tensor atg_replication_pad3d_out(gc_tensor out, gc_tensor self, int64_t *padding_data, int padding_len);
+raw_tensor atg_requires_grad_(gc_tensor self, int requires_grad);
+raw_tensor atg_reshape(gc_tensor self, int64_t *shape_data, int shape_len);
+raw_tensor atg_reshape_as(gc_tensor self, gc_tensor other);
+raw_tensor atg_resize(gc_tensor self, int64_t *size_data, int size_len);
+raw_tensor atg_resize_(gc_tensor self, int64_t *size_data, int size_len);
+raw_tensor atg_resize_as(gc_tensor self, gc_tensor the_template);
+raw_tensor atg_resize_as_(gc_tensor self, gc_tensor the_template);
+raw_tensor atg_resize_as_out(gc_tensor out, gc_tensor self, gc_tensor the_template);
+raw_tensor atg_resize_as_sparse(gc_tensor self, gc_tensor the_template);
+raw_tensor atg_resize_as_sparse_(gc_tensor self, gc_tensor the_template);
+raw_tensor atg_resize_as_sparse_out(gc_tensor out, gc_tensor self, gc_tensor the_template);
+raw_tensor atg_resize_out(gc_tensor out, gc_tensor self, int64_t *size_data, int size_len);
+raw_tensor atg_resolve_conj(gc_tensor self);
+raw_tensor atg_resolve_neg(gc_tensor self);
+int atg_retains_grad(gc_tensor self);
+void atg_rnn_relu(raw_tensor *, gc_tensor input, gc_tensor hx, gc_tensor *params_data, int params_len, int has_biases, int64_t num_layers, double dropout, int train, int bidirectional, int batch_first);
+raw_tensor atg_rnn_relu_cell(gc_tensor input, gc_tensor hx, gc_tensor w_ih, gc_tensor w_hh, gc_tensor b_ih, gc_tensor b_hh);
+void atg_rnn_relu_data(raw_tensor *, gc_tensor data, gc_tensor batch_sizes, gc_tensor hx, gc_tensor *params_data, int params_len, int has_biases, int64_t num_layers, double dropout, int train, int bidirectional);
+void atg_rnn_tanh(raw_tensor *, gc_tensor input, gc_tensor hx, gc_tensor *params_data, int params_len, int has_biases, int64_t num_layers, double dropout, int train, int bidirectional, int batch_first);
+raw_tensor atg_rnn_tanh_cell(gc_tensor input, gc_tensor hx, gc_tensor w_ih, gc_tensor w_hh, gc_tensor b_ih, gc_tensor b_hh);
+void atg_rnn_tanh_data(raw_tensor *, gc_tensor data, gc_tensor batch_sizes, gc_tensor hx, gc_tensor *params_data, int params_len, int has_biases, int64_t num_layers, double dropout, int train, int bidirectional);
+raw_tensor atg_roll(gc_tensor self, int64_t *shifts_data, int shifts_len, int64_t *dims_data, int dims_len);
+raw_tensor atg_roll_out(gc_tensor out, gc_tensor self, int64_t *shifts_data, int shifts_len, int64_t *dims_data, int dims_len);
+raw_tensor atg_rot90(gc_tensor self, int64_t k, int64_t *dims_data, int dims_len);
+raw_tensor atg_rot90_out(gc_tensor out, gc_tensor self, int64_t k, int64_t *dims_data, int dims_len);
+raw_tensor atg_round(gc_tensor self);
+raw_tensor atg_round_(gc_tensor self);
+raw_tensor atg_round_decimals(gc_tensor self, int64_t decimals);
+raw_tensor atg_round_decimals_(gc_tensor self, int64_t decimals);
+raw_tensor atg_round_decimals_out(gc_tensor out, gc_tensor self, int64_t decimals);
+raw_tensor atg_round_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_row_indices(gc_tensor self);
+raw_tensor atg_row_indices_copy(gc_tensor self);
+raw_tensor atg_row_indices_copy_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_row_stack(gc_tensor *tensors_data, int tensors_len);
+raw_tensor atg_row_stack_out(gc_tensor out, gc_tensor *tensors_data, int tensors_len);
+raw_tensor atg_rrelu(gc_tensor self, int training);
+raw_tensor atg_rrelu_(gc_tensor self, int training);
+raw_tensor atg_rrelu_with_noise(gc_tensor self, gc_tensor noise, int training);
+raw_tensor atg_rrelu_with_noise_(gc_tensor self, gc_tensor noise, int training);
+raw_tensor atg_rrelu_with_noise_backward(gc_tensor grad_output, gc_tensor self, gc_tensor noise, scalar lower, scalar upper, int training, int self_is_result);
+raw_tensor atg_rrelu_with_noise_backward_out(gc_tensor out, gc_tensor grad_output, gc_tensor self, gc_tensor noise, scalar lower, scalar upper, int training, int self_is_result);
+raw_tensor atg_rrelu_with_noise_out(gc_tensor out, gc_tensor self, gc_tensor noise, int training);
+raw_tensor atg_rsqrt(gc_tensor self);
+raw_tensor atg_rsqrt_(gc_tensor self);
+raw_tensor atg_rsqrt_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_rsub(gc_tensor self, gc_tensor other);
+raw_tensor atg_rsub_scalar(gc_tensor self, scalar other);
+raw_tensor atg_rsub_scalar_out(gc_tensor out, gc_tensor self, scalar other);
+raw_tensor atg_rsub_tensor_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_scalar_tensor(scalar s, int options_kind, int options_device);
+raw_tensor atg_scalar_tensor_out(gc_tensor out, scalar s);
+raw_tensor atg_scaled_dot_product_attention(gc_tensor query, gc_tensor key, gc_tensor value, gc_tensor attn_mask, double dropout_p, int is_causal, double scale_v, int scale_null);
+raw_tensor atg_scatter(gc_tensor self, int64_t dim, gc_tensor index, gc_tensor src);
+raw_tensor atg_scatter_(gc_tensor self, int64_t dim, gc_tensor index, gc_tensor src);
+raw_tensor atg_scatter_add(gc_tensor self, int64_t dim, gc_tensor index, gc_tensor src);
+raw_tensor atg_scatter_add_(gc_tensor self, int64_t dim, gc_tensor index, gc_tensor src);
+raw_tensor atg_scatter_add_out(gc_tensor out, gc_tensor self, int64_t dim, gc_tensor index, gc_tensor src);
+raw_tensor atg_scatter_reduce(gc_tensor self, int64_t dim, gc_tensor index, gc_tensor src, char * reduce);
+raw_tensor atg_scatter_reduce_(gc_tensor self, int64_t dim, gc_tensor index, gc_tensor src, char * reduce);
+raw_tensor atg_scatter_reduce_out(gc_tensor out, gc_tensor self, int64_t dim, gc_tensor index, gc_tensor src, char * reduce);
+raw_tensor atg_scatter_src_out(gc_tensor out, gc_tensor self, int64_t dim, gc_tensor index, gc_tensor src);
+raw_tensor atg_scatter_value(gc_tensor self, int64_t dim, gc_tensor index, scalar value);
+raw_tensor atg_scatter_value_(gc_tensor self, int64_t dim, gc_tensor index, scalar value);
+raw_tensor atg_scatter_value_out(gc_tensor out, gc_tensor self, int64_t dim, gc_tensor index, scalar value);
+raw_tensor atg_scatter_value_reduce(gc_tensor self, int64_t dim, gc_tensor index, scalar value, char * reduce);
+raw_tensor atg_scatter_value_reduce_(gc_tensor self, int64_t dim, gc_tensor index, scalar value, char * reduce);
+raw_tensor atg_scatter_value_reduce_out(gc_tensor out, gc_tensor self, int64_t dim, gc_tensor index, scalar value, char * reduce);
+raw_tensor atg_searchsorted(gc_tensor sorted_sequence, gc_tensor self, int out_int32, int right, char * side, gc_tensor sorter);
+raw_tensor atg_searchsorted_scalar(gc_tensor sorted_sequence, scalar self, int out_int32, int right, char * side, gc_tensor sorter);
+raw_tensor atg_searchsorted_scalar_out(gc_tensor out, gc_tensor sorted_sequence, scalar self, int out_int32, int right, char * side, gc_tensor sorter);
+raw_tensor atg_searchsorted_tensor_out(gc_tensor out, gc_tensor sorted_sequence, gc_tensor self, int out_int32, int right, char * side, gc_tensor sorter);
+raw_tensor atg_segment_reduce(gc_tensor data, char * reduce, gc_tensor lengths, gc_tensor indices, gc_tensor offsets, int64_t axis, int unsafe, scalar initial);
+raw_tensor atg_segment_reduce_out(gc_tensor out, gc_tensor data, char * reduce, gc_tensor lengths, gc_tensor indices, gc_tensor offsets, int64_t axis, int unsafe, scalar initial);
+raw_tensor atg_select(gc_tensor self, int64_t dim, int64_t index);
+raw_tensor atg_select_backward(gc_tensor grad_output, int64_t *input_sizes_data, int input_sizes_len, int64_t dim, int64_t index);
+raw_tensor atg_select_backward_out(gc_tensor out, gc_tensor grad_output, int64_t *input_sizes_data, int input_sizes_len, int64_t dim, int64_t index);
+raw_tensor atg_select_copy(gc_tensor self, int64_t dim, int64_t index);
+raw_tensor atg_select_copy_int_out(gc_tensor out, gc_tensor self, int64_t dim, int64_t index);
+raw_tensor atg_select_scatter(gc_tensor self, gc_tensor src, int64_t dim, int64_t index);
+raw_tensor atg_select_scatter_out(gc_tensor out, gc_tensor self, gc_tensor src, int64_t dim, int64_t index);
+raw_tensor atg_selu(gc_tensor self);
+raw_tensor atg_selu_(gc_tensor self);
+raw_tensor atg_set(gc_tensor self);
+raw_tensor atg_set_(gc_tensor self);
+raw_tensor atg_set_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_set_requires_grad(gc_tensor self, int r);
+raw_tensor atg_set_source_tensor(gc_tensor self, gc_tensor source);
+raw_tensor atg_set_source_tensor_(gc_tensor self, gc_tensor source);
+raw_tensor atg_set_source_tensor_out(gc_tensor out, gc_tensor self, gc_tensor source);
+raw_tensor atg_set_source_tensor_storage_offset_(gc_tensor self, gc_tensor source, int64_t storage_offset, int64_t *size_data, int size_len, int64_t *stride_data, int stride_len);
+raw_tensor atg_sgn(gc_tensor self);
+raw_tensor atg_sgn_(gc_tensor self);
+raw_tensor atg_sgn_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_sigmoid(gc_tensor self);
+raw_tensor atg_sigmoid_(gc_tensor self);
+raw_tensor atg_sigmoid_backward(gc_tensor grad_output, gc_tensor output);
+raw_tensor atg_sigmoid_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor output);
+raw_tensor atg_sigmoid_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_sign(gc_tensor self);
+raw_tensor atg_sign_(gc_tensor self);
+raw_tensor atg_sign_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_signbit(gc_tensor self);
+raw_tensor atg_signbit_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_silu(gc_tensor self);
+raw_tensor atg_silu_(gc_tensor self);
+raw_tensor atg_silu_backward(gc_tensor grad_output, gc_tensor self);
+raw_tensor atg_silu_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self);
+raw_tensor atg_silu_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_sin(gc_tensor self);
+raw_tensor atg_sin_(gc_tensor self);
+raw_tensor atg_sin_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_sinc(gc_tensor self);
+raw_tensor atg_sinc_(gc_tensor self);
+raw_tensor atg_sinc_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_sinh(gc_tensor self);
+raw_tensor atg_sinh_(gc_tensor self);
+raw_tensor atg_sinh_out(gc_tensor out, gc_tensor self);
+int64_t atg_size(gc_tensor self, int64_t dim);
+raw_tensor atg_slice(gc_tensor self, int64_t dim, int64_t start_v, int start_null, int64_t end_v, int end_null, int64_t step);
+raw_tensor atg_slice_backward(gc_tensor grad_output, int64_t *input_sizes_data, int input_sizes_len, int64_t dim, int64_t start, int64_t end, int64_t step);
+raw_tensor atg_slice_backward_out(gc_tensor out, gc_tensor grad_output, int64_t *input_sizes_data, int input_sizes_len, int64_t dim, int64_t start, int64_t end, int64_t step);
+raw_tensor atg_slice_copy(gc_tensor self, int64_t dim, int64_t start_v, int start_null, int64_t end_v, int end_null, int64_t step);
+raw_tensor atg_slice_copy_tensor_out(gc_tensor out, gc_tensor self, int64_t dim, int64_t start_v, int start_null, int64_t end_v, int end_null, int64_t step);
+raw_tensor atg_slice_scatter(gc_tensor self, gc_tensor src, int64_t dim, int64_t start_v, int start_null, int64_t end_v, int end_null, int64_t step);
+raw_tensor atg_slice_scatter_out(gc_tensor out, gc_tensor self, gc_tensor src, int64_t dim, int64_t start_v, int start_null, int64_t end_v, int end_null, int64_t step);
+void atg_slogdet(raw_tensor *, gc_tensor self);
+void atg_slogdet_out(raw_tensor *, gc_tensor sign, gc_tensor logabsdet, gc_tensor self);
+raw_tensor atg_slow_conv3d(gc_tensor self, gc_tensor weight, int64_t *kernel_size_data, int kernel_size_len, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len);
+raw_tensor atg_slow_conv3d_out(gc_tensor out, gc_tensor self, gc_tensor weight, int64_t *kernel_size_data, int kernel_size_len, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len);
+raw_tensor atg_slow_conv_dilated2d(gc_tensor self, gc_tensor weight, int64_t *kernel_size_data, int kernel_size_len, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len);
+raw_tensor atg_slow_conv_dilated2d_out(gc_tensor out, gc_tensor self, gc_tensor weight, int64_t *kernel_size_data, int kernel_size_len, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len);
+raw_tensor atg_slow_conv_dilated3d(gc_tensor self, gc_tensor weight, int64_t *kernel_size_data, int kernel_size_len, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len);
+raw_tensor atg_slow_conv_dilated3d_out(gc_tensor out, gc_tensor self, gc_tensor weight, int64_t *kernel_size_data, int kernel_size_len, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *dilation_data, int dilation_len);
+raw_tensor atg_slow_conv_transpose2d(gc_tensor self, gc_tensor weight, int64_t *kernel_size_data, int kernel_size_len, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t *dilation_data, int dilation_len);
+raw_tensor atg_slow_conv_transpose2d_out(gc_tensor out, gc_tensor self, gc_tensor weight, int64_t *kernel_size_data, int kernel_size_len, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t *dilation_data, int dilation_len);
+raw_tensor atg_slow_conv_transpose3d(gc_tensor self, gc_tensor weight, int64_t *kernel_size_data, int kernel_size_len, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t *dilation_data, int dilation_len);
+raw_tensor atg_slow_conv_transpose3d_out(gc_tensor out, gc_tensor self, gc_tensor weight, int64_t *kernel_size_data, int kernel_size_len, gc_tensor bias, int64_t *stride_data, int stride_len, int64_t *padding_data, int padding_len, int64_t *output_padding_data, int output_padding_len, int64_t *dilation_data, int dilation_len);
+raw_tensor atg_smm(gc_tensor self, gc_tensor mat2);
+raw_tensor atg_smooth_l1_loss(gc_tensor self, gc_tensor target, int64_t reduction, double beta);
+raw_tensor atg_smooth_l1_loss_backward(gc_tensor grad_output, gc_tensor self, gc_tensor target, int64_t reduction, double beta);
+raw_tensor atg_smooth_l1_loss_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, gc_tensor target, int64_t reduction, double beta);
+raw_tensor atg_smooth_l1_loss_out(gc_tensor out, gc_tensor self, gc_tensor target, int64_t reduction, double beta);
+raw_tensor atg_soft_margin_loss(gc_tensor self, gc_tensor target, int64_t reduction);
+raw_tensor atg_soft_margin_loss_backward(gc_tensor grad_output, gc_tensor self, gc_tensor target, int64_t reduction);
+raw_tensor atg_soft_margin_loss_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, gc_tensor target, int64_t reduction);
+raw_tensor atg_soft_margin_loss_out(gc_tensor out, gc_tensor self, gc_tensor target, int64_t reduction);
+raw_tensor atg_softmax(gc_tensor self, int64_t dim, int dtype);
+raw_tensor atg_softmax_int_out(gc_tensor out, gc_tensor self, int64_t dim, int dtype);
+raw_tensor atg_softplus(gc_tensor self);
+raw_tensor atg_softplus_backward(gc_tensor grad_output, gc_tensor self, scalar beta, scalar threshold);
+raw_tensor atg_softplus_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, scalar beta, scalar threshold);
+raw_tensor atg_softplus_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_softshrink(gc_tensor self);
+raw_tensor atg_softshrink_backward(gc_tensor grad_output, gc_tensor self, scalar lambd);
+raw_tensor atg_softshrink_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, scalar lambd);
+raw_tensor atg_softshrink_out(gc_tensor out, gc_tensor self);
+void atg_sort(raw_tensor *, gc_tensor self, int64_t dim, int descending);
+void atg_sort_stable(raw_tensor *, gc_tensor self, int stable, int64_t dim, int descending);
+void atg_sort_values(raw_tensor *, gc_tensor values, gc_tensor indices, gc_tensor self, int64_t dim, int descending);
+void atg_sort_values_stable(raw_tensor *, gc_tensor values, gc_tensor indices, gc_tensor self, int stable, int64_t dim, int descending);
+raw_tensor atg_sparse_bsc_tensor(gc_tensor ccol_indices, gc_tensor row_indices, gc_tensor values, int options_kind, int options_device);
+raw_tensor atg_sparse_bsc_tensor_ccol_row_value_size(gc_tensor ccol_indices, gc_tensor row_indices, gc_tensor values, int64_t *size_data, int size_len, int options_kind, int options_device);
+raw_tensor atg_sparse_bsr_tensor(gc_tensor crow_indices, gc_tensor col_indices, gc_tensor values, int options_kind, int options_device);
+raw_tensor atg_sparse_bsr_tensor_crow_col_value_size(gc_tensor crow_indices, gc_tensor col_indices, gc_tensor values, int64_t *size_data, int size_len, int options_kind, int options_device);
+raw_tensor atg_sparse_compressed_tensor(gc_tensor compressed_indices, gc_tensor plain_indices, gc_tensor values, int options_kind, int options_device);
+raw_tensor atg_sparse_compressed_tensor_comp_plain_value_size(gc_tensor compressed_indices, gc_tensor plain_indices, gc_tensor values, int64_t *size_data, int size_len, int options_kind, int options_device);
+raw_tensor atg_sparse_coo_tensor(int64_t *size_data, int size_len, int options_kind, int options_device);
+raw_tensor atg_sparse_coo_tensor_indices(gc_tensor indices, gc_tensor values, int options_kind, int options_device, int is_coalesced);
+raw_tensor atg_sparse_coo_tensor_indices_size(gc_tensor indices, gc_tensor values, int64_t *size_data, int size_len, int options_kind, int options_device, int is_coalesced);
+raw_tensor atg_sparse_coo_tensor_size_out(gc_tensor out, int64_t *size_data, int size_len);
+raw_tensor atg_sparse_csc_tensor(gc_tensor ccol_indices, gc_tensor row_indices, gc_tensor values, int options_kind, int options_device);
+raw_tensor atg_sparse_csc_tensor_ccol_row_value_size(gc_tensor ccol_indices, gc_tensor row_indices, gc_tensor values, int64_t *size_data, int size_len, int options_kind, int options_device);
+raw_tensor atg_sparse_csr_tensor(gc_tensor crow_indices, gc_tensor col_indices, gc_tensor values, int options_kind, int options_device);
+raw_tensor atg_sparse_csr_tensor_crow_col_value_size(gc_tensor crow_indices, gc_tensor col_indices, gc_tensor values, int64_t *size_data, int size_len, int options_kind, int options_device);
+int64_t atg_sparse_dim(gc_tensor self);
+raw_tensor atg_sparse_mask(gc_tensor self, gc_tensor mask);
+raw_tensor atg_sparse_mask_out(gc_tensor out, gc_tensor self, gc_tensor mask);
+raw_tensor atg_sparse_resize(gc_tensor self, int64_t *size_data, int size_len, int64_t sparse_dim, int64_t dense_dim);
+raw_tensor atg_sparse_resize_(gc_tensor self, int64_t *size_data, int size_len, int64_t sparse_dim, int64_t dense_dim);
+raw_tensor atg_sparse_resize_and_clear(gc_tensor self, int64_t *size_data, int size_len, int64_t sparse_dim, int64_t dense_dim);
+raw_tensor atg_sparse_resize_and_clear_(gc_tensor self, int64_t *size_data, int size_len, int64_t sparse_dim, int64_t dense_dim);
+raw_tensor atg_sparse_resize_and_clear_out(gc_tensor out, gc_tensor self, int64_t *size_data, int size_len, int64_t sparse_dim, int64_t dense_dim);
+raw_tensor atg_sparse_resize_out(gc_tensor out, gc_tensor self, int64_t *size_data, int size_len, int64_t sparse_dim, int64_t dense_dim);
+raw_tensor atg_sparse_sampled_addmm(gc_tensor self, gc_tensor mat1, gc_tensor mat2);
+raw_tensor atg_sparse_sampled_addmm_out(gc_tensor out, gc_tensor self, gc_tensor mat1, gc_tensor mat2);
+raw_tensor atg_special_airy_ai(gc_tensor x);
+raw_tensor atg_special_airy_ai_out(gc_tensor out, gc_tensor x);
+raw_tensor atg_special_bessel_j0(gc_tensor self);
+raw_tensor atg_special_bessel_j0_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_special_bessel_j1(gc_tensor self);
+raw_tensor atg_special_bessel_j1_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_special_bessel_y0(gc_tensor self);
+raw_tensor atg_special_bessel_y0_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_special_bessel_y1(gc_tensor self);
+raw_tensor atg_special_bessel_y1_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_special_chebyshev_polynomial_t(gc_tensor x, gc_tensor n);
+raw_tensor atg_special_chebyshev_polynomial_t_n_scalar(gc_tensor x, scalar n);
+raw_tensor atg_special_chebyshev_polynomial_t_n_scalar_out(gc_tensor out, gc_tensor x, scalar n);
+raw_tensor atg_special_chebyshev_polynomial_t_out(gc_tensor out, gc_tensor x, gc_tensor n);
+raw_tensor atg_special_chebyshev_polynomial_t_x_scalar(scalar x, gc_tensor n);
+raw_tensor atg_special_chebyshev_polynomial_t_x_scalar_out(gc_tensor out, scalar x, gc_tensor n);
+raw_tensor atg_special_chebyshev_polynomial_u(gc_tensor x, gc_tensor n);
+raw_tensor atg_special_chebyshev_polynomial_u_n_scalar(gc_tensor x, scalar n);
+raw_tensor atg_special_chebyshev_polynomial_u_n_scalar_out(gc_tensor out, gc_tensor x, scalar n);
+raw_tensor atg_special_chebyshev_polynomial_u_out(gc_tensor out, gc_tensor x, gc_tensor n);
+raw_tensor atg_special_chebyshev_polynomial_u_x_scalar(scalar x, gc_tensor n);
+raw_tensor atg_special_chebyshev_polynomial_u_x_scalar_out(gc_tensor out, scalar x, gc_tensor n);
+raw_tensor atg_special_chebyshev_polynomial_v(gc_tensor x, gc_tensor n);
+raw_tensor atg_special_chebyshev_polynomial_v_n_scalar(gc_tensor x, scalar n);
+raw_tensor atg_special_chebyshev_polynomial_v_n_scalar_out(gc_tensor out, gc_tensor x, scalar n);
+raw_tensor atg_special_chebyshev_polynomial_v_out(gc_tensor out, gc_tensor x, gc_tensor n);
+raw_tensor atg_special_chebyshev_polynomial_v_x_scalar(scalar x, gc_tensor n);
+raw_tensor atg_special_chebyshev_polynomial_v_x_scalar_out(gc_tensor out, scalar x, gc_tensor n);
+raw_tensor atg_special_chebyshev_polynomial_w(gc_tensor x, gc_tensor n);
+raw_tensor atg_special_chebyshev_polynomial_w_n_scalar(gc_tensor x, scalar n);
+raw_tensor atg_special_chebyshev_polynomial_w_n_scalar_out(gc_tensor out, gc_tensor x, scalar n);
+raw_tensor atg_special_chebyshev_polynomial_w_out(gc_tensor out, gc_tensor x, gc_tensor n);
+raw_tensor atg_special_chebyshev_polynomial_w_x_scalar(scalar x, gc_tensor n);
+raw_tensor atg_special_chebyshev_polynomial_w_x_scalar_out(gc_tensor out, scalar x, gc_tensor n);
+raw_tensor atg_special_digamma(gc_tensor self);
+raw_tensor atg_special_digamma_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_special_entr(gc_tensor self);
+raw_tensor atg_special_entr_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_special_erf(gc_tensor self);
+raw_tensor atg_special_erf_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_special_erfc(gc_tensor self);
+raw_tensor atg_special_erfc_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_special_erfcx(gc_tensor self);
+raw_tensor atg_special_erfcx_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_special_erfinv(gc_tensor self);
+raw_tensor atg_special_erfinv_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_special_exp2(gc_tensor self);
+raw_tensor atg_special_exp2_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_special_expit(gc_tensor self);
+raw_tensor atg_special_expit_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_special_expm1(gc_tensor self);
+raw_tensor atg_special_expm1_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_special_gammainc(gc_tensor self, gc_tensor other);
+raw_tensor atg_special_gammainc_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_special_gammaincc(gc_tensor self, gc_tensor other);
+raw_tensor atg_special_gammaincc_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_special_gammaln(gc_tensor self);
+raw_tensor atg_special_gammaln_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_special_hermite_polynomial_h(gc_tensor x, gc_tensor n);
+raw_tensor atg_special_hermite_polynomial_h_n_scalar(gc_tensor x, scalar n);
+raw_tensor atg_special_hermite_polynomial_h_n_scalar_out(gc_tensor out, gc_tensor x, scalar n);
+raw_tensor atg_special_hermite_polynomial_h_out(gc_tensor out, gc_tensor x, gc_tensor n);
+raw_tensor atg_special_hermite_polynomial_h_x_scalar(scalar x, gc_tensor n);
+raw_tensor atg_special_hermite_polynomial_h_x_scalar_out(gc_tensor out, scalar x, gc_tensor n);
+raw_tensor atg_special_hermite_polynomial_he(gc_tensor x, gc_tensor n);
+raw_tensor atg_special_hermite_polynomial_he_n_scalar(gc_tensor x, scalar n);
+raw_tensor atg_special_hermite_polynomial_he_n_scalar_out(gc_tensor out, gc_tensor x, scalar n);
+raw_tensor atg_special_hermite_polynomial_he_out(gc_tensor out, gc_tensor x, gc_tensor n);
+raw_tensor atg_special_hermite_polynomial_he_x_scalar(scalar x, gc_tensor n);
+raw_tensor atg_special_hermite_polynomial_he_x_scalar_out(gc_tensor out, scalar x, gc_tensor n);
+raw_tensor atg_special_i0(gc_tensor self);
+raw_tensor atg_special_i0_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_special_i0e(gc_tensor self);
+raw_tensor atg_special_i0e_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_special_i1(gc_tensor self);
+raw_tensor atg_special_i1_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_special_i1e(gc_tensor self);
+raw_tensor atg_special_i1e_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_special_laguerre_polynomial_l(gc_tensor x, gc_tensor n);
+raw_tensor atg_special_laguerre_polynomial_l_n_scalar(gc_tensor x, scalar n);
+raw_tensor atg_special_laguerre_polynomial_l_n_scalar_out(gc_tensor out, gc_tensor x, scalar n);
+raw_tensor atg_special_laguerre_polynomial_l_out(gc_tensor out, gc_tensor x, gc_tensor n);
+raw_tensor atg_special_laguerre_polynomial_l_x_scalar(scalar x, gc_tensor n);
+raw_tensor atg_special_laguerre_polynomial_l_x_scalar_out(gc_tensor out, scalar x, gc_tensor n);
+raw_tensor atg_special_legendre_polynomial_p(gc_tensor x, gc_tensor n);
+raw_tensor atg_special_legendre_polynomial_p_n_scalar(gc_tensor x, scalar n);
+raw_tensor atg_special_legendre_polynomial_p_n_scalar_out(gc_tensor out, gc_tensor x, scalar n);
+raw_tensor atg_special_legendre_polynomial_p_out(gc_tensor out, gc_tensor x, gc_tensor n);
+raw_tensor atg_special_legendre_polynomial_p_x_scalar(scalar x, gc_tensor n);
+raw_tensor atg_special_legendre_polynomial_p_x_scalar_out(gc_tensor out, scalar x, gc_tensor n);
+raw_tensor atg_special_log1p(gc_tensor self);
+raw_tensor atg_special_log1p_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_special_log_ndtr(gc_tensor self);
+raw_tensor atg_special_log_ndtr_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_special_log_softmax(gc_tensor self, int64_t dim, int dtype);
+raw_tensor atg_special_logit(gc_tensor self, double eps_v, int eps_null);
+raw_tensor atg_special_logit_out(gc_tensor out, gc_tensor self, double eps_v, int eps_null);
+raw_tensor atg_special_logsumexp(gc_tensor self, int64_t *dim_data, int dim_len, int keepdim);
+raw_tensor atg_special_logsumexp_out(gc_tensor out, gc_tensor self, int64_t *dim_data, int dim_len, int keepdim);
+raw_tensor atg_special_modified_bessel_i0(gc_tensor self);
+raw_tensor atg_special_modified_bessel_i0_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_special_modified_bessel_i1(gc_tensor self);
+raw_tensor atg_special_modified_bessel_i1_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_special_modified_bessel_k0(gc_tensor self);
+raw_tensor atg_special_modified_bessel_k0_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_special_modified_bessel_k1(gc_tensor self);
+raw_tensor atg_special_modified_bessel_k1_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_special_multigammaln(gc_tensor self, int64_t p);
+raw_tensor atg_special_multigammaln_out(gc_tensor out, gc_tensor self, int64_t p);
+raw_tensor atg_special_ndtr(gc_tensor self);
+raw_tensor atg_special_ndtr_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_special_ndtri(gc_tensor self);
+raw_tensor atg_special_ndtri_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_special_polygamma(int64_t n, gc_tensor self);
+raw_tensor atg_special_polygamma_out(gc_tensor out, int64_t n, gc_tensor self);
+raw_tensor atg_special_psi(gc_tensor self);
+raw_tensor atg_special_psi_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_special_round(gc_tensor self, int64_t decimals);
+raw_tensor atg_special_round_out(gc_tensor out, gc_tensor self, int64_t decimals);
+raw_tensor atg_special_scaled_modified_bessel_k0(gc_tensor x);
+raw_tensor atg_special_scaled_modified_bessel_k0_out(gc_tensor out, gc_tensor x);
+raw_tensor atg_special_scaled_modified_bessel_k1(gc_tensor x);
+raw_tensor atg_special_scaled_modified_bessel_k1_out(gc_tensor out, gc_tensor x);
+raw_tensor atg_special_shifted_chebyshev_polynomial_t(gc_tensor x, gc_tensor n);
+raw_tensor atg_special_shifted_chebyshev_polynomial_t_n_scalar(gc_tensor x, scalar n);
+raw_tensor atg_special_shifted_chebyshev_polynomial_t_n_scalar_out(gc_tensor out, gc_tensor x, scalar n);
+raw_tensor atg_special_shifted_chebyshev_polynomial_t_out(gc_tensor out, gc_tensor x, gc_tensor n);
+raw_tensor atg_special_shifted_chebyshev_polynomial_t_x_scalar(scalar x, gc_tensor n);
+raw_tensor atg_special_shifted_chebyshev_polynomial_t_x_scalar_out(gc_tensor out, scalar x, gc_tensor n);
+raw_tensor atg_special_shifted_chebyshev_polynomial_u(gc_tensor x, gc_tensor n);
+raw_tensor atg_special_shifted_chebyshev_polynomial_u_n_scalar(gc_tensor x, scalar n);
+raw_tensor atg_special_shifted_chebyshev_polynomial_u_n_scalar_out(gc_tensor out, gc_tensor x, scalar n);
+raw_tensor atg_special_shifted_chebyshev_polynomial_u_out(gc_tensor out, gc_tensor x, gc_tensor n);
+raw_tensor atg_special_shifted_chebyshev_polynomial_u_x_scalar(scalar x, gc_tensor n);
+raw_tensor atg_special_shifted_chebyshev_polynomial_u_x_scalar_out(gc_tensor out, scalar x, gc_tensor n);
+raw_tensor atg_special_shifted_chebyshev_polynomial_v(gc_tensor x, gc_tensor n);
+raw_tensor atg_special_shifted_chebyshev_polynomial_v_n_scalar(gc_tensor x, scalar n);
+raw_tensor atg_special_shifted_chebyshev_polynomial_v_n_scalar_out(gc_tensor out, gc_tensor x, scalar n);
+raw_tensor atg_special_shifted_chebyshev_polynomial_v_out(gc_tensor out, gc_tensor x, gc_tensor n);
+raw_tensor atg_special_shifted_chebyshev_polynomial_v_x_scalar(scalar x, gc_tensor n);
+raw_tensor atg_special_shifted_chebyshev_polynomial_v_x_scalar_out(gc_tensor out, scalar x, gc_tensor n);
+raw_tensor atg_special_shifted_chebyshev_polynomial_w(gc_tensor x, gc_tensor n);
+raw_tensor atg_special_shifted_chebyshev_polynomial_w_n_scalar(gc_tensor x, scalar n);
+raw_tensor atg_special_shifted_chebyshev_polynomial_w_n_scalar_out(gc_tensor out, gc_tensor x, scalar n);
+raw_tensor atg_special_shifted_chebyshev_polynomial_w_out(gc_tensor out, gc_tensor x, gc_tensor n);
+raw_tensor atg_special_shifted_chebyshev_polynomial_w_x_scalar(scalar x, gc_tensor n);
+raw_tensor atg_special_shifted_chebyshev_polynomial_w_x_scalar_out(gc_tensor out, scalar x, gc_tensor n);
+raw_tensor atg_special_sinc(gc_tensor self);
+raw_tensor atg_special_sinc_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_special_softmax(gc_tensor self, int64_t dim, int dtype);
+raw_tensor atg_special_spherical_bessel_j0(gc_tensor x);
+raw_tensor atg_special_spherical_bessel_j0_out(gc_tensor out, gc_tensor x);
+raw_tensor atg_special_xlog1py(gc_tensor self, gc_tensor other);
+raw_tensor atg_special_xlog1py_other_scalar(gc_tensor self, scalar other);
+raw_tensor atg_special_xlog1py_other_scalar_out(gc_tensor out, gc_tensor self, scalar other);
+raw_tensor atg_special_xlog1py_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_special_xlog1py_self_scalar(scalar self, gc_tensor other);
+raw_tensor atg_special_xlog1py_self_scalar_out(gc_tensor out, scalar self, gc_tensor other);
+raw_tensor atg_special_xlogy(gc_tensor self, gc_tensor other);
+raw_tensor atg_special_xlogy_other_scalar(gc_tensor self, scalar other);
+raw_tensor atg_special_xlogy_other_scalar_out(gc_tensor out, gc_tensor self, scalar other);
+raw_tensor atg_special_xlogy_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_special_xlogy_self_scalar(scalar self, gc_tensor other);
+raw_tensor atg_special_xlogy_self_scalar_out(gc_tensor out, scalar self, gc_tensor other);
+raw_tensor atg_special_zeta(gc_tensor self, gc_tensor other);
+raw_tensor atg_special_zeta_other_scalar(gc_tensor self, scalar other);
+raw_tensor atg_special_zeta_other_scalar_out(gc_tensor out, gc_tensor self, scalar other);
+raw_tensor atg_special_zeta_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_special_zeta_self_scalar(scalar self, gc_tensor other);
+raw_tensor atg_special_zeta_self_scalar_out(gc_tensor out, scalar self, gc_tensor other);
+raw_tensor *atg_split(gc_tensor self, int64_t split_size, int64_t dim);
+raw_tensor *atg_split_copy(gc_tensor self, int64_t split_size, int64_t dim);
+void atg_split_copy_tensor_out(gc_tensor *out_data, int out_len, gc_tensor self, int64_t split_size, int64_t dim);
+raw_tensor *atg_split_sizes(gc_tensor self, int64_t *split_size_data, int split_size_len, int64_t dim);
+raw_tensor *atg_split_with_sizes(gc_tensor self, int64_t *split_sizes_data, int split_sizes_len, int64_t dim);
+raw_tensor *atg_split_with_sizes_copy(gc_tensor self, int64_t *split_sizes_data, int split_sizes_len, int64_t dim);
+void atg_split_with_sizes_copy_out(gc_tensor *out_data, int out_len, gc_tensor self, int64_t *split_sizes_data, int split_sizes_len, int64_t dim);
+raw_tensor atg_sqrt(gc_tensor self);
+raw_tensor atg_sqrt_(gc_tensor self);
+raw_tensor atg_sqrt_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_square(gc_tensor self);
+raw_tensor atg_square_(gc_tensor self);
+raw_tensor atg_square_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_squeeze(gc_tensor self);
+raw_tensor atg_squeeze_(gc_tensor self);
+raw_tensor atg_squeeze_copy(gc_tensor self);
+raw_tensor atg_squeeze_copy_dim(gc_tensor self, int64_t dim);
+raw_tensor atg_squeeze_copy_dim_out(gc_tensor out, gc_tensor self, int64_t dim);
+raw_tensor atg_squeeze_copy_dims(gc_tensor self, int64_t *dim_data, int dim_len);
+raw_tensor atg_squeeze_copy_dims_out(gc_tensor out, gc_tensor self, int64_t *dim_data, int dim_len);
+raw_tensor atg_squeeze_copy_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_squeeze_dim(gc_tensor self, int64_t dim);
+raw_tensor atg_squeeze_dim_(gc_tensor self, int64_t dim);
+raw_tensor atg_squeeze_dims(gc_tensor self, int64_t *dim_data, int dim_len);
+raw_tensor atg_squeeze_dims_(gc_tensor self, int64_t *dim_data, int dim_len);
+raw_tensor atg_sspaddmm(gc_tensor self, gc_tensor mat1, gc_tensor mat2);
+raw_tensor atg_sspaddmm_out(gc_tensor out, gc_tensor self, gc_tensor mat1, gc_tensor mat2);
+raw_tensor atg_stack(gc_tensor *tensors_data, int tensors_len, int64_t dim);
+raw_tensor atg_stack_out(gc_tensor out, gc_tensor *tensors_data, int tensors_len, int64_t dim);
+raw_tensor atg_std(gc_tensor self, int unbiased);
+raw_tensor atg_std_correction(gc_tensor self, int64_t *dim_data, int dim_len, scalar correction, int keepdim);
+raw_tensor atg_std_correction_out(gc_tensor out, gc_tensor self, int64_t *dim_data, int dim_len, scalar correction, int keepdim);
+raw_tensor atg_std_dim(gc_tensor self, int64_t *dim_data, int dim_len, int unbiased, int keepdim);
+void atg_std_mean(raw_tensor *, gc_tensor self, int unbiased);
+void atg_std_mean_correction(raw_tensor *, gc_tensor self, int64_t *dim_data, int dim_len, scalar correction, int keepdim);
+void atg_std_mean_correction_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor self, int64_t *dim_data, int dim_len, scalar correction, int keepdim);
+void atg_std_mean_dim(raw_tensor *, gc_tensor self, int64_t *dim_data, int dim_len, int unbiased, int keepdim);
+raw_tensor atg_std_out(gc_tensor out, gc_tensor self, int64_t *dim_data, int dim_len, int unbiased, int keepdim);
+raw_tensor atg_stft(gc_tensor self, int64_t n_fft, int64_t hop_length_v, int hop_length_null, int64_t win_length_v, int win_length_null, gc_tensor window, int normalized, int onesided, int return_complex);
+raw_tensor atg_stft_center(gc_tensor self, int64_t n_fft, int64_t hop_length_v, int hop_length_null, int64_t win_length_v, int win_length_null, gc_tensor window, int center, char * pad_mode, int normalized, int onesided, int return_complex);
+int64_t atg_stride(gc_tensor self, int64_t dim);
+raw_tensor atg_sub(gc_tensor self, gc_tensor other);
+raw_tensor atg_sub_(gc_tensor self, gc_tensor other);
+raw_tensor atg_sub_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_sub_scalar(gc_tensor self, scalar other);
+raw_tensor atg_sub_scalar_(gc_tensor self, scalar other);
+raw_tensor atg_sub_scalar_out(gc_tensor out, gc_tensor self, scalar other);
+raw_tensor atg_subtract(gc_tensor self, gc_tensor other);
+raw_tensor atg_subtract_(gc_tensor self, gc_tensor other);
+raw_tensor atg_subtract_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_subtract_scalar(gc_tensor self, scalar other);
+raw_tensor atg_subtract_scalar_(gc_tensor self, scalar other);
+raw_tensor atg_sum(gc_tensor self, int dtype);
+raw_tensor atg_sum_dim_intlist(gc_tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype);
+raw_tensor atg_sum_intlist_out(gc_tensor out, gc_tensor self, int64_t *dim_data, int dim_len, int keepdim, int dtype);
+raw_tensor atg_sum_out(gc_tensor out, gc_tensor self, int dtype);
+raw_tensor atg_sum_to_size(gc_tensor self, int64_t *size_data, int size_len);
+void atg_svd(raw_tensor *, gc_tensor self, int some, int compute_uv);
+void atg_svd_u(raw_tensor *, gc_tensor U, gc_tensor S, gc_tensor V, gc_tensor self, int some, int compute_uv);
+raw_tensor atg_swapaxes(gc_tensor self, int64_t axis0, int64_t axis1);
+raw_tensor atg_swapaxes_(gc_tensor self, int64_t axis0, int64_t axis1);
+raw_tensor atg_swapdims(gc_tensor self, int64_t dim0, int64_t dim1);
+raw_tensor atg_swapdims_(gc_tensor self, int64_t dim0, int64_t dim1);
+raw_tensor atg_t(gc_tensor self);
+raw_tensor atg_t_(gc_tensor self);
+raw_tensor atg_t_copy(gc_tensor self);
+raw_tensor atg_t_copy_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_take(gc_tensor self, gc_tensor index);
+raw_tensor atg_take_along_dim(gc_tensor self, gc_tensor indices, int64_t dim_v, int dim_null);
+raw_tensor atg_take_along_dim_out(gc_tensor out, gc_tensor self, gc_tensor indices, int64_t dim_v, int dim_null);
+raw_tensor atg_take_out(gc_tensor out, gc_tensor self, gc_tensor index);
+raw_tensor atg_tan(gc_tensor self);
+raw_tensor atg_tan_(gc_tensor self);
+raw_tensor atg_tan_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_tanh(gc_tensor self);
+raw_tensor atg_tanh_(gc_tensor self);
+raw_tensor atg_tanh_backward(gc_tensor grad_output, gc_tensor output);
+raw_tensor atg_tanh_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor output);
+raw_tensor atg_tanh_out(gc_tensor out, gc_tensor self);
+raw_tensor *atg_tensor_split(gc_tensor self, int64_t sections, int64_t dim);
+raw_tensor *atg_tensor_split_indices(gc_tensor self, int64_t *indices_data, int indices_len, int64_t dim);
+raw_tensor *atg_tensor_split_tensor_indices_or_sections(gc_tensor self, gc_tensor tensor_indices_or_sections, int64_t dim);
+raw_tensor atg_tensordot(gc_tensor self, gc_tensor other, int64_t *dims_self_data, int dims_self_len, int64_t *dims_other_data, int dims_other_len);
+raw_tensor atg_tensordot_out(gc_tensor out, gc_tensor self, gc_tensor other, int64_t *dims_self_data, int dims_self_len, int64_t *dims_other_data, int dims_other_len);
+raw_tensor atg_threshold(gc_tensor self, scalar threshold, scalar value);
+raw_tensor atg_threshold_(gc_tensor self, scalar threshold, scalar value);
+raw_tensor atg_threshold_backward(gc_tensor grad_output, gc_tensor self, scalar threshold);
+raw_tensor atg_threshold_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, gc_tensor self, scalar threshold);
+raw_tensor atg_threshold_out(gc_tensor out, gc_tensor self, scalar threshold, scalar value);
+raw_tensor atg_tile(gc_tensor self, int64_t *dims_data, int dims_len);
+raw_tensor atg_to(gc_tensor self, int device);
+raw_tensor atg_to_dense(gc_tensor self, int dtype, int masked_grad);
+raw_tensor atg_to_dense_backward(gc_tensor grad, gc_tensor input, int masked_grad);
+raw_tensor atg_to_device(gc_tensor self, int device, int dtype, int non_blocking, int copy);
+raw_tensor atg_to_dtype(gc_tensor self, int dtype, int non_blocking, int copy);
+raw_tensor atg_to_dtype_layout(gc_tensor self, int options_kind, int options_device, int non_blocking, int copy);
+raw_tensor atg_to_mkldnn(gc_tensor self, int dtype);
+raw_tensor atg_to_mkldnn_backward(gc_tensor grad, gc_tensor input);
+raw_tensor atg_to_mkldnn_out(gc_tensor out, gc_tensor self, int dtype);
+raw_tensor atg_to_other(gc_tensor self, gc_tensor other, int non_blocking, int copy);
+raw_tensor atg_to_padded_tensor(gc_tensor self, double padding, int64_t *output_size_data, int output_size_len);
+raw_tensor atg_to_padded_tensor_out(gc_tensor out, gc_tensor self, double padding, int64_t *output_size_data, int output_size_len);
+void atg_topk(raw_tensor *, gc_tensor self, int64_t k, int64_t dim, int largest, int sorted);
+void atg_topk_values(raw_tensor *, gc_tensor values, gc_tensor indices, gc_tensor self, int64_t k, int64_t dim, int largest, int sorted);
+raw_tensor atg_totype(gc_tensor self, int scalar_type);
+raw_tensor atg_trace(gc_tensor self);
+raw_tensor atg_trace_backward(gc_tensor grad, int64_t *sizes_data, int sizes_len);
+raw_tensor atg_trace_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_transpose(gc_tensor self, int64_t dim0, int64_t dim1);
+raw_tensor atg_transpose_(gc_tensor self, int64_t dim0, int64_t dim1);
+raw_tensor atg_transpose_copy(gc_tensor self, int64_t dim0, int64_t dim1);
+raw_tensor atg_transpose_copy_int_out(gc_tensor out, gc_tensor self, int64_t dim0, int64_t dim1);
+raw_tensor atg_trapezoid(gc_tensor y, int64_t dim);
+raw_tensor atg_trapezoid_x(gc_tensor y, gc_tensor x, int64_t dim);
+raw_tensor atg_trapz(gc_tensor y, gc_tensor x, int64_t dim);
+raw_tensor atg_trapz_dx(gc_tensor y, double dx, int64_t dim);
+void atg_triangular_solve(raw_tensor *, gc_tensor self, gc_tensor A, int upper, int transpose, int unitriangular);
+void atg_triangular_solve_x(raw_tensor *, gc_tensor X, gc_tensor M, gc_tensor self, gc_tensor A, int upper, int transpose, int unitriangular);
+raw_tensor atg_tril(gc_tensor self, int64_t diagonal);
+raw_tensor atg_tril_(gc_tensor self, int64_t diagonal);
+raw_tensor atg_tril_indices(int64_t row, int64_t col, int64_t offset, int options_kind, int options_device);
+raw_tensor atg_tril_indices_out(gc_tensor out, int64_t row, int64_t col, int64_t offset);
+raw_tensor atg_tril_out(gc_tensor out, gc_tensor self, int64_t diagonal);
+raw_tensor atg_triplet_margin_loss(gc_tensor anchor, gc_tensor positive, gc_tensor negative, double margin, double p, double eps, int swap, int64_t reduction);
+raw_tensor atg_triu(gc_tensor self, int64_t diagonal);
+raw_tensor atg_triu_(gc_tensor self, int64_t diagonal);
+raw_tensor atg_triu_indices(int64_t row, int64_t col, int64_t offset, int options_kind, int options_device);
+raw_tensor atg_triu_indices_out(gc_tensor out, int64_t row, int64_t col, int64_t offset);
+raw_tensor atg_triu_out(gc_tensor out, gc_tensor self, int64_t diagonal);
+raw_tensor atg_true_divide(gc_tensor self, gc_tensor other);
+raw_tensor atg_true_divide_(gc_tensor self, gc_tensor other);
+raw_tensor atg_true_divide_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_true_divide_scalar(gc_tensor self, scalar other);
+raw_tensor atg_true_divide_scalar_(gc_tensor self, scalar other);
+raw_tensor atg_trunc(gc_tensor self);
+raw_tensor atg_trunc_(gc_tensor self);
+raw_tensor atg_trunc_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_type_as(gc_tensor self, gc_tensor other);
+raw_tensor *atg_unbind(gc_tensor self, int64_t dim);
+raw_tensor *atg_unbind_copy(gc_tensor self, int64_t dim);
+void atg_unbind_copy_int_out(gc_tensor *out_data, int out_len, gc_tensor self, int64_t dim);
+raw_tensor atg_unflatten(gc_tensor self, int64_t dim, int64_t *sizes_data, int sizes_len);
+raw_tensor *atg_unflatten_dense_tensors(gc_tensor flat, gc_tensor *tensors_data, int tensors_len);
+raw_tensor atg_unfold(gc_tensor self, int64_t dimension, int64_t size, int64_t step);
+raw_tensor atg_unfold_backward(gc_tensor grad_in, int64_t *input_sizes_data, int input_sizes_len, int64_t dim, int64_t size, int64_t step);
+raw_tensor atg_unfold_backward_out(gc_tensor out, gc_tensor grad_in, int64_t *input_sizes_data, int input_sizes_len, int64_t dim, int64_t size, int64_t step);
+raw_tensor atg_unfold_copy(gc_tensor self, int64_t dimension, int64_t size, int64_t step);
+raw_tensor atg_unfold_copy_out(gc_tensor out, gc_tensor self, int64_t dimension, int64_t size, int64_t step);
+raw_tensor atg_uniform(gc_tensor self, double from, double to);
+raw_tensor atg_uniform_(gc_tensor self, double from, double to);
+raw_tensor atg_uniform_out(gc_tensor out, gc_tensor self, double from, double to);
+void atg_unique_consecutive(raw_tensor *, gc_tensor self, int return_inverse, int return_counts, int64_t dim_v, int dim_null);
+void atg_unique_consecutive_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor self, int return_inverse, int return_counts, int64_t dim_v, int dim_null);
+void atg_unique_dim(raw_tensor *, gc_tensor self, int64_t dim, int sorted, int return_inverse, int return_counts);
+void atg_unique_dim_consecutive(raw_tensor *, gc_tensor self, int64_t dim, int return_inverse, int return_counts);
+void atg_unique_dim_consecutive_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor self, int64_t dim, int return_inverse, int return_counts);
+void atg_unique_dim_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor out2, gc_tensor self, int64_t dim, int sorted, int return_inverse, int return_counts);
+raw_tensor *atg_unsafe_chunk(gc_tensor self, int64_t chunks, int64_t dim);
+raw_tensor *atg_unsafe_split(gc_tensor self, int64_t split_size, int64_t dim);
+void atg_unsafe_split_tensor_out(gc_tensor *out_data, int out_len, gc_tensor self, int64_t split_size, int64_t dim);
+raw_tensor *atg_unsafe_split_with_sizes(gc_tensor self, int64_t *split_sizes_data, int split_sizes_len, int64_t dim);
+void atg_unsafe_split_with_sizes_out(gc_tensor *out_data, int out_len, gc_tensor self, int64_t *split_sizes_data, int split_sizes_len, int64_t dim);
+raw_tensor atg_unsqueeze(gc_tensor self, int64_t dim);
+raw_tensor atg_unsqueeze_(gc_tensor self, int64_t dim);
+raw_tensor atg_unsqueeze_copy(gc_tensor self, int64_t dim);
+raw_tensor atg_unsqueeze_copy_out(gc_tensor out, gc_tensor self, int64_t dim);
+raw_tensor atg_upsample_bicubic2d(gc_tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
+raw_tensor atg_upsample_bicubic2d_backward(gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
+raw_tensor atg_upsample_bicubic2d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
+raw_tensor atg_upsample_bicubic2d_out(gc_tensor out, gc_tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
+raw_tensor atg_upsample_bicubic2d_vec(gc_tensor input, int64_t *output_size_data, int output_size_len, int align_corners, double *scale_factors_data, int scale_factors_len);
+raw_tensor atg_upsample_bilinear2d(gc_tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
+raw_tensor atg_upsample_bilinear2d_backward(gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
+raw_tensor atg_upsample_bilinear2d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
+raw_tensor atg_upsample_bilinear2d_out(gc_tensor out, gc_tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
+raw_tensor atg_upsample_bilinear2d_vec(gc_tensor input, int64_t *output_size_data, int output_size_len, int align_corners, double *scale_factors_data, int scale_factors_len);
+raw_tensor atg_upsample_linear1d(gc_tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_v, int scales_null);
+raw_tensor atg_upsample_linear1d_backward(gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_v, int scales_null);
+raw_tensor atg_upsample_linear1d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_v, int scales_null);
+raw_tensor atg_upsample_linear1d_out(gc_tensor out, gc_tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_v, int scales_null);
+raw_tensor atg_upsample_linear1d_vec(gc_tensor input, int64_t *output_size_data, int output_size_len, int align_corners, double *scale_factors_data, int scale_factors_len);
+raw_tensor atg_upsample_nearest1d(gc_tensor self, int64_t *output_size_data, int output_size_len, double scales_v, int scales_null);
+raw_tensor atg_upsample_nearest1d_backward(gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_v, int scales_null);
+raw_tensor atg_upsample_nearest1d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_v, int scales_null);
+raw_tensor atg_upsample_nearest1d_out(gc_tensor out, gc_tensor self, int64_t *output_size_data, int output_size_len, double scales_v, int scales_null);
+raw_tensor atg_upsample_nearest1d_vec(gc_tensor input, int64_t *output_size_data, int output_size_len, double *scale_factors_data, int scale_factors_len);
+raw_tensor atg_upsample_nearest2d(gc_tensor self, int64_t *output_size_data, int output_size_len, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
+raw_tensor atg_upsample_nearest2d_backward(gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
+raw_tensor atg_upsample_nearest2d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
+raw_tensor atg_upsample_nearest2d_out(gc_tensor out, gc_tensor self, int64_t *output_size_data, int output_size_len, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
+raw_tensor atg_upsample_nearest2d_vec(gc_tensor input, int64_t *output_size_data, int output_size_len, double *scale_factors_data, int scale_factors_len);
+raw_tensor atg_upsample_nearest3d(gc_tensor self, int64_t *output_size_data, int output_size_len, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
+raw_tensor atg_upsample_nearest3d_backward(gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
+raw_tensor atg_upsample_nearest3d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
+raw_tensor atg_upsample_nearest3d_out(gc_tensor out, gc_tensor self, int64_t *output_size_data, int output_size_len, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
+raw_tensor atg_upsample_nearest3d_vec(gc_tensor input, int64_t *output_size_data, int output_size_len, double *scale_factors_data, int scale_factors_len);
+raw_tensor atg_upsample_trilinear3d(gc_tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
+raw_tensor atg_upsample_trilinear3d_backward(gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
+raw_tensor atg_upsample_trilinear3d_backward_grad_input(gc_tensor grad_input, gc_tensor grad_output, int64_t *output_size_data, int output_size_len, int64_t *input_size_data, int input_size_len, int align_corners, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
+raw_tensor atg_upsample_trilinear3d_out(gc_tensor out, gc_tensor self, int64_t *output_size_data, int output_size_len, int align_corners, double scales_d_v, int scales_d_null, double scales_h_v, int scales_h_null, double scales_w_v, int scales_w_null);
+raw_tensor atg_upsample_trilinear3d_vec(gc_tensor input, int64_t *output_size_data, int output_size_len, int align_corners, double *scale_factors_data, int scale_factors_len);
+raw_tensor atg_value_selecting_reduction_backward(gc_tensor grad, int64_t dim, gc_tensor indices, int64_t *sizes_data, int sizes_len, int keepdim);
+raw_tensor atg_values(gc_tensor self);
+raw_tensor atg_values_copy(gc_tensor self);
+raw_tensor atg_values_copy_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_vander(gc_tensor x, int64_t n_v, int n_null, int increasing);
+raw_tensor atg_var(gc_tensor self, int unbiased);
+raw_tensor atg_var_correction(gc_tensor self, int64_t *dim_data, int dim_len, scalar correction, int keepdim);
+raw_tensor atg_var_correction_out(gc_tensor out, gc_tensor self, int64_t *dim_data, int dim_len, scalar correction, int keepdim);
+raw_tensor atg_var_dim(gc_tensor self, int64_t *dim_data, int dim_len, int unbiased, int keepdim);
+void atg_var_mean(raw_tensor *, gc_tensor self, int unbiased);
+void atg_var_mean_correction(raw_tensor *, gc_tensor self, int64_t *dim_data, int dim_len, scalar correction, int keepdim);
+void atg_var_mean_correction_out(raw_tensor *, gc_tensor out0, gc_tensor out1, gc_tensor self, int64_t *dim_data, int dim_len, scalar correction, int keepdim);
+void atg_var_mean_dim(raw_tensor *, gc_tensor self, int64_t *dim_data, int dim_len, int unbiased, int keepdim);
+raw_tensor atg_var_out(gc_tensor out, gc_tensor self, int64_t *dim_data, int dim_len, int unbiased, int keepdim);
+raw_tensor atg_vdot(gc_tensor self, gc_tensor other);
+raw_tensor atg_vdot_out(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_view(gc_tensor self, int64_t *size_data, int size_len);
+raw_tensor atg_view_as(gc_tensor self, gc_tensor other);
+raw_tensor atg_view_as_complex(gc_tensor self);
+raw_tensor atg_view_as_complex_copy(gc_tensor self);
+raw_tensor atg_view_as_complex_copy_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_view_as_real(gc_tensor self);
+raw_tensor atg_view_as_real_copy(gc_tensor self);
+raw_tensor atg_view_as_real_copy_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_view_copy(gc_tensor self, int64_t *size_data, int size_len);
+raw_tensor atg_view_copy_dtype(gc_tensor self, int dtype);
+raw_tensor atg_view_copy_dtype_out(gc_tensor out, gc_tensor self, int dtype);
+raw_tensor atg_view_copy_out(gc_tensor out, gc_tensor self, int64_t *size_data, int size_len);
+raw_tensor atg_view_dtype(gc_tensor self, int dtype);
+raw_tensor *atg_vsplit(gc_tensor self, int64_t sections);
+raw_tensor *atg_vsplit_array(gc_tensor self, int64_t *indices_data, int indices_len);
+raw_tensor atg_vstack(gc_tensor *tensors_data, int tensors_len);
+raw_tensor atg_vstack_out(gc_tensor out, gc_tensor *tensors_data, int tensors_len);
+raw_tensor *atg_where(gc_tensor condition);
+raw_tensor atg_where_scalar(gc_tensor condition, scalar self, scalar other);
+raw_tensor atg_where_scalarother(gc_tensor condition, gc_tensor self, scalar other);
+raw_tensor atg_where_scalarself(gc_tensor condition, scalar self, gc_tensor other);
+raw_tensor atg_where_self(gc_tensor condition, gc_tensor self, gc_tensor other);
+raw_tensor atg_where_self_out(gc_tensor out, gc_tensor condition, gc_tensor self, gc_tensor other);
+raw_tensor atg_xlogy(gc_tensor self, gc_tensor other);
+raw_tensor atg_xlogy_(gc_tensor self, gc_tensor other);
+raw_tensor atg_xlogy_outscalar_other(gc_tensor out, gc_tensor self, scalar other);
+raw_tensor atg_xlogy_outscalar_self(gc_tensor out, scalar self, gc_tensor other);
+raw_tensor atg_xlogy_outtensor(gc_tensor out, gc_tensor self, gc_tensor other);
+raw_tensor atg_xlogy_scalar_other(gc_tensor self, scalar other);
+raw_tensor atg_xlogy_scalar_other_(gc_tensor self, scalar other);
+raw_tensor atg_xlogy_scalar_self(scalar self, gc_tensor other);
+raw_tensor atg_zero(gc_tensor self);
+raw_tensor atg_zero_(gc_tensor self);
+raw_tensor atg_zero_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_zeros(int64_t *size_data, int size_len, int options_kind, int options_device);
+raw_tensor atg_zeros_like(gc_tensor self);
+raw_tensor atg_zeros_like_out(gc_tensor out, gc_tensor self);
+raw_tensor atg_zeros_out(gc_tensor out, int64_t *size_data, int size_len);
diff --git a/src/wrapper/torch_stubs.c b/src/wrapper/torch_stubs.c
new file mode 100644
index 0000000..d28461f
--- /dev/null
+++ b/src/wrapper/torch_stubs.c
@@ -0,0 +1,10 @@
+#include "torch_api.h"
+#include "ctypes_cstubs_internals.h"
+#include <caml/memory.h>
+
+// returns a voidp (ocaml block storing a void pointer)
+CAMLprim value with_tensor_gc(value raw_tensor_value) {
+  CAMLparam1(raw_tensor_value);
+  raw_tensor raw = CTYPES_ADDR_OF_FATPTR(raw_tensor_value);
+  CAMLreturn(with_tensor_gc_internal(raw));
+}
diff --git a/src/wrapper/torch_stubs.ml b/src/wrapper/torch_stubs.ml
new file mode 100644
index 0000000..9f27908
--- /dev/null
+++ b/src/wrapper/torch_stubs.ml
@@ -0,0 +1,21 @@
+open Ctypes
+open Torch_bindings.Type_defs
+module C = Torch_bindings.C (Torch_stubs_generated)
+
+external with_tensor_gc : _ Cstubs_internals.fatptr -> Ctypes_ptr.voidp = "with_tensor_gc"
+
+let with_tensor_gc (raw : raw_tensor) : gc_tensor =
+  fatptr_of_raw_tensor raw |> with_tensor_gc |> gc_tensor_of_voidp
+;;
+
+let to_tensor_list (ptr : raw_tensor ptr) =
+  let rec loop ptr acc =
+    let tensor : raw_tensor = !@ptr in
+    if is_none_raw_tensor tensor
+    then acc
+    else loop (ptr +@ 1) (with_tensor_gc tensor :: acc)
+  in
+  let result = loop ptr [] in
+  C.free (to_voidp ptr);
+  List.rev result
+;;
diff --git a/src/wrapper/wrapper.ml b/src/wrapper/wrapper.ml
index 838e17f..c9d0248 100644
--- a/src/wrapper/wrapper.ml
+++ b/src/wrapper/wrapper.ml
@@ -1,4 +1,6 @@
 open Ctypes
+open Torch_stubs
+open Torch_bindings.Type_defs
 
 let ptr_of_string str =
   let len = String.length str in
@@ -19,7 +21,9 @@ module Tensor = struct
   include Wrapper_generated
   open! C.Tensor
 
-  type nonrec t = t
+  type t = gc_tensor
+
+  let new_tensor () = new_tensor () |> with_tensor_gc
 
   let float_vec ?(kind = `float) values =
     let values_len = List.length values in
@@ -31,8 +35,7 @@ module Tensor = struct
       | `half -> Kind.T Half
     in
     let t = float_vec values values_len (Kind.packed_to_int kind) in
-    Gc.finalise free t;
-    t
+    with_tensor_gc t
   ;;
 
   let int_vec ?(kind = `int) values =
@@ -47,8 +50,7 @@ module Tensor = struct
       | `int64 -> Kind.T Int64
     in
     let t = int_vec values values_len (Kind.packed_to_int kind) in
-    Gc.finalise free t;
-    t
+    with_tensor_gc t
   ;;
 
   let of_bigarray (type a b) (ga : (b, a, Bigarray.c_layout) Bigarray.Genarray.t) =
@@ -78,8 +80,7 @@ module Tensor = struct
         (Bigarray.kind_size_in_bytes kind)
         (Kind.packed_to_int tensor_kind)
     in
-    Gc.finalise free t;
-    t
+    with_tensor_gc t
   ;;
 
   let copy_to_bigarray (type a b) t (ga : (b, a, Bigarray.c_layout) Bigarray.Genarray.t) =
@@ -135,8 +136,7 @@ module Tensor = struct
 
   let get t index =
     let t = get t index in
-    Gc.finalise free t;
-    t
+    with_tensor_gc t
   ;;
 
   let float_value t = double_value t (from_voidp int null) 0
@@ -184,30 +184,22 @@ module Tensor = struct
   let defined = defined
   let device t = device t |> Device.of_int
 
-  let new_tensor () =
-    let t = new_tensor () in
-    Gc.finalise free t;
-    t
-  ;;
-
   let run_backward ?keep_graph ?(create_graph = false) tensors inputs =
     let keep_graph =
       match keep_graph with
       | None -> create_graph
       | Some keep_graph -> keep_graph
     in
-    let out_ = CArray.make t (List.length inputs) in
+    let out_ = CArray.make raw_tensor (List.length inputs) in
     run_backward
-      (CArray.of_list t tensors |> CArray.start)
+      (CArray.of_list gc_tensor tensors |> CArray.start)
       (List.length tensors)
-      (CArray.of_list t inputs |> CArray.start)
+      (CArray.of_list gc_tensor inputs |> CArray.start)
       (List.length inputs)
       (CArray.start out_)
       (if keep_graph then 1 else 0)
       (if create_graph then 1 else 0);
-    let out_ = CArray.to_list out_ in
-    List.iter (Gc.finalise free) out_;
-    out_
+    List.map with_tensor_gc (CArray.to_list out_)
   ;;
 
   let sum t = sum t ~dtype:(kind t)
@@ -215,16 +207,10 @@ module Tensor = struct
 end
 
 module Scalar = struct
-  module S = Wrapper_generated.C.Scalar
-
-  include (
-    S :
-      module type of struct
-        include S
-      end
-      with type t := S.t)
+  module S = C.Scalar
+  include (S : module type of S)
 
-  type nonrec _ t = S.t
+  type nonrec _ t = scalar
 
   let int i =
     let t = int (Int64.of_int i) in
@@ -240,7 +226,9 @@ module Scalar = struct
 end
 
 module Optimizer = struct
-  include Wrapper_generated.C.Optimizer
+  include C.Optimizer
+
+  type t = optimizer
 
   let adam ~learning_rate ~beta1 ~beta2 ~weight_decay ~eps =
     let t = adam learning_rate beta1 beta2 weight_decay eps in
@@ -262,15 +250,12 @@ module Optimizer = struct
   ;;
 
   let add_parameters t tensors =
-    add_parameters
-      t
-      CArray.(of_list Wrapper_generated.C.Tensor.t tensors |> start)
-      (List.length tensors)
+    add_parameters t CArray.(of_list gc_tensor tensors |> start) (List.length tensors)
   ;;
 end
 
 module Serialize = struct
-  include Wrapper_generated.C.Serialize
+  include C.Serialize
 
   let save t ~filename = save t filename
 
@@ -290,17 +275,13 @@ module Serialize = struct
       s
   ;;
 
-  let load ~filename =
-    let t = load filename in
-    Gc.finalise Wrapper_generated.C.Tensor.free t;
-    t
-  ;;
+  let load ~filename = load filename |> with_tensor_gc
 
   let save_multi ~named_tensors ~filename =
     let names, tensors = List.split named_tensors in
     let names = List.map escape names in
     save_multi
-      CArray.(of_list Wrapper_generated.C.Tensor.t tensors |> start)
+      CArray.(of_list gc_tensor tensors |> start)
       (ptr_of_strings names)
       (List.length named_tensors)
       filename
@@ -309,10 +290,9 @@ module Serialize = struct
   let load_multi ~names ~filename =
     let names = List.map escape names in
     let ntensors = List.length names in
-    let tensors = CArray.make Wrapper_generated.C.Tensor.t ntensors in
+    let tensors = CArray.make raw_tensor ntensors in
     load_multi (CArray.start tensors) (ptr_of_strings names) ntensors filename;
-    let tensors = CArray.to_list tensors in
-    List.iter (Gc.finalise Wrapper_generated.C.Tensor.free) tensors;
+    let tensors = List.map with_tensor_gc (CArray.to_list tensors) in
     tensors
   ;;
 
@@ -320,7 +300,7 @@ module Serialize = struct
     let names, tensors = List.split named_tensors in
     let names = List.map escape names in
     load_multi_
-      CArray.(of_list Wrapper_generated.C.Tensor.t tensors |> start)
+      CArray.(of_list gc_tensor tensors |> start)
       (ptr_of_strings names)
       (List.length named_tensors)
       filename
@@ -330,11 +310,10 @@ module Serialize = struct
     let all_tensors = ref [] in
     let callback =
       coerce
-        (Foreign.funptr (string @-> Wrapper_generated.C.Tensor.t @-> returning void))
-        (static_funptr (string @-> Wrapper_generated.C.Tensor.t @-> returning void))
+        (Foreign.funptr (string @-> raw_tensor @-> returning void))
+        (static_funptr (string @-> raw_tensor @-> returning void))
         (fun tensor_name tensor ->
-          Gc.finalise Wrapper_generated.C.Tensor.free tensor;
-          all_tensors := (unescape tensor_name, tensor) :: !all_tensors)
+          all_tensors := (unescape tensor_name, with_tensor_gc tensor) :: !all_tensors)
       [@alert "-deprecated"]
     in
     load_callback filename callback;
@@ -343,7 +322,7 @@ module Serialize = struct
 end
 
 module Cuda = struct
-  include Wrapper_generated.C.Cuda
+  include C.Cuda
 
   let is_available () = is_available () <> 0
   let cudnn_is_available () = cudnn_is_available () <> 0
@@ -368,7 +347,9 @@ module Ivalue = struct
       | GenericDict
   end
 
-  include Wrapper_generated.C.Ivalue
+  include C.Ivalue
+
+  type t = ivalue
 
   let none () =
     let t = none () in
@@ -401,17 +382,13 @@ module Ivalue = struct
   ;;
 
   let tuple ts =
-    let t = tuple CArray.(of_list t ts |> start) (List.length ts) in
+    let t = tuple CArray.(of_list ivalue ts |> start) (List.length ts) in
     Gc.finalise free t;
     t
   ;;
 
   let tensor_list ts =
-    let t =
-      tensor_list
-        CArray.(of_list Wrapper_generated.C.Tensor.t ts |> start)
-        (List.length ts)
-    in
+    let t = tensor_list CArray.(of_list gc_tensor ts |> start) (List.length ts) in
     Gc.finalise free t;
     t
   ;;
@@ -441,16 +418,11 @@ module Ivalue = struct
   ;;
 
   let to_bool t = to_bool t <> 0
-
-  let to_tensor t =
-    let tensor = to_tensor t in
-    Gc.finalise Wrapper_generated.C.Tensor.free tensor;
-    tensor
-  ;;
+  let to_tensor t = to_tensor t |> with_tensor_gc
 
   let to_tuple t =
     let noutputs = tuple_length t in
-    let outputs = CArray.make Wrapper_generated.C.Tensor.t noutputs in
+    let outputs = CArray.make ivalue noutputs in
     to_tuple t (CArray.start outputs) noutputs;
     let outputs = CArray.to_list outputs in
     List.iter (Gc.finalise free) outputs;
@@ -459,55 +431,43 @@ module Ivalue = struct
 
   let to_tensor_list t =
     let noutputs = list_length t in
-    let outputs = CArray.make Wrapper_generated.C.Tensor.t noutputs in
-    Wrapper_generated.C.Ivalue.to_tensor_list t (CArray.start outputs) noutputs;
-    let outputs = CArray.to_list outputs in
-    (* free calls ati_free which destructs an Ivalue. Here we need to destruct a Tensor. *)
-    List.iter (Gc.finalise Wrapper_generated.C.Tensor.free) outputs;
-    outputs
+    let outputs = CArray.make raw_tensor noutputs in
+    C.Ivalue.to_tensor_list t (CArray.start outputs) noutputs;
+    CArray.to_list outputs |> List.map with_tensor_gc
   ;;
 end
 
 module Module = struct
-  include Wrapper_generated.C.Module
+  include C.Module
+
+  type t = module_
 
   let forward t tensors =
-    let tensor =
-      forward
-        t
-        CArray.(of_list Wrapper_generated.C.Tensor.t tensors |> start)
-        (List.length tensors)
-    in
-    Gc.finalise Wrapper_generated.C.Tensor.free tensor;
-    tensor
+    forward t CArray.(of_list gc_tensor tensors |> start) (List.length tensors)
+    |> with_tensor_gc
   ;;
 
   let forward_ t ivalues =
     let ivalue =
-      forward_
-        t
-        CArray.(of_list Wrapper_generated.C.Ivalue.t ivalues |> start)
-        (List.length ivalues)
+      forward_ t CArray.(of_list ivalue ivalues |> start) (List.length ivalues)
     in
-    Gc.finalise Wrapper_generated.C.Ivalue.free ivalue;
+    Gc.finalise C.Ivalue.free ivalue;
     ivalue
   ;;
 
   let named_buffers t =
     let wrapper_ivalue = named_buffers t in
-    let names_and_tensors = CArray.make Wrapper_generated.C.Ivalue.t 2 in
-    Wrapper_generated.C.Ivalue.to_tuple wrapper_ivalue (CArray.start names_and_tensors) 2;
+    let names_and_tensors = CArray.make ivalue 2 in
+    C.Ivalue.to_tuple wrapper_ivalue (CArray.start names_and_tensors) 2;
     let names_ivalue = CArray.get names_and_tensors 0
     and tensors_ivalue = CArray.get names_and_tensors 1 in
-    let len = Wrapper_generated.C.Ivalue.list_length tensors_ivalue in
-    let names = CArray.make Wrapper_generated.C.Ivalue.t len
-    and tensors = CArray.make Wrapper_generated.C.Ivalue.t len in
-    Wrapper_generated.C.Ivalue.to_generic_list names_ivalue (CArray.start names) len;
-    Wrapper_generated.C.Ivalue.to_generic_list tensors_ivalue (CArray.start tensors) len;
-    let names = names |> CArray.to_list |> List.map Wrapper_generated.C.Ivalue.to_string
-    and tensors =
-      tensors |> CArray.to_list |> List.map Wrapper_generated.C.Ivalue.to_tensor
-    in
+    let len = C.Ivalue.list_length tensors_ivalue in
+    let names = CArray.make ivalue len
+    and tensors = CArray.make ivalue len in
+    C.Ivalue.to_generic_list names_ivalue (CArray.start names) len;
+    C.Ivalue.to_generic_list tensors_ivalue (CArray.start tensors) len;
+    let names = names |> CArray.to_list |> List.map C.Ivalue.to_string
+    and tensors = tensors |> CArray.to_list |> List.map Ivalue.to_tensor in
     List.combine names tensors |> Base.Map.of_alist_exn (module Base.String)
   ;;
 
@@ -524,6 +484,6 @@ module Module = struct
   ;;
 end
 
-let manual_seed seed = Wrapper_generated.C.manual_seed (Int64.of_int seed)
-let set_num_threads = Wrapper_generated.C.set_num_threads
-let get_num_threads = Wrapper_generated.C.get_num_threads
+let manual_seed seed = C.manual_seed (Int64.of_int seed)
+let set_num_threads = C.set_num_threads
+let get_num_threads = C.get_num_threads
diff --git a/src/wrapper/wrapper_generated.ml b/src/wrapper/wrapper_generated.ml
index 5bb7c75..71915e4 100644
--- a/src/wrapper/wrapper_generated.ml
+++ b/src/wrapper/wrapper_generated.ml
@@ -1,405 +1,147 @@
 (* THIS FILE IS AUTOMATICALLY GENERATED, DO NOT EDIT BY HAND! *)
 
 open Ctypes
-module C = Torch_bindings.C (Torch_generated)
-open C.TensorG
-
-let to_tensor_list ptr =
-  let rec loop ptr acc =
-    let tensor = !@ptr in
-    if is_null tensor
-    then acc
-    else (
-      Gc.finalise C.Tensor.free tensor;
-      loop (ptr +@ 1) (tensor :: acc))
-  in
-  let result = loop ptr [] in
-  C.free (to_voidp ptr);
-  List.rev result
-;;
-
-let __and__ self other =
-  let out__ = CArray.make t 1 in
-  stubs___and__ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let __and__tensor_ self other =
-  let out__ = CArray.make t 1 in
-  stubs___and__tensor_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let __iand__ self other =
-  let out__ = CArray.make t 1 in
-  stubs___iand__ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let __iand__tensor_ self other =
-  let out__ = CArray.make t 1 in
-  stubs___iand__tensor_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let __ilshift__ self other =
-  let out__ = CArray.make t 1 in
-  stubs___ilshift__ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let __ilshift__tensor_ self other =
-  let out__ = CArray.make t 1 in
-  stubs___ilshift__tensor_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let __ior__ self other =
-  let out__ = CArray.make t 1 in
-  stubs___ior__ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let __ior__tensor_ self other =
-  let out__ = CArray.make t 1 in
-  stubs___ior__tensor_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let __irshift__ self other =
-  let out__ = CArray.make t 1 in
-  stubs___irshift__ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let __irshift__tensor_ self other =
-  let out__ = CArray.make t 1 in
-  stubs___irshift__tensor_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let __ixor__ self other =
-  let out__ = CArray.make t 1 in
-  stubs___ixor__ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let __ixor__tensor_ self other =
-  let out__ = CArray.make t 1 in
-  stubs___ixor__tensor_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let __lshift__ self other =
-  let out__ = CArray.make t 1 in
-  stubs___lshift__ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+open Torch_bindings.Type_defs
+open Torch_stubs
+open C.Generated
+
+let __and__ self other = stubs___and__ self other |> with_tensor_gc
+let __and__tensor_ self other = stubs___and__tensor_ self other |> with_tensor_gc
+let __iand__ self other = stubs___iand__ self other |> with_tensor_gc
+let __iand__tensor_ self other = stubs___iand__tensor_ self other |> with_tensor_gc
+let __ilshift__ self other = stubs___ilshift__ self other |> with_tensor_gc
+let __ilshift__tensor_ self other = stubs___ilshift__tensor_ self other |> with_tensor_gc
+let __ior__ self other = stubs___ior__ self other |> with_tensor_gc
+let __ior__tensor_ self other = stubs___ior__tensor_ self other |> with_tensor_gc
+let __irshift__ self other = stubs___irshift__ self other |> with_tensor_gc
+let __irshift__tensor_ self other = stubs___irshift__tensor_ self other |> with_tensor_gc
+let __ixor__ self other = stubs___ixor__ self other |> with_tensor_gc
+let __ixor__tensor_ self other = stubs___ixor__tensor_ self other |> with_tensor_gc
+let __lshift__ self other = stubs___lshift__ self other |> with_tensor_gc
 
 let __lshift__scalar_out_ ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs___lshift__scalar_out_ (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs___lshift__scalar_out_ out self other |> with_tensor_gc
 ;;
 
-let __lshift__tensor_ self other =
-  let out__ = CArray.make t 1 in
-  stubs___lshift__tensor_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let __lshift__tensor_ self other = stubs___lshift__tensor_ self other |> with_tensor_gc
 
 let __lshift__tensor_out_ ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs___lshift__tensor_out_ (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let __or__ self other =
-  let out__ = CArray.make t 1 in
-  stubs___or__ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let __or__tensor_ self other =
-  let out__ = CArray.make t 1 in
-  stubs___or__tensor_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs___lshift__tensor_out_ out self other |> with_tensor_gc
 ;;
 
-let __rshift__ self other =
-  let out__ = CArray.make t 1 in
-  stubs___rshift__ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let __or__ self other = stubs___or__ self other |> with_tensor_gc
+let __or__tensor_ self other = stubs___or__tensor_ self other |> with_tensor_gc
+let __rshift__ self other = stubs___rshift__ self other |> with_tensor_gc
 
 let __rshift__scalar_out_ ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs___rshift__scalar_out_ (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs___rshift__scalar_out_ out self other |> with_tensor_gc
 ;;
 
-let __rshift__tensor_ self other =
-  let out__ = CArray.make t 1 in
-  stubs___rshift__tensor_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let __rshift__tensor_ self other = stubs___rshift__tensor_ self other |> with_tensor_gc
 
 let __rshift__tensor_out_ ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs___rshift__tensor_out_ (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let __xor__ self other =
-  let out__ = CArray.make t 1 in
-  stubs___xor__ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs___rshift__tensor_out_ out self other |> with_tensor_gc
 ;;
 
-let __xor__tensor_ self other =
-  let out__ = CArray.make t 1 in
-  stubs___xor__tensor_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let __xor__ self other = stubs___xor__ self other |> with_tensor_gc
+let __xor__tensor_ self other = stubs___xor__tensor_ self other |> with_tensor_gc
 
 let _adaptive_avg_pool2d self ~output_size =
-  let out__ = CArray.make t 1 in
   stubs__adaptive_avg_pool2d
-    (CArray.start out__)
     self
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
-    (List.length output_size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length output_size)
+  |> with_tensor_gc
 ;;
 
 let _adaptive_avg_pool2d_backward ~grad_output self =
-  let out__ = CArray.make t 1 in
-  stubs__adaptive_avg_pool2d_backward (CArray.start out__) grad_output self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__adaptive_avg_pool2d_backward grad_output self |> with_tensor_gc
 ;;
 
 let _adaptive_avg_pool2d_backward_out ~out ~grad_output self =
-  let out__ = CArray.make t 1 in
-  stubs__adaptive_avg_pool2d_backward_out (CArray.start out__) out grad_output self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__adaptive_avg_pool2d_backward_out out grad_output self |> with_tensor_gc
 ;;
 
 let _adaptive_avg_pool2d_out ~out self ~output_size =
-  let out__ = CArray.make t 1 in
   stubs__adaptive_avg_pool2d_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
-    (List.length output_size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length output_size)
+  |> with_tensor_gc
 ;;
 
 let _adaptive_avg_pool3d self ~output_size =
-  let out__ = CArray.make t 1 in
   stubs__adaptive_avg_pool3d
-    (CArray.start out__)
     self
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
-    (List.length output_size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length output_size)
+  |> with_tensor_gc
 ;;
 
 let _adaptive_avg_pool3d_backward ~grad_output self =
-  let out__ = CArray.make t 1 in
-  stubs__adaptive_avg_pool3d_backward (CArray.start out__) grad_output self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__adaptive_avg_pool3d_backward grad_output self |> with_tensor_gc
 ;;
 
 let _adaptive_avg_pool3d_backward_out ~out ~grad_output self =
-  let out__ = CArray.make t 1 in
-  stubs__adaptive_avg_pool3d_backward_out (CArray.start out__) out grad_output self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__adaptive_avg_pool3d_backward_out out grad_output self |> with_tensor_gc
 ;;
 
 let _adaptive_avg_pool3d_out ~out self ~output_size =
-  let out__ = CArray.make t 1 in
   stubs__adaptive_avg_pool3d_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
-    (List.length output_size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length output_size)
+  |> with_tensor_gc
 ;;
 
 let _add_batch_dim self ~batch_dim ~level =
-  let out__ = CArray.make t 1 in
-  stubs__add_batch_dim
-    (CArray.start out__)
-    self
-    (Int64.of_int batch_dim)
-    (Int64.of_int level);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let _add_relu self other =
-  let out__ = CArray.make t 1 in
-  stubs__add_relu (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let _add_relu_ self other =
-  let out__ = CArray.make t 1 in
-  stubs__add_relu_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let _add_relu_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs__add_relu_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let _add_relu_scalar self other =
-  let out__ = CArray.make t 1 in
-  stubs__add_relu_scalar (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__add_batch_dim self (Int64.of_int batch_dim) (Int64.of_int level)
+  |> with_tensor_gc
 ;;
 
-let _add_relu_scalar_ self other =
-  let out__ = CArray.make t 1 in
-  stubs__add_relu_scalar_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let _add_relu self other = stubs__add_relu self other |> with_tensor_gc
+let _add_relu_ self other = stubs__add_relu_ self other |> with_tensor_gc
+let _add_relu_out ~out self other = stubs__add_relu_out out self other |> with_tensor_gc
+let _add_relu_scalar self other = stubs__add_relu_scalar self other |> with_tensor_gc
+let _add_relu_scalar_ self other = stubs__add_relu_scalar_ self other |> with_tensor_gc
 
 let _add_relu_scalar_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs__add_relu_scalar_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__add_relu_scalar_out out self other |> with_tensor_gc
 ;;
 
 let _addmm_activation self ~mat1 ~mat2 ~use_gelu =
-  let out__ = CArray.make t 1 in
-  stubs__addmm_activation (CArray.start out__) self mat1 mat2 (if use_gelu then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__addmm_activation self mat1 mat2 (if use_gelu then 1 else 0) |> with_tensor_gc
 ;;
 
 let _addmm_activation_out ~out self ~mat1 ~mat2 ~use_gelu =
-  let out__ = CArray.make t 1 in
-  stubs__addmm_activation_out
-    (CArray.start out__)
-    out
-    self
-    mat1
-    mat2
-    (if use_gelu then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__addmm_activation_out out self mat1 mat2 (if use_gelu then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _aminmax self =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs__aminmax (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let _aminmax_dim self ~dim ~keepdim =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs__aminmax_dim
     (CArray.start out__)
     self
     (Int64.of_int dim)
     (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let _aminmax_dim_out ~out0 ~out1 self ~dim ~keepdim =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs__aminmax_dim_out
     (CArray.start out__)
     out0
@@ -407,20 +149,16 @@ let _aminmax_dim_out ~out0 ~out1 self ~dim ~keepdim =
     self
     (Int64.of_int dim)
     (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let _aminmax_out ~out0 ~out1 self =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs__aminmax_out (CArray.start out__) out0 out1 self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
@@ -432,7 +170,7 @@ let _amp_update_scale
   ~scale_backoff_factor
   ~growth_interval
   =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs__amp_update_scale
     (CArray.start out__)
     self
@@ -441,10 +179,8 @@ let _amp_update_scale
     scale_growth_factor
     scale_backoff_factor
     (Int64.of_int growth_interval);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
@@ -456,18 +192,14 @@ let _amp_update_scale_
   ~scale_backoff_factor
   ~growth_interval
   =
-  let out__ = CArray.make t 1 in
   stubs__amp_update_scale_
-    (CArray.start out__)
     self
     growth_tracker
     found_inf
     scale_growth_factor
     scale_backoff_factor
-    (Int64.of_int growth_interval);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int growth_interval)
+  |> with_tensor_gc
 ;;
 
 let _amp_update_scale_out
@@ -479,19 +211,15 @@ let _amp_update_scale_out
   ~scale_backoff_factor
   ~growth_interval
   =
-  let out__ = CArray.make t 1 in
   stubs__amp_update_scale_out
-    (CArray.start out__)
     out
     self
     growth_tracker
     found_inf
     scale_growth_factor
     scale_backoff_factor
-    (Int64.of_int growth_interval);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int growth_interval)
+  |> with_tensor_gc
 ;;
 
 let _assert_tensor_metadata ~a ~size ~stride ~dtype =
@@ -513,254 +241,120 @@ let _assert_tensor_metadata ~a ~size ~stride ~dtype =
 ;;
 
 let _autocast_to_full_precision self ~cuda_enabled ~cpu_enabled =
-  let out__ = CArray.make t 1 in
   stubs__autocast_to_full_precision
-    (CArray.start out__)
     self
     (if cuda_enabled then 1 else 0)
-    (if cpu_enabled then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if cpu_enabled then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _autocast_to_reduced_precision self ~cuda_enabled ~cpu_enabled ~cuda_dtype ~cpu_dtype =
-  let out__ = CArray.make t 1 in
   stubs__autocast_to_reduced_precision
-    (CArray.start out__)
     self
     (if cuda_enabled then 1 else 0)
     (if cpu_enabled then 1 else 0)
     (Kind.packed_to_int cuda_dtype)
-    (Kind.packed_to_int cpu_dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Kind.packed_to_int cpu_dtype)
+  |> with_tensor_gc
 ;;
 
 let _cast_byte self ~non_blocking =
-  let out__ = CArray.make t 1 in
-  stubs__cast_byte (CArray.start out__) self (if non_blocking then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__cast_byte self (if non_blocking then 1 else 0) |> with_tensor_gc
 ;;
 
 let _cast_char self ~non_blocking =
-  let out__ = CArray.make t 1 in
-  stubs__cast_char (CArray.start out__) self (if non_blocking then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__cast_char self (if non_blocking then 1 else 0) |> with_tensor_gc
 ;;
 
 let _cast_double self ~non_blocking =
-  let out__ = CArray.make t 1 in
-  stubs__cast_double (CArray.start out__) self (if non_blocking then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__cast_double self (if non_blocking then 1 else 0) |> with_tensor_gc
 ;;
 
 let _cast_float self ~non_blocking =
-  let out__ = CArray.make t 1 in
-  stubs__cast_float (CArray.start out__) self (if non_blocking then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__cast_float self (if non_blocking then 1 else 0) |> with_tensor_gc
 ;;
 
 let _cast_half self ~non_blocking =
-  let out__ = CArray.make t 1 in
-  stubs__cast_half (CArray.start out__) self (if non_blocking then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__cast_half self (if non_blocking then 1 else 0) |> with_tensor_gc
 ;;
 
 let _cast_int self ~non_blocking =
-  let out__ = CArray.make t 1 in
-  stubs__cast_int (CArray.start out__) self (if non_blocking then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__cast_int self (if non_blocking then 1 else 0) |> with_tensor_gc
 ;;
 
 let _cast_long self ~non_blocking =
-  let out__ = CArray.make t 1 in
-  stubs__cast_long (CArray.start out__) self (if non_blocking then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__cast_long self (if non_blocking then 1 else 0) |> with_tensor_gc
 ;;
 
 let _cast_short self ~non_blocking =
-  let out__ = CArray.make t 1 in
-  stubs__cast_short (CArray.start out__) self (if non_blocking then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__cast_short self (if non_blocking then 1 else 0) |> with_tensor_gc
 ;;
 
 let _cdist_backward ~grad ~x1 ~x2 ~p ~cdist =
-  let out__ = CArray.make t 1 in
-  stubs__cdist_backward (CArray.start out__) grad x1 x2 p cdist;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__cdist_backward grad x1 x2 p cdist |> with_tensor_gc
 ;;
 
 let _cdist_backward_out ~out ~grad ~x1 ~x2 ~p ~cdist =
-  let out__ = CArray.make t 1 in
-  stubs__cdist_backward_out (CArray.start out__) out grad x1 x2 p cdist;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__cdist_backward_out out grad x1 x2 p cdist |> with_tensor_gc
 ;;
 
 let _cholesky_solve_helper self ~a ~upper =
-  let out__ = CArray.make t 1 in
-  stubs__cholesky_solve_helper (CArray.start out__) self a (if upper then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__cholesky_solve_helper self a (if upper then 1 else 0) |> with_tensor_gc
 ;;
 
 let _cholesky_solve_helper_out ~out self ~a ~upper =
-  let out__ = CArray.make t 1 in
-  stubs__cholesky_solve_helper_out
-    (CArray.start out__)
-    out
-    self
-    a
-    (if upper then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let _coalesce self =
-  let out__ = CArray.make t 1 in
-  stubs__coalesce (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__cholesky_solve_helper_out out self a (if upper then 1 else 0) |> with_tensor_gc
 ;;
 
-let _coalesce_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs__coalesce_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let _coalesce self = stubs__coalesce self |> with_tensor_gc
+let _coalesce_out ~out self = stubs__coalesce_out out self |> with_tensor_gc
 
 let _coalesced self ~coalesced =
-  let out__ = CArray.make t 1 in
-  stubs__coalesced (CArray.start out__) self (if coalesced then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__coalesced self (if coalesced then 1 else 0) |> with_tensor_gc
 ;;
 
 let _coalesced_ self ~coalesced =
-  let out__ = CArray.make t 1 in
-  stubs__coalesced_ (CArray.start out__) self (if coalesced then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__coalesced_ self (if coalesced then 1 else 0) |> with_tensor_gc
 ;;
 
 let _coalesced_out ~out self ~coalesced =
-  let out__ = CArray.make t 1 in
-  stubs__coalesced_out (CArray.start out__) out self (if coalesced then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__coalesced_out out self (if coalesced then 1 else 0) |> with_tensor_gc
 ;;
 
 let _compute_linear_combination input ~coefficients =
-  let out__ = CArray.make t 1 in
-  stubs__compute_linear_combination (CArray.start out__) input coefficients;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__compute_linear_combination input coefficients |> with_tensor_gc
 ;;
 
 let _compute_linear_combination_out ~out input ~coefficients =
-  let out__ = CArray.make t 1 in
-  stubs__compute_linear_combination_out (CArray.start out__) out input coefficients;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let _conj self =
-  let out__ = CArray.make t 1 in
-  stubs__conj (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__compute_linear_combination_out out input coefficients |> with_tensor_gc
 ;;
 
-let _conj_copy self =
-  let out__ = CArray.make t 1 in
-  stubs__conj_copy (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let _conj_copy_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs__conj_copy_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let _conj_physical self =
-  let out__ = CArray.make t 1 in
-  stubs__conj_physical (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let _conj_physical_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs__conj_physical_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let _conj self = stubs__conj self |> with_tensor_gc
+let _conj_copy self = stubs__conj_copy self |> with_tensor_gc
+let _conj_copy_out ~out self = stubs__conj_copy_out out self |> with_tensor_gc
+let _conj_physical self = stubs__conj_physical self |> with_tensor_gc
+let _conj_physical_out ~out self = stubs__conj_physical_out out self |> with_tensor_gc
 
 let _conv_depthwise2d self ~weight ~kernel_size ~bias ~stride ~padding ~dilation =
-  let out__ = CArray.make t 1 in
   stubs__conv_depthwise2d
-    (CArray.start out__)
     self
     weight
     (List.map Int64.of_int kernel_size |> CArray.of_list int64_t |> CArray.start)
     (List.length kernel_size)
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
     (List.length stride)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
     (List.length padding)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
-    (List.length dilation);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length dilation)
+  |> with_tensor_gc
 ;;
 
 let _conv_depthwise2d_out ~out self ~weight ~kernel_size ~bias ~stride ~padding ~dilation =
-  let out__ = CArray.make t 1 in
   stubs__conv_depthwise2d_out
-    (CArray.start out__)
     out
     self
     weight
@@ -768,54 +362,40 @@ let _conv_depthwise2d_out ~out self ~weight ~kernel_size ~bias ~stride ~padding
     (List.length kernel_size)
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
     (List.length stride)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
     (List.length padding)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
-    (List.length dilation);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length dilation)
+  |> with_tensor_gc
 ;;
 
 let _convert_indices_from_coo_to_csr self ~size ~out_int32 =
-  let out__ = CArray.make t 1 in
   stubs__convert_indices_from_coo_to_csr
-    (CArray.start out__)
     self
     (Int64.of_int size)
-    (if out_int32 then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if out_int32 then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _convert_indices_from_coo_to_csr_out ~out self ~size ~out_int32 =
-  let out__ = CArray.make t 1 in
   stubs__convert_indices_from_coo_to_csr_out
-    (CArray.start out__)
     out
     self
     (Int64.of_int size)
-    (if out_int32 then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if out_int32 then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _convert_indices_from_csr_to_coo ~crow_indices ~col_indices ~out_int32 ~transpose =
-  let out__ = CArray.make t 1 in
   stubs__convert_indices_from_csr_to_coo
-    (CArray.start out__)
     crow_indices
     col_indices
     (if out_int32 then 1 else 0)
-    (if transpose then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if transpose then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _convert_indices_from_csr_to_coo_out
@@ -825,17 +405,13 @@ let _convert_indices_from_csr_to_coo_out
   ~out_int32
   ~transpose
   =
-  let out__ = CArray.make t 1 in
   stubs__convert_indices_from_csr_to_coo_out
-    (CArray.start out__)
     out
     crow_indices
     col_indices
     (if out_int32 then 1 else 0)
-    (if transpose then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if transpose then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _convolution
@@ -853,14 +429,12 @@ let _convolution
   ~cudnn_enabled
   ~allow_tf32
   =
-  let out__ = CArray.make t 1 in
   stubs__convolution
-    (CArray.start out__)
     input
     weight
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
     (List.length stride)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
@@ -874,10 +448,8 @@ let _convolution
     (if benchmark then 1 else 0)
     (if deterministic then 1 else 0)
     (if cudnn_enabled then 1 else 0)
-    (if allow_tf32 then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if allow_tf32 then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _convolution_deprecated
@@ -894,14 +466,12 @@ let _convolution_deprecated
   ~deterministic
   ~cudnn_enabled
   =
-  let out__ = CArray.make t 1 in
   stubs__convolution_deprecated
-    (CArray.start out__)
     input
     weight
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
     (List.length stride)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
@@ -914,30 +484,24 @@ let _convolution_deprecated
     (Int64.of_int groups)
     (if benchmark then 1 else 0)
     (if deterministic then 1 else 0)
-    (if cudnn_enabled then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if cudnn_enabled then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _convolution_mode input ~weight ~bias ~stride ~padding ~dilation ~groups =
-  let out__ = CArray.make t 1 in
   stubs__convolution_mode
-    (CArray.start out__)
     input
     weight
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
     (List.length stride)
     padding
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
-    (Int64.of_int groups);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int groups)
+  |> with_tensor_gc
 ;;
 
 let _convolution_out
@@ -956,15 +520,13 @@ let _convolution_out
   ~cudnn_enabled
   ~allow_tf32
   =
-  let out__ = CArray.make t 1 in
   stubs__convolution_out
-    (CArray.start out__)
     out
     input
     weight
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
     (List.length stride)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
@@ -978,69 +540,41 @@ let _convolution_out
     (if benchmark then 1 else 0)
     (if deterministic then 1 else 0)
     (if cudnn_enabled then 1 else 0)
-    (if allow_tf32 then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if allow_tf32 then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _copy_from self ~dst ~non_blocking =
-  let out__ = CArray.make t 1 in
-  stubs__copy_from (CArray.start out__) self dst (if non_blocking then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__copy_from self dst (if non_blocking then 1 else 0) |> with_tensor_gc
 ;;
 
 let _copy_from_and_resize self ~dst =
-  let out__ = CArray.make t 1 in
-  stubs__copy_from_and_resize (CArray.start out__) self dst;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__copy_from_and_resize self dst |> with_tensor_gc
 ;;
 
 let _copy_from_and_resize_out ~out self ~dst =
-  let out__ = CArray.make t 1 in
-  stubs__copy_from_and_resize_out (CArray.start out__) out self dst;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__copy_from_and_resize_out out self dst |> with_tensor_gc
 ;;
 
 let _copy_from_out ~out self ~dst ~non_blocking =
-  let out__ = CArray.make t 1 in
-  stubs__copy_from_out (CArray.start out__) out self dst (if non_blocking then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__copy_from_out out self dst (if non_blocking then 1 else 0) |> with_tensor_gc
 ;;
 
-let _cslt_compress input =
-  let out__ = CArray.make t 1 in
-  stubs__cslt_compress (CArray.start out__) input;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let _cslt_compress input = stubs__cslt_compress input |> with_tensor_gc
 
 let _cslt_sparse_mm ~compressed_a ~dense_b ~bias ~transpose_result =
-  let out__ = CArray.make t 1 in
   stubs__cslt_sparse_mm
-    (CArray.start out__)
     compressed_a
     dense_b
     (match bias with
      | Some v -> v
-     | None -> null)
-    (if transpose_result then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+    (if transpose_result then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _ctc_loss ~log_probs ~targets ~input_lengths ~target_lengths ~blank ~zero_infinity =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs__ctc_loss
     (CArray.start out__)
     log_probs
@@ -1051,10 +585,8 @@ let _ctc_loss ~log_probs ~targets ~input_lengths ~target_lengths ~blank ~zero_in
     (List.length target_lengths)
     (Int64.of_int blank)
     (if zero_infinity then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
@@ -1069,9 +601,7 @@ let _ctc_loss_backward
   ~blank
   ~zero_infinity
   =
-  let out__ = CArray.make t 1 in
   stubs__ctc_loss_backward
-    (CArray.start out__)
     grad
     log_probs
     targets
@@ -1082,10 +612,8 @@ let _ctc_loss_backward
     neg_log_likelihood
     log_alpha
     (Int64.of_int blank)
-    (if zero_infinity then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if zero_infinity then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _ctc_loss_backward_out
@@ -1100,9 +628,7 @@ let _ctc_loss_backward_out
   ~blank
   ~zero_infinity
   =
-  let out__ = CArray.make t 1 in
   stubs__ctc_loss_backward_out
-    (CArray.start out__)
     out
     grad
     log_probs
@@ -1114,10 +640,8 @@ let _ctc_loss_backward_out
     neg_log_likelihood
     log_alpha
     (Int64.of_int blank)
-    (if zero_infinity then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if zero_infinity then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _ctc_loss_backward_tensor
@@ -1131,9 +655,7 @@ let _ctc_loss_backward_tensor
   ~blank
   ~zero_infinity
   =
-  let out__ = CArray.make t 1 in
   stubs__ctc_loss_backward_tensor
-    (CArray.start out__)
     grad
     log_probs
     targets
@@ -1142,10 +664,8 @@ let _ctc_loss_backward_tensor
     neg_log_likelihood
     log_alpha
     (Int64.of_int blank)
-    (if zero_infinity then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if zero_infinity then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _ctc_loss_out
@@ -1158,7 +678,7 @@ let _ctc_loss_out
   ~blank
   ~zero_infinity
   =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs__ctc_loss_out
     (CArray.start out__)
     out0
@@ -1171,10 +691,8 @@ let _ctc_loss_out
     (List.length target_lengths)
     (Int64.of_int blank)
     (if zero_infinity then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
@@ -1186,7 +704,7 @@ let _ctc_loss_tensor
   ~blank
   ~zero_infinity
   =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs__ctc_loss_tensor
     (CArray.start out__)
     log_probs
@@ -1195,10 +713,8 @@ let _ctc_loss_tensor
     target_lengths
     (Int64.of_int blank)
     (if zero_infinity then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
@@ -1212,7 +728,7 @@ let _ctc_loss_tensor_out
   ~blank
   ~zero_infinity
   =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs__ctc_loss_tensor_out
     (CArray.start out__)
     out0
@@ -1223,10 +739,8 @@ let _ctc_loss_tensor_out
     target_lengths
     (Int64.of_int blank)
     (if zero_infinity then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
@@ -1239,7 +753,7 @@ let _cudnn_ctc_loss
   ~deterministic
   ~zero_infinity
   =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs__cudnn_ctc_loss
     (CArray.start out__)
     log_probs
@@ -1251,10 +765,8 @@ let _cudnn_ctc_loss
     (Int64.of_int blank)
     (if deterministic then 1 else 0)
     (if zero_infinity then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
@@ -1269,7 +781,7 @@ let _cudnn_ctc_loss_out
   ~deterministic
   ~zero_infinity
   =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs__cudnn_ctc_loss_out
     (CArray.start out__)
     out0
@@ -1283,10 +795,8 @@ let _cudnn_ctc_loss_out
     (Int64.of_int blank)
     (if deterministic then 1 else 0)
     (if zero_infinity then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
@@ -1299,7 +809,7 @@ let _cudnn_ctc_loss_tensor
   ~deterministic
   ~zero_infinity
   =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs__cudnn_ctc_loss_tensor
     (CArray.start out__)
     log_probs
@@ -1309,38 +819,28 @@ let _cudnn_ctc_loss_tensor
     (Int64.of_int blank)
     (if deterministic then 1 else 0)
     (if zero_infinity then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let _cudnn_init_dropout_state ~dropout ~train ~dropout_seed ~options =
-  let out__ = CArray.make t 1 in
   stubs__cudnn_init_dropout_state
-    (CArray.start out__)
     dropout
     (if train then 1 else 0)
     (Int64.of_int dropout_seed)
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let _cudnn_init_dropout_state_out ~out ~dropout ~train ~dropout_seed =
-  let out__ = CArray.make t 1 in
   stubs__cudnn_init_dropout_state_out
-    (CArray.start out__)
     out
     dropout
     (if train then 1 else 0)
-    (Int64.of_int dropout_seed);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int dropout_seed)
+  |> with_tensor_gc
 ;;
 
 let _cudnn_rnn
@@ -1361,20 +861,20 @@ let _cudnn_rnn
   ~batch_sizes
   ~dropout_state
   =
-  let out__ = CArray.make t 5 in
+  let out__ = CArray.make raw_tensor 5 in
   stubs__cudnn_rnn
     (CArray.start out__)
     input
-    (CArray.of_list t weight |> CArray.start)
+    (CArray.of_list gc_tensor weight |> CArray.start)
     (List.length weight)
     (Int64.of_int weight_stride0)
     (match weight_buf with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     hx
     (match cx with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (Int64.of_int mode)
     (Int64.of_int hidden_size)
     (Int64.of_int proj_size)
@@ -1387,17 +887,12 @@ let _cudnn_rnn
     (List.length batch_sizes)
     (match dropout_state with
      | Some v -> v
-     | None -> null);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
-  let t3 = CArray.get out__ 3 in
-  Gc.finalise C.Tensor.free t3;
-  let t4 = CArray.get out__ 4 in
-  Gc.finalise C.Tensor.free t4;
+     | None -> none_gc_tensor);
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
+  let t3 = CArray.get out__ 3 |> with_tensor_gc in
+  let t4 = CArray.get out__ 4 |> with_tensor_gc in
   t0, t1, t2, t3, t4
 ;;
 
@@ -1412,10 +907,8 @@ let _cudnn_rnn_flatten_weight
   ~batch_first
   ~bidirectional
   =
-  let out__ = CArray.make t 1 in
   stubs__cudnn_rnn_flatten_weight
-    (CArray.start out__)
-    (CArray.of_list t weight_arr |> CArray.start)
+    (CArray.of_list gc_tensor weight_arr |> CArray.start)
     (List.length weight_arr)
     (Int64.of_int weight_stride0)
     (Int64.of_int input_size)
@@ -1424,10 +917,8 @@ let _cudnn_rnn_flatten_weight
     (Int64.of_int proj_size)
     (Int64.of_int num_layers)
     (if batch_first then 1 else 0)
-    (if bidirectional then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if bidirectional then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _cudnn_rnn_flatten_weight_out
@@ -1442,11 +933,9 @@ let _cudnn_rnn_flatten_weight_out
   ~batch_first
   ~bidirectional
   =
-  let out__ = CArray.make t 1 in
   stubs__cudnn_rnn_flatten_weight_out
-    (CArray.start out__)
     out
-    (CArray.of_list t weight_arr |> CArray.start)
+    (CArray.of_list gc_tensor weight_arr |> CArray.start)
     (List.length weight_arr)
     (Int64.of_int weight_stride0)
     (Int64.of_int input_size)
@@ -1455,10 +944,8 @@ let _cudnn_rnn_flatten_weight_out
     (Int64.of_int proj_size)
     (Int64.of_int num_layers)
     (if batch_first then 1 else 0)
-    (if bidirectional then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if bidirectional then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _cudnn_rnn_out
@@ -1484,7 +971,7 @@ let _cudnn_rnn_out
   ~batch_sizes
   ~dropout_state
   =
-  let out__ = CArray.make t 5 in
+  let out__ = CArray.make raw_tensor 5 in
   stubs__cudnn_rnn_out
     (CArray.start out__)
     out0
@@ -1493,16 +980,16 @@ let _cudnn_rnn_out
     out3
     out4
     input
-    (CArray.of_list t weight |> CArray.start)
+    (CArray.of_list gc_tensor weight |> CArray.start)
     (List.length weight)
     (Int64.of_int weight_stride0)
     (match weight_buf with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     hx
     (match cx with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (Int64.of_int mode)
     (Int64.of_int hidden_size)
     (Int64.of_int proj_size)
@@ -1515,47 +1002,26 @@ let _cudnn_rnn_out
     (List.length batch_sizes)
     (match dropout_state with
      | Some v -> v
-     | None -> null);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
-  let t3 = CArray.get out__ 3 in
-  Gc.finalise C.Tensor.free t3;
-  let t4 = CArray.get out__ 4 in
-  Gc.finalise C.Tensor.free t4;
+     | None -> none_gc_tensor);
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
+  let t3 = CArray.get out__ 3 |> with_tensor_gc in
+  let t4 = CArray.get out__ 4 |> with_tensor_gc in
   t0, t1, t2, t3, t4
 ;;
 
 let _debug_has_internal_overlap self = stubs__debug_has_internal_overlap self
-
-let _dim_arange ~like ~dim =
-  let out__ = CArray.make t 1 in
-  stubs__dim_arange (CArray.start out__) like (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
+let _dim_arange ~like ~dim = stubs__dim_arange like (Int64.of_int dim) |> with_tensor_gc
 let _dimi self = stubs__dimi self
 let _dimv self = stubs__dimv self
 
 let _dirichlet_grad ~x ~alpha ~total =
-  let out__ = CArray.make t 1 in
-  stubs__dirichlet_grad (CArray.start out__) x alpha total;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__dirichlet_grad x alpha total |> with_tensor_gc
 ;;
 
 let _dirichlet_grad_out ~out ~x ~alpha ~total =
-  let out__ = CArray.make t 1 in
-  stubs__dirichlet_grad_out (CArray.start out__) out x alpha total;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__dirichlet_grad_out out x alpha total |> with_tensor_gc
 ;;
 
 let _efficient_attention_backward
@@ -1578,7 +1044,7 @@ let _efficient_attention_backward
   ~scale
   ~num_splits_key
   =
-  let out__ = CArray.make t 4 in
+  let out__ = CArray.make raw_tensor 4 in
   stubs__efficient_attention_backward
     (CArray.start out__)
     grad_out_
@@ -1587,14 +1053,14 @@ let _efficient_attention_backward
     value
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     out
     (match cu_seqlens_q with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match cu_seqlens_k with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (Int64.of_int max_seqlen_k)
     (Int64.of_int max_seqlen_q)
     logsumexp
@@ -1613,40 +1079,28 @@ let _efficient_attention_backward
     (match num_splits_key with
      | Some _ -> 0
      | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
-  let t3 = CArray.get out__ 3 in
-  Gc.finalise C.Tensor.free t3;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
+  let t3 = CArray.get out__ 3 |> with_tensor_gc in
   t0, t1, t2, t3
 ;;
 
 let _efficientzerotensor ~size ~options =
-  let out__ = CArray.make t 1 in
   stubs__efficientzerotensor
-    (CArray.start out__)
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let _efficientzerotensor_out ~out ~size =
-  let out__ = CArray.make t 1 in
   stubs__efficientzerotensor_out
-    (CArray.start out__)
     out
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
-    (List.length size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length size)
+  |> with_tensor_gc
 ;;
 
 let _embedding_bag
@@ -1660,7 +1114,7 @@ let _embedding_bag
   ~include_last_offset
   ~padding_idx
   =
-  let out__ = CArray.make t 4 in
+  let out__ = CArray.make raw_tensor 4 in
   stubs__embedding_bag
     (CArray.start out__)
     weight
@@ -1671,17 +1125,13 @@ let _embedding_bag
     (if sparse then 1 else 0)
     (match per_sample_weights with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (if include_last_offset then 1 else 0)
     (Int64.of_int padding_idx);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
-  let t3 = CArray.get out__ 3 in
-  Gc.finalise C.Tensor.free t3;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
+  let t3 = CArray.get out__ 3 |> with_tensor_gc in
   t0, t1, t2, t3
 ;;
 
@@ -1699,9 +1149,7 @@ let _embedding_bag_backward
   ~per_sample_weights
   ~padding_idx
   =
-  let out__ = CArray.make t 1 in
   stubs__embedding_bag_backward
-    (CArray.start out__)
     grad
     indices
     offsets
@@ -1714,11 +1162,9 @@ let _embedding_bag_backward
     (if sparse then 1 else 0)
     (match per_sample_weights with
      | Some v -> v
-     | None -> null)
-    (Int64.of_int padding_idx);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+    (Int64.of_int padding_idx)
+  |> with_tensor_gc
 ;;
 
 let _embedding_bag_dense_backward
@@ -1733,9 +1179,7 @@ let _embedding_bag_dense_backward
   ~per_sample_weights
   ~padding_idx
   =
-  let out__ = CArray.make t 1 in
   stubs__embedding_bag_dense_backward
-    (CArray.start out__)
     grad
     indices
     offset2bag
@@ -1746,11 +1190,9 @@ let _embedding_bag_dense_backward
     (Int64.of_int mode)
     (match per_sample_weights with
      | Some v -> v
-     | None -> null)
-    (Int64.of_int padding_idx);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+    (Int64.of_int padding_idx)
+  |> with_tensor_gc
 ;;
 
 let _embedding_bag_dense_backward_out
@@ -1766,9 +1208,7 @@ let _embedding_bag_dense_backward_out
   ~per_sample_weights
   ~padding_idx
   =
-  let out__ = CArray.make t 1 in
   stubs__embedding_bag_dense_backward_out
-    (CArray.start out__)
     out
     grad
     indices
@@ -1780,11 +1220,9 @@ let _embedding_bag_dense_backward_out
     (Int64.of_int mode)
     (match per_sample_weights with
      | Some v -> v
-     | None -> null)
-    (Int64.of_int padding_idx);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+    (Int64.of_int padding_idx)
+  |> with_tensor_gc
 ;;
 
 let _embedding_bag_forward_only
@@ -1798,7 +1236,7 @@ let _embedding_bag_forward_only
   ~include_last_offset
   ~padding_idx
   =
-  let out__ = CArray.make t 4 in
+  let out__ = CArray.make raw_tensor 4 in
   stubs__embedding_bag_forward_only
     (CArray.start out__)
     weight
@@ -1809,17 +1247,13 @@ let _embedding_bag_forward_only
     (if sparse then 1 else 0)
     (match per_sample_weights with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (if include_last_offset then 1 else 0)
     (Int64.of_int padding_idx);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
-  let t3 = CArray.get out__ 3 in
-  Gc.finalise C.Tensor.free t3;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
+  let t3 = CArray.get out__ 3 |> with_tensor_gc in
   t0, t1, t2, t3
 ;;
 
@@ -1838,7 +1272,7 @@ let _embedding_bag_forward_only_out
   ~include_last_offset
   ~padding_idx
   =
-  let out__ = CArray.make t 4 in
+  let out__ = CArray.make raw_tensor 4 in
   stubs__embedding_bag_forward_only_out
     (CArray.start out__)
     out0
@@ -1853,17 +1287,13 @@ let _embedding_bag_forward_only_out
     (if sparse then 1 else 0)
     (match per_sample_weights with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (if include_last_offset then 1 else 0)
     (Int64.of_int padding_idx);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
-  let t3 = CArray.get out__ 3 in
-  Gc.finalise C.Tensor.free t3;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
+  let t3 = CArray.get out__ 3 |> with_tensor_gc in
   t0, t1, t2, t3
 ;;
 
@@ -1882,7 +1312,7 @@ let _embedding_bag_out
   ~include_last_offset
   ~padding_idx
   =
-  let out__ = CArray.make t 4 in
+  let out__ = CArray.make raw_tensor 4 in
   stubs__embedding_bag_out
     (CArray.start out__)
     out0
@@ -1897,17 +1327,13 @@ let _embedding_bag_out
     (if sparse then 1 else 0)
     (match per_sample_weights with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (if include_last_offset then 1 else 0)
     (Int64.of_int padding_idx);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
-  let t3 = CArray.get out__ 3 in
-  Gc.finalise C.Tensor.free t3;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
+  let t3 = CArray.get out__ 3 |> with_tensor_gc in
   t0, t1, t2, t3
 ;;
 
@@ -1920,19 +1346,15 @@ let _embedding_bag_per_sample_weights_backward
   ~mode
   ~padding_idx
   =
-  let out__ = CArray.make t 1 in
   stubs__embedding_bag_per_sample_weights_backward
-    (CArray.start out__)
     grad
     weight
     indices
     offsets
     offset2bag
     (Int64.of_int mode)
-    (Int64.of_int padding_idx);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int padding_idx)
+  |> with_tensor_gc
 ;;
 
 let _embedding_bag_per_sample_weights_backward_out
@@ -1945,9 +1367,7 @@ let _embedding_bag_per_sample_weights_backward_out
   ~mode
   ~padding_idx
   =
-  let out__ = CArray.make t 1 in
   stubs__embedding_bag_per_sample_weights_backward_out
-    (CArray.start out__)
     out
     grad
     weight
@@ -1955,10 +1375,8 @@ let _embedding_bag_per_sample_weights_backward_out
     offsets
     offset2bag
     (Int64.of_int mode)
-    (Int64.of_int padding_idx);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int padding_idx)
+  |> with_tensor_gc
 ;;
 
 let _embedding_bag_sparse_backward
@@ -1973,9 +1391,7 @@ let _embedding_bag_sparse_backward
   ~per_sample_weights
   ~padding_idx
   =
-  let out__ = CArray.make t 1 in
   stubs__embedding_bag_sparse_backward
-    (CArray.start out__)
     grad
     indices
     offsets
@@ -1986,87 +1402,59 @@ let _embedding_bag_sparse_backward
     (Int64.of_int mode)
     (match per_sample_weights with
      | Some v -> v
-     | None -> null)
-    (Int64.of_int padding_idx);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+    (Int64.of_int padding_idx)
+  |> with_tensor_gc
 ;;
 
 let _empty_affine_quantized ~size ~options ~scale ~zero_point =
-  let out__ = CArray.make t 1 in
   stubs__empty_affine_quantized
-    (CArray.start out__)
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     (Kind.packed_to_int (fst options))
     (Device.to_int (snd options))
     scale
-    (Int64.of_int zero_point);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int zero_point)
+  |> with_tensor_gc
 ;;
 
 let _empty_affine_quantized_out ~out ~size ~scale ~zero_point =
-  let out__ = CArray.make t 1 in
   stubs__empty_affine_quantized_out
-    (CArray.start out__)
     out
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     scale
-    (Int64.of_int zero_point);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int zero_point)
+  |> with_tensor_gc
 ;;
 
 let _empty_per_channel_affine_quantized ~size ~scales ~zero_points ~axis ~options =
-  let out__ = CArray.make t 1 in
   stubs__empty_per_channel_affine_quantized
-    (CArray.start out__)
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     scales
     zero_points
     (Int64.of_int axis)
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let _empty_per_channel_affine_quantized_out ~out ~size ~scales ~zero_points ~axis =
-  let out__ = CArray.make t 1 in
   stubs__empty_per_channel_affine_quantized_out
-    (CArray.start out__)
     out
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     scales
     zero_points
-    (Int64.of_int axis);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int axis)
+  |> with_tensor_gc
 ;;
 
-let _euclidean_dist ~x1 ~x2 =
-  let out__ = CArray.make t 1 in
-  stubs__euclidean_dist (CArray.start out__) x1 x2;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let _euclidean_dist ~x1 ~x2 = stubs__euclidean_dist x1 x2 |> with_tensor_gc
 
 let _euclidean_dist_out ~out ~x1 ~x2 =
-  let out__ = CArray.make t 1 in
-  stubs__euclidean_dist_out (CArray.start out__) out x1 x2;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__euclidean_dist_out out x1 x2 |> with_tensor_gc
 ;;
 
 let _fake_quantize_learnable_per_channel_affine
@@ -2078,19 +1466,15 @@ let _fake_quantize_learnable_per_channel_affine
   ~quant_max
   ~grad_factor
   =
-  let out__ = CArray.make t 1 in
   stubs__fake_quantize_learnable_per_channel_affine
-    (CArray.start out__)
     self
     scale
     zero_point
     (Int64.of_int axis)
     (Int64.of_int quant_min)
     (Int64.of_int quant_max)
-    grad_factor;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    grad_factor
+  |> with_tensor_gc
 ;;
 
 let _fake_quantize_learnable_per_channel_affine_backward
@@ -2103,7 +1487,7 @@ let _fake_quantize_learnable_per_channel_affine_backward
   ~quant_max
   ~grad_factor
   =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs__fake_quantize_learnable_per_channel_affine_backward
     (CArray.start out__)
     grad
@@ -2114,12 +1498,9 @@ let _fake_quantize_learnable_per_channel_affine_backward
     (Int64.of_int quant_min)
     (Int64.of_int quant_max)
     grad_factor;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
@@ -2133,9 +1514,7 @@ let _fake_quantize_learnable_per_channel_affine_out
   ~quant_max
   ~grad_factor
   =
-  let out__ = CArray.make t 1 in
   stubs__fake_quantize_learnable_per_channel_affine_out
-    (CArray.start out__)
     out
     self
     scale
@@ -2143,10 +1522,8 @@ let _fake_quantize_learnable_per_channel_affine_out
     (Int64.of_int axis)
     (Int64.of_int quant_min)
     (Int64.of_int quant_max)
-    grad_factor;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    grad_factor
+  |> with_tensor_gc
 ;;
 
 let _fake_quantize_learnable_per_tensor_affine
@@ -2157,18 +1534,14 @@ let _fake_quantize_learnable_per_tensor_affine
   ~quant_max
   ~grad_factor
   =
-  let out__ = CArray.make t 1 in
   stubs__fake_quantize_learnable_per_tensor_affine
-    (CArray.start out__)
     self
     scale
     zero_point
     (Int64.of_int quant_min)
     (Int64.of_int quant_max)
-    grad_factor;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    grad_factor
+  |> with_tensor_gc
 ;;
 
 let _fake_quantize_learnable_per_tensor_affine_backward
@@ -2180,7 +1553,7 @@ let _fake_quantize_learnable_per_tensor_affine_backward
   ~quant_max
   ~grad_factor
   =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs__fake_quantize_learnable_per_tensor_affine_backward
     (CArray.start out__)
     grad
@@ -2190,12 +1563,9 @@ let _fake_quantize_learnable_per_tensor_affine_backward
     (Int64.of_int quant_min)
     (Int64.of_int quant_max)
     grad_factor;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
@@ -2208,19 +1578,15 @@ let _fake_quantize_learnable_per_tensor_affine_out
   ~quant_max
   ~grad_factor
   =
-  let out__ = CArray.make t 1 in
   stubs__fake_quantize_learnable_per_tensor_affine_out
-    (CArray.start out__)
     out
     self
     scale
     zero_point
     (Int64.of_int quant_min)
     (Int64.of_int quant_max)
-    grad_factor;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    grad_factor
+  |> with_tensor_gc
 ;;
 
 let _fake_quantize_per_tensor_affine_cachemask_tensor_qparams
@@ -2231,7 +1597,7 @@ let _fake_quantize_per_tensor_affine_cachemask_tensor_qparams
   ~quant_min
   ~quant_max
   =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs__fake_quantize_per_tensor_affine_cachemask_tensor_qparams
     (CArray.start out__)
     self
@@ -2240,10 +1606,8 @@ let _fake_quantize_per_tensor_affine_cachemask_tensor_qparams
     fake_quant_enabled
     (Int64.of_int quant_min)
     (Int64.of_int quant_max);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
@@ -2257,7 +1621,7 @@ let _fake_quantize_per_tensor_affine_cachemask_tensor_qparams_out
   ~quant_min
   ~quant_max
   =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs__fake_quantize_per_tensor_affine_cachemask_tensor_qparams_out
     (CArray.start out__)
     out0
@@ -2268,111 +1632,81 @@ let _fake_quantize_per_tensor_affine_cachemask_tensor_qparams_out
     fake_quant_enabled
     (Int64.of_int quant_min)
     (Int64.of_int quant_max);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let _fft_c2c self ~dim ~normalization ~forward =
-  let out__ = CArray.make t 1 in
   stubs__fft_c2c
-    (CArray.start out__)
     self
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
     (List.length dim)
     (Int64.of_int normalization)
-    (if forward then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if forward then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _fft_c2c_out ~out self ~dim ~normalization ~forward =
-  let out__ = CArray.make t 1 in
   stubs__fft_c2c_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
     (List.length dim)
     (Int64.of_int normalization)
-    (if forward then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if forward then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _fft_c2r self ~dim ~normalization ~last_dim_size =
-  let out__ = CArray.make t 1 in
   stubs__fft_c2r
-    (CArray.start out__)
     self
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
     (List.length dim)
     (Int64.of_int normalization)
-    (Int64.of_int last_dim_size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int last_dim_size)
+  |> with_tensor_gc
 ;;
 
 let _fft_c2r_out ~out self ~dim ~normalization ~last_dim_size =
-  let out__ = CArray.make t 1 in
   stubs__fft_c2r_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
     (List.length dim)
     (Int64.of_int normalization)
-    (Int64.of_int last_dim_size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int last_dim_size)
+  |> with_tensor_gc
 ;;
 
 let _fft_r2c self ~dim ~normalization ~onesided =
-  let out__ = CArray.make t 1 in
   stubs__fft_r2c
-    (CArray.start out__)
     self
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
     (List.length dim)
     (Int64.of_int normalization)
-    (if onesided then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if onesided then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _fft_r2c_out ~out self ~dim ~normalization ~onesided =
-  let out__ = CArray.make t 1 in
   stubs__fft_r2c_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
     (List.length dim)
     (Int64.of_int normalization)
-    (if onesided then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if onesided then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _fill_mem_eff_dropout_mask_ self ~dropout_p ~seed ~offset =
-  let out__ = CArray.make t 1 in
   stubs__fill_mem_eff_dropout_mask_
-    (CArray.start out__)
     self
     dropout_p
     (Int64.of_int seed)
-    (Int64.of_int offset);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int offset)
+  |> with_tensor_gc
 ;;
 
 let _flash_attention_backward
@@ -2392,7 +1726,7 @@ let _flash_attention_backward
   ~philox_offset
   ~scale
   =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs__flash_attention_backward
     (CArray.start out__)
     grad_out
@@ -2413,54 +1747,37 @@ let _flash_attention_backward
     (match scale with
      | Some _ -> 0
      | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
 let _foobar self ~arg1 ~arg2 ~arg3 =
-  let out__ = CArray.make t 1 in
   stubs__foobar
-    (CArray.start out__)
     self
     (if arg1 then 1 else 0)
     (if arg2 then 1 else 0)
-    (if arg3 then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if arg3 then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _foobar_out ~out self ~arg1 ~arg2 ~arg3 =
-  let out__ = CArray.make t 1 in
   stubs__foobar_out
-    (CArray.start out__)
     out
     self
     (if arg1 then 1 else 0)
     (if arg2 then 1 else 0)
-    (if arg3 then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if arg3 then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _functional_assert_async self ~assert_msg ~dep_token =
-  let out__ = CArray.make t 1 in
-  stubs__functional_assert_async (CArray.start out__) self assert_msg dep_token;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__functional_assert_async self assert_msg dep_token |> with_tensor_gc
 ;;
 
 let _functional_sym_constrain_range ~size ~min ~max ~dep_token =
-  let out__ = CArray.make t 1 in
   stubs__functional_sym_constrain_range
-    (CArray.start out__)
     size
     (match min with
      | None -> Int64.zero
@@ -2474,16 +1791,12 @@ let _functional_sym_constrain_range ~size ~min ~max ~dep_token =
     (match max with
      | Some _ -> 0
      | None -> 1)
-    dep_token;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    dep_token
+  |> with_tensor_gc
 ;;
 
 let _functional_sym_constrain_range_for_size ~size ~min ~max ~dep_token =
-  let out__ = CArray.make t 1 in
   stubs__functional_sym_constrain_range_for_size
-    (CArray.start out__)
     size
     (match min with
      | None -> Int64.zero
@@ -2497,10 +1810,8 @@ let _functional_sym_constrain_range_for_size ~size ~min ~max ~dep_token =
     (match max with
      | Some _ -> 0
      | None -> 1)
-    dep_token;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    dep_token
+  |> with_tensor_gc
 ;;
 
 let _fused_adam
@@ -2522,19 +1833,19 @@ let _fused_adam
   ~found_inf
   =
   stubs__fused_adam
-    (CArray.of_list t out |> CArray.start)
+    (CArray.of_list gc_tensor out |> CArray.start)
     (List.length out)
-    (CArray.of_list t self |> CArray.start)
+    (CArray.of_list gc_tensor self |> CArray.start)
     (List.length self)
-    (CArray.of_list t grads |> CArray.start)
+    (CArray.of_list gc_tensor grads |> CArray.start)
     (List.length grads)
-    (CArray.of_list t exp_avgs |> CArray.start)
+    (CArray.of_list gc_tensor exp_avgs |> CArray.start)
     (List.length exp_avgs)
-    (CArray.of_list t exp_avg_sqs |> CArray.start)
+    (CArray.of_list gc_tensor exp_avg_sqs |> CArray.start)
     (List.length exp_avg_sqs)
-    (CArray.of_list t max_exp_avg_sqs |> CArray.start)
+    (CArray.of_list gc_tensor max_exp_avg_sqs |> CArray.start)
     (List.length max_exp_avg_sqs)
-    (CArray.of_list t state_steps |> CArray.start)
+    (CArray.of_list gc_tensor state_steps |> CArray.start)
     (List.length state_steps)
     lr
     beta1
@@ -2545,10 +1856,10 @@ let _fused_adam
     (if maximize then 1 else 0)
     (match grad_scale with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match found_inf with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
 ;;
 
 let _fused_adam_
@@ -2569,17 +1880,17 @@ let _fused_adam_
   ~found_inf
   =
   stubs__fused_adam_
-    (CArray.of_list t self |> CArray.start)
+    (CArray.of_list gc_tensor self |> CArray.start)
     (List.length self)
-    (CArray.of_list t grads |> CArray.start)
+    (CArray.of_list gc_tensor grads |> CArray.start)
     (List.length grads)
-    (CArray.of_list t exp_avgs |> CArray.start)
+    (CArray.of_list gc_tensor exp_avgs |> CArray.start)
     (List.length exp_avgs)
-    (CArray.of_list t exp_avg_sqs |> CArray.start)
+    (CArray.of_list gc_tensor exp_avg_sqs |> CArray.start)
     (List.length exp_avg_sqs)
-    (CArray.of_list t max_exp_avg_sqs |> CArray.start)
+    (CArray.of_list gc_tensor max_exp_avg_sqs |> CArray.start)
     (List.length max_exp_avg_sqs)
-    (CArray.of_list t state_steps |> CArray.start)
+    (CArray.of_list gc_tensor state_steps |> CArray.start)
     (List.length state_steps)
     lr
     beta1
@@ -2590,10 +1901,10 @@ let _fused_adam_
     (if maximize then 1 else 0)
     (match grad_scale with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match found_inf with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
 ;;
 
 let _fused_adam_tensor_lr_
@@ -2614,17 +1925,17 @@ let _fused_adam_tensor_lr_
   ~found_inf
   =
   stubs__fused_adam_tensor_lr_
-    (CArray.of_list t self |> CArray.start)
+    (CArray.of_list gc_tensor self |> CArray.start)
     (List.length self)
-    (CArray.of_list t grads |> CArray.start)
+    (CArray.of_list gc_tensor grads |> CArray.start)
     (List.length grads)
-    (CArray.of_list t exp_avgs |> CArray.start)
+    (CArray.of_list gc_tensor exp_avgs |> CArray.start)
     (List.length exp_avgs)
-    (CArray.of_list t exp_avg_sqs |> CArray.start)
+    (CArray.of_list gc_tensor exp_avg_sqs |> CArray.start)
     (List.length exp_avg_sqs)
-    (CArray.of_list t max_exp_avg_sqs |> CArray.start)
+    (CArray.of_list gc_tensor max_exp_avg_sqs |> CArray.start)
     (List.length max_exp_avg_sqs)
-    (CArray.of_list t state_steps |> CArray.start)
+    (CArray.of_list gc_tensor state_steps |> CArray.start)
     (List.length state_steps)
     lr
     beta1
@@ -2635,10 +1946,10 @@ let _fused_adam_tensor_lr_
     (if maximize then 1 else 0)
     (match grad_scale with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match found_inf with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
 ;;
 
 let _fused_adam_tensor_lr_out
@@ -2660,19 +1971,19 @@ let _fused_adam_tensor_lr_out
   ~found_inf
   =
   stubs__fused_adam_tensor_lr_out
-    (CArray.of_list t out |> CArray.start)
+    (CArray.of_list gc_tensor out |> CArray.start)
     (List.length out)
-    (CArray.of_list t self |> CArray.start)
+    (CArray.of_list gc_tensor self |> CArray.start)
     (List.length self)
-    (CArray.of_list t grads |> CArray.start)
+    (CArray.of_list gc_tensor grads |> CArray.start)
     (List.length grads)
-    (CArray.of_list t exp_avgs |> CArray.start)
+    (CArray.of_list gc_tensor exp_avgs |> CArray.start)
     (List.length exp_avgs)
-    (CArray.of_list t exp_avg_sqs |> CArray.start)
+    (CArray.of_list gc_tensor exp_avg_sqs |> CArray.start)
     (List.length exp_avg_sqs)
-    (CArray.of_list t max_exp_avg_sqs |> CArray.start)
+    (CArray.of_list gc_tensor max_exp_avg_sqs |> CArray.start)
     (List.length max_exp_avg_sqs)
-    (CArray.of_list t state_steps |> CArray.start)
+    (CArray.of_list gc_tensor state_steps |> CArray.start)
     (List.length state_steps)
     lr
     beta1
@@ -2683,10 +1994,10 @@ let _fused_adam_tensor_lr_out
     (if maximize then 1 else 0)
     (match grad_scale with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match found_inf with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
 ;;
 
 let _fused_adamw
@@ -2708,19 +2019,19 @@ let _fused_adamw
   ~found_inf
   =
   stubs__fused_adamw
-    (CArray.of_list t out |> CArray.start)
+    (CArray.of_list gc_tensor out |> CArray.start)
     (List.length out)
-    (CArray.of_list t self |> CArray.start)
+    (CArray.of_list gc_tensor self |> CArray.start)
     (List.length self)
-    (CArray.of_list t grads |> CArray.start)
+    (CArray.of_list gc_tensor grads |> CArray.start)
     (List.length grads)
-    (CArray.of_list t exp_avgs |> CArray.start)
+    (CArray.of_list gc_tensor exp_avgs |> CArray.start)
     (List.length exp_avgs)
-    (CArray.of_list t exp_avg_sqs |> CArray.start)
+    (CArray.of_list gc_tensor exp_avg_sqs |> CArray.start)
     (List.length exp_avg_sqs)
-    (CArray.of_list t max_exp_avg_sqs |> CArray.start)
+    (CArray.of_list gc_tensor max_exp_avg_sqs |> CArray.start)
     (List.length max_exp_avg_sqs)
-    (CArray.of_list t state_steps |> CArray.start)
+    (CArray.of_list gc_tensor state_steps |> CArray.start)
     (List.length state_steps)
     lr
     beta1
@@ -2731,10 +2042,10 @@ let _fused_adamw
     (if maximize then 1 else 0)
     (match grad_scale with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match found_inf with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
 ;;
 
 let _fused_adamw_
@@ -2755,17 +2066,17 @@ let _fused_adamw_
   ~found_inf
   =
   stubs__fused_adamw_
-    (CArray.of_list t self |> CArray.start)
+    (CArray.of_list gc_tensor self |> CArray.start)
     (List.length self)
-    (CArray.of_list t grads |> CArray.start)
+    (CArray.of_list gc_tensor grads |> CArray.start)
     (List.length grads)
-    (CArray.of_list t exp_avgs |> CArray.start)
+    (CArray.of_list gc_tensor exp_avgs |> CArray.start)
     (List.length exp_avgs)
-    (CArray.of_list t exp_avg_sqs |> CArray.start)
+    (CArray.of_list gc_tensor exp_avg_sqs |> CArray.start)
     (List.length exp_avg_sqs)
-    (CArray.of_list t max_exp_avg_sqs |> CArray.start)
+    (CArray.of_list gc_tensor max_exp_avg_sqs |> CArray.start)
     (List.length max_exp_avg_sqs)
-    (CArray.of_list t state_steps |> CArray.start)
+    (CArray.of_list gc_tensor state_steps |> CArray.start)
     (List.length state_steps)
     lr
     beta1
@@ -2776,10 +2087,10 @@ let _fused_adamw_
     (if maximize then 1 else 0)
     (match grad_scale with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match found_inf with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
 ;;
 
 let _fused_adamw_tensor_lr_
@@ -2800,17 +2111,17 @@ let _fused_adamw_tensor_lr_
   ~found_inf
   =
   stubs__fused_adamw_tensor_lr_
-    (CArray.of_list t self |> CArray.start)
+    (CArray.of_list gc_tensor self |> CArray.start)
     (List.length self)
-    (CArray.of_list t grads |> CArray.start)
+    (CArray.of_list gc_tensor grads |> CArray.start)
     (List.length grads)
-    (CArray.of_list t exp_avgs |> CArray.start)
+    (CArray.of_list gc_tensor exp_avgs |> CArray.start)
     (List.length exp_avgs)
-    (CArray.of_list t exp_avg_sqs |> CArray.start)
+    (CArray.of_list gc_tensor exp_avg_sqs |> CArray.start)
     (List.length exp_avg_sqs)
-    (CArray.of_list t max_exp_avg_sqs |> CArray.start)
+    (CArray.of_list gc_tensor max_exp_avg_sqs |> CArray.start)
     (List.length max_exp_avg_sqs)
-    (CArray.of_list t state_steps |> CArray.start)
+    (CArray.of_list gc_tensor state_steps |> CArray.start)
     (List.length state_steps)
     lr
     beta1
@@ -2821,10 +2132,10 @@ let _fused_adamw_tensor_lr_
     (if maximize then 1 else 0)
     (match grad_scale with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match found_inf with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
 ;;
 
 let _fused_adamw_tensor_lr_out
@@ -2846,19 +2157,19 @@ let _fused_adamw_tensor_lr_out
   ~found_inf
   =
   stubs__fused_adamw_tensor_lr_out
-    (CArray.of_list t out |> CArray.start)
+    (CArray.of_list gc_tensor out |> CArray.start)
     (List.length out)
-    (CArray.of_list t self |> CArray.start)
+    (CArray.of_list gc_tensor self |> CArray.start)
     (List.length self)
-    (CArray.of_list t grads |> CArray.start)
+    (CArray.of_list gc_tensor grads |> CArray.start)
     (List.length grads)
-    (CArray.of_list t exp_avgs |> CArray.start)
+    (CArray.of_list gc_tensor exp_avgs |> CArray.start)
     (List.length exp_avgs)
-    (CArray.of_list t exp_avg_sqs |> CArray.start)
+    (CArray.of_list gc_tensor exp_avg_sqs |> CArray.start)
     (List.length exp_avg_sqs)
-    (CArray.of_list t max_exp_avg_sqs |> CArray.start)
+    (CArray.of_list gc_tensor max_exp_avg_sqs |> CArray.start)
     (List.length max_exp_avg_sqs)
-    (CArray.of_list t state_steps |> CArray.start)
+    (CArray.of_list gc_tensor state_steps |> CArray.start)
     (List.length state_steps)
     lr
     beta1
@@ -2869,29 +2180,25 @@ let _fused_adamw_tensor_lr_out
     (if maximize then 1 else 0)
     (match grad_scale with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match found_inf with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
 ;;
 
 let _fused_dropout self ~p =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs__fused_dropout (CArray.start out__) self p;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let _fused_dropout_out ~out0 ~out1 self ~p =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs__fused_dropout_out (CArray.start out__) out0 out1 self p;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
@@ -2910,7 +2217,7 @@ let _fused_moving_avg_obs_fq_helper
   ~per_row_fake_quant
   ~symmetric_quant
   =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs__fused_moving_avg_obs_fq_helper
     (CArray.start out__)
     self
@@ -2926,10 +2233,8 @@ let _fused_moving_avg_obs_fq_helper
     (Int64.of_int ch_axis)
     (if per_row_fake_quant then 1 else 0)
     (if symmetric_quant then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
@@ -2948,7 +2253,7 @@ let _fused_moving_avg_obs_fq_helper_functional
   ~per_row_fake_quant
   ~symmetric_quant
   =
-  let out__ = CArray.make t 6 in
+  let out__ = CArray.make raw_tensor 6 in
   stubs__fused_moving_avg_obs_fq_helper_functional
     (CArray.start out__)
     self
@@ -2964,18 +2269,12 @@ let _fused_moving_avg_obs_fq_helper_functional
     (Int64.of_int ch_axis)
     (if per_row_fake_quant then 1 else 0)
     (if symmetric_quant then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
-  let t3 = CArray.get out__ 3 in
-  Gc.finalise C.Tensor.free t3;
-  let t4 = CArray.get out__ 4 in
-  Gc.finalise C.Tensor.free t4;
-  let t5 = CArray.get out__ 5 in
-  Gc.finalise C.Tensor.free t5;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
+  let t3 = CArray.get out__ 3 |> with_tensor_gc in
+  let t4 = CArray.get out__ 4 |> with_tensor_gc in
+  let t5 = CArray.get out__ 5 |> with_tensor_gc in
   t0, t1, t2, t3, t4, t5
 ;;
 
@@ -2996,7 +2295,7 @@ let _fused_moving_avg_obs_fq_helper_out
   ~per_row_fake_quant
   ~symmetric_quant
   =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs__fused_moving_avg_obs_fq_helper_out
     (CArray.start out__)
     out0
@@ -3014,10 +2313,8 @@ let _fused_moving_avg_obs_fq_helper_out
     (Int64.of_int ch_axis)
     (if per_row_fake_quant then 1 else 0)
     (if symmetric_quant then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
@@ -3028,7 +2325,7 @@ let _fused_sdp_choice ~query ~key ~value ~attn_mask ~dropout_p ~is_causal ~scale
     value
     (match attn_mask with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     dropout_p
     (if is_causal then 1 else 0)
     (Option.value scale ~default:0.0)
@@ -3037,36 +2334,18 @@ let _fused_sdp_choice ~query ~key ~value ~attn_mask ~dropout_p ~is_causal ~scale
      | None -> 1)
 ;;
 
-let _fw_primal self ~level =
-  let out__ = CArray.make t 1 in
-  stubs__fw_primal (CArray.start out__) self (Int64.of_int level);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let _fw_primal self ~level = stubs__fw_primal self (Int64.of_int level) |> with_tensor_gc
 
 let _fw_primal_copy self ~level =
-  let out__ = CArray.make t 1 in
-  stubs__fw_primal_copy (CArray.start out__) self (Int64.of_int level);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__fw_primal_copy self (Int64.of_int level) |> with_tensor_gc
 ;;
 
 let _fw_primal_copy_out ~out self ~level =
-  let out__ = CArray.make t 1 in
-  stubs__fw_primal_copy_out (CArray.start out__) out self (Int64.of_int level);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__fw_primal_copy_out out self (Int64.of_int level) |> with_tensor_gc
 ;;
 
 let _gather_sparse_backward self ~dim ~index ~grad =
-  let out__ = CArray.make t 1 in
-  stubs__gather_sparse_backward (CArray.start out__) self (Int64.of_int dim) index grad;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__gather_sparse_backward self (Int64.of_int dim) index grad |> with_tensor_gc
 ;;
 
 let _grid_sampler_2d_cpu_fallback
@@ -3076,17 +2355,13 @@ let _grid_sampler_2d_cpu_fallback
   ~padding_mode
   ~align_corners
   =
-  let out__ = CArray.make t 1 in
   stubs__grid_sampler_2d_cpu_fallback
-    (CArray.start out__)
     input
     grid
     (Int64.of_int interpolation_mode)
     (Int64.of_int padding_mode)
-    (if align_corners then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if align_corners then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _grid_sampler_2d_cpu_fallback_backward
@@ -3097,7 +2372,7 @@ let _grid_sampler_2d_cpu_fallback_backward
   ~padding_mode
   ~align_corners
   =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs__grid_sampler_2d_cpu_fallback_backward
     (CArray.start out__)
     grad_output
@@ -3106,10 +2381,8 @@ let _grid_sampler_2d_cpu_fallback_backward
     (Int64.of_int interpolation_mode)
     (Int64.of_int padding_mode)
     (if align_corners then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
@@ -3121,18 +2394,14 @@ let _grid_sampler_2d_cpu_fallback_out
   ~padding_mode
   ~align_corners
   =
-  let out__ = CArray.make t 1 in
   stubs__grid_sampler_2d_cpu_fallback_out
-    (CArray.start out__)
     out
     input
     grid
     (Int64.of_int interpolation_mode)
     (Int64.of_int padding_mode)
-    (if align_corners then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if align_corners then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _has_compatible_shallow_copy_type self ~from =
@@ -3150,14 +2419,14 @@ let _histogramdd_bin_edges self ~bins ~range ~weight ~density =
     (List.length range)
     (match weight with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (if density then 1 else 0)
   |> to_tensor_list
 ;;
 
 let _histogramdd_bin_edges_out ~out self ~bins ~range ~weight ~density =
   stubs__histogramdd_bin_edges_out
-    (CArray.of_list t out |> CArray.start)
+    (CArray.of_list gc_tensor out |> CArray.start)
     (List.length out)
     self
     (List.map Int64.of_int bins |> CArray.of_list int64_t |> CArray.start)
@@ -3166,14 +2435,12 @@ let _histogramdd_bin_edges_out ~out self ~bins ~range ~weight ~density =
     (List.length range)
     (match weight with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (if density then 1 else 0)
 ;;
 
 let _histogramdd_from_bin_cts self ~bins ~range ~weight ~density =
-  let out__ = CArray.make t 1 in
   stubs__histogramdd_from_bin_cts
-    (CArray.start out__)
     self
     (List.map Int64.of_int bins |> CArray.of_list int64_t |> CArray.start)
     (List.length bins)
@@ -3181,17 +2448,13 @@ let _histogramdd_from_bin_cts self ~bins ~range ~weight ~density =
     (List.length range)
     (match weight with
      | Some v -> v
-     | None -> null)
-    (if density then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+    (if density then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _histogramdd_from_bin_cts_out ~out self ~bins ~range ~weight ~density =
-  let out__ = CArray.make t 1 in
   stubs__histogramdd_from_bin_cts_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int bins |> CArray.of_list int64_t |> CArray.start)
@@ -3200,166 +2463,95 @@ let _histogramdd_from_bin_cts_out ~out self ~bins ~range ~weight ~density =
     (List.length range)
     (match weight with
      | Some v -> v
-     | None -> null)
-    (if density then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+    (if density then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _histogramdd_from_bin_tensors self ~bins ~weight ~density =
-  let out__ = CArray.make t 1 in
   stubs__histogramdd_from_bin_tensors
-    (CArray.start out__)
     self
-    (CArray.of_list t bins |> CArray.start)
+    (CArray.of_list gc_tensor bins |> CArray.start)
     (List.length bins)
     (match weight with
      | Some v -> v
-     | None -> null)
-    (if density then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+    (if density then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _histogramdd_from_bin_tensors_out ~out self ~bins ~weight ~density =
-  let out__ = CArray.make t 1 in
   stubs__histogramdd_from_bin_tensors_out
-    (CArray.start out__)
     out
     self
-    (CArray.of_list t bins |> CArray.start)
+    (CArray.of_list gc_tensor bins |> CArray.start)
     (List.length bins)
     (match weight with
      | Some v -> v
-     | None -> null)
-    (if density then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+    (if density then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _index_put_impl self ~indices ~values ~accumulate ~unsafe =
-  let out__ = CArray.make t 1 in
   stubs__index_put_impl
-    (CArray.start out__)
     self
     (List.map
        (function
         | Some x -> x
-        | None -> null)
+        | None -> none_gc_tensor)
        indices
-     |> CArray.of_list t
+     |> CArray.of_list gc_tensor
      |> CArray.start)
     (List.length indices)
     values
     (if accumulate then 1 else 0)
-    (if unsafe then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if unsafe then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _index_put_impl_ self ~indices ~values ~accumulate ~unsafe =
-  let out__ = CArray.make t 1 in
   stubs__index_put_impl_
-    (CArray.start out__)
     self
     (List.map
        (function
         | Some x -> x
-        | None -> null)
+        | None -> none_gc_tensor)
        indices
-     |> CArray.of_list t
+     |> CArray.of_list gc_tensor
      |> CArray.start)
     (List.length indices)
     values
     (if accumulate then 1 else 0)
-    (if unsafe then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if unsafe then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _index_put_impl_out ~out self ~indices ~values ~accumulate ~unsafe =
-  let out__ = CArray.make t 1 in
   stubs__index_put_impl_out
-    (CArray.start out__)
     out
     self
     (List.map
        (function
         | Some x -> x
-        | None -> null)
+        | None -> none_gc_tensor)
        indices
-     |> CArray.of_list t
+     |> CArray.of_list gc_tensor
      |> CArray.start)
     (List.length indices)
     values
     (if accumulate then 1 else 0)
-    (if unsafe then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let _indices self =
-  let out__ = CArray.make t 1 in
-  stubs__indices (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let _indices_copy self =
-  let out__ = CArray.make t 1 in
-  stubs__indices_copy (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let _indices_copy_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs__indices_copy_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let _int_mm self ~mat2 =
-  let out__ = CArray.make t 1 in
-  stubs__int_mm (CArray.start out__) self mat2;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let _int_mm_out ~out self ~mat2 =
-  let out__ = CArray.make t 1 in
-  stubs__int_mm_out (CArray.start out__) out self mat2;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let _is_all_true self =
-  let out__ = CArray.make t 1 in
-  stubs__is_all_true (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let _is_any_true self =
-  let out__ = CArray.make t 1 in
-  stubs__is_any_true (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if unsafe then 1 else 0)
+  |> with_tensor_gc
 ;;
 
+let _indices self = stubs__indices self |> with_tensor_gc
+let _indices_copy self = stubs__indices_copy self |> with_tensor_gc
+let _indices_copy_out ~out self = stubs__indices_copy_out out self |> with_tensor_gc
+let _int_mm self ~mat2 = stubs__int_mm self mat2 |> with_tensor_gc
+let _int_mm_out ~out self ~mat2 = stubs__int_mm_out out self mat2 |> with_tensor_gc
+let _is_all_true self = stubs__is_all_true self |> with_tensor_gc
+let _is_any_true self = stubs__is_any_true self |> with_tensor_gc
 let _is_zerotensor self = stubs__is_zerotensor self
 
 let _linalg_check_errors ~info ~api_name ~is_matrix =
@@ -3367,41 +2559,33 @@ let _linalg_check_errors ~info ~api_name ~is_matrix =
 ;;
 
 let _linalg_det ~a =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs__linalg_det (CArray.start out__) a;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
 let _linalg_det_result result ~lu ~pivots ~a =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs__linalg_det_result (CArray.start out__) result lu pivots a;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
 let _linalg_eigh ~a ~uplo ~compute_v =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs__linalg_eigh (CArray.start out__) a uplo (if compute_v then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let _linalg_eigh_eigenvalues ~eigenvalues ~eigenvectors ~a ~uplo ~compute_v =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs__linalg_eigh_eigenvalues
     (CArray.start out__)
     eigenvalues
@@ -3409,62 +2593,48 @@ let _linalg_eigh_eigenvalues ~eigenvalues ~eigenvectors ~a ~uplo ~compute_v =
     a
     uplo
     (if compute_v then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let _linalg_slogdet ~a =
-  let out__ = CArray.make t 4 in
+  let out__ = CArray.make raw_tensor 4 in
   stubs__linalg_slogdet (CArray.start out__) a;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
-  let t3 = CArray.get out__ 3 in
-  Gc.finalise C.Tensor.free t3;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
+  let t3 = CArray.get out__ 3 |> with_tensor_gc in
   t0, t1, t2, t3
 ;;
 
 let _linalg_slogdet_sign ~sign ~logabsdet ~lu ~pivots ~a =
-  let out__ = CArray.make t 4 in
+  let out__ = CArray.make raw_tensor 4 in
   stubs__linalg_slogdet_sign (CArray.start out__) sign logabsdet lu pivots a;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
-  let t3 = CArray.get out__ 3 in
-  Gc.finalise C.Tensor.free t3;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
+  let t3 = CArray.get out__ 3 |> with_tensor_gc in
   t0, t1, t2, t3
 ;;
 
 let _linalg_solve_ex ~a ~b ~left ~check_errors =
-  let out__ = CArray.make t 4 in
+  let out__ = CArray.make raw_tensor 4 in
   stubs__linalg_solve_ex
     (CArray.start out__)
     a
     b
     (if left then 1 else 0)
     (if check_errors then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
-  let t3 = CArray.get out__ 3 in
-  Gc.finalise C.Tensor.free t3;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
+  let t3 = CArray.get out__ 3 |> with_tensor_gc in
   t0, t1, t2, t3
 ;;
 
 let _linalg_solve_ex_result result ~lu ~pivots ~info ~a ~b ~left ~check_errors =
-  let out__ = CArray.make t 4 in
+  let out__ = CArray.make raw_tensor 4 in
   stubs__linalg_solve_ex_result
     (CArray.start out__)
     result
@@ -3475,36 +2645,29 @@ let _linalg_solve_ex_result result ~lu ~pivots ~info ~a ~b ~left ~check_errors =
     b
     (if left then 1 else 0)
     (if check_errors then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
-  let t3 = CArray.get out__ 3 in
-  Gc.finalise C.Tensor.free t3;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
+  let t3 = CArray.get out__ 3 |> with_tensor_gc in
   t0, t1, t2, t3
 ;;
 
 let _linalg_svd ~a ~full_matrices ~compute_uv ~driver =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs__linalg_svd
     (CArray.start out__)
     a
     (if full_matrices then 1 else 0)
     (if compute_uv then 1 else 0)
     driver;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
 let _linalg_svd_u ~u ~s ~vh ~a ~full_matrices ~compute_uv ~driver =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs__linalg_svd_u
     (CArray.start out__)
     u
@@ -3514,81 +2677,47 @@ let _linalg_svd_u ~u ~s ~vh ~a ~full_matrices ~compute_uv ~driver =
     (if full_matrices then 1 else 0)
     (if compute_uv then 1 else 0)
     driver;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
 let _log_softmax self ~dim ~half_to_float =
-  let out__ = CArray.make t 1 in
-  stubs__log_softmax
-    (CArray.start out__)
-    self
-    (Int64.of_int dim)
-    (if half_to_float then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__log_softmax self (Int64.of_int dim) (if half_to_float then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _log_softmax_backward_data ~grad_output ~output ~dim ~input_dtype =
-  let out__ = CArray.make t 1 in
   stubs__log_softmax_backward_data
-    (CArray.start out__)
     grad_output
     output
     (Int64.of_int dim)
-    (Kind.packed_to_int input_dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Kind.packed_to_int input_dtype)
+  |> with_tensor_gc
 ;;
 
 let _log_softmax_backward_data_out ~out ~grad_output ~output ~dim ~input_dtype =
-  let out__ = CArray.make t 1 in
   stubs__log_softmax_backward_data_out
-    (CArray.start out__)
     out
     grad_output
     output
     (Int64.of_int dim)
-    (Kind.packed_to_int input_dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Kind.packed_to_int input_dtype)
+  |> with_tensor_gc
 ;;
 
 let _log_softmax_out ~out self ~dim ~half_to_float =
-  let out__ = CArray.make t 1 in
-  stubs__log_softmax_out
-    (CArray.start out__)
-    out
-    self
-    (Int64.of_int dim)
-    (if half_to_float then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__log_softmax_out out self (Int64.of_int dim) (if half_to_float then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _logcumsumexp self ~dim =
-  let out__ = CArray.make t 1 in
-  stubs__logcumsumexp (CArray.start out__) self (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__logcumsumexp self (Int64.of_int dim) |> with_tensor_gc
 ;;
 
 let _logcumsumexp_out ~out self ~dim =
-  let out__ = CArray.make t 1 in
-  stubs__logcumsumexp_out (CArray.start out__) out self (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__logcumsumexp_out out self (Int64.of_int dim) |> with_tensor_gc
 ;;
 
 let _lstm_mps
@@ -3602,13 +2731,13 @@ let _lstm_mps
   ~bidirectional
   ~batch_first
   =
-  let out__ = CArray.make t 6 in
+  let out__ = CArray.make raw_tensor 6 in
   stubs__lstm_mps
     (CArray.start out__)
     input
-    (CArray.of_list t hx |> CArray.start)
+    (CArray.of_list gc_tensor hx |> CArray.start)
     (List.length hx)
-    (CArray.of_list t params |> CArray.start)
+    (CArray.of_list gc_tensor params |> CArray.start)
     (List.length params)
     (if has_biases then 1 else 0)
     (Int64.of_int num_layers)
@@ -3616,18 +2745,12 @@ let _lstm_mps
     (if train then 1 else 0)
     (if bidirectional then 1 else 0)
     (if batch_first then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
-  let t3 = CArray.get out__ 3 in
-  Gc.finalise C.Tensor.free t3;
-  let t4 = CArray.get out__ 4 in
-  Gc.finalise C.Tensor.free t4;
-  let t5 = CArray.get out__ 5 in
-  Gc.finalise C.Tensor.free t5;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
+  let t3 = CArray.get out__ 3 |> with_tensor_gc in
+  let t4 = CArray.get out__ 4 |> with_tensor_gc in
+  let t5 = CArray.get out__ 5 |> with_tensor_gc in
   t0, t1, t2, t3, t4, t5
 ;;
 
@@ -3648,7 +2771,7 @@ let _lstm_mps_out
   ~bidirectional
   ~batch_first
   =
-  let out__ = CArray.make t 6 in
+  let out__ = CArray.make raw_tensor 6 in
   stubs__lstm_mps_out
     (CArray.start out__)
     out0
@@ -3658,9 +2781,9 @@ let _lstm_mps_out
     out4
     out5
     input
-    (CArray.of_list t hx |> CArray.start)
+    (CArray.of_list gc_tensor hx |> CArray.start)
     (List.length hx)
-    (CArray.of_list t params |> CArray.start)
+    (CArray.of_list gc_tensor params |> CArray.start)
     (List.length params)
     (if has_biases then 1 else 0)
     (Int64.of_int num_layers)
@@ -3668,144 +2791,80 @@ let _lstm_mps_out
     (if train then 1 else 0)
     (if bidirectional then 1 else 0)
     (if batch_first then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
-  let t3 = CArray.get out__ 3 in
-  Gc.finalise C.Tensor.free t3;
-  let t4 = CArray.get out__ 4 in
-  Gc.finalise C.Tensor.free t4;
-  let t5 = CArray.get out__ 5 in
-  Gc.finalise C.Tensor.free t5;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
+  let t3 = CArray.get out__ 3 |> with_tensor_gc in
+  let t4 = CArray.get out__ 4 |> with_tensor_gc in
+  let t5 = CArray.get out__ 5 |> with_tensor_gc in
   t0, t1, t2, t3, t4, t5
 ;;
 
 let _lu_with_info self ~pivot ~check_errors =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs__lu_with_info
     (CArray.start out__)
     self
     (if pivot then 1 else 0)
     (if check_errors then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
 let _make_dep_token ~options =
-  let out__ = CArray.make t 1 in
-  stubs__make_dep_token
-    (CArray.start out__)
-    (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__make_dep_token (Kind.packed_to_int (fst options)) (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let _make_dual ~primal ~tangent ~level =
-  let out__ = CArray.make t 1 in
-  stubs__make_dual (CArray.start out__) primal tangent (Int64.of_int level);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__make_dual primal tangent (Int64.of_int level) |> with_tensor_gc
 ;;
 
 let _make_dual_copy ~primal ~tangent ~level =
-  let out__ = CArray.make t 1 in
-  stubs__make_dual_copy (CArray.start out__) primal tangent (Int64.of_int level);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__make_dual_copy primal tangent (Int64.of_int level) |> with_tensor_gc
 ;;
 
 let _make_dual_copy_out ~out ~primal ~tangent ~level =
-  let out__ = CArray.make t 1 in
-  stubs__make_dual_copy_out (CArray.start out__) out primal tangent (Int64.of_int level);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__make_dual_copy_out out primal tangent (Int64.of_int level) |> with_tensor_gc
 ;;
 
 let _make_per_channel_quantized_tensor self ~scale ~zero_point ~axis =
-  let out__ = CArray.make t 1 in
-  stubs__make_per_channel_quantized_tensor
-    (CArray.start out__)
-    self
-    scale
-    zero_point
-    (Int64.of_int axis);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__make_per_channel_quantized_tensor self scale zero_point (Int64.of_int axis)
+  |> with_tensor_gc
 ;;
 
 let _make_per_channel_quantized_tensor_out ~out self ~scale ~zero_point ~axis =
-  let out__ = CArray.make t 1 in
   stubs__make_per_channel_quantized_tensor_out
-    (CArray.start out__)
     out
     self
     scale
     zero_point
-    (Int64.of_int axis);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int axis)
+  |> with_tensor_gc
 ;;
 
 let _make_per_tensor_quantized_tensor self ~scale ~zero_point =
-  let out__ = CArray.make t 1 in
-  stubs__make_per_tensor_quantized_tensor
-    (CArray.start out__)
-    self
-    scale
-    (Int64.of_int zero_point);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__make_per_tensor_quantized_tensor self scale (Int64.of_int zero_point)
+  |> with_tensor_gc
 ;;
 
 let _make_per_tensor_quantized_tensor_out ~out self ~scale ~zero_point =
-  let out__ = CArray.make t 1 in
-  stubs__make_per_tensor_quantized_tensor_out
-    (CArray.start out__)
-    out
-    self
-    scale
-    (Int64.of_int zero_point);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__make_per_tensor_quantized_tensor_out out self scale (Int64.of_int zero_point)
+  |> with_tensor_gc
 ;;
 
 let _masked_scale self ~mask ~scale =
-  let out__ = CArray.make t 1 in
-  stubs__masked_scale (CArray.start out__) self mask scale;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__masked_scale self mask scale |> with_tensor_gc
 ;;
 
 let _masked_scale_out ~out self ~mask ~scale =
-  let out__ = CArray.make t 1 in
-  stubs__masked_scale_out (CArray.start out__) out self mask scale;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__masked_scale_out out self mask scale |> with_tensor_gc
 ;;
 
 let _masked_softmax self ~mask ~dim ~mask_type =
-  let out__ = CArray.make t 1 in
   stubs__masked_softmax
-    (CArray.start out__)
     self
     mask
     (match dim with
@@ -3819,16 +2878,12 @@ let _masked_softmax self ~mask ~dim ~mask_type =
      | Some v -> Int64.of_int v)
     (match mask_type with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let _masked_softmax_backward ~grad_output ~output ~mask ~dim =
-  let out__ = CArray.make t 1 in
   stubs__masked_softmax_backward
-    (CArray.start out__)
     grad_output
     output
     mask
@@ -3837,16 +2892,12 @@ let _masked_softmax_backward ~grad_output ~output ~mask ~dim =
      | Some v -> Int64.of_int v)
     (match dim with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let _masked_softmax_backward_out ~out ~grad_output ~output ~mask ~dim =
-  let out__ = CArray.make t 1 in
   stubs__masked_softmax_backward_out
-    (CArray.start out__)
     out
     grad_output
     output
@@ -3856,16 +2907,12 @@ let _masked_softmax_backward_out ~out ~grad_output ~output ~mask ~dim =
      | Some v -> Int64.of_int v)
     (match dim with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let _masked_softmax_out ~out self ~mask ~dim ~mask_type =
-  let out__ = CArray.make t 1 in
   stubs__masked_softmax_out
-    (CArray.start out__)
     out
     self
     mask
@@ -3880,115 +2927,73 @@ let _masked_softmax_out ~out self ~mask ~dim ~mask_type =
      | Some v -> Int64.of_int v)
     (match mask_type with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let _mkldnn_reshape self ~shape =
-  let out__ = CArray.make t 1 in
   stubs__mkldnn_reshape
-    (CArray.start out__)
     self
     (List.map Int64.of_int shape |> CArray.of_list int64_t |> CArray.start)
-    (List.length shape);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length shape)
+  |> with_tensor_gc
 ;;
 
 let _mkldnn_reshape_out ~out self ~shape =
-  let out__ = CArray.make t 1 in
   stubs__mkldnn_reshape_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int shape |> CArray.of_list int64_t |> CArray.start)
-    (List.length shape);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length shape)
+  |> with_tensor_gc
 ;;
 
 let _mkldnn_transpose self ~dim0 ~dim1 =
-  let out__ = CArray.make t 1 in
-  stubs__mkldnn_transpose
-    (CArray.start out__)
-    self
-    (Int64.of_int dim0)
-    (Int64.of_int dim1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__mkldnn_transpose self (Int64.of_int dim0) (Int64.of_int dim1) |> with_tensor_gc
 ;;
 
 let _mkldnn_transpose_ self ~dim0 ~dim1 =
-  let out__ = CArray.make t 1 in
-  stubs__mkldnn_transpose_
-    (CArray.start out__)
-    self
-    (Int64.of_int dim0)
-    (Int64.of_int dim1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__mkldnn_transpose_ self (Int64.of_int dim0) (Int64.of_int dim1) |> with_tensor_gc
 ;;
 
 let _mkldnn_transpose_out ~out self ~dim0 ~dim1 =
-  let out__ = CArray.make t 1 in
-  stubs__mkldnn_transpose_out
-    (CArray.start out__)
-    out
-    self
-    (Int64.of_int dim0)
-    (Int64.of_int dim1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__mkldnn_transpose_out out self (Int64.of_int dim0) (Int64.of_int dim1)
+  |> with_tensor_gc
 ;;
 
 let _mps_convolution self ~weight ~bias ~padding ~stride ~dilation ~groups =
-  let out__ = CArray.make t 1 in
   stubs__mps_convolution
-    (CArray.start out__)
     self
     weight
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
     (List.length padding)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
     (List.length stride)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
-    (Int64.of_int groups);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int groups)
+  |> with_tensor_gc
 ;;
 
 let _mps_convolution_out ~out self ~weight ~bias ~padding ~stride ~dilation ~groups =
-  let out__ = CArray.make t 1 in
   stubs__mps_convolution_out
-    (CArray.start out__)
     out
     self
     weight
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
     (List.length padding)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
     (List.length stride)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
-    (Int64.of_int groups);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int groups)
+  |> with_tensor_gc
 ;;
 
 let _mps_convolution_transpose
@@ -4000,9 +3005,7 @@ let _mps_convolution_transpose
   ~dilation
   ~groups
   =
-  let out__ = CArray.make t 1 in
   stubs__mps_convolution_transpose
-    (CArray.start out__)
     self
     weight
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
@@ -4013,10 +3016,8 @@ let _mps_convolution_transpose
     (List.length stride)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
-    (Int64.of_int groups);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int groups)
+  |> with_tensor_gc
 ;;
 
 let _mps_convolution_transpose_out
@@ -4029,9 +3030,7 @@ let _mps_convolution_transpose_out
   ~dilation
   ~groups
   =
-  let out__ = CArray.make t 1 in
   stubs__mps_convolution_transpose_out
-    (CArray.start out__)
     out
     self
     weight
@@ -4043,10 +3042,8 @@ let _mps_convolution_transpose_out
     (List.length stride)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
-    (Int64.of_int groups);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int groups)
+  |> with_tensor_gc
 ;;
 
 let _native_batch_norm_legit
@@ -4059,27 +3056,24 @@ let _native_batch_norm_legit
   ~momentum
   ~eps
   =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs__native_batch_norm_legit
     (CArray.start out__)
     input
     (match weight with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     running_mean
     running_var
     (if training then 1 else 0)
     momentum
     eps;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
@@ -4093,54 +3087,46 @@ let _native_batch_norm_legit_functional
   ~momentum
   ~eps
   =
-  let out__ = CArray.make t 5 in
+  let out__ = CArray.make raw_tensor 5 in
   stubs__native_batch_norm_legit_functional
     (CArray.start out__)
     input
     (match weight with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     running_mean
     running_var
     (if training then 1 else 0)
     momentum
     eps;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
-  let t3 = CArray.get out__ 3 in
-  Gc.finalise C.Tensor.free t3;
-  let t4 = CArray.get out__ 4 in
-  Gc.finalise C.Tensor.free t4;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
+  let t3 = CArray.get out__ 3 |> with_tensor_gc in
+  let t4 = CArray.get out__ 4 |> with_tensor_gc in
   t0, t1, t2, t3, t4
 ;;
 
 let _native_batch_norm_legit_no_stats input ~weight ~bias ~training ~momentum ~eps =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs__native_batch_norm_legit_no_stats
     (CArray.start out__)
     input
     (match weight with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (if training then 1 else 0)
     momentum
     eps;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
@@ -4155,7 +3141,7 @@ let _native_batch_norm_legit_no_stats_out
   ~momentum
   ~eps
   =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs__native_batch_norm_legit_no_stats_out
     (CArray.start out__)
     out
@@ -4164,19 +3150,16 @@ let _native_batch_norm_legit_no_stats_out
     input
     (match weight with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (if training then 1 else 0)
     momentum
     eps;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
@@ -4189,26 +3172,23 @@ let _native_batch_norm_legit_no_training
   ~momentum
   ~eps
   =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs__native_batch_norm_legit_no_training
     (CArray.start out__)
     input
     (match weight with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     running_mean
     running_var
     momentum
     eps;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
@@ -4224,7 +3204,7 @@ let _native_batch_norm_legit_no_training_out
   ~momentum
   ~eps
   =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs__native_batch_norm_legit_no_training_out
     (CArray.start out__)
     out0
@@ -4233,20 +3213,17 @@ let _native_batch_norm_legit_no_training_out
     input
     (match weight with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     running_mean
     running_var
     momentum
     eps;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
@@ -4263,7 +3240,7 @@ let _native_batch_norm_legit_out
   ~momentum
   ~eps
   =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs__native_batch_norm_legit_out
     (CArray.start out__)
     out
@@ -4272,21 +3249,18 @@ let _native_batch_norm_legit_out
     input
     (match weight with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     running_mean
     running_var
     (if training then 1 else 0)
     momentum
     eps;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
@@ -4305,7 +3279,7 @@ let _native_multi_head_attention
   ~average_attn_weights
   ~mask_type
   =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs__native_multi_head_attention
     (CArray.start out__)
     query
@@ -4319,7 +3293,7 @@ let _native_multi_head_attention
     proj_bias
     (match mask with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (if need_weights then 1 else 0)
     (if average_attn_weights then 1 else 0)
     (match mask_type with
@@ -4328,10 +3302,8 @@ let _native_multi_head_attention
     (match mask_type with
      | Some _ -> 0
      | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
@@ -4352,7 +3324,7 @@ let _native_multi_head_attention_out
   ~average_attn_weights
   ~mask_type
   =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs__native_multi_head_attention_out
     (CArray.start out__)
     out0
@@ -4368,7 +3340,7 @@ let _native_multi_head_attention_out
     proj_bias
     (match mask with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (if need_weights then 1 else 0)
     (if average_attn_weights then 1 else 0)
     (match mask_type with
@@ -4377,99 +3349,47 @@ let _native_multi_head_attention_out
     (match mask_type with
      | Some _ -> 0
      | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
-let _neg_view self =
-  let out__ = CArray.make t 1 in
-  stubs__neg_view (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let _neg_view_copy self =
-  let out__ = CArray.make t 1 in
-  stubs__neg_view_copy (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let _neg_view_copy_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs__neg_view_copy_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let _neg_view self = stubs__neg_view self |> with_tensor_gc
+let _neg_view_copy self = stubs__neg_view_copy self |> with_tensor_gc
+let _neg_view_copy_out ~out self = stubs__neg_view_copy_out out self |> with_tensor_gc
 
 let _nested_from_padded ~padded ~cpu_nested_shape_example ~fuse_transform_0213 =
-  let out__ = CArray.make t 1 in
   stubs__nested_from_padded
-    (CArray.start out__)
     padded
     cpu_nested_shape_example
-    (if fuse_transform_0213 then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if fuse_transform_0213 then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _nested_from_padded_and_nested_example ~padded ~nt_example =
-  let out__ = CArray.make t 1 in
-  stubs__nested_from_padded_and_nested_example (CArray.start out__) padded nt_example;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__nested_from_padded_and_nested_example padded nt_example |> with_tensor_gc
 ;;
 
 let _nested_from_padded_and_nested_example_out ~out ~padded ~nt_example =
-  let out__ = CArray.make t 1 in
-  stubs__nested_from_padded_and_nested_example_out
-    (CArray.start out__)
-    out
-    padded
-    nt_example;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__nested_from_padded_and_nested_example_out out padded nt_example |> with_tensor_gc
 ;;
 
 let _nested_from_padded_out ~out ~padded ~cpu_nested_shape_example ~fuse_transform_0213 =
-  let out__ = CArray.make t 1 in
   stubs__nested_from_padded_out
-    (CArray.start out__)
     out
     padded
     cpu_nested_shape_example
-    (if fuse_transform_0213 then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if fuse_transform_0213 then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _nested_select_backward ~grad_output self ~dim ~index =
-  let out__ = CArray.make t 1 in
-  stubs__nested_select_backward
-    (CArray.start out__)
-    grad_output
-    self
-    (Int64.of_int dim)
-    (Int64.of_int index);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__nested_select_backward grad_output self (Int64.of_int dim) (Int64.of_int index)
+  |> with_tensor_gc
 ;;
 
 let _nested_sum_backward ~grad self ~dim ~keepdim =
-  let out__ = CArray.make t 1 in
   stubs__nested_sum_backward
-    (CArray.start out__)
     grad
     self
     (match dim with
@@ -4478,148 +3398,95 @@ let _nested_sum_backward ~grad self ~dim ~keepdim =
     (match dim with
      | None -> -1
      | Some v -> List.length v)
-    (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if keepdim then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _nested_view_from_buffer self ~nested_size ~nested_strides ~offsets =
-  let out__ = CArray.make t 1 in
-  stubs__nested_view_from_buffer
-    (CArray.start out__)
-    self
-    nested_size
-    nested_strides
-    offsets;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__nested_view_from_buffer self nested_size nested_strides offsets |> with_tensor_gc
 ;;
 
 let _nested_view_from_buffer_copy self ~nested_size ~nested_strides ~offsets =
-  let out__ = CArray.make t 1 in
-  stubs__nested_view_from_buffer_copy
-    (CArray.start out__)
-    self
-    nested_size
-    nested_strides
-    offsets;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__nested_view_from_buffer_copy self nested_size nested_strides offsets
+  |> with_tensor_gc
 ;;
 
 let _nested_view_from_buffer_copy_out ~out self ~nested_size ~nested_strides ~offsets =
-  let out__ = CArray.make t 1 in
-  stubs__nested_view_from_buffer_copy_out
-    (CArray.start out__)
-    out
-    self
-    nested_size
-    nested_strides
-    offsets;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__nested_view_from_buffer_copy_out out self nested_size nested_strides offsets
+  |> with_tensor_gc
 ;;
 
 let _new_zeros_with_same_feature_meta self other ~self_num_batch_dims =
-  let out__ = CArray.make t 1 in
-  stubs__new_zeros_with_same_feature_meta
-    (CArray.start out__)
-    self
-    other
-    (Int64.of_int self_num_batch_dims);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__new_zeros_with_same_feature_meta self other (Int64.of_int self_num_batch_dims)
+  |> with_tensor_gc
 ;;
 
 let _new_zeros_with_same_feature_meta_out ~out self other ~self_num_batch_dims =
-  let out__ = CArray.make t 1 in
   stubs__new_zeros_with_same_feature_meta_out
-    (CArray.start out__)
     out
     self
     other
-    (Int64.of_int self_num_batch_dims);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int self_num_batch_dims)
+  |> with_tensor_gc
 ;;
 
 let _nnpack_available = stubs__nnpack_available
 
 let _nnpack_spatial_convolution input ~weight ~bias ~padding ~stride =
-  let out__ = CArray.make t 1 in
   stubs__nnpack_spatial_convolution
-    (CArray.start out__)
     input
     weight
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
     (List.length padding)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
-    (List.length stride);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length stride)
+  |> with_tensor_gc
 ;;
 
 let _nnpack_spatial_convolution_out ~out input ~weight ~bias ~padding ~stride =
-  let out__ = CArray.make t 1 in
   stubs__nnpack_spatial_convolution_out
-    (CArray.start out__)
     out
     input
     weight
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
     (List.length padding)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
-    (List.length stride);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length stride)
+  |> with_tensor_gc
 ;;
 
 let _nnz self = stubs__nnz self
 
 let _pack_padded_sequence input ~lengths ~batch_first =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs__pack_padded_sequence
     (CArray.start out__)
     input
     lengths
     (if batch_first then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let _pack_padded_sequence_backward ~grad ~input_size ~batch_sizes ~batch_first =
-  let out__ = CArray.make t 1 in
   stubs__pack_padded_sequence_backward
-    (CArray.start out__)
     grad
     (List.map Int64.of_int input_size |> CArray.of_list int64_t |> CArray.start)
     (List.length input_size)
     batch_sizes
-    (if batch_first then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if batch_first then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _pack_padded_sequence_out ~out0 ~out1 input ~lengths ~batch_first =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs__pack_padded_sequence_out
     (CArray.start out__)
     out0
@@ -4627,29 +3494,21 @@ let _pack_padded_sequence_out ~out0 ~out1 input ~lengths ~batch_first =
     input
     lengths
     (if batch_first then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let _pad_circular self ~pad =
-  let out__ = CArray.make t 1 in
   stubs__pad_circular
-    (CArray.start out__)
     self
     (List.map Int64.of_int pad |> CArray.of_list int64_t |> CArray.start)
-    (List.length pad);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length pad)
+  |> with_tensor_gc
 ;;
 
 let _pad_enum self ~pad ~mode ~value =
-  let out__ = CArray.make t 1 in
   stubs__pad_enum
-    (CArray.start out__)
     self
     (List.map Int64.of_int pad |> CArray.of_list int64_t |> CArray.start)
     (List.length pad)
@@ -4657,14 +3516,12 @@ let _pad_enum self ~pad ~mode ~value =
     (Option.value value ~default:0.0)
     (match value with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let _pad_packed_sequence ~data ~batch_sizes ~batch_first ~padding_value ~total_length =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs__pad_packed_sequence
     (CArray.start out__)
     data
@@ -4672,217 +3529,139 @@ let _pad_packed_sequence ~data ~batch_sizes ~batch_first ~padding_value ~total_l
     (if batch_first then 1 else 0)
     padding_value
     (Int64.of_int total_length);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let _pdist_backward ~grad self ~p ~pdist =
-  let out__ = CArray.make t 1 in
-  stubs__pdist_backward (CArray.start out__) grad self p pdist;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__pdist_backward grad self p pdist |> with_tensor_gc
 ;;
 
 let _pdist_backward_out ~out ~grad self ~p ~pdist =
-  let out__ = CArray.make t 1 in
-  stubs__pdist_backward_out (CArray.start out__) out grad self p pdist;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__pdist_backward_out out grad self p pdist |> with_tensor_gc
 ;;
 
 let _pin_memory self ~device =
-  let out__ = CArray.make t 1 in
-  stubs__pin_memory (CArray.start out__) self (Device.to_int device);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__pin_memory self (Device.to_int device) |> with_tensor_gc
 ;;
 
 let _pin_memory_out ~out self ~device =
-  let out__ = CArray.make t 1 in
-  stubs__pin_memory_out (CArray.start out__) out self (Device.to_int device);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__pin_memory_out out self (Device.to_int device) |> with_tensor_gc
 ;;
 
-let _prelu_kernel self ~weight =
-  let out__ = CArray.make t 1 in
-  stubs__prelu_kernel (CArray.start out__) self weight;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let _prelu_kernel self ~weight = stubs__prelu_kernel self weight |> with_tensor_gc
 
 let _prelu_kernel_backward ~grad_output self ~weight =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs__prelu_kernel_backward (CArray.start out__) grad_output self weight;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let _propagate_xla_data input ~output = stubs__propagate_xla_data input output
 
 let _remove_batch_dim self ~level ~batch_size ~out_dim =
-  let out__ = CArray.make t 1 in
   stubs__remove_batch_dim
-    (CArray.start out__)
     self
     (Int64.of_int level)
     (Int64.of_int batch_size)
-    (Int64.of_int out_dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int out_dim)
+  |> with_tensor_gc
 ;;
 
 let _reshape_alias self ~size ~stride =
-  let out__ = CArray.make t 1 in
   stubs__reshape_alias
-    (CArray.start out__)
     self
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
-    (List.length stride);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length stride)
+  |> with_tensor_gc
 ;;
 
 let _reshape_alias_copy self ~size ~stride =
-  let out__ = CArray.make t 1 in
   stubs__reshape_alias_copy
-    (CArray.start out__)
     self
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
-    (List.length stride);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length stride)
+  |> with_tensor_gc
 ;;
 
 let _reshape_alias_copy_out ~out self ~size ~stride =
-  let out__ = CArray.make t 1 in
   stubs__reshape_alias_copy_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
-    (List.length stride);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length stride)
+  |> with_tensor_gc
 ;;
 
 let _reshape_copy self ~size =
-  let out__ = CArray.make t 1 in
   stubs__reshape_copy
-    (CArray.start out__)
     self
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
-    (List.length size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length size)
+  |> with_tensor_gc
 ;;
 
 let _reshape_from_tensor self ~shape =
-  let out__ = CArray.make t 1 in
-  stubs__reshape_from_tensor (CArray.start out__) self shape;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__reshape_from_tensor self shape |> with_tensor_gc
 ;;
 
 let _resize_output self ~size ~device =
-  let out__ = CArray.make t 1 in
   stubs__resize_output
-    (CArray.start out__)
     self
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
-    (Device.to_int device);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int device)
+  |> with_tensor_gc
 ;;
 
 let _resize_output_ self ~size ~device =
-  let out__ = CArray.make t 1 in
   stubs__resize_output_
-    (CArray.start out__)
     self
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
-    (Device.to_int device);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int device)
+  |> with_tensor_gc
 ;;
 
 let _resize_output_out ~out self ~size ~device =
-  let out__ = CArray.make t 1 in
   stubs__resize_output_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
-    (Device.to_int device);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int device)
+  |> with_tensor_gc
 ;;
 
 let _rowwise_prune ~weight ~mask ~compressed_indices_dtype =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs__rowwise_prune
     (CArray.start out__)
     weight
     mask
     (Kind.packed_to_int compressed_indices_dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
-let _sample_dirichlet self =
-  let out__ = CArray.make t 1 in
-  stubs__sample_dirichlet (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let _sample_dirichlet self = stubs__sample_dirichlet self |> with_tensor_gc
 
 let _sample_dirichlet_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs__sample_dirichlet_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__sample_dirichlet_out out self |> with_tensor_gc
 ;;
 
 let _saturate_weight_to_fp16 ~weight =
-  let out__ = CArray.make t 1 in
-  stubs__saturate_weight_to_fp16 (CArray.start out__) weight;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__saturate_weight_to_fp16 weight |> with_tensor_gc
 ;;
 
 let _scaled_dot_product_attention_math
@@ -4895,7 +3674,7 @@ let _scaled_dot_product_attention_math
   ~dropout_mask
   ~scale
   =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs__scaled_dot_product_attention_math
     (CArray.start out__)
     query
@@ -4903,20 +3682,18 @@ let _scaled_dot_product_attention_math
     value
     (match attn_mask with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     dropout_p
     (if is_causal then 1 else 0)
     (match dropout_mask with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (Option.value scale ~default:0.0)
     (match scale with
      | Some _ -> 0
      | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
@@ -4930,7 +3707,7 @@ let _scaled_dot_product_efficient_attention
   ~is_causal
   ~scale
   =
-  let out__ = CArray.make t 4 in
+  let out__ = CArray.make raw_tensor 4 in
   stubs__scaled_dot_product_efficient_attention
     (CArray.start out__)
     query
@@ -4938,7 +3715,7 @@ let _scaled_dot_product_efficient_attention
     value
     (match attn_bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (if compute_log_sumexp then 1 else 0)
     dropout_p
     (if is_causal then 1 else 0)
@@ -4946,14 +3723,10 @@ let _scaled_dot_product_efficient_attention
     (match scale with
      | Some _ -> 0
      | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
-  let t3 = CArray.get out__ 3 in
-  Gc.finalise C.Tensor.free t3;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
+  let t3 = CArray.get out__ 3 |> with_tensor_gc in
   t0, t1, t2, t3
 ;;
 
@@ -4974,7 +3747,7 @@ let _scaled_dot_product_flash_attention_backward
   ~philox_offset
   ~scale
   =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs__scaled_dot_product_flash_attention_backward
     (CArray.start out__)
     grad_out
@@ -4995,38 +3768,33 @@ let _scaled_dot_product_flash_attention_backward
     (match scale with
      | Some _ -> 0
      | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
 let _scaled_mm self ~mat2 ~bias ~out_dtype ~scale_a ~scale_b ~scale_result =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs__scaled_mm
     (CArray.start out__)
     self
     mat2
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (Kind.packed_to_int out_dtype)
     (match scale_a with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match scale_b with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match scale_result with
      | Some v -> v
-     | None -> null);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+     | None -> none_gc_tensor);
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
@@ -5041,7 +3809,7 @@ let _scaled_mm_out
   ~scale_b
   ~scale_result
   =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs__scaled_mm_out
     (CArray.start out__)
     out
@@ -5050,89 +3818,71 @@ let _scaled_mm_out
     mat2
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (Kind.packed_to_int out_dtype)
     (match scale_a with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match scale_b with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match scale_result with
      | Some v -> v
-     | None -> null);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+     | None -> none_gc_tensor);
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let _scatter_reduce self ~dim ~index ~src ~reduce ~include_self =
-  let out__ = CArray.make t 1 in
   stubs__scatter_reduce
-    (CArray.start out__)
     self
     (Int64.of_int dim)
     index
     src
     reduce
-    (if include_self then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if include_self then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _scatter_reduce_ self ~dim ~index ~src ~reduce ~include_self =
-  let out__ = CArray.make t 1 in
   stubs__scatter_reduce_
-    (CArray.start out__)
     self
     (Int64.of_int dim)
     index
     src
     reduce
-    (if include_self then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if include_self then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _scatter_reduce_two_out ~out self ~dim ~index ~src ~reduce ~include_self =
-  let out__ = CArray.make t 1 in
   stubs__scatter_reduce_two_out
-    (CArray.start out__)
     out
     self
     (Int64.of_int dim)
     index
     src
     reduce
-    (if include_self then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if include_self then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _segment_reduce_backward ~grad ~output ~data ~reduce ~lengths ~offsets ~axis ~initial =
-  let out__ = CArray.make t 1 in
   stubs__segment_reduce_backward
-    (CArray.start out__)
     grad
     output
     data
     reduce
     (match lengths with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match offsets with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (Int64.of_int axis)
-    initial;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    initial
+  |> with_tensor_gc
 ;;
 
 let _segment_reduce_backward_out
@@ -5146,9 +3896,7 @@ let _segment_reduce_backward_out
   ~axis
   ~initial
   =
-  let out__ = CArray.make t 1 in
   stubs__segment_reduce_backward_out
-    (CArray.start out__)
     out
     grad
     output
@@ -5156,24 +3904,16 @@ let _segment_reduce_backward_out
     reduce
     (match lengths with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match offsets with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (Int64.of_int axis)
-    initial;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    initial
+  |> with_tensor_gc
 ;;
 
-let _shape_as_tensor self =
-  let out__ = CArray.make t 1 in
-  stubs__shape_as_tensor (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let _shape_as_tensor self = stubs__shape_as_tensor self |> with_tensor_gc
 
 let _slow_conv2d_backward
   ~grad_input
@@ -5186,7 +3926,7 @@ let _slow_conv2d_backward
   ~stride
   ~padding
   =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs__slow_conv2d_backward
     (CArray.start out__)
     grad_input
@@ -5201,17 +3941,14 @@ let _slow_conv2d_backward
     (List.length stride)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
     (List.length padding);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
 let _sobol_engine_draw ~quasi ~n ~sobolstate ~dimension ~num_generated ~dtype =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs__sobol_engine_draw
     (CArray.start out__)
     quasi
@@ -5220,178 +3957,111 @@ let _sobol_engine_draw ~quasi ~n ~sobolstate ~dimension ~num_generated ~dtype =
     (Int64.of_int dimension)
     (Int64.of_int num_generated)
     (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let _sobol_engine_ff_ self ~n ~sobolstate ~dimension ~num_generated =
-  let out__ = CArray.make t 1 in
   stubs__sobol_engine_ff_
-    (CArray.start out__)
     self
     (Int64.of_int n)
     sobolstate
     (Int64.of_int dimension)
-    (Int64.of_int num_generated);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int num_generated)
+  |> with_tensor_gc
 ;;
 
 let _sobol_engine_initialize_state_ self ~dimension =
-  let out__ = CArray.make t 1 in
-  stubs__sobol_engine_initialize_state_ (CArray.start out__) self (Int64.of_int dimension);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__sobol_engine_initialize_state_ self (Int64.of_int dimension) |> with_tensor_gc
 ;;
 
 let _sobol_engine_scramble_ self ~ltm ~dimension =
-  let out__ = CArray.make t 1 in
-  stubs__sobol_engine_scramble_ (CArray.start out__) self ltm (Int64.of_int dimension);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__sobol_engine_scramble_ self ltm (Int64.of_int dimension) |> with_tensor_gc
 ;;
 
 let _softmax self ~dim ~half_to_float =
-  let out__ = CArray.make t 1 in
-  stubs__softmax
-    (CArray.start out__)
-    self
-    (Int64.of_int dim)
-    (if half_to_float then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__softmax self (Int64.of_int dim) (if half_to_float then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _softmax_backward_data ~grad_output ~output ~dim ~input_dtype =
-  let out__ = CArray.make t 1 in
   stubs__softmax_backward_data
-    (CArray.start out__)
     grad_output
     output
     (Int64.of_int dim)
-    (Kind.packed_to_int input_dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Kind.packed_to_int input_dtype)
+  |> with_tensor_gc
 ;;
 
 let _softmax_backward_data_out ~grad_input ~grad_output ~output ~dim ~input_dtype =
-  let out__ = CArray.make t 1 in
   stubs__softmax_backward_data_out
-    (CArray.start out__)
     grad_input
     grad_output
     output
     (Int64.of_int dim)
-    (Kind.packed_to_int input_dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Kind.packed_to_int input_dtype)
+  |> with_tensor_gc
 ;;
 
 let _softmax_out ~out self ~dim ~half_to_float =
-  let out__ = CArray.make t 1 in
-  stubs__softmax_out
-    (CArray.start out__)
-    out
-    self
-    (Int64.of_int dim)
-    (if half_to_float then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__softmax_out out self (Int64.of_int dim) (if half_to_float then 1 else 0)
+  |> with_tensor_gc
 ;;
 
-let _sparse_addmm self ~mat1 ~mat2 =
-  let out__ = CArray.make t 1 in
-  stubs__sparse_addmm (CArray.start out__) self mat1 mat2;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let _sparse_addmm self ~mat1 ~mat2 = stubs__sparse_addmm self mat1 mat2 |> with_tensor_gc
 
 let _sparse_addmm_out ~out self ~mat1 ~mat2 =
-  let out__ = CArray.make t 1 in
-  stubs__sparse_addmm_out (CArray.start out__) out self mat1 mat2;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__sparse_addmm_out out self mat1 mat2 |> with_tensor_gc
 ;;
 
 let _sparse_broadcast_to self ~size =
-  let out__ = CArray.make t 1 in
   stubs__sparse_broadcast_to
-    (CArray.start out__)
     self
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
-    (List.length size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length size)
+  |> with_tensor_gc
 ;;
 
 let _sparse_broadcast_to_copy self ~size =
-  let out__ = CArray.make t 1 in
   stubs__sparse_broadcast_to_copy
-    (CArray.start out__)
     self
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
-    (List.length size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length size)
+  |> with_tensor_gc
 ;;
 
 let _sparse_broadcast_to_copy_out ~out self ~size =
-  let out__ = CArray.make t 1 in
   stubs__sparse_broadcast_to_copy_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
-    (List.length size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length size)
+  |> with_tensor_gc
 ;;
 
 let _sparse_bsc_tensor_unsafe ~ccol_indices ~row_indices ~values ~size ~options =
-  let out__ = CArray.make t 1 in
   stubs__sparse_bsc_tensor_unsafe
-    (CArray.start out__)
     ccol_indices
     row_indices
     values
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let _sparse_bsr_tensor_unsafe ~crow_indices ~col_indices ~values ~size ~options =
-  let out__ = CArray.make t 1 in
   stubs__sparse_bsr_tensor_unsafe
-    (CArray.start out__)
     crow_indices
     col_indices
     values
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let _sparse_compressed_tensor_unsafe
@@ -5401,50 +4071,38 @@ let _sparse_compressed_tensor_unsafe
   ~size
   ~options
   =
-  let out__ = CArray.make t 1 in
   stubs__sparse_compressed_tensor_unsafe
-    (CArray.start out__)
     compressed_indices
     plain_indices
     values
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let _sparse_coo_tensor_unsafe ~indices ~values ~size ~options ~is_coalesced =
-  let out__ = CArray.make t 1 in
   stubs__sparse_coo_tensor_unsafe
-    (CArray.start out__)
     indices
     values
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     (Kind.packed_to_int (fst options))
     (Device.to_int (snd options))
-    (if is_coalesced then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if is_coalesced then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _sparse_coo_tensor_with_dims ~sparse_dim ~dense_dim ~size ~options =
-  let out__ = CArray.make t 1 in
   stubs__sparse_coo_tensor_with_dims
-    (CArray.start out__)
     (Int64.of_int sparse_dim)
     (Int64.of_int dense_dim)
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let _sparse_coo_tensor_with_dims_and_tensors
@@ -5456,9 +4114,7 @@ let _sparse_coo_tensor_with_dims_and_tensors
   ~options
   ~is_coalesced
   =
-  let out__ = CArray.make t 1 in
   stubs__sparse_coo_tensor_with_dims_and_tensors
-    (CArray.start out__)
     (Int64.of_int sparse_dim)
     (Int64.of_int dense_dim)
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
@@ -5467,10 +4123,8 @@ let _sparse_coo_tensor_with_dims_and_tensors
     values
     (Kind.packed_to_int (fst options))
     (Device.to_int (snd options))
-    (if is_coalesced then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if is_coalesced then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _sparse_coo_tensor_with_dims_and_tensors_out
@@ -5482,9 +4136,7 @@ let _sparse_coo_tensor_with_dims_and_tensors_out
   ~values
   ~is_coalesced
   =
-  let out__ = CArray.make t 1 in
   stubs__sparse_coo_tensor_with_dims_and_tensors_out
-    (CArray.start out__)
     out
     (Int64.of_int sparse_dim)
     (Int64.of_int dense_dim)
@@ -5492,607 +4144,345 @@ let _sparse_coo_tensor_with_dims_and_tensors_out
     (List.length size)
     indices
     values
-    (if is_coalesced then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if is_coalesced then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _sparse_coo_tensor_with_dims_out ~out ~sparse_dim ~dense_dim ~size =
-  let out__ = CArray.make t 1 in
   stubs__sparse_coo_tensor_with_dims_out
-    (CArray.start out__)
     out
     (Int64.of_int sparse_dim)
     (Int64.of_int dense_dim)
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
-    (List.length size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length size)
+  |> with_tensor_gc
 ;;
 
 let _sparse_csc_tensor_unsafe ~ccol_indices ~row_indices ~values ~size ~options =
-  let out__ = CArray.make t 1 in
   stubs__sparse_csc_tensor_unsafe
-    (CArray.start out__)
     ccol_indices
     row_indices
     values
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let _sparse_csr_prod self ~dim ~keepdim ~dtype =
-  let out__ = CArray.make t 1 in
   stubs__sparse_csr_prod
-    (CArray.start out__)
     self
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
     (List.length dim)
     (if keepdim then 1 else 0)
-    (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Kind.packed_to_int dtype)
+  |> with_tensor_gc
 ;;
 
 let _sparse_csr_prod_dim_dtype_out ~out self ~dim ~keepdim ~dtype =
-  let out__ = CArray.make t 1 in
   stubs__sparse_csr_prod_dim_dtype_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
     (List.length dim)
     (if keepdim then 1 else 0)
-    (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Kind.packed_to_int dtype)
+  |> with_tensor_gc
 ;;
 
 let _sparse_csr_sum self ~dim ~keepdim ~dtype =
-  let out__ = CArray.make t 1 in
   stubs__sparse_csr_sum
-    (CArray.start out__)
     self
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
     (List.length dim)
     (if keepdim then 1 else 0)
-    (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Kind.packed_to_int dtype)
+  |> with_tensor_gc
 ;;
 
 let _sparse_csr_sum_dim_dtype_out ~out self ~dim ~keepdim ~dtype =
-  let out__ = CArray.make t 1 in
   stubs__sparse_csr_sum_dim_dtype_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
     (List.length dim)
     (if keepdim then 1 else 0)
-    (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Kind.packed_to_int dtype)
+  |> with_tensor_gc
 ;;
 
 let _sparse_csr_tensor_unsafe ~crow_indices ~col_indices ~values ~size ~options =
-  let out__ = CArray.make t 1 in
   stubs__sparse_csr_tensor_unsafe
-    (CArray.start out__)
     crow_indices
     col_indices
     values
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let _sparse_log_softmax self ~dim ~half_to_float =
-  let out__ = CArray.make t 1 in
-  stubs__sparse_log_softmax
-    (CArray.start out__)
-    self
-    (Int64.of_int dim)
-    (if half_to_float then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__sparse_log_softmax self (Int64.of_int dim) (if half_to_float then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _sparse_log_softmax_backward_data ~grad_output ~output ~dim self =
-  let out__ = CArray.make t 1 in
-  stubs__sparse_log_softmax_backward_data
-    (CArray.start out__)
-    grad_output
-    output
-    (Int64.of_int dim)
-    self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__sparse_log_softmax_backward_data grad_output output (Int64.of_int dim) self
+  |> with_tensor_gc
 ;;
 
 let _sparse_log_softmax_backward_data_out ~out ~grad_output ~output ~dim self =
-  let out__ = CArray.make t 1 in
   stubs__sparse_log_softmax_backward_data_out
-    (CArray.start out__)
     out
     grad_output
     output
     (Int64.of_int dim)
-    self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    self
+  |> with_tensor_gc
 ;;
 
 let _sparse_log_softmax_int self ~dim ~dtype =
-  let out__ = CArray.make t 1 in
-  stubs__sparse_log_softmax_int
-    (CArray.start out__)
-    self
-    (Int64.of_int dim)
-    (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__sparse_log_softmax_int self (Int64.of_int dim) (Kind.packed_to_int dtype)
+  |> with_tensor_gc
 ;;
 
 let _sparse_log_softmax_out ~out self ~dim ~half_to_float =
-  let out__ = CArray.make t 1 in
   stubs__sparse_log_softmax_out
-    (CArray.start out__)
     out
     self
     (Int64.of_int dim)
-    (if half_to_float then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if half_to_float then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _sparse_mask_projection self ~mask ~accumulate_matches =
-  let out__ = CArray.make t 1 in
-  stubs__sparse_mask_projection
-    (CArray.start out__)
-    self
-    mask
-    (if accumulate_matches then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__sparse_mask_projection self mask (if accumulate_matches then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _sparse_mask_projection_out ~out self ~mask ~accumulate_matches =
-  let out__ = CArray.make t 1 in
-  stubs__sparse_mask_projection_out
-    (CArray.start out__)
-    out
-    self
-    mask
-    (if accumulate_matches then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__sparse_mask_projection_out out self mask (if accumulate_matches then 1 else 0)
+  |> with_tensor_gc
 ;;
 
-let _sparse_mm ~sparse ~dense =
-  let out__ = CArray.make t 1 in
-  stubs__sparse_mm (CArray.start out__) sparse dense;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let _sparse_mm ~sparse ~dense = stubs__sparse_mm sparse dense |> with_tensor_gc
 
 let _sparse_mm_reduce ~sparse ~dense ~reduce =
-  let out__ = CArray.make t 1 in
-  stubs__sparse_mm_reduce (CArray.start out__) sparse dense reduce;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__sparse_mm_reduce sparse dense reduce |> with_tensor_gc
 ;;
 
 let _sparse_mm_reduce_impl self other ~reduce =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs__sparse_mm_reduce_impl (CArray.start out__) self other reduce;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let _sparse_semi_structured_linear input ~weight ~meta ~bias ~activation =
-  let out__ = CArray.make t 1 in
   stubs__sparse_semi_structured_linear
-    (CArray.start out__)
     input
     weight
     meta
     (match bias with
      | Some v -> v
-     | None -> null)
-    activation;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+    activation
+  |> with_tensor_gc
 ;;
 
 let _sparse_softmax self ~dim ~half_to_float =
-  let out__ = CArray.make t 1 in
-  stubs__sparse_softmax
-    (CArray.start out__)
-    self
-    (Int64.of_int dim)
-    (if half_to_float then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__sparse_softmax self (Int64.of_int dim) (if half_to_float then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _sparse_softmax_backward_data ~grad_output ~output ~dim self =
-  let out__ = CArray.make t 1 in
-  stubs__sparse_softmax_backward_data
-    (CArray.start out__)
-    grad_output
-    output
-    (Int64.of_int dim)
-    self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__sparse_softmax_backward_data grad_output output (Int64.of_int dim) self
+  |> with_tensor_gc
 ;;
 
 let _sparse_softmax_backward_data_out ~out ~grad_output ~output ~dim self =
-  let out__ = CArray.make t 1 in
-  stubs__sparse_softmax_backward_data_out
-    (CArray.start out__)
-    out
-    grad_output
-    output
-    (Int64.of_int dim)
-    self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__sparse_softmax_backward_data_out out grad_output output (Int64.of_int dim) self
+  |> with_tensor_gc
 ;;
 
 let _sparse_softmax_int self ~dim ~dtype =
-  let out__ = CArray.make t 1 in
-  stubs__sparse_softmax_int
-    (CArray.start out__)
-    self
-    (Int64.of_int dim)
-    (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__sparse_softmax_int self (Int64.of_int dim) (Kind.packed_to_int dtype)
+  |> with_tensor_gc
 ;;
 
 let _sparse_softmax_out ~out self ~dim ~half_to_float =
-  let out__ = CArray.make t 1 in
-  stubs__sparse_softmax_out
-    (CArray.start out__)
-    out
-    self
-    (Int64.of_int dim)
-    (if half_to_float then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__sparse_softmax_out out self (Int64.of_int dim) (if half_to_float then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _sparse_sparse_matmul self other =
-  let out__ = CArray.make t 1 in
-  stubs__sparse_sparse_matmul (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__sparse_sparse_matmul self other |> with_tensor_gc
 ;;
 
 let _sparse_sparse_matmul_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs__sparse_sparse_matmul_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__sparse_sparse_matmul_out out self other |> with_tensor_gc
 ;;
 
-let _sparse_sum self =
-  let out__ = CArray.make t 1 in
-  stubs__sparse_sum (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let _sparse_sum self = stubs__sparse_sum self |> with_tensor_gc
 
 let _sparse_sum_backward ~grad self ~dim =
-  let out__ = CArray.make t 1 in
   stubs__sparse_sum_backward
-    (CArray.start out__)
     grad
     self
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
-    (List.length dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length dim)
+  |> with_tensor_gc
 ;;
 
 let _sparse_sum_backward_out ~out ~grad self ~dim =
-  let out__ = CArray.make t 1 in
   stubs__sparse_sum_backward_out
-    (CArray.start out__)
     out
     grad
     self
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
-    (List.length dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length dim)
+  |> with_tensor_gc
 ;;
 
 let _sparse_sum_dim self ~dim =
-  let out__ = CArray.make t 1 in
   stubs__sparse_sum_dim
-    (CArray.start out__)
     self
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
-    (List.length dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length dim)
+  |> with_tensor_gc
 ;;
 
 let _sparse_sum_dim_dtype self ~dim ~dtype =
-  let out__ = CArray.make t 1 in
   stubs__sparse_sum_dim_dtype
-    (CArray.start out__)
     self
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
     (List.length dim)
-    (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Kind.packed_to_int dtype)
+  |> with_tensor_gc
 ;;
 
 let _sparse_sum_dim_out ~out self ~dim =
-  let out__ = CArray.make t 1 in
   stubs__sparse_sum_dim_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
-    (List.length dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length dim)
+  |> with_tensor_gc
 ;;
 
 let _sparse_sum_dtype self ~dtype =
-  let out__ = CArray.make t 1 in
-  stubs__sparse_sum_dtype (CArray.start out__) self (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__sparse_sum_dtype self (Kind.packed_to_int dtype) |> with_tensor_gc
 ;;
 
 let _spdiags ~diagonals ~offsets ~shape =
-  let out__ = CArray.make t 1 in
   stubs__spdiags
-    (CArray.start out__)
     diagonals
     offsets
     (List.map Int64.of_int shape |> CArray.of_list int64_t |> CArray.start)
-    (List.length shape);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length shape)
+  |> with_tensor_gc
 ;;
 
 let _spdiags_out ~out ~diagonals ~offsets ~shape =
-  let out__ = CArray.make t 1 in
   stubs__spdiags_out
-    (CArray.start out__)
     out
     diagonals
     offsets
     (List.map Int64.of_int shape |> CArray.of_list int64_t |> CArray.start)
-    (List.length shape);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length shape)
+  |> with_tensor_gc
 ;;
 
 let _stack tensors ~dim =
-  let out__ = CArray.make t 1 in
   stubs__stack
-    (CArray.start out__)
-    (CArray.of_list t tensors |> CArray.start)
+    (CArray.of_list gc_tensor tensors |> CArray.start)
     (List.length tensors)
-    (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int dim)
+  |> with_tensor_gc
 ;;
 
 let _stack_out ~out tensors ~dim =
-  let out__ = CArray.make t 1 in
   stubs__stack_out
-    (CArray.start out__)
     out
-    (CArray.of_list t tensors |> CArray.start)
+    (CArray.of_list gc_tensor tensors |> CArray.start)
     (List.length tensors)
-    (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int dim)
+  |> with_tensor_gc
 ;;
 
-let _standard_gamma self =
-  let out__ = CArray.make t 1 in
-  stubs__standard_gamma (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let _standard_gamma self = stubs__standard_gamma self |> with_tensor_gc
 
 let _standard_gamma_grad self ~output =
-  let out__ = CArray.make t 1 in
-  stubs__standard_gamma_grad (CArray.start out__) self output;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__standard_gamma_grad self output |> with_tensor_gc
 ;;
 
 let _standard_gamma_grad_out ~out self ~output =
-  let out__ = CArray.make t 1 in
-  stubs__standard_gamma_grad_out (CArray.start out__) out self output;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__standard_gamma_grad_out out self output |> with_tensor_gc
 ;;
 
-let _standard_gamma_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs__standard_gamma_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let _standard_gamma_out ~out self = stubs__standard_gamma_out out self |> with_tensor_gc
 
 let _test_ambiguous_defaults ~dummy ~a ~b =
-  let out__ = CArray.make t 1 in
-  stubs__test_ambiguous_defaults
-    (CArray.start out__)
-    dummy
-    (Int64.of_int a)
-    (Int64.of_int b);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__test_ambiguous_defaults dummy (Int64.of_int a) (Int64.of_int b) |> with_tensor_gc
 ;;
 
 let _test_ambiguous_defaults_b ~dummy ~a ~b =
-  let out__ = CArray.make t 1 in
-  stubs__test_ambiguous_defaults_b (CArray.start out__) dummy (Int64.of_int a) b;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__test_ambiguous_defaults_b dummy (Int64.of_int a) b |> with_tensor_gc
 ;;
 
 let _test_autograd_multiple_dispatch self =
-  let out__ = CArray.make t 1 in
-  stubs__test_autograd_multiple_dispatch (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__test_autograd_multiple_dispatch self |> with_tensor_gc
 ;;
 
 let _test_autograd_multiple_dispatch_fullcoverage_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs__test_autograd_multiple_dispatch_fullcoverage_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__test_autograd_multiple_dispatch_fullcoverage_out out self |> with_tensor_gc
 ;;
 
 let _test_autograd_multiple_dispatch_ntonly self ~b =
-  let out__ = CArray.make t 1 in
-  stubs__test_autograd_multiple_dispatch_ntonly
-    (CArray.start out__)
-    self
-    (if b then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__test_autograd_multiple_dispatch_ntonly self (if b then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _test_autograd_multiple_dispatch_view self =
-  let out__ = CArray.make t 1 in
-  stubs__test_autograd_multiple_dispatch_view (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__test_autograd_multiple_dispatch_view self |> with_tensor_gc
 ;;
 
 let _test_autograd_multiple_dispatch_view_copy self =
-  let out__ = CArray.make t 1 in
-  stubs__test_autograd_multiple_dispatch_view_copy (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__test_autograd_multiple_dispatch_view_copy self |> with_tensor_gc
 ;;
 
 let _test_autograd_multiple_dispatch_view_copy_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs__test_autograd_multiple_dispatch_view_copy_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__test_autograd_multiple_dispatch_view_copy_out out self |> with_tensor_gc
 ;;
 
-let _test_check_tensor self =
-  let out__ = CArray.make t 1 in
-  stubs__test_check_tensor (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let _test_check_tensor self = stubs__test_check_tensor self |> with_tensor_gc
 
 let _test_functorch_fallback self other =
-  let out__ = CArray.make t 1 in
-  stubs__test_functorch_fallback (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__test_functorch_fallback self other |> with_tensor_gc
 ;;
 
 let _test_functorch_fallback_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs__test_functorch_fallback_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__test_functorch_fallback_out out self other |> with_tensor_gc
 ;;
 
 let _test_optional_filled_intlist ~values ~addends =
-  let out__ = CArray.make t 1 in
   stubs__test_optional_filled_intlist
-    (CArray.start out__)
     values
     (match addends with
      | None -> from_voidp int64_t null
      | Some v -> List.map Int64.of_int v |> CArray.of_list int64_t |> CArray.start)
     (match addends with
      | None -> -1
-     | Some v -> List.length v);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | Some v -> List.length v)
+  |> with_tensor_gc
 ;;
 
 let _test_optional_filled_intlist_out ~out ~values ~addends =
-  let out__ = CArray.make t 1 in
   stubs__test_optional_filled_intlist_out
-    (CArray.start out__)
     out
     values
     (match addends with
@@ -6100,57 +4490,41 @@ let _test_optional_filled_intlist_out ~out ~values ~addends =
      | Some v -> List.map Int64.of_int v |> CArray.of_list int64_t |> CArray.start)
     (match addends with
      | None -> -1
-     | Some v -> List.length v);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | Some v -> List.length v)
+  |> with_tensor_gc
 ;;
 
 let _test_optional_floatlist ~values ~addends =
-  let out__ = CArray.make t 1 in
   stubs__test_optional_floatlist
-    (CArray.start out__)
     values
     (addends |> CArray.of_list double |> CArray.start)
-    (List.length addends);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length addends)
+  |> with_tensor_gc
 ;;
 
 let _test_optional_floatlist_out ~out ~values ~addends =
-  let out__ = CArray.make t 1 in
   stubs__test_optional_floatlist_out
-    (CArray.start out__)
     out
     values
     (addends |> CArray.of_list double |> CArray.start)
-    (List.length addends);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length addends)
+  |> with_tensor_gc
 ;;
 
 let _test_optional_intlist ~values ~addends =
-  let out__ = CArray.make t 1 in
   stubs__test_optional_intlist
-    (CArray.start out__)
     values
     (match addends with
      | None -> from_voidp int64_t null
      | Some v -> List.map Int64.of_int v |> CArray.of_list int64_t |> CArray.start)
     (match addends with
      | None -> -1
-     | Some v -> List.length v);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | Some v -> List.length v)
+  |> with_tensor_gc
 ;;
 
 let _test_optional_intlist_out ~out ~values ~addends =
-  let out__ = CArray.make t 1 in
   stubs__test_optional_intlist_out
-    (CArray.start out__)
     out
     values
     (match addends with
@@ -6158,42 +4532,22 @@ let _test_optional_intlist_out ~out ~values ~addends =
      | Some v -> List.map Int64.of_int v |> CArray.of_list int64_t |> CArray.start)
     (match addends with
      | None -> -1
-     | Some v -> List.length v);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | Some v -> List.length v)
+  |> with_tensor_gc
 ;;
 
 let _test_serialization_subcmul self other =
-  let out__ = CArray.make t 1 in
-  stubs__test_serialization_subcmul (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__test_serialization_subcmul self other |> with_tensor_gc
 ;;
 
 let _test_string_default ~dummy ~a ~b =
-  let out__ = CArray.make t 1 in
-  stubs__test_string_default (CArray.start out__) dummy a b;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__test_string_default dummy a b |> with_tensor_gc
 ;;
 
-let _test_warn_in_autograd self =
-  let out__ = CArray.make t 1 in
-  stubs__test_warn_in_autograd (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let _test_warn_in_autograd self = stubs__test_warn_in_autograd self |> with_tensor_gc
 
 let _test_warn_in_autograd_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs__test_warn_in_autograd_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__test_warn_in_autograd_out out self |> with_tensor_gc
 ;;
 
 let _thnn_differentiable_gru_cell_backward
@@ -6204,7 +4558,7 @@ let _thnn_differentiable_gru_cell_backward
   ~input_bias
   ~hidden_bias
   =
-  let out__ = CArray.make t 5 in
+  let out__ = CArray.make raw_tensor 5 in
   stubs__thnn_differentiable_gru_cell_backward
     (CArray.start out__)
     grad_hy
@@ -6213,20 +4567,15 @@ let _thnn_differentiable_gru_cell_backward
     hx
     (match input_bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match hidden_bias with
      | Some v -> v
-     | None -> null);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
-  let t3 = CArray.get out__ 3 in
-  Gc.finalise C.Tensor.free t3;
-  let t4 = CArray.get out__ 4 in
-  Gc.finalise C.Tensor.free t4;
+     | None -> none_gc_tensor);
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
+  let t3 = CArray.get out__ 3 |> with_tensor_gc in
+  let t4 = CArray.get out__ 4 |> with_tensor_gc in
   t0, t1, t2, t3, t4
 ;;
 
@@ -6240,40 +4589,35 @@ let _thnn_differentiable_lstm_cell_backward
   ~cx
   ~cy
   =
-  let out__ = CArray.make t 5 in
+  let out__ = CArray.make raw_tensor 5 in
   stubs__thnn_differentiable_lstm_cell_backward
     (CArray.start out__)
     (match grad_hy with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match grad_cy with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     input_gates
     hidden_gates
     (match input_bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match hidden_bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     cx
     cy;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
-  let t3 = CArray.get out__ 3 in
-  Gc.finalise C.Tensor.free t3;
-  let t4 = CArray.get out__ 4 in
-  Gc.finalise C.Tensor.free t4;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
+  let t3 = CArray.get out__ 3 |> with_tensor_gc in
+  let t4 = CArray.get out__ 4 |> with_tensor_gc in
   t0, t1, t2, t3, t4
 ;;
 
 let _thnn_fused_gru_cell ~input_gates ~hidden_gates ~hx ~input_bias ~hidden_bias =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs__thnn_fused_gru_cell
     (CArray.start out__)
     input_gates
@@ -6281,34 +4625,27 @@ let _thnn_fused_gru_cell ~input_gates ~hidden_gates ~hx ~input_bias ~hidden_bias
     hx
     (match input_bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match hidden_bias with
      | Some v -> v
-     | None -> null);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+     | None -> none_gc_tensor);
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let _thnn_fused_gru_cell_backward ~grad_hy ~workspace ~has_bias =
-  let out__ = CArray.make t 5 in
+  let out__ = CArray.make raw_tensor 5 in
   stubs__thnn_fused_gru_cell_backward
     (CArray.start out__)
     grad_hy
     workspace
     (if has_bias then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
-  let t3 = CArray.get out__ 3 in
-  Gc.finalise C.Tensor.free t3;
-  let t4 = CArray.get out__ 4 in
-  Gc.finalise C.Tensor.free t4;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
+  let t3 = CArray.get out__ 3 |> with_tensor_gc in
+  let t4 = CArray.get out__ 4 |> with_tensor_gc in
   t0, t1, t2, t3, t4
 ;;
 
@@ -6322,7 +4659,7 @@ let _thnn_fused_gru_cell_backward_out
   ~workspace
   ~has_bias
   =
-  let out__ = CArray.make t 5 in
+  let out__ = CArray.make raw_tensor 5 in
   stubs__thnn_fused_gru_cell_backward_out
     (CArray.start out__)
     out0
@@ -6333,16 +4670,11 @@ let _thnn_fused_gru_cell_backward_out
     grad_hy
     workspace
     (if has_bias then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
-  let t3 = CArray.get out__ 3 in
-  Gc.finalise C.Tensor.free t3;
-  let t4 = CArray.get out__ 4 in
-  Gc.finalise C.Tensor.free t4;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
+  let t3 = CArray.get out__ 3 |> with_tensor_gc in
+  let t4 = CArray.get out__ 4 |> with_tensor_gc in
   t0, t1, t2, t3, t4
 ;;
 
@@ -6355,7 +4687,7 @@ let _thnn_fused_gru_cell_out
   ~input_bias
   ~hidden_bias
   =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs__thnn_fused_gru_cell_out
     (CArray.start out__)
     out0
@@ -6365,19 +4697,17 @@ let _thnn_fused_gru_cell_out
     hx
     (match input_bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match hidden_bias with
      | Some v -> v
-     | None -> null);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+     | None -> none_gc_tensor);
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let _thnn_fused_lstm_cell ~input_gates ~hidden_gates ~cx ~input_bias ~hidden_bias =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs__thnn_fused_lstm_cell
     (CArray.start out__)
     input_gates
@@ -6385,66 +4715,55 @@ let _thnn_fused_lstm_cell ~input_gates ~hidden_gates ~cx ~input_bias ~hidden_bia
     cx
     (match input_bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match hidden_bias with
      | Some v -> v
-     | None -> null);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+     | None -> none_gc_tensor);
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
 let _thnn_fused_lstm_cell_backward ~grad_hy ~grad_cy ~cx ~cy ~workspace ~has_bias =
-  let out__ = CArray.make t 5 in
+  let out__ = CArray.make raw_tensor 5 in
   stubs__thnn_fused_lstm_cell_backward
     (CArray.start out__)
     (match grad_hy with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match grad_cy with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     cx
     cy
     workspace
     (if has_bias then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
-  let t3 = CArray.get out__ 3 in
-  Gc.finalise C.Tensor.free t3;
-  let t4 = CArray.get out__ 4 in
-  Gc.finalise C.Tensor.free t4;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
+  let t3 = CArray.get out__ 3 |> with_tensor_gc in
+  let t4 = CArray.get out__ 4 |> with_tensor_gc in
   t0, t1, t2, t3, t4
 ;;
 
 let _thnn_fused_lstm_cell_backward_impl ~grad_hy ~grad_cy ~cx ~cy ~workspace ~has_bias =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs__thnn_fused_lstm_cell_backward_impl
     (CArray.start out__)
     (match grad_hy with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match grad_cy with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     cx
     cy
     workspace
     (if has_bias then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
@@ -6459,7 +4778,7 @@ let _thnn_fused_lstm_cell_backward_impl_out
   ~workspace
   ~has_bias
   =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs__thnn_fused_lstm_cell_backward_impl_out
     (CArray.start out__)
     out0
@@ -6467,20 +4786,17 @@ let _thnn_fused_lstm_cell_backward_impl_out
     out2
     (match grad_hy with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match grad_cy with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     cx
     cy
     workspace
     (if has_bias then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
@@ -6494,7 +4810,7 @@ let _thnn_fused_lstm_cell_out
   ~input_bias
   ~hidden_bias
   =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs__thnn_fused_lstm_cell_out
     (CArray.start out__)
     out0
@@ -6505,74 +4821,46 @@ let _thnn_fused_lstm_cell_out
     cx
     (match input_bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match hidden_bias with
      | Some v -> v
-     | None -> null);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+     | None -> none_gc_tensor);
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
 let _to_copy self ~options ~non_blocking =
-  let out__ = CArray.make t 1 in
   stubs__to_copy
-    (CArray.start out__)
     self
     (Kind.packed_to_int (fst options))
     (Device.to_int (snd options))
-    (if non_blocking then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if non_blocking then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _to_copy_out ~out self ~non_blocking =
-  let out__ = CArray.make t 1 in
-  stubs__to_copy_out (CArray.start out__) out self (if non_blocking then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__to_copy_out out self (if non_blocking then 1 else 0) |> with_tensor_gc
 ;;
 
 let _to_cpu tensors =
-  stubs__to_cpu (CArray.of_list t tensors |> CArray.start) (List.length tensors)
+  stubs__to_cpu (CArray.of_list gc_tensor tensors |> CArray.start) (List.length tensors)
   |> to_tensor_list
 ;;
 
 let _to_dense self ~dtype ~masked_grad =
-  let out__ = CArray.make t 1 in
-  stubs__to_dense
-    (CArray.start out__)
-    self
-    (Kind.packed_to_int dtype)
-    (if masked_grad then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__to_dense self (Kind.packed_to_int dtype) (if masked_grad then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _to_dense_out ~out self ~dtype ~masked_grad =
-  let out__ = CArray.make t 1 in
-  stubs__to_dense_out
-    (CArray.start out__)
-    out
-    self
-    (Kind.packed_to_int dtype)
-    (if masked_grad then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__to_dense_out out self (Kind.packed_to_int dtype) (if masked_grad then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _to_sparse_bsc self ~blocksize ~dense_dim =
-  let out__ = CArray.make t 1 in
   stubs__to_sparse_bsc
-    (CArray.start out__)
     self
     (List.map Int64.of_int blocksize |> CArray.of_list int64_t |> CArray.start)
     (List.length blocksize)
@@ -6581,16 +4869,12 @@ let _to_sparse_bsc self ~blocksize ~dense_dim =
      | Some v -> Int64.of_int v)
     (match dense_dim with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let _to_sparse_bsc_out ~out self ~blocksize ~dense_dim =
-  let out__ = CArray.make t 1 in
   stubs__to_sparse_bsc_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int blocksize |> CArray.of_list int64_t |> CArray.start)
@@ -6600,16 +4884,12 @@ let _to_sparse_bsc_out ~out self ~blocksize ~dense_dim =
      | Some v -> Int64.of_int v)
     (match dense_dim with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let _to_sparse_bsr self ~blocksize ~dense_dim =
-  let out__ = CArray.make t 1 in
   stubs__to_sparse_bsr
-    (CArray.start out__)
     self
     (List.map Int64.of_int blocksize |> CArray.of_list int64_t |> CArray.start)
     (List.length blocksize)
@@ -6618,16 +4898,12 @@ let _to_sparse_bsr self ~blocksize ~dense_dim =
      | Some v -> Int64.of_int v)
     (match dense_dim with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let _to_sparse_bsr_out ~out self ~blocksize ~dense_dim =
-  let out__ = CArray.make t 1 in
   stubs__to_sparse_bsr_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int blocksize |> CArray.of_list int64_t |> CArray.start)
@@ -6637,32 +4913,24 @@ let _to_sparse_bsr_out ~out self ~blocksize ~dense_dim =
      | Some v -> Int64.of_int v)
     (match dense_dim with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let _to_sparse_csc self ~dense_dim =
-  let out__ = CArray.make t 1 in
   stubs__to_sparse_csc
-    (CArray.start out__)
     self
     (match dense_dim with
      | None -> Int64.zero
      | Some v -> Int64.of_int v)
     (match dense_dim with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let _to_sparse_csc_out ~out self ~dense_dim =
-  let out__ = CArray.make t 1 in
   stubs__to_sparse_csc_out
-    (CArray.start out__)
     out
     self
     (match dense_dim with
@@ -6670,32 +4938,24 @@ let _to_sparse_csc_out ~out self ~dense_dim =
      | Some v -> Int64.of_int v)
     (match dense_dim with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let _to_sparse_csr self ~dense_dim =
-  let out__ = CArray.make t 1 in
   stubs__to_sparse_csr
-    (CArray.start out__)
     self
     (match dense_dim with
      | None -> Int64.zero
      | Some v -> Int64.of_int v)
     (match dense_dim with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let _to_sparse_csr_out ~out self ~dense_dim =
-  let out__ = CArray.make t 1 in
   stubs__to_sparse_csr_out
-    (CArray.start out__)
     out
     self
     (match dense_dim with
@@ -6703,40 +4963,33 @@ let _to_sparse_csr_out ~out self ~dense_dim =
      | Some v -> Int64.of_int v)
     (match dense_dim with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let _to_sparse_semi_structured ~dense =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs__to_sparse_semi_structured (CArray.start out__) dense;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let _transform_bias_rescale_qkv ~qkv ~qkv_bias ~num_heads =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs__transform_bias_rescale_qkv
     (CArray.start out__)
     qkv
     qkv_bias
     (Int64.of_int num_heads);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
 let _transform_bias_rescale_qkv_out ~out0 ~out1 ~out2 ~qkv ~qkv_bias ~num_heads =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs__transform_bias_rescale_qkv_out
     (CArray.start out__)
     out0
@@ -6745,12 +4998,9 @@ let _transform_bias_rescale_qkv_out ~out0 ~out1 ~out2 ~qkv ~qkv_bias ~num_heads
     qkv
     qkv_bias
     (Int64.of_int num_heads);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
@@ -6776,9 +5026,7 @@ let _transformer_encoder_layer_fwd
   ~mask
   ~mask_type
   =
-  let out__ = CArray.make t 1 in
   stubs__transformer_encoder_layer_fwd
-    (CArray.start out__)
     src
     (Int64.of_int embed_dim)
     (Int64.of_int num_heads)
@@ -6799,16 +5047,14 @@ let _transformer_encoder_layer_fwd
     ffn_bias_2
     (match mask with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match mask_type with
      | None -> Int64.zero
      | Some v -> Int64.of_int v)
     (match mask_type with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let _transformer_encoder_layer_fwd_out
@@ -6834,9 +5080,7 @@ let _transformer_encoder_layer_fwd_out
   ~mask
   ~mask_type
   =
-  let out__ = CArray.make t 1 in
   stubs__transformer_encoder_layer_fwd_out
-    (CArray.start out__)
     out
     src
     (Int64.of_int embed_dim)
@@ -6858,22 +5102,18 @@ let _transformer_encoder_layer_fwd_out
     ffn_bias_2
     (match mask with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match mask_type with
      | None -> Int64.zero
      | Some v -> Int64.of_int v)
     (match mask_type with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let _trilinear ~i1 ~i2 ~i3 ~expand1 ~expand2 ~expand3 ~sumdim ~unroll_dim =
-  let out__ = CArray.make t 1 in
   stubs__trilinear
-    (CArray.start out__)
     i1
     i2
     i3
@@ -6885,16 +5125,12 @@ let _trilinear ~i1 ~i2 ~i3 ~expand1 ~expand2 ~expand3 ~sumdim ~unroll_dim =
     (List.length expand3)
     (List.map Int64.of_int sumdim |> CArray.of_list int64_t |> CArray.start)
     (List.length sumdim)
-    (Int64.of_int unroll_dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int unroll_dim)
+  |> with_tensor_gc
 ;;
 
 let _trilinear_out ~out ~i1 ~i2 ~i3 ~expand1 ~expand2 ~expand3 ~sumdim ~unroll_dim =
-  let out__ = CArray.make t 1 in
   stubs__trilinear_out
-    (CArray.start out__)
     out
     i1
     i2
@@ -6907,10 +5143,8 @@ let _trilinear_out ~out ~i1 ~i2 ~i3 ~expand1 ~expand2 ~expand3 ~sumdim ~unroll_d
     (List.length expand3)
     (List.map Int64.of_int sumdim |> CArray.of_list int64_t |> CArray.start)
     (List.length sumdim)
-    (Int64.of_int unroll_dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int unroll_dim)
+  |> with_tensor_gc
 ;;
 
 let _triton_multi_head_attention
@@ -6925,9 +5159,7 @@ let _triton_multi_head_attention
   ~proj_bias
   ~mask
   =
-  let out__ = CArray.make t 1 in
   stubs__triton_multi_head_attention
-    (CArray.start out__)
     query
     key
     value
@@ -6939,10 +5171,8 @@ let _triton_multi_head_attention
     proj_bias
     (match mask with
      | Some v -> v
-     | None -> null);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+  |> with_tensor_gc
 ;;
 
 let _triton_multi_head_attention_out
@@ -6958,9 +5188,7 @@ let _triton_multi_head_attention_out
   ~proj_bias
   ~mask
   =
-  let out__ = CArray.make t 1 in
   stubs__triton_multi_head_attention_out
-    (CArray.start out__)
     out
     query
     key
@@ -6973,61 +5201,46 @@ let _triton_multi_head_attention_out
     proj_bias
     (match mask with
      | Some v -> v
-     | None -> null);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+  |> with_tensor_gc
 ;;
 
 let _triton_scaled_dot_attention ~q ~k ~v ~dropout_p =
-  let out__ = CArray.make t 1 in
-  stubs__triton_scaled_dot_attention (CArray.start out__) q k v dropout_p;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__triton_scaled_dot_attention q k v dropout_p |> with_tensor_gc
 ;;
 
 let _triton_scaled_dot_attention_out ~out ~q ~k ~v ~dropout_p =
-  let out__ = CArray.make t 1 in
-  stubs__triton_scaled_dot_attention_out (CArray.start out__) out q k v dropout_p;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs__triton_scaled_dot_attention_out out q k v dropout_p |> with_tensor_gc
 ;;
 
 let _unique self ~sorted ~return_inverse =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs__unique
     (CArray.start out__)
     self
     (if sorted then 1 else 0)
     (if return_inverse then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let _unique2 self ~sorted ~return_inverse ~return_counts =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs__unique2
     (CArray.start out__)
     self
     (if sorted then 1 else 0)
     (if return_inverse then 1 else 0)
     (if return_counts then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
 let _unique2_out ~out0 ~out1 ~out2 self ~sorted ~return_inverse ~return_counts =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs__unique2_out
     (CArray.start out__)
     out0
@@ -7037,17 +5250,14 @@ let _unique2_out ~out0 ~out1 ~out2 self ~sorted ~return_inverse ~return_counts =
     (if sorted then 1 else 0)
     (if return_inverse then 1 else 0)
     (if return_counts then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
 let _unique_out ~out0 ~out1 self ~sorted ~return_inverse =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs__unique_out
     (CArray.start out__)
     out0
@@ -7055,90 +5265,68 @@ let _unique_out ~out0 ~out1 self ~sorted ~return_inverse =
     self
     (if sorted then 1 else 0)
     (if return_inverse then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let _unpack_dual ~dual ~level =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs__unpack_dual (CArray.start out__) dual (Int64.of_int level);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let _unsafe_index self ~indices =
-  let out__ = CArray.make t 1 in
   stubs__unsafe_index
-    (CArray.start out__)
     self
     (List.map
        (function
         | Some x -> x
-        | None -> null)
+        | None -> none_gc_tensor)
        indices
-     |> CArray.of_list t
+     |> CArray.of_list gc_tensor
      |> CArray.start)
-    (List.length indices);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length indices)
+  |> with_tensor_gc
 ;;
 
 let _unsafe_index_put self ~indices ~values ~accumulate =
-  let out__ = CArray.make t 1 in
   stubs__unsafe_index_put
-    (CArray.start out__)
     self
     (List.map
        (function
         | Some x -> x
-        | None -> null)
+        | None -> none_gc_tensor)
        indices
-     |> CArray.of_list t
+     |> CArray.of_list gc_tensor
      |> CArray.start)
     (List.length indices)
     values
-    (if accumulate then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if accumulate then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let _unsafe_view self ~size =
-  let out__ = CArray.make t 1 in
   stubs__unsafe_view
-    (CArray.start out__)
     self
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
-    (List.length size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length size)
+  |> with_tensor_gc
 ;;
 
 let _unsafe_view_out ~out self ~size =
-  let out__ = CArray.make t 1 in
   stubs__unsafe_view_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
-    (List.length size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length size)
+  |> with_tensor_gc
 ;;
 
 let _upsample_bicubic2d_aa self ~output_size ~align_corners ~scales_h ~scales_w =
-  let out__ = CArray.make t 1 in
   stubs__upsample_bicubic2d_aa
-    (CArray.start out__)
     self
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
     (List.length output_size)
@@ -7150,10 +5338,8 @@ let _upsample_bicubic2d_aa self ~output_size ~align_corners ~scales_h ~scales_w
     (Option.value scales_w ~default:0.0)
     (match scales_w with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let _upsample_bicubic2d_aa_backward
@@ -7164,9 +5350,7 @@ let _upsample_bicubic2d_aa_backward
   ~scales_h
   ~scales_w
   =
-  let out__ = CArray.make t 1 in
   stubs__upsample_bicubic2d_aa_backward
-    (CArray.start out__)
     grad_output
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
     (List.length output_size)
@@ -7180,10 +5364,8 @@ let _upsample_bicubic2d_aa_backward
     (Option.value scales_w ~default:0.0)
     (match scales_w with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let _upsample_bicubic2d_aa_backward_grad_input
@@ -7195,9 +5377,7 @@ let _upsample_bicubic2d_aa_backward_grad_input
   ~scales_h
   ~scales_w
   =
-  let out__ = CArray.make t 1 in
   stubs__upsample_bicubic2d_aa_backward_grad_input
-    (CArray.start out__)
     grad_input
     grad_output
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
@@ -7212,16 +5392,12 @@ let _upsample_bicubic2d_aa_backward_grad_input
     (Option.value scales_w ~default:0.0)
     (match scales_w with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let _upsample_bicubic2d_aa_out ~out self ~output_size ~align_corners ~scales_h ~scales_w =
-  let out__ = CArray.make t 1 in
   stubs__upsample_bicubic2d_aa_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
@@ -7234,16 +5410,12 @@ let _upsample_bicubic2d_aa_out ~out self ~output_size ~align_corners ~scales_h ~
     (Option.value scales_w ~default:0.0)
     (match scales_w with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let _upsample_bicubic2d_aa_vec input ~output_size ~align_corners ~scale_factors =
-  let out__ = CArray.make t 1 in
   stubs__upsample_bicubic2d_aa_vec
-    (CArray.start out__)
     input
     (match output_size with
      | None -> from_voidp int64_t null
@@ -7253,16 +5425,12 @@ let _upsample_bicubic2d_aa_vec input ~output_size ~align_corners ~scale_factors
      | Some v -> List.length v)
     (if align_corners then 1 else 0)
     (scale_factors |> CArray.of_list double |> CArray.start)
-    (List.length scale_factors);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length scale_factors)
+  |> with_tensor_gc
 ;;
 
 let _upsample_bilinear2d_aa self ~output_size ~align_corners ~scales_h ~scales_w =
-  let out__ = CArray.make t 1 in
   stubs__upsample_bilinear2d_aa
-    (CArray.start out__)
     self
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
     (List.length output_size)
@@ -7274,10 +5442,8 @@ let _upsample_bilinear2d_aa self ~output_size ~align_corners ~scales_h ~scales_w
     (Option.value scales_w ~default:0.0)
     (match scales_w with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let _upsample_bilinear2d_aa_backward
@@ -7288,9 +5454,7 @@ let _upsample_bilinear2d_aa_backward
   ~scales_h
   ~scales_w
   =
-  let out__ = CArray.make t 1 in
   stubs__upsample_bilinear2d_aa_backward
-    (CArray.start out__)
     grad_output
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
     (List.length output_size)
@@ -7304,10 +5468,8 @@ let _upsample_bilinear2d_aa_backward
     (Option.value scales_w ~default:0.0)
     (match scales_w with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let _upsample_bilinear2d_aa_backward_grad_input
@@ -7319,9 +5481,7 @@ let _upsample_bilinear2d_aa_backward_grad_input
   ~scales_h
   ~scales_w
   =
-  let out__ = CArray.make t 1 in
   stubs__upsample_bilinear2d_aa_backward_grad_input
-    (CArray.start out__)
     grad_input
     grad_output
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
@@ -7336,16 +5496,12 @@ let _upsample_bilinear2d_aa_backward_grad_input
     (Option.value scales_w ~default:0.0)
     (match scales_w with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let _upsample_bilinear2d_aa_out ~out self ~output_size ~align_corners ~scales_h ~scales_w =
-  let out__ = CArray.make t 1 in
   stubs__upsample_bilinear2d_aa_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
@@ -7358,16 +5514,12 @@ let _upsample_bilinear2d_aa_out ~out self ~output_size ~align_corners ~scales_h
     (Option.value scales_w ~default:0.0)
     (match scales_w with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let _upsample_bilinear2d_aa_vec input ~output_size ~align_corners ~scale_factors =
-  let out__ = CArray.make t 1 in
   stubs__upsample_bilinear2d_aa_vec
-    (CArray.start out__)
     input
     (match output_size with
      | None -> from_voidp int64_t null
@@ -7377,32 +5529,24 @@ let _upsample_bilinear2d_aa_vec input ~output_size ~align_corners ~scale_factors
      | Some v -> List.length v)
     (if align_corners then 1 else 0)
     (scale_factors |> CArray.of_list double |> CArray.start)
-    (List.length scale_factors);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length scale_factors)
+  |> with_tensor_gc
 ;;
 
 let _upsample_nearest_exact1d self ~output_size ~scales =
-  let out__ = CArray.make t 1 in
   stubs__upsample_nearest_exact1d
-    (CArray.start out__)
     self
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
     (List.length output_size)
     (Option.value scales ~default:0.0)
     (match scales with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let _upsample_nearest_exact1d_backward ~grad_output ~output_size ~input_size ~scales =
-  let out__ = CArray.make t 1 in
   stubs__upsample_nearest_exact1d_backward
-    (CArray.start out__)
     grad_output
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
     (List.length output_size)
@@ -7411,10 +5555,8 @@ let _upsample_nearest_exact1d_backward ~grad_output ~output_size ~input_size ~sc
     (Option.value scales ~default:0.0)
     (match scales with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let _upsample_nearest_exact1d_backward_grad_input
@@ -7424,9 +5566,7 @@ let _upsample_nearest_exact1d_backward_grad_input
   ~input_size
   ~scales
   =
-  let out__ = CArray.make t 1 in
   stubs__upsample_nearest_exact1d_backward_grad_input
-    (CArray.start out__)
     grad_input
     grad_output
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
@@ -7436,16 +5576,12 @@ let _upsample_nearest_exact1d_backward_grad_input
     (Option.value scales ~default:0.0)
     (match scales with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let _upsample_nearest_exact1d_out ~out self ~output_size ~scales =
-  let out__ = CArray.make t 1 in
   stubs__upsample_nearest_exact1d_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
@@ -7453,16 +5589,12 @@ let _upsample_nearest_exact1d_out ~out self ~output_size ~scales =
     (Option.value scales ~default:0.0)
     (match scales with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let _upsample_nearest_exact1d_vec input ~output_size ~scale_factors =
-  let out__ = CArray.make t 1 in
   stubs__upsample_nearest_exact1d_vec
-    (CArray.start out__)
     input
     (match output_size with
      | None -> from_voidp int64_t null
@@ -7471,16 +5603,12 @@ let _upsample_nearest_exact1d_vec input ~output_size ~scale_factors =
      | None -> -1
      | Some v -> List.length v)
     (scale_factors |> CArray.of_list double |> CArray.start)
-    (List.length scale_factors);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length scale_factors)
+  |> with_tensor_gc
 ;;
 
 let _upsample_nearest_exact2d self ~output_size ~scales_h ~scales_w =
-  let out__ = CArray.make t 1 in
   stubs__upsample_nearest_exact2d
-    (CArray.start out__)
     self
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
     (List.length output_size)
@@ -7491,10 +5619,8 @@ let _upsample_nearest_exact2d self ~output_size ~scales_h ~scales_w =
     (Option.value scales_w ~default:0.0)
     (match scales_w with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let _upsample_nearest_exact2d_backward
@@ -7504,9 +5630,7 @@ let _upsample_nearest_exact2d_backward
   ~scales_h
   ~scales_w
   =
-  let out__ = CArray.make t 1 in
   stubs__upsample_nearest_exact2d_backward
-    (CArray.start out__)
     grad_output
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
     (List.length output_size)
@@ -7519,10 +5643,8 @@ let _upsample_nearest_exact2d_backward
     (Option.value scales_w ~default:0.0)
     (match scales_w with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let _upsample_nearest_exact2d_backward_grad_input
@@ -7533,9 +5655,7 @@ let _upsample_nearest_exact2d_backward_grad_input
   ~scales_h
   ~scales_w
   =
-  let out__ = CArray.make t 1 in
   stubs__upsample_nearest_exact2d_backward_grad_input
-    (CArray.start out__)
     grad_input
     grad_output
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
@@ -7549,16 +5669,12 @@ let _upsample_nearest_exact2d_backward_grad_input
     (Option.value scales_w ~default:0.0)
     (match scales_w with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let _upsample_nearest_exact2d_out ~out self ~output_size ~scales_h ~scales_w =
-  let out__ = CArray.make t 1 in
   stubs__upsample_nearest_exact2d_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
@@ -7570,16 +5686,12 @@ let _upsample_nearest_exact2d_out ~out self ~output_size ~scales_h ~scales_w =
     (Option.value scales_w ~default:0.0)
     (match scales_w with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let _upsample_nearest_exact2d_vec input ~output_size ~scale_factors =
-  let out__ = CArray.make t 1 in
   stubs__upsample_nearest_exact2d_vec
-    (CArray.start out__)
     input
     (match output_size with
      | None -> from_voidp int64_t null
@@ -7588,16 +5700,12 @@ let _upsample_nearest_exact2d_vec input ~output_size ~scale_factors =
      | None -> -1
      | Some v -> List.length v)
     (scale_factors |> CArray.of_list double |> CArray.start)
-    (List.length scale_factors);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length scale_factors)
+  |> with_tensor_gc
 ;;
 
 let _upsample_nearest_exact3d self ~output_size ~scales_d ~scales_h ~scales_w =
-  let out__ = CArray.make t 1 in
   stubs__upsample_nearest_exact3d
-    (CArray.start out__)
     self
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
     (List.length output_size)
@@ -7612,10 +5720,8 @@ let _upsample_nearest_exact3d self ~output_size ~scales_d ~scales_h ~scales_w =
     (Option.value scales_w ~default:0.0)
     (match scales_w with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let _upsample_nearest_exact3d_backward
@@ -7626,9 +5732,7 @@ let _upsample_nearest_exact3d_backward
   ~scales_h
   ~scales_w
   =
-  let out__ = CArray.make t 1 in
   stubs__upsample_nearest_exact3d_backward
-    (CArray.start out__)
     grad_output
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
     (List.length output_size)
@@ -7645,10 +5749,8 @@ let _upsample_nearest_exact3d_backward
     (Option.value scales_w ~default:0.0)
     (match scales_w with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let _upsample_nearest_exact3d_backward_grad_input
@@ -7660,9 +5762,7 @@ let _upsample_nearest_exact3d_backward_grad_input
   ~scales_h
   ~scales_w
   =
-  let out__ = CArray.make t 1 in
   stubs__upsample_nearest_exact3d_backward_grad_input
-    (CArray.start out__)
     grad_input
     grad_output
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
@@ -7680,16 +5780,12 @@ let _upsample_nearest_exact3d_backward_grad_input
     (Option.value scales_w ~default:0.0)
     (match scales_w with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let _upsample_nearest_exact3d_out ~out self ~output_size ~scales_d ~scales_h ~scales_w =
-  let out__ = CArray.make t 1 in
   stubs__upsample_nearest_exact3d_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
@@ -7705,16 +5801,12 @@ let _upsample_nearest_exact3d_out ~out self ~output_size ~scales_d ~scales_h ~sc
     (Option.value scales_w ~default:0.0)
     (match scales_w with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let _upsample_nearest_exact3d_vec input ~output_size ~scale_factors =
-  let out__ = CArray.make t 1 in
   stubs__upsample_nearest_exact3d_vec
-    (CArray.start out__)
     input
     (match output_size with
      | None -> from_voidp int64_t null
@@ -7723,10 +5815,8 @@ let _upsample_nearest_exact3d_vec input ~output_size ~scale_factors =
      | None -> -1
      | Some v -> List.length v)
     (scale_factors |> CArray.of_list double |> CArray.start)
-    (List.length scale_factors);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length scale_factors)
+  |> with_tensor_gc
 ;;
 
 let _use_cudnn_ctc_loss ~log_probs ~targets ~input_lengths ~target_lengths ~blank =
@@ -7795,42 +5885,14 @@ let _validate_sparse_csc_tensor_args ~ccol_indices ~row_indices ~values ~size =
     (List.length size)
 ;;
 
-let _values self =
-  let out__ = CArray.make t 1 in
-  stubs__values (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let _values_copy self =
-  let out__ = CArray.make t 1 in
-  stubs__values_copy (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let _values_copy_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs__values_copy_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
+let _values self = stubs__values self |> with_tensor_gc
+let _values_copy self = stubs__values_copy self |> with_tensor_gc
+let _values_copy_out ~out self = stubs__values_copy_out out self |> with_tensor_gc
 let _version self = stubs__version self
-
-let _weight_norm ~v ~g ~dim =
-  let out__ = CArray.make t 1 in
-  stubs__weight_norm (CArray.start out__) v g (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let _weight_norm ~v ~g ~dim = stubs__weight_norm v g (Int64.of_int dim) |> with_tensor_gc
 
 let _weight_norm_differentiable_backward ~grad_w ~saved_v ~saved_g ~saved_norms ~dim =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs__weight_norm_differentiable_backward
     (CArray.start out__)
     grad_w
@@ -7838,25 +5900,21 @@ let _weight_norm_differentiable_backward ~grad_w ~saved_v ~saved_g ~saved_norms
     saved_g
     saved_norms
     (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let _weight_norm_interface ~v ~g ~dim =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs__weight_norm_interface (CArray.start out__) v g (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let _weight_norm_interface_backward ~grad_w ~saved_v ~saved_g ~saved_norms ~dim =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs__weight_norm_interface_backward
     (CArray.start out__)
     grad_w
@@ -7864,10 +5922,8 @@ let _weight_norm_interface_backward ~grad_w ~saved_v ~saved_g ~saved_norms ~dim
     saved_g
     saved_norms
     (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
@@ -7880,7 +5936,7 @@ let _weight_norm_interface_backward_out
   ~saved_norms
   ~dim
   =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs__weight_norm_interface_backward_out
     (CArray.start out__)
     out0
@@ -7890,240 +5946,113 @@ let _weight_norm_interface_backward_out
     saved_g
     saved_norms
     (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let _weight_norm_interface_out ~out0 ~out1 ~v ~g ~dim =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs__weight_norm_interface_out (CArray.start out__) out0 out1 v g (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
-let abs self =
-  let out__ = CArray.make t 1 in
-  stubs_abs (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let abs_ self =
-  let out__ = CArray.make t 1 in
-  stubs_abs_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let abs_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_abs_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let absolute self =
-  let out__ = CArray.make t 1 in
-  stubs_absolute (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let absolute_ self =
-  let out__ = CArray.make t 1 in
-  stubs_absolute_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let absolute_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_absolute_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let acos self =
-  let out__ = CArray.make t 1 in
-  stubs_acos (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let acos_ self =
-  let out__ = CArray.make t 1 in
-  stubs_acos_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let acos_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_acos_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let acosh self =
-  let out__ = CArray.make t 1 in
-  stubs_acosh (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let acosh_ self =
-  let out__ = CArray.make t 1 in
-  stubs_acosh_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let acosh_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_acosh_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let abs self = stubs_abs self |> with_tensor_gc
+let abs_ self = stubs_abs_ self |> with_tensor_gc
+let abs_out ~out self = stubs_abs_out out self |> with_tensor_gc
+let absolute self = stubs_absolute self |> with_tensor_gc
+let absolute_ self = stubs_absolute_ self |> with_tensor_gc
+let absolute_out ~out self = stubs_absolute_out out self |> with_tensor_gc
+let acos self = stubs_acos self |> with_tensor_gc
+let acos_ self = stubs_acos_ self |> with_tensor_gc
+let acos_out ~out self = stubs_acos_out out self |> with_tensor_gc
+let acosh self = stubs_acosh self |> with_tensor_gc
+let acosh_ self = stubs_acosh_ self |> with_tensor_gc
+let acosh_out ~out self = stubs_acosh_out out self |> with_tensor_gc
 
 let adaptive_avg_pool1d self ~output_size =
-  let out__ = CArray.make t 1 in
   stubs_adaptive_avg_pool1d
-    (CArray.start out__)
     self
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
-    (List.length output_size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length output_size)
+  |> with_tensor_gc
 ;;
 
 let adaptive_avg_pool2d self ~output_size =
-  let out__ = CArray.make t 1 in
   stubs_adaptive_avg_pool2d
-    (CArray.start out__)
     self
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
-    (List.length output_size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length output_size)
+  |> with_tensor_gc
 ;;
 
 let adaptive_avg_pool2d_out ~out self ~output_size =
-  let out__ = CArray.make t 1 in
   stubs_adaptive_avg_pool2d_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
-    (List.length output_size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length output_size)
+  |> with_tensor_gc
 ;;
 
 let adaptive_avg_pool3d self ~output_size =
-  let out__ = CArray.make t 1 in
   stubs_adaptive_avg_pool3d
-    (CArray.start out__)
     self
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
-    (List.length output_size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length output_size)
+  |> with_tensor_gc
 ;;
 
 let adaptive_avg_pool3d_backward ~grad_input ~grad_output self =
-  let out__ = CArray.make t 1 in
-  stubs_adaptive_avg_pool3d_backward (CArray.start out__) grad_input grad_output self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_adaptive_avg_pool3d_backward grad_input grad_output self |> with_tensor_gc
 ;;
 
 let adaptive_avg_pool3d_out ~out self ~output_size =
-  let out__ = CArray.make t 1 in
   stubs_adaptive_avg_pool3d_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
-    (List.length output_size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length output_size)
+  |> with_tensor_gc
 ;;
 
 let adaptive_max_pool1d self ~output_size =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_adaptive_max_pool1d
     (CArray.start out__)
     self
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
     (List.length output_size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let adaptive_max_pool2d self ~output_size =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_adaptive_max_pool2d
     (CArray.start out__)
     self
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
     (List.length output_size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let adaptive_max_pool2d_backward ~grad_output self ~indices =
-  let out__ = CArray.make t 1 in
-  stubs_adaptive_max_pool2d_backward (CArray.start out__) grad_output self indices;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_adaptive_max_pool2d_backward grad_output self indices |> with_tensor_gc
 ;;
 
 let adaptive_max_pool2d_backward_grad_input ~grad_input ~grad_output self ~indices =
-  let out__ = CArray.make t 1 in
-  stubs_adaptive_max_pool2d_backward_grad_input
-    (CArray.start out__)
-    grad_input
-    grad_output
-    self
-    indices;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_adaptive_max_pool2d_backward_grad_input grad_input grad_output self indices
+  |> with_tensor_gc
 ;;
 
 let adaptive_max_pool2d_out ~out ~indices self ~output_size =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_adaptive_max_pool2d_out
     (CArray.start out__)
     out
@@ -8131,50 +6060,34 @@ let adaptive_max_pool2d_out ~out ~indices self ~output_size =
     self
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
     (List.length output_size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let adaptive_max_pool3d self ~output_size =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_adaptive_max_pool3d
     (CArray.start out__)
     self
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
     (List.length output_size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let adaptive_max_pool3d_backward ~grad_output self ~indices =
-  let out__ = CArray.make t 1 in
-  stubs_adaptive_max_pool3d_backward (CArray.start out__) grad_output self indices;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_adaptive_max_pool3d_backward grad_output self indices |> with_tensor_gc
 ;;
 
 let adaptive_max_pool3d_backward_grad_input ~grad_input ~grad_output self ~indices =
-  let out__ = CArray.make t 1 in
-  stubs_adaptive_max_pool3d_backward_grad_input
-    (CArray.start out__)
-    grad_input
-    grad_output
-    self
-    indices;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_adaptive_max_pool3d_backward_grad_input grad_input grad_output self indices
+  |> with_tensor_gc
 ;;
 
 let adaptive_max_pool3d_out ~out ~indices self ~output_size =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_adaptive_max_pool3d_out
     (CArray.start out__)
     out
@@ -8182,325 +6095,104 @@ let adaptive_max_pool3d_out ~out ~indices self ~output_size =
     self
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
     (List.length output_size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
-let add self other =
-  let out__ = CArray.make t 1 in
-  stubs_add (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let add_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_add_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let add_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_add_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let add_scalar self other =
-  let out__ = CArray.make t 1 in
-  stubs_add_scalar (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let add_scalar_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_add_scalar_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let add_scalar_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_add_scalar_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let addbmm self ~batch1 ~batch2 =
-  let out__ = CArray.make t 1 in
-  stubs_addbmm (CArray.start out__) self batch1 batch2;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let addbmm_ self ~batch1 ~batch2 =
-  let out__ = CArray.make t 1 in
-  stubs_addbmm_ (CArray.start out__) self batch1 batch2;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let add self other = stubs_add self other |> with_tensor_gc
+let add_ self other = stubs_add_ self other |> with_tensor_gc
+let add_out ~out self other = stubs_add_out out self other |> with_tensor_gc
+let add_scalar self other = stubs_add_scalar self other |> with_tensor_gc
+let add_scalar_ self other = stubs_add_scalar_ self other |> with_tensor_gc
+let add_scalar_out ~out self other = stubs_add_scalar_out out self other |> with_tensor_gc
+let addbmm self ~batch1 ~batch2 = stubs_addbmm self batch1 batch2 |> with_tensor_gc
+let addbmm_ self ~batch1 ~batch2 = stubs_addbmm_ self batch1 batch2 |> with_tensor_gc
 
 let addbmm_out ~out self ~batch1 ~batch2 =
-  let out__ = CArray.make t 1 in
-  stubs_addbmm_out (CArray.start out__) out self batch1 batch2;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_addbmm_out out self batch1 batch2 |> with_tensor_gc
 ;;
 
-let addcdiv self ~tensor1 ~tensor2 =
-  let out__ = CArray.make t 1 in
-  stubs_addcdiv (CArray.start out__) self tensor1 tensor2;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let addcdiv self ~tensor1 ~tensor2 = stubs_addcdiv self tensor1 tensor2 |> with_tensor_gc
 
 let addcdiv_ self ~tensor1 ~tensor2 =
-  let out__ = CArray.make t 1 in
-  stubs_addcdiv_ (CArray.start out__) self tensor1 tensor2;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_addcdiv_ self tensor1 tensor2 |> with_tensor_gc
 ;;
 
 let addcdiv_out ~out self ~tensor1 ~tensor2 =
-  let out__ = CArray.make t 1 in
-  stubs_addcdiv_out (CArray.start out__) out self tensor1 tensor2;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_addcdiv_out out self tensor1 tensor2 |> with_tensor_gc
 ;;
 
-let addcmul self ~tensor1 ~tensor2 =
-  let out__ = CArray.make t 1 in
-  stubs_addcmul (CArray.start out__) self tensor1 tensor2;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let addcmul self ~tensor1 ~tensor2 = stubs_addcmul self tensor1 tensor2 |> with_tensor_gc
 
 let addcmul_ self ~tensor1 ~tensor2 =
-  let out__ = CArray.make t 1 in
-  stubs_addcmul_ (CArray.start out__) self tensor1 tensor2;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_addcmul_ self tensor1 tensor2 |> with_tensor_gc
 ;;
 
 let addcmul_out ~out self ~tensor1 ~tensor2 =
-  let out__ = CArray.make t 1 in
-  stubs_addcmul_out (CArray.start out__) out self tensor1 tensor2;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let addmm self ~mat1 ~mat2 =
-  let out__ = CArray.make t 1 in
-  stubs_addmm (CArray.start out__) self mat1 mat2;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let addmm_ self ~mat1 ~mat2 =
-  let out__ = CArray.make t 1 in
-  stubs_addmm_ (CArray.start out__) self mat1 mat2;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_addcmul_out out self tensor1 tensor2 |> with_tensor_gc
 ;;
 
-let addmm_out ~out self ~mat1 ~mat2 =
-  let out__ = CArray.make t 1 in
-  stubs_addmm_out (CArray.start out__) out self mat1 mat2;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let addmv self ~mat ~vec =
-  let out__ = CArray.make t 1 in
-  stubs_addmv (CArray.start out__) self mat vec;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let addmv_ self ~mat ~vec =
-  let out__ = CArray.make t 1 in
-  stubs_addmv_ (CArray.start out__) self mat vec;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let addmv_out ~out self ~mat ~vec =
-  let out__ = CArray.make t 1 in
-  stubs_addmv_out (CArray.start out__) out self mat vec;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let addr self ~vec1 ~vec2 =
-  let out__ = CArray.make t 1 in
-  stubs_addr (CArray.start out__) self vec1 vec2;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let addr_ self ~vec1 ~vec2 =
-  let out__ = CArray.make t 1 in
-  stubs_addr_ (CArray.start out__) self vec1 vec2;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let addr_out ~out self ~vec1 ~vec2 =
-  let out__ = CArray.make t 1 in
-  stubs_addr_out (CArray.start out__) out self vec1 vec2;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let adjoint self =
-  let out__ = CArray.make t 1 in
-  stubs_adjoint (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let addmm self ~mat1 ~mat2 = stubs_addmm self mat1 mat2 |> with_tensor_gc
+let addmm_ self ~mat1 ~mat2 = stubs_addmm_ self mat1 mat2 |> with_tensor_gc
+let addmm_out ~out self ~mat1 ~mat2 = stubs_addmm_out out self mat1 mat2 |> with_tensor_gc
+let addmv self ~mat ~vec = stubs_addmv self mat vec |> with_tensor_gc
+let addmv_ self ~mat ~vec = stubs_addmv_ self mat vec |> with_tensor_gc
+let addmv_out ~out self ~mat ~vec = stubs_addmv_out out self mat vec |> with_tensor_gc
+let addr self ~vec1 ~vec2 = stubs_addr self vec1 vec2 |> with_tensor_gc
+let addr_ self ~vec1 ~vec2 = stubs_addr_ self vec1 vec2 |> with_tensor_gc
+let addr_out ~out self ~vec1 ~vec2 = stubs_addr_out out self vec1 vec2 |> with_tensor_gc
+let adjoint self = stubs_adjoint self |> with_tensor_gc
 
 let affine_grid_generator ~theta ~size ~align_corners =
-  let out__ = CArray.make t 1 in
   stubs_affine_grid_generator
-    (CArray.start out__)
     theta
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
-    (if align_corners then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if align_corners then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let affine_grid_generator_backward ~grad ~size ~align_corners =
-  let out__ = CArray.make t 1 in
   stubs_affine_grid_generator_backward
-    (CArray.start out__)
     grad
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
-    (if align_corners then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if align_corners then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let affine_grid_generator_out ~out ~theta ~size ~align_corners =
-  let out__ = CArray.make t 1 in
   stubs_affine_grid_generator_out
-    (CArray.start out__)
     out
     theta
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
-    (if align_corners then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let alias self =
-  let out__ = CArray.make t 1 in
-  stubs_alias (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let alias_copy self =
-  let out__ = CArray.make t 1 in
-  stubs_alias_copy (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let alias_copy_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_alias_copy_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if align_corners then 1 else 0)
+  |> with_tensor_gc
 ;;
 
-let align_as self other =
-  let out__ = CArray.make t 1 in
-  stubs_align_as (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let alias self = stubs_alias self |> with_tensor_gc
+let alias_copy self = stubs_alias_copy self |> with_tensor_gc
+let alias_copy_out ~out self = stubs_alias_copy_out out self |> with_tensor_gc
+let align_as self other = stubs_align_as self other |> with_tensor_gc
 
 let align_tensors tensors =
-  stubs_align_tensors (CArray.of_list t tensors |> CArray.start) (List.length tensors)
+  stubs_align_tensors
+    (CArray.of_list gc_tensor tensors |> CArray.start)
+    (List.length tensors)
   |> to_tensor_list
 ;;
 
-let all self =
-  let out__ = CArray.make t 1 in
-  stubs_all (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let all_all_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_all_all_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let all self = stubs_all self |> with_tensor_gc
+let all_all_out ~out self = stubs_all_all_out out self |> with_tensor_gc
 
 let all_dim self ~dim ~keepdim =
-  let out__ = CArray.make t 1 in
-  stubs_all_dim (CArray.start out__) self (Int64.of_int dim) (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_all_dim self (Int64.of_int dim) (if keepdim then 1 else 0) |> with_tensor_gc
 ;;
 
 let all_out ~out self ~dim ~keepdim =
-  let out__ = CArray.make t 1 in
-  stubs_all_out
-    (CArray.start out__)
-    out
-    self
-    (Int64.of_int dim)
-    (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_all_out out self (Int64.of_int dim) (if keepdim then 1 else 0) |> with_tensor_gc
 ;;
 
 let allclose self other ~rtol ~atol ~equal_nan =
@@ -8508,77 +6200,53 @@ let allclose self other ~rtol ~atol ~equal_nan =
 ;;
 
 let alpha_dropout input ~p ~train =
-  let out__ = CArray.make t 1 in
-  stubs_alpha_dropout (CArray.start out__) input p (if train then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_alpha_dropout input p (if train then 1 else 0) |> with_tensor_gc
 ;;
 
 let alpha_dropout_ self ~p ~train =
-  let out__ = CArray.make t 1 in
-  stubs_alpha_dropout_ (CArray.start out__) self p (if train then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_alpha_dropout_ self p (if train then 1 else 0) |> with_tensor_gc
 ;;
 
 let amax self ~dim ~keepdim =
-  let out__ = CArray.make t 1 in
   stubs_amax
-    (CArray.start out__)
     self
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
     (List.length dim)
-    (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if keepdim then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let amax_out ~out self ~dim ~keepdim =
-  let out__ = CArray.make t 1 in
   stubs_amax_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
     (List.length dim)
-    (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if keepdim then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let amin self ~dim ~keepdim =
-  let out__ = CArray.make t 1 in
   stubs_amin
-    (CArray.start out__)
     self
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
     (List.length dim)
-    (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if keepdim then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let amin_out ~out self ~dim ~keepdim =
-  let out__ = CArray.make t 1 in
   stubs_amin_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
     (List.length dim)
-    (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if keepdim then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let aminmax self ~dim ~keepdim =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_aminmax
     (CArray.start out__)
     self
@@ -8589,15 +6257,13 @@ let aminmax self ~dim ~keepdim =
      | Some _ -> 0
      | None -> 1)
     (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let aminmax_out ~min ~max self ~dim ~keepdim =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_aminmax_out
     (CArray.start out__)
     min
@@ -8610,276 +6276,71 @@ let aminmax_out ~min ~max self ~dim ~keepdim =
      | Some _ -> 0
      | None -> 1)
     (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
-let angle self =
-  let out__ = CArray.make t 1 in
-  stubs_angle (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let angle_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_angle_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let any self =
-  let out__ = CArray.make t 1 in
-  stubs_any (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let any_all_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_any_all_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let angle self = stubs_angle self |> with_tensor_gc
+let angle_out ~out self = stubs_angle_out out self |> with_tensor_gc
+let any self = stubs_any self |> with_tensor_gc
+let any_all_out ~out self = stubs_any_all_out out self |> with_tensor_gc
 
 let any_dim self ~dim ~keepdim =
-  let out__ = CArray.make t 1 in
-  stubs_any_dim (CArray.start out__) self (Int64.of_int dim) (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_any_dim self (Int64.of_int dim) (if keepdim then 1 else 0) |> with_tensor_gc
 ;;
 
 let any_out ~out self ~dim ~keepdim =
-  let out__ = CArray.make t 1 in
-  stubs_any_out
-    (CArray.start out__)
-    out
-    self
-    (Int64.of_int dim)
-    (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_any_out out self (Int64.of_int dim) (if keepdim then 1 else 0) |> with_tensor_gc
 ;;
 
 let arange ~end_ ~options =
-  let out__ = CArray.make t 1 in
-  stubs_arange
-    (CArray.start out__)
-    end_
-    (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_arange end_ (Kind.packed_to_int (fst options)) (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let arange_start ~start ~end_ ~options =
-  let out__ = CArray.make t 1 in
   stubs_arange_start
-    (CArray.start out__)
     start
     end_
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let arange_start_step ~start ~end_ ~options =
-  let out__ = CArray.make t 1 in
   stubs_arange_start_step
-    (CArray.start out__)
     start
     end_
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let arccos self =
-  let out__ = CArray.make t 1 in
-  stubs_arccos (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let arccos_ self =
-  let out__ = CArray.make t 1 in
-  stubs_arccos_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let arccos_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_arccos_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let arccosh self =
-  let out__ = CArray.make t 1 in
-  stubs_arccosh (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let arccosh_ self =
-  let out__ = CArray.make t 1 in
-  stubs_arccosh_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let arccosh_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_arccosh_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let arcsin self =
-  let out__ = CArray.make t 1 in
-  stubs_arcsin (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let arcsin_ self =
-  let out__ = CArray.make t 1 in
-  stubs_arcsin_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let arcsin_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_arcsin_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let arcsinh self =
-  let out__ = CArray.make t 1 in
-  stubs_arcsinh (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let arcsinh_ self =
-  let out__ = CArray.make t 1 in
-  stubs_arcsinh_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let arcsinh_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_arcsinh_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let arctan self =
-  let out__ = CArray.make t 1 in
-  stubs_arctan (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let arctan2 self other =
-  let out__ = CArray.make t 1 in
-  stubs_arctan2 (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let arctan2_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_arctan2_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let arctan2_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_arctan2_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let arctan_ self =
-  let out__ = CArray.make t 1 in
-  stubs_arctan_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let arctan_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_arctan_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let arctanh self =
-  let out__ = CArray.make t 1 in
-  stubs_arctanh (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let arctanh_ self =
-  let out__ = CArray.make t 1 in
-  stubs_arctanh_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let arctanh_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_arctanh_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+    (Device.to_int (snd options))
+  |> with_tensor_gc
+;;
+
+let arccos self = stubs_arccos self |> with_tensor_gc
+let arccos_ self = stubs_arccos_ self |> with_tensor_gc
+let arccos_out ~out self = stubs_arccos_out out self |> with_tensor_gc
+let arccosh self = stubs_arccosh self |> with_tensor_gc
+let arccosh_ self = stubs_arccosh_ self |> with_tensor_gc
+let arccosh_out ~out self = stubs_arccosh_out out self |> with_tensor_gc
+let arcsin self = stubs_arcsin self |> with_tensor_gc
+let arcsin_ self = stubs_arcsin_ self |> with_tensor_gc
+let arcsin_out ~out self = stubs_arcsin_out out self |> with_tensor_gc
+let arcsinh self = stubs_arcsinh self |> with_tensor_gc
+let arcsinh_ self = stubs_arcsinh_ self |> with_tensor_gc
+let arcsinh_out ~out self = stubs_arcsinh_out out self |> with_tensor_gc
+let arctan self = stubs_arctan self |> with_tensor_gc
+let arctan2 self other = stubs_arctan2 self other |> with_tensor_gc
+let arctan2_ self other = stubs_arctan2_ self other |> with_tensor_gc
+let arctan2_out ~out self other = stubs_arctan2_out out self other |> with_tensor_gc
+let arctan_ self = stubs_arctan_ self |> with_tensor_gc
+let arctan_out ~out self = stubs_arctan_out out self |> with_tensor_gc
+let arctanh self = stubs_arctanh self |> with_tensor_gc
+let arctanh_ self = stubs_arctanh_ self |> with_tensor_gc
+let arctanh_out ~out self = stubs_arctanh_out out self |> with_tensor_gc
 
 let argmax self ~dim ~keepdim =
-  let out__ = CArray.make t 1 in
   stubs_argmax
-    (CArray.start out__)
     self
     (match dim with
      | None -> Int64.zero
@@ -8887,16 +6348,12 @@ let argmax self ~dim ~keepdim =
     (match dim with
      | Some _ -> 0
      | None -> 1)
-    (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if keepdim then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let argmax_out ~out self ~dim ~keepdim =
-  let out__ = CArray.make t 1 in
   stubs_argmax_out
-    (CArray.start out__)
     out
     self
     (match dim with
@@ -8905,16 +6362,12 @@ let argmax_out ~out self ~dim ~keepdim =
     (match dim with
      | Some _ -> 0
      | None -> 1)
-    (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if keepdim then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let argmin self ~dim ~keepdim =
-  let out__ = CArray.make t 1 in
   stubs_argmin
-    (CArray.start out__)
     self
     (match dim with
      | None -> Int64.zero
@@ -8922,16 +6375,12 @@ let argmin self ~dim ~keepdim =
     (match dim with
      | Some _ -> 0
      | None -> 1)
-    (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if keepdim then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let argmin_out ~out self ~dim ~keepdim =
-  let out__ = CArray.make t 1 in
   stubs_argmin_out
-    (CArray.start out__)
     out
     self
     (match dim with
@@ -8940,59 +6389,37 @@ let argmin_out ~out self ~dim ~keepdim =
     (match dim with
      | Some _ -> 0
      | None -> 1)
-    (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if keepdim then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let argsort self ~dim ~descending =
-  let out__ = CArray.make t 1 in
-  stubs_argsort (CArray.start out__) self (Int64.of_int dim) (if descending then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_argsort self (Int64.of_int dim) (if descending then 1 else 0) |> with_tensor_gc
 ;;
 
 let argsort_stable self ~stable ~dim ~descending =
-  let out__ = CArray.make t 1 in
   stubs_argsort_stable
-    (CArray.start out__)
     self
     (if stable then 1 else 0)
     (Int64.of_int dim)
-    (if descending then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if descending then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let argsort_stable_out ~out self ~stable ~dim ~descending =
-  let out__ = CArray.make t 1 in
   stubs_argsort_stable_out
-    (CArray.start out__)
     out
     self
     (if stable then 1 else 0)
     (Int64.of_int dim)
-    (if descending then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if descending then 1 else 0)
+  |> with_tensor_gc
 ;;
 
-let argwhere self =
-  let out__ = CArray.make t 1 in
-  stubs_argwhere (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let argwhere self = stubs_argwhere self |> with_tensor_gc
 
 let as_strided self ~size ~stride ~storage_offset =
-  let out__ = CArray.make t 1 in
   stubs_as_strided
-    (CArray.start out__)
     self
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
@@ -9003,16 +6430,12 @@ let as_strided self ~size ~stride ~storage_offset =
      | Some v -> Int64.of_int v)
     (match storage_offset with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let as_strided_ self ~size ~stride ~storage_offset =
-  let out__ = CArray.make t 1 in
   stubs_as_strided_
-    (CArray.start out__)
     self
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
@@ -9023,16 +6446,12 @@ let as_strided_ self ~size ~stride ~storage_offset =
      | Some v -> Int64.of_int v)
     (match storage_offset with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let as_strided_copy self ~size ~stride ~storage_offset =
-  let out__ = CArray.make t 1 in
   stubs_as_strided_copy
-    (CArray.start out__)
     self
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
@@ -9043,16 +6462,12 @@ let as_strided_copy self ~size ~stride ~storage_offset =
      | Some v -> Int64.of_int v)
     (match storage_offset with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let as_strided_copy_out ~out self ~size ~stride ~storage_offset =
-  let out__ = CArray.make t 1 in
   stubs_as_strided_copy_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
@@ -9064,16 +6479,12 @@ let as_strided_copy_out ~out self ~size ~stride ~storage_offset =
      | Some v -> Int64.of_int v)
     (match storage_offset with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let as_strided_scatter self ~src ~size ~stride ~storage_offset =
-  let out__ = CArray.make t 1 in
   stubs_as_strided_scatter
-    (CArray.start out__)
     self
     src
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
@@ -9085,16 +6496,12 @@ let as_strided_scatter self ~src ~size ~stride ~storage_offset =
      | Some v -> Int64.of_int v)
     (match storage_offset with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let as_strided_scatter_out ~out self ~src ~size ~stride ~storage_offset =
-  let out__ = CArray.make t 1 in
   stubs_as_strided_scatter_out
-    (CArray.start out__)
     out
     self
     src
@@ -9107,181 +6514,54 @@ let as_strided_scatter_out ~out self ~src ~size ~stride ~storage_offset =
      | Some v -> Int64.of_int v)
     (match storage_offset with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let asin self =
-  let out__ = CArray.make t 1 in
-  stubs_asin (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let asin_ self =
-  let out__ = CArray.make t 1 in
-  stubs_asin_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let asin_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_asin_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let asinh self =
-  let out__ = CArray.make t 1 in
-  stubs_asinh (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let asinh_ self =
-  let out__ = CArray.make t 1 in
-  stubs_asinh_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let asinh_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_asinh_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let atan self =
-  let out__ = CArray.make t 1 in
-  stubs_atan (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let atan2 self other =
-  let out__ = CArray.make t 1 in
-  stubs_atan2 (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let atan2_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_atan2_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let atan2_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_atan2_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let atan_ self =
-  let out__ = CArray.make t 1 in
-  stubs_atan_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let atan_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_atan_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let atanh self =
-  let out__ = CArray.make t 1 in
-  stubs_atanh (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let atanh_ self =
-  let out__ = CArray.make t 1 in
-  stubs_atanh_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let atanh_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_atanh_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let atleast_1d self =
-  let out__ = CArray.make t 1 in
-  stubs_atleast_1d (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+     | None -> 1)
+  |> with_tensor_gc
+;;
+
+let asin self = stubs_asin self |> with_tensor_gc
+let asin_ self = stubs_asin_ self |> with_tensor_gc
+let asin_out ~out self = stubs_asin_out out self |> with_tensor_gc
+let asinh self = stubs_asinh self |> with_tensor_gc
+let asinh_ self = stubs_asinh_ self |> with_tensor_gc
+let asinh_out ~out self = stubs_asinh_out out self |> with_tensor_gc
+let atan self = stubs_atan self |> with_tensor_gc
+let atan2 self other = stubs_atan2 self other |> with_tensor_gc
+let atan2_ self other = stubs_atan2_ self other |> with_tensor_gc
+let atan2_out ~out self other = stubs_atan2_out out self other |> with_tensor_gc
+let atan_ self = stubs_atan_ self |> with_tensor_gc
+let atan_out ~out self = stubs_atan_out out self |> with_tensor_gc
+let atanh self = stubs_atanh self |> with_tensor_gc
+let atanh_ self = stubs_atanh_ self |> with_tensor_gc
+let atanh_out ~out self = stubs_atanh_out out self |> with_tensor_gc
+let atleast_1d self = stubs_atleast_1d self |> with_tensor_gc
 
 let atleast_1d_sequence tensors =
   stubs_atleast_1d_sequence
-    (CArray.of_list t tensors |> CArray.start)
+    (CArray.of_list gc_tensor tensors |> CArray.start)
     (List.length tensors)
   |> to_tensor_list
 ;;
 
-let atleast_2d self =
-  let out__ = CArray.make t 1 in
-  stubs_atleast_2d (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let atleast_2d self = stubs_atleast_2d self |> with_tensor_gc
 
 let atleast_2d_sequence tensors =
   stubs_atleast_2d_sequence
-    (CArray.of_list t tensors |> CArray.start)
+    (CArray.of_list gc_tensor tensors |> CArray.start)
     (List.length tensors)
   |> to_tensor_list
 ;;
 
-let atleast_3d self =
-  let out__ = CArray.make t 1 in
-  stubs_atleast_3d (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let atleast_3d self = stubs_atleast_3d self |> with_tensor_gc
 
 let atleast_3d_sequence tensors =
   stubs_atleast_3d_sequence
-    (CArray.of_list t tensors |> CArray.start)
+    (CArray.of_list gc_tensor tensors |> CArray.start)
     (List.length tensors)
   |> to_tensor_list
 ;;
 
 let avg_pool1d self ~kernel_size ~stride ~padding ~ceil_mode ~count_include_pad =
-  let out__ = CArray.make t 1 in
   stubs_avg_pool1d
-    (CArray.start out__)
     self
     (List.map Int64.of_int kernel_size |> CArray.of_list int64_t |> CArray.start)
     (List.length kernel_size)
@@ -9290,10 +6570,8 @@ let avg_pool1d self ~kernel_size ~stride ~padding ~ceil_mode ~count_include_pad
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
     (List.length padding)
     (if ceil_mode then 1 else 0)
-    (if count_include_pad then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if count_include_pad then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let avg_pool2d
@@ -9305,9 +6583,7 @@ let avg_pool2d
   ~count_include_pad
   ~divisor_override
   =
-  let out__ = CArray.make t 1 in
   stubs_avg_pool2d
-    (CArray.start out__)
     self
     (List.map Int64.of_int kernel_size |> CArray.of_list int64_t |> CArray.start)
     (List.length kernel_size)
@@ -9322,10 +6598,8 @@ let avg_pool2d
      | Some v -> Int64.of_int v)
     (match divisor_override with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let avg_pool2d_backward
@@ -9338,9 +6612,7 @@ let avg_pool2d_backward
   ~count_include_pad
   ~divisor_override
   =
-  let out__ = CArray.make t 1 in
   stubs_avg_pool2d_backward
-    (CArray.start out__)
     grad_output
     self
     (List.map Int64.of_int kernel_size |> CArray.of_list int64_t |> CArray.start)
@@ -9356,10 +6628,8 @@ let avg_pool2d_backward
      | Some v -> Int64.of_int v)
     (match divisor_override with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let avg_pool2d_backward_grad_input
@@ -9373,9 +6643,7 @@ let avg_pool2d_backward_grad_input
   ~count_include_pad
   ~divisor_override
   =
-  let out__ = CArray.make t 1 in
   stubs_avg_pool2d_backward_grad_input
-    (CArray.start out__)
     grad_input
     grad_output
     self
@@ -9392,10 +6660,8 @@ let avg_pool2d_backward_grad_input
      | Some v -> Int64.of_int v)
     (match divisor_override with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let avg_pool2d_out
@@ -9408,9 +6674,7 @@ let avg_pool2d_out
   ~count_include_pad
   ~divisor_override
   =
-  let out__ = CArray.make t 1 in
   stubs_avg_pool2d_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int kernel_size |> CArray.of_list int64_t |> CArray.start)
@@ -9426,10 +6690,8 @@ let avg_pool2d_out
      | Some v -> Int64.of_int v)
     (match divisor_override with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let avg_pool3d
@@ -9441,9 +6703,7 @@ let avg_pool3d
   ~count_include_pad
   ~divisor_override
   =
-  let out__ = CArray.make t 1 in
   stubs_avg_pool3d
-    (CArray.start out__)
     self
     (List.map Int64.of_int kernel_size |> CArray.of_list int64_t |> CArray.start)
     (List.length kernel_size)
@@ -9458,10 +6718,8 @@ let avg_pool3d
      | Some v -> Int64.of_int v)
     (match divisor_override with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let avg_pool3d_backward
@@ -9474,9 +6732,7 @@ let avg_pool3d_backward
   ~count_include_pad
   ~divisor_override
   =
-  let out__ = CArray.make t 1 in
   stubs_avg_pool3d_backward
-    (CArray.start out__)
     grad_output
     self
     (List.map Int64.of_int kernel_size |> CArray.of_list int64_t |> CArray.start)
@@ -9492,10 +6748,8 @@ let avg_pool3d_backward
      | Some v -> Int64.of_int v)
     (match divisor_override with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let avg_pool3d_backward_grad_input
@@ -9509,9 +6763,7 @@ let avg_pool3d_backward_grad_input
   ~count_include_pad
   ~divisor_override
   =
-  let out__ = CArray.make t 1 in
   stubs_avg_pool3d_backward_grad_input
-    (CArray.start out__)
     grad_input
     grad_output
     self
@@ -9528,10 +6780,8 @@ let avg_pool3d_backward_grad_input
      | Some v -> Int64.of_int v)
     (match divisor_override with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let avg_pool3d_out
@@ -9544,9 +6794,7 @@ let avg_pool3d_out
   ~count_include_pad
   ~divisor_override
   =
-  let out__ = CArray.make t 1 in
   stubs_avg_pool3d_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int kernel_size |> CArray.of_list int64_t |> CArray.start)
@@ -9562,79 +6810,44 @@ let avg_pool3d_out
      | Some v -> Int64.of_int v)
     (match divisor_override with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let baddbmm self ~batch1 ~batch2 =
-  let out__ = CArray.make t 1 in
-  stubs_baddbmm (CArray.start out__) self batch1 batch2;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
-let baddbmm_ self ~batch1 ~batch2 =
-  let out__ = CArray.make t 1 in
-  stubs_baddbmm_ (CArray.start out__) self batch1 batch2;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let baddbmm self ~batch1 ~batch2 = stubs_baddbmm self batch1 batch2 |> with_tensor_gc
+let baddbmm_ self ~batch1 ~batch2 = stubs_baddbmm_ self batch1 batch2 |> with_tensor_gc
 
 let baddbmm_out ~out self ~batch1 ~batch2 =
-  let out__ = CArray.make t 1 in
-  stubs_baddbmm_out (CArray.start out__) out self batch1 batch2;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_baddbmm_out out self batch1 batch2 |> with_tensor_gc
 ;;
 
 let bartlett_window ~window_length ~options =
-  let out__ = CArray.make t 1 in
   stubs_bartlett_window
-    (CArray.start out__)
     (Int64.of_int window_length)
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let bartlett_window_out ~out ~window_length =
-  let out__ = CArray.make t 1 in
-  stubs_bartlett_window_out (CArray.start out__) out (Int64.of_int window_length);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_bartlett_window_out out (Int64.of_int window_length) |> with_tensor_gc
 ;;
 
 let bartlett_window_periodic ~window_length ~periodic ~options =
-  let out__ = CArray.make t 1 in
   stubs_bartlett_window_periodic
-    (CArray.start out__)
     (Int64.of_int window_length)
     (if periodic then 1 else 0)
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let bartlett_window_periodic_out ~out ~window_length ~periodic =
-  let out__ = CArray.make t 1 in
   stubs_bartlett_window_periodic_out
-    (CArray.start out__)
     out
     (Int64.of_int window_length)
-    (if periodic then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if periodic then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let batch_norm
@@ -9648,29 +6861,25 @@ let batch_norm
   ~eps
   ~cudnn_enabled
   =
-  let out__ = CArray.make t 1 in
   stubs_batch_norm
-    (CArray.start out__)
     input
     (match weight with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match running_mean with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match running_var with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (if training then 1 else 0)
     momentum
     eps
-    (if cudnn_enabled then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if cudnn_enabled then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let batch_norm_backward_elemt
@@ -9683,22 +6892,18 @@ let batch_norm_backward_elemt
   ~sum_dy_xmu
   ~count
   =
-  let out__ = CArray.make t 1 in
   stubs_batch_norm_backward_elemt
-    (CArray.start out__)
     grad_out
     input
     mean
     invstd
     (match weight with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     sum_dy
     sum_dy_xmu
-    count;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    count
+  |> with_tensor_gc
 ;;
 
 let batch_norm_backward_elemt_out
@@ -9712,9 +6917,7 @@ let batch_norm_backward_elemt_out
   ~sum_dy_xmu
   ~count
   =
-  let out__ = CArray.make t 1 in
   stubs_batch_norm_backward_elemt_out
-    (CArray.start out__)
     out
     grad_out
     input
@@ -9722,13 +6925,11 @@ let batch_norm_backward_elemt_out
     invstd
     (match weight with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     sum_dy
     sum_dy_xmu
-    count;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    count
+  |> with_tensor_gc
 ;;
 
 let batch_norm_backward_reduce
@@ -9741,7 +6942,7 @@ let batch_norm_backward_reduce
   ~weight_g
   ~bias_g
   =
-  let out__ = CArray.make t 4 in
+  let out__ = CArray.make raw_tensor 4 in
   stubs_batch_norm_backward_reduce
     (CArray.start out__)
     grad_out
@@ -9750,18 +6951,14 @@ let batch_norm_backward_reduce
     invstd
     (match weight with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (if input_g then 1 else 0)
     (if weight_g then 1 else 0)
     (if bias_g then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
-  let t3 = CArray.get out__ 3 in
-  Gc.finalise C.Tensor.free t3;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
+  let t3 = CArray.get out__ 3 |> with_tensor_gc in
   t0, t1, t2, t3
 ;;
 
@@ -9779,7 +6976,7 @@ let batch_norm_backward_reduce_out
   ~weight_g
   ~bias_g
   =
-  let out__ = CArray.make t 4 in
+  let out__ = CArray.make raw_tensor 4 in
   stubs_batch_norm_backward_reduce_out
     (CArray.start out__)
     out0
@@ -9792,58 +6989,46 @@ let batch_norm_backward_reduce_out
     invstd
     (match weight with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (if input_g then 1 else 0)
     (if weight_g then 1 else 0)
     (if bias_g then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
-  let t3 = CArray.get out__ 3 in
-  Gc.finalise C.Tensor.free t3;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
+  let t3 = CArray.get out__ 3 |> with_tensor_gc in
   t0, t1, t2, t3
 ;;
 
 let batch_norm_elemt input ~weight ~bias ~mean ~invstd ~eps =
-  let out__ = CArray.make t 1 in
   stubs_batch_norm_elemt
-    (CArray.start out__)
     input
     (match weight with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     mean
     invstd
-    eps;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    eps
+  |> with_tensor_gc
 ;;
 
 let batch_norm_elemt_out ~out input ~weight ~bias ~mean ~invstd ~eps =
-  let out__ = CArray.make t 1 in
   stubs_batch_norm_elemt_out
-    (CArray.start out__)
     out
     input
     (match weight with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     mean
     invstd
-    eps;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    eps
+  |> with_tensor_gc
 ;;
 
 let batch_norm_gather_stats
@@ -9856,7 +7041,7 @@ let batch_norm_gather_stats
   ~eps
   ~count
   =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_batch_norm_gather_stats
     (CArray.start out__)
     input
@@ -9864,17 +7049,15 @@ let batch_norm_gather_stats
     invstd
     (match running_mean with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match running_var with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     momentum
     eps
     (Int64.of_int count);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
@@ -9890,7 +7073,7 @@ let batch_norm_gather_stats_out
   ~eps
   ~count
   =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_batch_norm_gather_stats_out
     (CArray.start out__)
     out0
@@ -9900,17 +7083,15 @@ let batch_norm_gather_stats_out
     invstd
     (match running_mean with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match running_var with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     momentum
     eps
     (Int64.of_int count);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
@@ -9924,7 +7105,7 @@ let batch_norm_gather_stats_with_counts
   ~eps
   ~counts
   =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_batch_norm_gather_stats_with_counts
     (CArray.start out__)
     input
@@ -9932,17 +7113,15 @@ let batch_norm_gather_stats_with_counts
     invstd
     (match running_mean with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match running_var with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     momentum
     eps
     counts;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
@@ -9958,7 +7137,7 @@ let batch_norm_gather_stats_with_counts_out
   ~eps
   ~counts
   =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_batch_norm_gather_stats_with_counts_out
     (CArray.start out__)
     out0
@@ -9968,61 +7147,53 @@ let batch_norm_gather_stats_with_counts_out
     invstd
     (match running_mean with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match running_var with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     momentum
     eps
     counts;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let batch_norm_stats input ~eps =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_batch_norm_stats (CArray.start out__) input eps;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let batch_norm_stats_out ~out0 ~out1 input ~eps =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_batch_norm_stats_out (CArray.start out__) out0 out1 input eps;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let batch_norm_update_stats input ~running_mean ~running_var ~momentum =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_batch_norm_update_stats
     (CArray.start out__)
     input
     (match running_mean with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match running_var with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     momentum;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let batch_norm_update_stats_out ~out0 ~out1 input ~running_mean ~running_var ~momentum =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_batch_norm_update_stats_out
     (CArray.start out__)
     out0
@@ -10030,102 +7201,54 @@ let batch_norm_update_stats_out ~out0 ~out1 input ~running_mean ~running_var ~mo
     input
     (match running_mean with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match running_var with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     momentum;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
-let bernoulli self =
-  let out__ = CArray.make t 1 in
-  stubs_bernoulli (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let bernoulli_ self ~p =
-  let out__ = CArray.make t 1 in
-  stubs_bernoulli_ (CArray.start out__) self p;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let bernoulli_float_ self ~p =
-  let out__ = CArray.make t 1 in
-  stubs_bernoulli_float_ (CArray.start out__) self p;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let bernoulli_p self ~p =
-  let out__ = CArray.make t 1 in
-  stubs_bernoulli_p (CArray.start out__) self p;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let bernoulli_tensor self ~p =
-  let out__ = CArray.make t 1 in
-  stubs_bernoulli_tensor (CArray.start out__) self p;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let bernoulli self = stubs_bernoulli self |> with_tensor_gc
+let bernoulli_ self ~p = stubs_bernoulli_ self p |> with_tensor_gc
+let bernoulli_float_ self ~p = stubs_bernoulli_float_ self p |> with_tensor_gc
+let bernoulli_p self ~p = stubs_bernoulli_p self p |> with_tensor_gc
+let bernoulli_tensor self ~p = stubs_bernoulli_tensor self p |> with_tensor_gc
 
 let bilinear ~input1 ~input2 ~weight ~bias =
-  let out__ = CArray.make t 1 in
   stubs_bilinear
-    (CArray.start out__)
     input1
     input2
     weight
     (match bias with
      | Some v -> v
-     | None -> null);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+  |> with_tensor_gc
 ;;
 
 let binary_cross_entropy self ~target ~weight ~reduction =
-  let out__ = CArray.make t 1 in
   stubs_binary_cross_entropy
-    (CArray.start out__)
     self
     target
     (match weight with
      | Some v -> v
-     | None -> null)
-    (Reduction.to_int reduction |> Int64.of_int);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+    (Reduction.to_int reduction |> Int64.of_int)
+  |> with_tensor_gc
 ;;
 
 let binary_cross_entropy_backward ~grad_output self ~target ~weight ~reduction =
-  let out__ = CArray.make t 1 in
   stubs_binary_cross_entropy_backward
-    (CArray.start out__)
     grad_output
     self
     target
     (match weight with
      | Some v -> v
-     | None -> null)
-    (Reduction.to_int reduction |> Int64.of_int);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+    (Reduction.to_int reduction |> Int64.of_int)
+  |> with_tensor_gc
 ;;
 
 let binary_cross_entropy_backward_grad_input
@@ -10136,617 +7259,313 @@ let binary_cross_entropy_backward_grad_input
   ~weight
   ~reduction
   =
-  let out__ = CArray.make t 1 in
   stubs_binary_cross_entropy_backward_grad_input
-    (CArray.start out__)
     grad_input
     grad_output
     self
     target
     (match weight with
      | Some v -> v
-     | None -> null)
-    (Reduction.to_int reduction |> Int64.of_int);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+    (Reduction.to_int reduction |> Int64.of_int)
+  |> with_tensor_gc
 ;;
 
 let binary_cross_entropy_out ~out self ~target ~weight ~reduction =
-  let out__ = CArray.make t 1 in
   stubs_binary_cross_entropy_out
-    (CArray.start out__)
     out
     self
     target
     (match weight with
      | Some v -> v
-     | None -> null)
-    (Reduction.to_int reduction |> Int64.of_int);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+    (Reduction.to_int reduction |> Int64.of_int)
+  |> with_tensor_gc
 ;;
 
 let binary_cross_entropy_with_logits self ~target ~weight ~pos_weight ~reduction =
-  let out__ = CArray.make t 1 in
   stubs_binary_cross_entropy_with_logits
-    (CArray.start out__)
     self
     target
     (match weight with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match pos_weight with
      | Some v -> v
-     | None -> null)
-    (Reduction.to_int reduction |> Int64.of_int);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+    (Reduction.to_int reduction |> Int64.of_int)
+  |> with_tensor_gc
 ;;
 
 let binary_cross_entropy_with_logits_out ~out self ~target ~weight ~pos_weight ~reduction =
-  let out__ = CArray.make t 1 in
   stubs_binary_cross_entropy_with_logits_out
-    (CArray.start out__)
     out
     self
     target
     (match weight with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match pos_weight with
      | Some v -> v
-     | None -> null)
-    (Reduction.to_int reduction |> Int64.of_int);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+    (Reduction.to_int reduction |> Int64.of_int)
+  |> with_tensor_gc
 ;;
 
 let bincount self ~weights ~minlength =
-  let out__ = CArray.make t 1 in
   stubs_bincount
-    (CArray.start out__)
     self
     (match weights with
      | Some v -> v
-     | None -> null)
-    (Int64.of_int minlength);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+    (Int64.of_int minlength)
+  |> with_tensor_gc
 ;;
 
 let bincount_out ~out self ~weights ~minlength =
-  let out__ = CArray.make t 1 in
   stubs_bincount_out
-    (CArray.start out__)
     out
     self
     (match weights with
      | Some v -> v
-     | None -> null)
-    (Int64.of_int minlength);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let binomial ~count ~prob =
-  let out__ = CArray.make t 1 in
-  stubs_binomial (CArray.start out__) count prob;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let binomial_out ~out ~count ~prob =
-  let out__ = CArray.make t 1 in
-  stubs_binomial_out (CArray.start out__) out count prob;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let bitwise_and self other =
-  let out__ = CArray.make t 1 in
-  stubs_bitwise_and (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+    (Int64.of_int minlength)
+  |> with_tensor_gc
 ;;
 
-let bitwise_and_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_bitwise_and_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let binomial ~count ~prob = stubs_binomial count prob |> with_tensor_gc
+let binomial_out ~out ~count ~prob = stubs_binomial_out out count prob |> with_tensor_gc
+let bitwise_and self other = stubs_bitwise_and self other |> with_tensor_gc
+let bitwise_and_ self other = stubs_bitwise_and_ self other |> with_tensor_gc
 
 let bitwise_and_scalar_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_bitwise_and_scalar_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_bitwise_and_scalar_out out self other |> with_tensor_gc
 ;;
 
 let bitwise_and_scalar_tensor self other =
-  let out__ = CArray.make t 1 in
-  stubs_bitwise_and_scalar_tensor (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_bitwise_and_scalar_tensor self other |> with_tensor_gc
 ;;
 
 let bitwise_and_scalar_tensor_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_bitwise_and_scalar_tensor_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_bitwise_and_scalar_tensor_out out self other |> with_tensor_gc
 ;;
 
-let bitwise_and_tensor self other =
-  let out__ = CArray.make t 1 in
-  stubs_bitwise_and_tensor (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let bitwise_and_tensor self other = stubs_bitwise_and_tensor self other |> with_tensor_gc
 
 let bitwise_and_tensor_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_bitwise_and_tensor_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_bitwise_and_tensor_ self other |> with_tensor_gc
 ;;
 
 let bitwise_and_tensor_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_bitwise_and_tensor_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_bitwise_and_tensor_out out self other |> with_tensor_gc
 ;;
 
-let bitwise_left_shift self other =
-  let out__ = CArray.make t 1 in
-  stubs_bitwise_left_shift (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let bitwise_left_shift self other = stubs_bitwise_left_shift self other |> with_tensor_gc
 
 let bitwise_left_shift_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_bitwise_left_shift_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_bitwise_left_shift_ self other |> with_tensor_gc
 ;;
 
 let bitwise_left_shift_scalar_tensor self other =
-  let out__ = CArray.make t 1 in
-  stubs_bitwise_left_shift_scalar_tensor (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_bitwise_left_shift_scalar_tensor self other |> with_tensor_gc
 ;;
 
 let bitwise_left_shift_scalar_tensor_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_bitwise_left_shift_scalar_tensor_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_bitwise_left_shift_scalar_tensor_out out self other |> with_tensor_gc
 ;;
 
 let bitwise_left_shift_tensor_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_bitwise_left_shift_tensor_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_bitwise_left_shift_tensor_out out self other |> with_tensor_gc
 ;;
 
 let bitwise_left_shift_tensor_scalar self other =
-  let out__ = CArray.make t 1 in
-  stubs_bitwise_left_shift_tensor_scalar (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_bitwise_left_shift_tensor_scalar self other |> with_tensor_gc
 ;;
 
 let bitwise_left_shift_tensor_scalar_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_bitwise_left_shift_tensor_scalar_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_bitwise_left_shift_tensor_scalar_ self other |> with_tensor_gc
 ;;
 
 let bitwise_left_shift_tensor_scalar_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_bitwise_left_shift_tensor_scalar_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let bitwise_not self =
-  let out__ = CArray.make t 1 in
-  stubs_bitwise_not (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_bitwise_left_shift_tensor_scalar_out out self other |> with_tensor_gc
 ;;
 
-let bitwise_not_ self =
-  let out__ = CArray.make t 1 in
-  stubs_bitwise_not_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let bitwise_not_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_bitwise_not_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let bitwise_or self other =
-  let out__ = CArray.make t 1 in
-  stubs_bitwise_or (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let bitwise_or_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_bitwise_or_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let bitwise_not self = stubs_bitwise_not self |> with_tensor_gc
+let bitwise_not_ self = stubs_bitwise_not_ self |> with_tensor_gc
+let bitwise_not_out ~out self = stubs_bitwise_not_out out self |> with_tensor_gc
+let bitwise_or self other = stubs_bitwise_or self other |> with_tensor_gc
+let bitwise_or_ self other = stubs_bitwise_or_ self other |> with_tensor_gc
 
 let bitwise_or_scalar_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_bitwise_or_scalar_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_bitwise_or_scalar_out out self other |> with_tensor_gc
 ;;
 
 let bitwise_or_scalar_tensor self other =
-  let out__ = CArray.make t 1 in
-  stubs_bitwise_or_scalar_tensor (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_bitwise_or_scalar_tensor self other |> with_tensor_gc
 ;;
 
 let bitwise_or_scalar_tensor_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_bitwise_or_scalar_tensor_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let bitwise_or_tensor self other =
-  let out__ = CArray.make t 1 in
-  stubs_bitwise_or_tensor (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_bitwise_or_scalar_tensor_out out self other |> with_tensor_gc
 ;;
 
-let bitwise_or_tensor_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_bitwise_or_tensor_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let bitwise_or_tensor self other = stubs_bitwise_or_tensor self other |> with_tensor_gc
+let bitwise_or_tensor_ self other = stubs_bitwise_or_tensor_ self other |> with_tensor_gc
 
 let bitwise_or_tensor_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_bitwise_or_tensor_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_bitwise_or_tensor_out out self other |> with_tensor_gc
 ;;
 
 let bitwise_right_shift self other =
-  let out__ = CArray.make t 1 in
-  stubs_bitwise_right_shift (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_bitwise_right_shift self other |> with_tensor_gc
 ;;
 
 let bitwise_right_shift_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_bitwise_right_shift_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_bitwise_right_shift_ self other |> with_tensor_gc
 ;;
 
 let bitwise_right_shift_scalar_tensor self other =
-  let out__ = CArray.make t 1 in
-  stubs_bitwise_right_shift_scalar_tensor (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_bitwise_right_shift_scalar_tensor self other |> with_tensor_gc
 ;;
 
 let bitwise_right_shift_scalar_tensor_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_bitwise_right_shift_scalar_tensor_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_bitwise_right_shift_scalar_tensor_out out self other |> with_tensor_gc
 ;;
 
 let bitwise_right_shift_tensor_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_bitwise_right_shift_tensor_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_bitwise_right_shift_tensor_out out self other |> with_tensor_gc
 ;;
 
 let bitwise_right_shift_tensor_scalar self other =
-  let out__ = CArray.make t 1 in
-  stubs_bitwise_right_shift_tensor_scalar (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_bitwise_right_shift_tensor_scalar self other |> with_tensor_gc
 ;;
 
 let bitwise_right_shift_tensor_scalar_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_bitwise_right_shift_tensor_scalar_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_bitwise_right_shift_tensor_scalar_ self other |> with_tensor_gc
 ;;
 
 let bitwise_right_shift_tensor_scalar_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_bitwise_right_shift_tensor_scalar_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let bitwise_xor self other =
-  let out__ = CArray.make t 1 in
-  stubs_bitwise_xor (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_bitwise_right_shift_tensor_scalar_out out self other |> with_tensor_gc
 ;;
 
-let bitwise_xor_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_bitwise_xor_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let bitwise_xor self other = stubs_bitwise_xor self other |> with_tensor_gc
+let bitwise_xor_ self other = stubs_bitwise_xor_ self other |> with_tensor_gc
 
 let bitwise_xor_scalar_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_bitwise_xor_scalar_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_bitwise_xor_scalar_out out self other |> with_tensor_gc
 ;;
 
 let bitwise_xor_scalar_tensor self other =
-  let out__ = CArray.make t 1 in
-  stubs_bitwise_xor_scalar_tensor (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_bitwise_xor_scalar_tensor self other |> with_tensor_gc
 ;;
 
 let bitwise_xor_scalar_tensor_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_bitwise_xor_scalar_tensor_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_bitwise_xor_scalar_tensor_out out self other |> with_tensor_gc
 ;;
 
-let bitwise_xor_tensor self other =
-  let out__ = CArray.make t 1 in
-  stubs_bitwise_xor_tensor (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let bitwise_xor_tensor self other = stubs_bitwise_xor_tensor self other |> with_tensor_gc
 
 let bitwise_xor_tensor_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_bitwise_xor_tensor_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_bitwise_xor_tensor_ self other |> with_tensor_gc
 ;;
 
 let bitwise_xor_tensor_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_bitwise_xor_tensor_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_bitwise_xor_tensor_out out self other |> with_tensor_gc
 ;;
 
 let blackman_window ~window_length ~options =
-  let out__ = CArray.make t 1 in
   stubs_blackman_window
-    (CArray.start out__)
     (Int64.of_int window_length)
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let blackman_window_out ~out ~window_length =
-  let out__ = CArray.make t 1 in
-  stubs_blackman_window_out (CArray.start out__) out (Int64.of_int window_length);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_blackman_window_out out (Int64.of_int window_length) |> with_tensor_gc
 ;;
 
 let blackman_window_periodic ~window_length ~periodic ~options =
-  let out__ = CArray.make t 1 in
   stubs_blackman_window_periodic
-    (CArray.start out__)
     (Int64.of_int window_length)
     (if periodic then 1 else 0)
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let blackman_window_periodic_out ~out ~window_length ~periodic =
-  let out__ = CArray.make t 1 in
   stubs_blackman_window_periodic_out
-    (CArray.start out__)
     out
     (Int64.of_int window_length)
-    (if periodic then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if periodic then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let block_diag tensors =
-  let out__ = CArray.make t 1 in
   stubs_block_diag
-    (CArray.start out__)
-    (CArray.of_list t tensors |> CArray.start)
-    (List.length tensors);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (CArray.of_list gc_tensor tensors |> CArray.start)
+    (List.length tensors)
+  |> with_tensor_gc
 ;;
 
 let block_diag_out ~out tensors =
-  let out__ = CArray.make t 1 in
   stubs_block_diag_out
-    (CArray.start out__)
     out
-    (CArray.of_list t tensors |> CArray.start)
-    (List.length tensors);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let bmm self ~mat2 =
-  let out__ = CArray.make t 1 in
-  stubs_bmm (CArray.start out__) self mat2;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (CArray.of_list gc_tensor tensors |> CArray.start)
+    (List.length tensors)
+  |> with_tensor_gc
 ;;
 
-let bmm_out ~out self ~mat2 =
-  let out__ = CArray.make t 1 in
-  stubs_bmm_out (CArray.start out__) out self mat2;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let bmm self ~mat2 = stubs_bmm self mat2 |> with_tensor_gc
+let bmm_out ~out self ~mat2 = stubs_bmm_out out self mat2 |> with_tensor_gc
 
 let broadcast_tensors tensors =
-  stubs_broadcast_tensors (CArray.of_list t tensors |> CArray.start) (List.length tensors)
+  stubs_broadcast_tensors
+    (CArray.of_list gc_tensor tensors |> CArray.start)
+    (List.length tensors)
   |> to_tensor_list
 ;;
 
 let broadcast_to self ~size =
-  let out__ = CArray.make t 1 in
   stubs_broadcast_to
-    (CArray.start out__)
     self
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
-    (List.length size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length size)
+  |> with_tensor_gc
 ;;
 
 let bucketize self ~boundaries ~out_int32 ~right =
-  let out__ = CArray.make t 1 in
-  stubs_bucketize
-    (CArray.start out__)
-    self
-    boundaries
-    (if out_int32 then 1 else 0)
-    (if right then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_bucketize self boundaries (if out_int32 then 1 else 0) (if right then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let bucketize_scalar self ~boundaries ~out_int32 ~right =
-  let out__ = CArray.make t 1 in
   stubs_bucketize_scalar
-    (CArray.start out__)
     self
     boundaries
     (if out_int32 then 1 else 0)
-    (if right then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if right then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let bucketize_scalar_out ~out self ~boundaries ~out_int32 ~right =
-  let out__ = CArray.make t 1 in
   stubs_bucketize_scalar_out
-    (CArray.start out__)
     out
     self
     boundaries
     (if out_int32 then 1 else 0)
-    (if right then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if right then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let bucketize_tensor_out ~out self ~boundaries ~out_int32 ~right =
-  let out__ = CArray.make t 1 in
   stubs_bucketize_tensor_out
-    (CArray.start out__)
     out
     self
     boundaries
     (if out_int32 then 1 else 0)
-    (if right then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if right then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let can_cast ~from ~to_ =
@@ -10754,93 +7573,45 @@ let can_cast ~from ~to_ =
 ;;
 
 let cartesian_prod tensors =
-  let out__ = CArray.make t 1 in
   stubs_cartesian_prod
-    (CArray.start out__)
-    (CArray.of_list t tensors |> CArray.start)
-    (List.length tensors);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (CArray.of_list gc_tensor tensors |> CArray.start)
+    (List.length tensors)
+  |> with_tensor_gc
 ;;
 
 let cat tensors ~dim =
-  let out__ = CArray.make t 1 in
   stubs_cat
-    (CArray.start out__)
-    (CArray.of_list t tensors |> CArray.start)
+    (CArray.of_list gc_tensor tensors |> CArray.start)
     (List.length tensors)
-    (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int dim)
+  |> with_tensor_gc
 ;;
 
 let cat_out ~out tensors ~dim =
-  let out__ = CArray.make t 1 in
   stubs_cat_out
-    (CArray.start out__)
     out
-    (CArray.of_list t tensors |> CArray.start)
+    (CArray.of_list gc_tensor tensors |> CArray.start)
     (List.length tensors)
-    (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let cauchy self ~median ~sigma =
-  let out__ = CArray.make t 1 in
-  stubs_cauchy (CArray.start out__) self median sigma;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int dim)
+  |> with_tensor_gc
 ;;
 
-let cauchy_ self ~median ~sigma =
-  let out__ = CArray.make t 1 in
-  stubs_cauchy_ (CArray.start out__) self median sigma;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let cauchy self ~median ~sigma = stubs_cauchy self median sigma |> with_tensor_gc
+let cauchy_ self ~median ~sigma = stubs_cauchy_ self median sigma |> with_tensor_gc
 
 let cauchy_out ~out self ~median ~sigma =
-  let out__ = CArray.make t 1 in
-  stubs_cauchy_out (CArray.start out__) out self median sigma;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let ccol_indices self =
-  let out__ = CArray.make t 1 in
-  stubs_ccol_indices (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_cauchy_out out self median sigma |> with_tensor_gc
 ;;
 
-let ccol_indices_copy self =
-  let out__ = CArray.make t 1 in
-  stubs_ccol_indices_copy (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let ccol_indices self = stubs_ccol_indices self |> with_tensor_gc
+let ccol_indices_copy self = stubs_ccol_indices_copy self |> with_tensor_gc
 
 let ccol_indices_copy_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_ccol_indices_copy_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_ccol_indices_copy_out out self |> with_tensor_gc
 ;;
 
 let cdist ~x1 ~x2 ~p ~compute_mode =
-  let out__ = CArray.make t 1 in
   stubs_cdist
-    (CArray.start out__)
     x1
     x2
     p
@@ -10849,157 +7620,66 @@ let cdist ~x1 ~x2 ~p ~compute_mode =
      | Some v -> Int64.of_int v)
     (match compute_mode with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let ceil self =
-  let out__ = CArray.make t 1 in
-  stubs_ceil (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let ceil_ self =
-  let out__ = CArray.make t 1 in
-  stubs_ceil_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let ceil_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_ceil_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let celu self =
-  let out__ = CArray.make t 1 in
-  stubs_celu (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let celu_ self =
-  let out__ = CArray.make t 1 in
-  stubs_celu_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
-let celu_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_celu_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let ceil self = stubs_ceil self |> with_tensor_gc
+let ceil_ self = stubs_ceil_ self |> with_tensor_gc
+let ceil_out ~out self = stubs_ceil_out out self |> with_tensor_gc
+let celu self = stubs_celu self |> with_tensor_gc
+let celu_ self = stubs_celu_ self |> with_tensor_gc
+let celu_out ~out self = stubs_celu_out out self |> with_tensor_gc
 
 let chain_matmul ~matrices =
-  let out__ = CArray.make t 1 in
   stubs_chain_matmul
-    (CArray.start out__)
-    (CArray.of_list t matrices |> CArray.start)
-    (List.length matrices);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (CArray.of_list gc_tensor matrices |> CArray.start)
+    (List.length matrices)
+  |> with_tensor_gc
 ;;
 
 let chain_matmul_out ~out ~matrices =
-  let out__ = CArray.make t 1 in
   stubs_chain_matmul_out
-    (CArray.start out__)
     out
-    (CArray.of_list t matrices |> CArray.start)
-    (List.length matrices);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (CArray.of_list gc_tensor matrices |> CArray.start)
+    (List.length matrices)
+  |> with_tensor_gc
 ;;
 
-let chalf self =
-  let out__ = CArray.make t 1 in
-  stubs_chalf (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let chalf self = stubs_chalf self |> with_tensor_gc
 
 let channel_shuffle self ~groups =
-  let out__ = CArray.make t 1 in
-  stubs_channel_shuffle (CArray.start out__) self (Int64.of_int groups);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_channel_shuffle self (Int64.of_int groups) |> with_tensor_gc
 ;;
 
 let channel_shuffle_out ~out self ~groups =
-  let out__ = CArray.make t 1 in
-  stubs_channel_shuffle_out (CArray.start out__) out self (Int64.of_int groups);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_channel_shuffle_out out self (Int64.of_int groups) |> with_tensor_gc
 ;;
 
-let cholesky self ~upper =
-  let out__ = CArray.make t 1 in
-  stubs_cholesky (CArray.start out__) self (if upper then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let cholesky self ~upper = stubs_cholesky self (if upper then 1 else 0) |> with_tensor_gc
 
 let cholesky_inverse self ~upper =
-  let out__ = CArray.make t 1 in
-  stubs_cholesky_inverse (CArray.start out__) self (if upper then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_cholesky_inverse self (if upper then 1 else 0) |> with_tensor_gc
 ;;
 
 let cholesky_inverse_out ~out self ~upper =
-  let out__ = CArray.make t 1 in
-  stubs_cholesky_inverse_out (CArray.start out__) out self (if upper then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_cholesky_inverse_out out self (if upper then 1 else 0) |> with_tensor_gc
 ;;
 
 let cholesky_out ~out self ~upper =
-  let out__ = CArray.make t 1 in
-  stubs_cholesky_out (CArray.start out__) out self (if upper then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_cholesky_out out self (if upper then 1 else 0) |> with_tensor_gc
 ;;
 
 let cholesky_solve self ~input2 ~upper =
-  let out__ = CArray.make t 1 in
-  stubs_cholesky_solve (CArray.start out__) self input2 (if upper then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_cholesky_solve self input2 (if upper then 1 else 0) |> with_tensor_gc
 ;;
 
 let cholesky_solve_out ~out self ~input2 ~upper =
-  let out__ = CArray.make t 1 in
-  stubs_cholesky_solve_out (CArray.start out__) out self input2 (if upper then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_cholesky_solve_out out self input2 (if upper then 1 else 0) |> with_tensor_gc
 ;;
 
 let choose_qparams_optimized input ~numel ~n_bins ~ratio ~bit_width =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_choose_qparams_optimized
     (CArray.start out__)
     input
@@ -11007,10 +7687,8 @@ let choose_qparams_optimized input ~numel ~n_bins ~ratio ~bit_width =
     (Int64.of_int n_bins)
     ratio
     (Int64.of_int bit_width);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
@@ -11018,276 +7696,114 @@ let chunk self ~chunks ~dim =
   stubs_chunk self (Int64.of_int chunks) (Int64.of_int dim) |> to_tensor_list
 ;;
 
-let clamp self ~min ~max =
-  let out__ = CArray.make t 1 in
-  stubs_clamp (CArray.start out__) self min max;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let clamp_ self ~min ~max =
-  let out__ = CArray.make t 1 in
-  stubs_clamp_ (CArray.start out__) self min max;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let clamp_max self ~max =
-  let out__ = CArray.make t 1 in
-  stubs_clamp_max (CArray.start out__) self max;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let clamp_max_ self ~max =
-  let out__ = CArray.make t 1 in
-  stubs_clamp_max_ (CArray.start out__) self max;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let clamp_max_out ~out self ~max =
-  let out__ = CArray.make t 1 in
-  stubs_clamp_max_out (CArray.start out__) out self max;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let clamp_max_tensor self ~max =
-  let out__ = CArray.make t 1 in
-  stubs_clamp_max_tensor (CArray.start out__) self max;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let clamp_max_tensor_ self ~max =
-  let out__ = CArray.make t 1 in
-  stubs_clamp_max_tensor_ (CArray.start out__) self max;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let clamp self ~min ~max = stubs_clamp self min max |> with_tensor_gc
+let clamp_ self ~min ~max = stubs_clamp_ self min max |> with_tensor_gc
+let clamp_max self ~max = stubs_clamp_max self max |> with_tensor_gc
+let clamp_max_ self ~max = stubs_clamp_max_ self max |> with_tensor_gc
+let clamp_max_out ~out self ~max = stubs_clamp_max_out out self max |> with_tensor_gc
+let clamp_max_tensor self ~max = stubs_clamp_max_tensor self max |> with_tensor_gc
+let clamp_max_tensor_ self ~max = stubs_clamp_max_tensor_ self max |> with_tensor_gc
 
 let clamp_max_tensor_out ~out self ~max =
-  let out__ = CArray.make t 1 in
-  stubs_clamp_max_tensor_out (CArray.start out__) out self max;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let clamp_min self ~min =
-  let out__ = CArray.make t 1 in
-  stubs_clamp_min (CArray.start out__) self min;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let clamp_min_ self ~min =
-  let out__ = CArray.make t 1 in
-  stubs_clamp_min_ (CArray.start out__) self min;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let clamp_min_out ~out self ~min =
-  let out__ = CArray.make t 1 in
-  stubs_clamp_min_out (CArray.start out__) out self min;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let clamp_min_tensor self ~min =
-  let out__ = CArray.make t 1 in
-  stubs_clamp_min_tensor (CArray.start out__) self min;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_clamp_max_tensor_out out self max |> with_tensor_gc
 ;;
 
-let clamp_min_tensor_ self ~min =
-  let out__ = CArray.make t 1 in
-  stubs_clamp_min_tensor_ (CArray.start out__) self min;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let clamp_min self ~min = stubs_clamp_min self min |> with_tensor_gc
+let clamp_min_ self ~min = stubs_clamp_min_ self min |> with_tensor_gc
+let clamp_min_out ~out self ~min = stubs_clamp_min_out out self min |> with_tensor_gc
+let clamp_min_tensor self ~min = stubs_clamp_min_tensor self min |> with_tensor_gc
+let clamp_min_tensor_ self ~min = stubs_clamp_min_tensor_ self min |> with_tensor_gc
 
 let clamp_min_tensor_out ~out self ~min =
-  let out__ = CArray.make t 1 in
-  stubs_clamp_min_tensor_out (CArray.start out__) out self min;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_clamp_min_tensor_out out self min |> with_tensor_gc
 ;;
 
-let clamp_out ~out self ~min ~max =
-  let out__ = CArray.make t 1 in
-  stubs_clamp_out (CArray.start out__) out self min max;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let clamp_out ~out self ~min ~max = stubs_clamp_out out self min max |> with_tensor_gc
 
 let clamp_tensor self ~min ~max =
-  let out__ = CArray.make t 1 in
   stubs_clamp_tensor
-    (CArray.start out__)
     self
     (match min with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match max with
      | Some v -> v
-     | None -> null);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+  |> with_tensor_gc
 ;;
 
 let clamp_tensor_ self ~min ~max =
-  let out__ = CArray.make t 1 in
   stubs_clamp_tensor_
-    (CArray.start out__)
     self
     (match min with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match max with
      | Some v -> v
-     | None -> null);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+  |> with_tensor_gc
 ;;
 
 let clamp_tensor_out ~out self ~min ~max =
-  let out__ = CArray.make t 1 in
   stubs_clamp_tensor_out
-    (CArray.start out__)
     out
     self
     (match min with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match max with
      | Some v -> v
-     | None -> null);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let clip self ~min ~max =
-  let out__ = CArray.make t 1 in
-  stubs_clip (CArray.start out__) self min max;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+  |> with_tensor_gc
 ;;
 
-let clip_ self ~min ~max =
-  let out__ = CArray.make t 1 in
-  stubs_clip_ (CArray.start out__) self min max;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let clip_out ~out self ~min ~max =
-  let out__ = CArray.make t 1 in
-  stubs_clip_out (CArray.start out__) out self min max;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let clip self ~min ~max = stubs_clip self min max |> with_tensor_gc
+let clip_ self ~min ~max = stubs_clip_ self min max |> with_tensor_gc
+let clip_out ~out self ~min ~max = stubs_clip_out out self min max |> with_tensor_gc
 
 let clip_tensor self ~min ~max =
-  let out__ = CArray.make t 1 in
   stubs_clip_tensor
-    (CArray.start out__)
     self
     (match min with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match max with
      | Some v -> v
-     | None -> null);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+  |> with_tensor_gc
 ;;
 
 let clip_tensor_ self ~min ~max =
-  let out__ = CArray.make t 1 in
   stubs_clip_tensor_
-    (CArray.start out__)
     self
     (match min with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match max with
      | Some v -> v
-     | None -> null);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+  |> with_tensor_gc
 ;;
 
 let clip_tensor_out ~out self ~min ~max =
-  let out__ = CArray.make t 1 in
   stubs_clip_tensor_out
-    (CArray.start out__)
     out
     self
     (match min with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match max with
      | Some v -> v
-     | None -> null);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let clone self =
-  let out__ = CArray.make t 1 in
-  stubs_clone (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+  |> with_tensor_gc
 ;;
 
-let clone_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_clone_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let coalesce self =
-  let out__ = CArray.make t 1 in
-  stubs_coalesce (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let clone self = stubs_clone self |> with_tensor_gc
+let clone_out ~out self = stubs_clone_out out self |> with_tensor_gc
+let coalesce self = stubs_coalesce self |> with_tensor_gc
 
 let col2im self ~output_size ~kernel_size ~dilation ~padding ~stride =
-  let out__ = CArray.make t 1 in
   stubs_col2im
-    (CArray.start out__)
     self
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
     (List.length output_size)
@@ -11298,16 +7814,12 @@ let col2im self ~output_size ~kernel_size ~dilation ~padding ~stride =
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
     (List.length padding)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
-    (List.length stride);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length stride)
+  |> with_tensor_gc
 ;;
 
 let col2im_out ~out self ~output_size ~kernel_size ~dilation ~padding ~stride =
-  let out__ = CArray.make t 1 in
   stubs_col2im_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
@@ -11319,351 +7831,214 @@ let col2im_out ~out self ~output_size ~kernel_size ~dilation ~padding ~stride =
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
     (List.length padding)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
-    (List.length stride);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let col_indices self =
-  let out__ = CArray.make t 1 in
-  stubs_col_indices (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let col_indices_copy self =
-  let out__ = CArray.make t 1 in
-  stubs_col_indices_copy (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length stride)
+  |> with_tensor_gc
 ;;
 
-let col_indices_copy_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_col_indices_copy_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let col_indices self = stubs_col_indices self |> with_tensor_gc
+let col_indices_copy self = stubs_col_indices_copy self |> with_tensor_gc
+let col_indices_copy_out ~out self = stubs_col_indices_copy_out out self |> with_tensor_gc
 
 let column_stack tensors =
-  let out__ = CArray.make t 1 in
   stubs_column_stack
-    (CArray.start out__)
-    (CArray.of_list t tensors |> CArray.start)
-    (List.length tensors);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (CArray.of_list gc_tensor tensors |> CArray.start)
+    (List.length tensors)
+  |> with_tensor_gc
 ;;
 
 let column_stack_out ~out tensors =
-  let out__ = CArray.make t 1 in
   stubs_column_stack_out
-    (CArray.start out__)
     out
-    (CArray.of_list t tensors |> CArray.start)
-    (List.length tensors);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (CArray.of_list gc_tensor tensors |> CArray.start)
+    (List.length tensors)
+  |> with_tensor_gc
 ;;
 
 let combinations self ~r ~with_replacement =
-  let out__ = CArray.make t 1 in
-  stubs_combinations
-    (CArray.start out__)
-    self
-    (Int64.of_int r)
-    (if with_replacement then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let complex ~real ~imag =
-  let out__ = CArray.make t 1 in
-  stubs_complex (CArray.start out__) real imag;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_combinations self (Int64.of_int r) (if with_replacement then 1 else 0)
+  |> with_tensor_gc
 ;;
 
-let complex_out ~out ~real ~imag =
-  let out__ = CArray.make t 1 in
-  stubs_complex_out (CArray.start out__) out real imag;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let complex ~real ~imag = stubs_complex real imag |> with_tensor_gc
+let complex_out ~out ~real ~imag = stubs_complex_out out real imag |> with_tensor_gc
 
 let concat tensors ~dim =
-  let out__ = CArray.make t 1 in
   stubs_concat
-    (CArray.start out__)
-    (CArray.of_list t tensors |> CArray.start)
+    (CArray.of_list gc_tensor tensors |> CArray.start)
     (List.length tensors)
-    (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int dim)
+  |> with_tensor_gc
 ;;
 
 let concat_out ~out tensors ~dim =
-  let out__ = CArray.make t 1 in
   stubs_concat_out
-    (CArray.start out__)
     out
-    (CArray.of_list t tensors |> CArray.start)
+    (CArray.of_list gc_tensor tensors |> CArray.start)
     (List.length tensors)
-    (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int dim)
+  |> with_tensor_gc
 ;;
 
 let concatenate tensors ~dim =
-  let out__ = CArray.make t 1 in
   stubs_concatenate
-    (CArray.start out__)
-    (CArray.of_list t tensors |> CArray.start)
+    (CArray.of_list gc_tensor tensors |> CArray.start)
     (List.length tensors)
-    (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int dim)
+  |> with_tensor_gc
 ;;
 
 let concatenate_out ~out tensors ~dim =
-  let out__ = CArray.make t 1 in
   stubs_concatenate_out
-    (CArray.start out__)
     out
-    (CArray.of_list t tensors |> CArray.start)
+    (CArray.of_list gc_tensor tensors |> CArray.start)
     (List.length tensors)
-    (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let conj self =
-  let out__ = CArray.make t 1 in
-  stubs_conj (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let conj_physical self =
-  let out__ = CArray.make t 1 in
-  stubs_conj_physical (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let conj_physical_ self =
-  let out__ = CArray.make t 1 in
-  stubs_conj_physical_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int dim)
+  |> with_tensor_gc
 ;;
 
-let conj_physical_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_conj_physical_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let conj self = stubs_conj self |> with_tensor_gc
+let conj_physical self = stubs_conj_physical self |> with_tensor_gc
+let conj_physical_ self = stubs_conj_physical_ self |> with_tensor_gc
+let conj_physical_out ~out self = stubs_conj_physical_out out self |> with_tensor_gc
 
 let constant_pad_nd self ~pad =
-  let out__ = CArray.make t 1 in
   stubs_constant_pad_nd
-    (CArray.start out__)
     self
     (List.map Int64.of_int pad |> CArray.of_list int64_t |> CArray.start)
-    (List.length pad);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length pad)
+  |> with_tensor_gc
 ;;
 
 let constant_pad_nd_out ~out self ~pad =
-  let out__ = CArray.make t 1 in
   stubs_constant_pad_nd_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int pad |> CArray.of_list int64_t |> CArray.start)
-    (List.length pad);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length pad)
+  |> with_tensor_gc
 ;;
 
-let contiguous self =
-  let out__ = CArray.make t 1 in
-  stubs_contiguous (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let contiguous self = stubs_contiguous self |> with_tensor_gc
 
 let conv1d input ~weight ~bias ~stride ~padding ~dilation ~groups =
-  let out__ = CArray.make t 1 in
   stubs_conv1d
-    (CArray.start out__)
     input
     weight
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
     (List.length stride)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
     (List.length padding)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
-    (Int64.of_int groups);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int groups)
+  |> with_tensor_gc
 ;;
 
 let conv1d_padding input ~weight ~bias ~stride ~padding ~dilation ~groups =
-  let out__ = CArray.make t 1 in
   stubs_conv1d_padding
-    (CArray.start out__)
     input
     weight
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
     (List.length stride)
     padding
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
-    (Int64.of_int groups);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int groups)
+  |> with_tensor_gc
 ;;
 
 let conv2d input ~weight ~bias ~stride ~padding ~dilation ~groups =
-  let out__ = CArray.make t 1 in
   stubs_conv2d
-    (CArray.start out__)
     input
     weight
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
     (List.length stride)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
     (List.length padding)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
-    (Int64.of_int groups);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int groups)
+  |> with_tensor_gc
 ;;
 
 let conv2d_padding input ~weight ~bias ~stride ~padding ~dilation ~groups =
-  let out__ = CArray.make t 1 in
   stubs_conv2d_padding
-    (CArray.start out__)
     input
     weight
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
     (List.length stride)
     padding
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
-    (Int64.of_int groups);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int groups)
+  |> with_tensor_gc
 ;;
 
 let conv3d input ~weight ~bias ~stride ~padding ~dilation ~groups =
-  let out__ = CArray.make t 1 in
   stubs_conv3d
-    (CArray.start out__)
     input
     weight
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
     (List.length stride)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
     (List.length padding)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
-    (Int64.of_int groups);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int groups)
+  |> with_tensor_gc
 ;;
 
 let conv3d_padding input ~weight ~bias ~stride ~padding ~dilation ~groups =
-  let out__ = CArray.make t 1 in
   stubs_conv3d_padding
-    (CArray.start out__)
     input
     weight
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
     (List.length stride)
     padding
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
-    (Int64.of_int groups);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int groups)
+  |> with_tensor_gc
 ;;
 
 let conv_depthwise3d self ~weight ~kernel_size ~bias ~stride ~padding ~dilation =
-  let out__ = CArray.make t 1 in
   stubs_conv_depthwise3d
-    (CArray.start out__)
     self
     weight
     (List.map Int64.of_int kernel_size |> CArray.of_list int64_t |> CArray.start)
     (List.length kernel_size)
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
     (List.length stride)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
     (List.length padding)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
-    (List.length dilation);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length dilation)
+  |> with_tensor_gc
 ;;
 
 let conv_depthwise3d_out ~out self ~weight ~kernel_size ~bias ~stride ~padding ~dilation =
-  let out__ = CArray.make t 1 in
   stubs_conv_depthwise3d_out
-    (CArray.start out__)
     out
     self
     weight
@@ -11671,44 +8046,31 @@ let conv_depthwise3d_out ~out self ~weight ~kernel_size ~bias ~stride ~padding ~
     (List.length kernel_size)
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
     (List.length stride)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
     (List.length padding)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
-    (List.length dilation);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length dilation)
+  |> with_tensor_gc
 ;;
 
 let conv_tbc self ~weight ~bias ~pad =
-  let out__ = CArray.make t 1 in
-  stubs_conv_tbc (CArray.start out__) self weight bias (Int64.of_int pad);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_conv_tbc self weight bias (Int64.of_int pad) |> with_tensor_gc
 ;;
 
 let conv_tbc_backward self input ~weight ~bias ~pad =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs_conv_tbc_backward (CArray.start out__) self input weight bias (Int64.of_int pad);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
 let conv_tbc_out ~out self ~weight ~bias ~pad =
-  let out__ = CArray.make t 1 in
-  stubs_conv_tbc_out (CArray.start out__) out self weight bias (Int64.of_int pad);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_conv_tbc_out out self weight bias (Int64.of_int pad) |> with_tensor_gc
 ;;
 
 let conv_transpose1d
@@ -11721,14 +8083,12 @@ let conv_transpose1d
   ~groups
   ~dilation
   =
-  let out__ = CArray.make t 1 in
   stubs_conv_transpose1d
-    (CArray.start out__)
     input
     weight
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
     (List.length stride)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
@@ -11737,10 +8097,8 @@ let conv_transpose1d
     (List.length output_padding)
     (Int64.of_int groups)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
-    (List.length dilation);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length dilation)
+  |> with_tensor_gc
 ;;
 
 let conv_transpose2d
@@ -11753,14 +8111,12 @@ let conv_transpose2d
   ~groups
   ~dilation
   =
-  let out__ = CArray.make t 1 in
   stubs_conv_transpose2d
-    (CArray.start out__)
     input
     weight
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
     (List.length stride)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
@@ -11769,10 +8125,8 @@ let conv_transpose2d
     (List.length output_padding)
     (Int64.of_int groups)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
-    (List.length dilation);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length dilation)
+  |> with_tensor_gc
 ;;
 
 let conv_transpose3d
@@ -11785,14 +8139,12 @@ let conv_transpose3d
   ~groups
   ~dilation
   =
-  let out__ = CArray.make t 1 in
   stubs_conv_transpose3d
-    (CArray.start out__)
     input
     weight
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
     (List.length stride)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
@@ -11801,10 +8153,8 @@ let conv_transpose3d
     (List.length output_padding)
     (Int64.of_int groups)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
-    (List.length dilation);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length dilation)
+  |> with_tensor_gc
 ;;
 
 let convolution
@@ -11818,14 +8168,12 @@ let convolution
   ~output_padding
   ~groups
   =
-  let out__ = CArray.make t 1 in
   stubs_convolution
-    (CArray.start out__)
     input
     weight
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
     (List.length stride)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
@@ -11835,10 +8183,8 @@ let convolution
     (if transposed then 1 else 0)
     (List.map Int64.of_int output_padding |> CArray.of_list int64_t |> CArray.start)
     (List.length output_padding)
-    (Int64.of_int groups);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int groups)
+  |> with_tensor_gc
 ;;
 
 let convolution_out
@@ -11853,15 +8199,13 @@ let convolution_out
   ~output_padding
   ~groups
   =
-  let out__ = CArray.make t 1 in
   stubs_convolution_out
-    (CArray.start out__)
     out
     input
     weight
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
     (List.length stride)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
@@ -11871,10 +8215,8 @@ let convolution_out
     (if transposed then 1 else 0)
     (List.map Int64.of_int output_padding |> CArray.of_list int64_t |> CArray.start)
     (List.length output_padding)
-    (Int64.of_int groups);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int groups)
+  |> with_tensor_gc
 ;;
 
 let convolution_overrideable
@@ -11888,14 +8230,12 @@ let convolution_overrideable
   ~output_padding
   ~groups
   =
-  let out__ = CArray.make t 1 in
   stubs_convolution_overrideable
-    (CArray.start out__)
     input
     weight
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
     (List.length stride)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
@@ -11905,10 +8245,8 @@ let convolution_overrideable
     (if transposed then 1 else 0)
     (List.map Int64.of_int output_padding |> CArray.of_list int64_t |> CArray.start)
     (List.length output_padding)
-    (Int64.of_int groups);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int groups)
+  |> with_tensor_gc
 ;;
 
 let convolution_overrideable_out
@@ -11923,15 +8261,13 @@ let convolution_overrideable_out
   ~output_padding
   ~groups
   =
-  let out__ = CArray.make t 1 in
   stubs_convolution_overrideable_out
-    (CArray.start out__)
     out
     input
     weight
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
     (List.length stride)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
@@ -11941,208 +8277,74 @@ let convolution_overrideable_out
     (if transposed then 1 else 0)
     (List.map Int64.of_int output_padding |> CArray.of_list int64_t |> CArray.start)
     (List.length output_padding)
-    (Int64.of_int groups);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int groups)
+  |> with_tensor_gc
 ;;
 
 let copy self ~src ~non_blocking =
-  let out__ = CArray.make t 1 in
-  stubs_copy (CArray.start out__) self src (if non_blocking then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_copy self src (if non_blocking then 1 else 0) |> with_tensor_gc
 ;;
 
 let copy_out ~out self ~src ~non_blocking =
-  let out__ = CArray.make t 1 in
-  stubs_copy_out (CArray.start out__) out self src (if non_blocking then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_copy_out out self src (if non_blocking then 1 else 0) |> with_tensor_gc
 ;;
 
 let copy_sparse_to_sparse self ~src ~non_blocking =
-  let out__ = CArray.make t 1 in
-  stubs_copy_sparse_to_sparse
-    (CArray.start out__)
-    self
-    src
-    (if non_blocking then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_copy_sparse_to_sparse self src (if non_blocking then 1 else 0) |> with_tensor_gc
 ;;
 
 let copy_sparse_to_sparse_ self ~src ~non_blocking =
-  let out__ = CArray.make t 1 in
-  stubs_copy_sparse_to_sparse_
-    (CArray.start out__)
-    self
-    src
-    (if non_blocking then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_copy_sparse_to_sparse_ self src (if non_blocking then 1 else 0) |> with_tensor_gc
 ;;
 
 let copy_sparse_to_sparse_out ~out self ~src ~non_blocking =
-  let out__ = CArray.make t 1 in
-  stubs_copy_sparse_to_sparse_out
-    (CArray.start out__)
-    out
-    self
-    src
-    (if non_blocking then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let copysign self other =
-  let out__ = CArray.make t 1 in
-  stubs_copysign (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let copysign_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_copysign_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let copysign_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_copysign_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let copysign_scalar self other =
-  let out__ = CArray.make t 1 in
-  stubs_copysign_scalar (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_copy_sparse_to_sparse_out out self src (if non_blocking then 1 else 0)
+  |> with_tensor_gc
 ;;
 
-let copysign_scalar_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_copysign_scalar_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let copysign self other = stubs_copysign self other |> with_tensor_gc
+let copysign_ self other = stubs_copysign_ self other |> with_tensor_gc
+let copysign_out ~out self other = stubs_copysign_out out self other |> with_tensor_gc
+let copysign_scalar self other = stubs_copysign_scalar self other |> with_tensor_gc
+let copysign_scalar_ self other = stubs_copysign_scalar_ self other |> with_tensor_gc
 
 let copysign_scalar_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_copysign_scalar_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let corrcoef self =
-  let out__ = CArray.make t 1 in
-  stubs_corrcoef (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let cos self =
-  let out__ = CArray.make t 1 in
-  stubs_cos (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let cos_ self =
-  let out__ = CArray.make t 1 in
-  stubs_cos_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_copysign_scalar_out out self other |> with_tensor_gc
 ;;
 
-let cos_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_cos_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let cosh self =
-  let out__ = CArray.make t 1 in
-  stubs_cosh (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let cosh_ self =
-  let out__ = CArray.make t 1 in
-  stubs_cosh_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let cosh_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_cosh_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let corrcoef self = stubs_corrcoef self |> with_tensor_gc
+let cos self = stubs_cos self |> with_tensor_gc
+let cos_ self = stubs_cos_ self |> with_tensor_gc
+let cos_out ~out self = stubs_cos_out out self |> with_tensor_gc
+let cosh self = stubs_cosh self |> with_tensor_gc
+let cosh_ self = stubs_cosh_ self |> with_tensor_gc
+let cosh_out ~out self = stubs_cosh_out out self |> with_tensor_gc
 
 let cosine_embedding_loss ~input1 ~input2 ~target ~margin ~reduction =
-  let out__ = CArray.make t 1 in
   stubs_cosine_embedding_loss
-    (CArray.start out__)
     input1
     input2
     target
     margin
-    (Reduction.to_int reduction |> Int64.of_int);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Reduction.to_int reduction |> Int64.of_int)
+  |> with_tensor_gc
 ;;
 
 let cosine_similarity ~x1 ~x2 ~dim ~eps =
-  let out__ = CArray.make t 1 in
-  stubs_cosine_similarity (CArray.start out__) x1 x2 (Int64.of_int dim) eps;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_cosine_similarity x1 x2 (Int64.of_int dim) eps |> with_tensor_gc
 ;;
 
 let count_nonzero ~out self ~dim =
-  let out__ = CArray.make t 1 in
   stubs_count_nonzero
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
-    (List.length dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length dim)
+  |> with_tensor_gc
 ;;
 
 let count_nonzero_out ~out self ~dim =
-  let out__ = CArray.make t 1 in
   stubs_count_nonzero_out
-    (CArray.start out__)
     out
     self
     (match dim with
@@ -12150,33 +8352,25 @@ let count_nonzero_out ~out self ~dim =
      | Some v -> Int64.of_int v)
     (match dim with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let cov self ~correction ~fweights ~aweights =
-  let out__ = CArray.make t 1 in
   stubs_cov
-    (CArray.start out__)
     self
     (Int64.of_int correction)
     (match fweights with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match aweights with
      | Some v -> v
-     | None -> null);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+  |> with_tensor_gc
 ;;
 
 let cross self other ~dim =
-  let out__ = CArray.make t 1 in
   stubs_cross
-    (CArray.start out__)
     self
     other
     (match dim with
@@ -12184,33 +8378,25 @@ let cross self other ~dim =
      | Some v -> Int64.of_int v)
     (match dim with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let cross_entropy_loss self ~target ~weight ~reduction ~ignore_index ~label_smoothing =
-  let out__ = CArray.make t 1 in
   stubs_cross_entropy_loss
-    (CArray.start out__)
     self
     target
     (match weight with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (Reduction.to_int reduction |> Int64.of_int)
     (Int64.of_int ignore_index)
-    label_smoothing;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    label_smoothing
+  |> with_tensor_gc
 ;;
 
 let cross_out ~out self other ~dim =
-  let out__ = CArray.make t 1 in
   stubs_cross_out
-    (CArray.start out__)
     out
     self
     other
@@ -12219,34 +8405,15 @@ let cross_out ~out self other ~dim =
      | Some v -> Int64.of_int v)
     (match dim with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let crow_indices self =
-  let out__ = CArray.make t 1 in
-  stubs_crow_indices (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
-let crow_indices_copy self =
-  let out__ = CArray.make t 1 in
-  stubs_crow_indices_copy (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let crow_indices self = stubs_crow_indices self |> with_tensor_gc
+let crow_indices_copy self = stubs_crow_indices_copy self |> with_tensor_gc
 
 let crow_indices_copy_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_crow_indices_copy_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_crow_indices_copy_out out self |> with_tensor_gc
 ;;
 
 let ctc_loss
@@ -12258,9 +8425,7 @@ let ctc_loss
   ~reduction
   ~zero_infinity
   =
-  let out__ = CArray.make t 1 in
   stubs_ctc_loss
-    (CArray.start out__)
     log_probs
     targets
     (List.map Int64.of_int input_lengths |> CArray.of_list int64_t |> CArray.start)
@@ -12269,10 +8434,8 @@ let ctc_loss
     (List.length target_lengths)
     (Int64.of_int blank)
     (Reduction.to_int reduction |> Int64.of_int)
-    (if zero_infinity then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if zero_infinity then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let ctc_loss_tensor
@@ -12284,77 +8447,57 @@ let ctc_loss_tensor
   ~reduction
   ~zero_infinity
   =
-  let out__ = CArray.make t 1 in
   stubs_ctc_loss_tensor
-    (CArray.start out__)
     log_probs
     targets
     input_lengths
     target_lengths
     (Int64.of_int blank)
     (Reduction.to_int reduction |> Int64.of_int)
-    (if zero_infinity then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if zero_infinity then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let cudnn_affine_grid_generator ~theta ~n ~c ~h ~w =
-  let out__ = CArray.make t 1 in
   stubs_cudnn_affine_grid_generator
-    (CArray.start out__)
     theta
     (Int64.of_int n)
     (Int64.of_int c)
     (Int64.of_int h)
-    (Int64.of_int w);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int w)
+  |> with_tensor_gc
 ;;
 
 let cudnn_affine_grid_generator_backward ~grad ~n ~c ~h ~w =
-  let out__ = CArray.make t 1 in
   stubs_cudnn_affine_grid_generator_backward
-    (CArray.start out__)
     grad
     (Int64.of_int n)
     (Int64.of_int c)
     (Int64.of_int h)
-    (Int64.of_int w);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int w)
+  |> with_tensor_gc
 ;;
 
 let cudnn_affine_grid_generator_backward_out ~out ~grad ~n ~c ~h ~w =
-  let out__ = CArray.make t 1 in
   stubs_cudnn_affine_grid_generator_backward_out
-    (CArray.start out__)
     out
     grad
     (Int64.of_int n)
     (Int64.of_int c)
     (Int64.of_int h)
-    (Int64.of_int w);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int w)
+  |> with_tensor_gc
 ;;
 
 let cudnn_affine_grid_generator_out ~out ~theta ~n ~c ~h ~w =
-  let out__ = CArray.make t 1 in
   stubs_cudnn_affine_grid_generator_out
-    (CArray.start out__)
     out
     theta
     (Int64.of_int n)
     (Int64.of_int c)
     (Int64.of_int h)
-    (Int64.of_int w);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int w)
+  |> with_tensor_gc
 ;;
 
 let cudnn_batch_norm
@@ -12367,31 +8510,27 @@ let cudnn_batch_norm
   ~exponential_average_factor
   ~epsilon
   =
-  let out__ = CArray.make t 4 in
+  let out__ = CArray.make raw_tensor 4 in
   stubs_cudnn_batch_norm
     (CArray.start out__)
     input
     weight
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match running_mean with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match running_var with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (if training then 1 else 0)
     exponential_average_factor
     epsilon;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
-  let t3 = CArray.get out__ 3 in
-  Gc.finalise C.Tensor.free t3;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
+  let t3 = CArray.get out__ 3 |> with_tensor_gc in
   t0, t1, t2, t3
 ;;
 
@@ -12406,7 +8545,7 @@ let cudnn_batch_norm_backward
   ~epsilon
   ~reservespace
   =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs_cudnn_batch_norm_backward
     (CArray.start out__)
     input
@@ -12414,24 +8553,21 @@ let cudnn_batch_norm_backward
     weight
     (match running_mean with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match running_var with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match save_mean with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match save_var with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     epsilon
     reservespace;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
@@ -12449,7 +8585,7 @@ let cudnn_batch_norm_backward_out
   ~epsilon
   ~reservespace
   =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs_cudnn_batch_norm_backward_out
     (CArray.start out__)
     out0
@@ -12460,24 +8596,21 @@ let cudnn_batch_norm_backward_out
     weight
     (match running_mean with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match running_var with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match save_mean with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match save_var with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     epsilon
     reservespace;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
@@ -12495,7 +8628,7 @@ let cudnn_batch_norm_out
   ~exponential_average_factor
   ~epsilon
   =
-  let out__ = CArray.make t 4 in
+  let out__ = CArray.make raw_tensor 4 in
   stubs_cudnn_batch_norm_out
     (CArray.start out__)
     out0
@@ -12506,24 +8639,20 @@ let cudnn_batch_norm_out
     weight
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match running_mean with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match running_var with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (if training then 1 else 0)
     exponential_average_factor
     epsilon;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
-  let t3 = CArray.get out__ 3 in
-  Gc.finalise C.Tensor.free t3;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
+  let t3 = CArray.get out__ 3 |> with_tensor_gc in
   t0, t1, t2, t3
 ;;
 
@@ -12538,9 +8667,7 @@ let cudnn_convolution
   ~deterministic
   ~allow_tf32
   =
-  let out__ = CArray.make t 1 in
   stubs_cudnn_convolution
-    (CArray.start out__)
     self
     weight
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
@@ -12552,10 +8679,8 @@ let cudnn_convolution
     (Int64.of_int groups)
     (if benchmark then 1 else 0)
     (if deterministic then 1 else 0)
-    (if allow_tf32 then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if allow_tf32 then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let cudnn_convolution_add_relu
@@ -12569,26 +8694,22 @@ let cudnn_convolution_add_relu
   ~dilation
   ~groups
   =
-  let out__ = CArray.make t 1 in
   stubs_cudnn_convolution_add_relu
-    (CArray.start out__)
     self
     weight
     z
     alpha
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
     (List.length stride)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
     (List.length padding)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
-    (Int64.of_int groups);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int groups)
+  |> with_tensor_gc
 ;;
 
 let cudnn_convolution_add_relu_out
@@ -12603,9 +8724,7 @@ let cudnn_convolution_add_relu_out
   ~dilation
   ~groups
   =
-  let out__ = CArray.make t 1 in
   stubs_cudnn_convolution_add_relu_out
-    (CArray.start out__)
     out
     self
     weight
@@ -12613,17 +8732,15 @@ let cudnn_convolution_add_relu_out
     alpha
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
     (List.length stride)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
     (List.length padding)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
-    (Int64.of_int groups);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int groups)
+  |> with_tensor_gc
 ;;
 
 let cudnn_convolution_out
@@ -12638,9 +8755,7 @@ let cudnn_convolution_out
   ~deterministic
   ~allow_tf32
   =
-  let out__ = CArray.make t 1 in
   stubs_cudnn_convolution_out
-    (CArray.start out__)
     out
     self
     weight
@@ -12653,53 +8768,43 @@ let cudnn_convolution_out
     (Int64.of_int groups)
     (if benchmark then 1 else 0)
     (if deterministic then 1 else 0)
-    (if allow_tf32 then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if allow_tf32 then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let cudnn_convolution_relu self ~weight ~bias ~stride ~padding ~dilation ~groups =
-  let out__ = CArray.make t 1 in
   stubs_cudnn_convolution_relu
-    (CArray.start out__)
     self
     weight
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
     (List.length stride)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
     (List.length padding)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
-    (Int64.of_int groups);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int groups)
+  |> with_tensor_gc
 ;;
 
 let cudnn_convolution_relu_out ~out self ~weight ~bias ~stride ~padding ~dilation ~groups =
-  let out__ = CArray.make t 1 in
   stubs_cudnn_convolution_relu_out
-    (CArray.start out__)
     out
     self
     weight
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
     (List.length stride)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
     (List.length padding)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
-    (Int64.of_int groups);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int groups)
+  |> with_tensor_gc
 ;;
 
 let cudnn_convolution_transpose
@@ -12714,9 +8819,7 @@ let cudnn_convolution_transpose
   ~deterministic
   ~allow_tf32
   =
-  let out__ = CArray.make t 1 in
   stubs_cudnn_convolution_transpose
-    (CArray.start out__)
     self
     weight
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
@@ -12730,10 +8833,8 @@ let cudnn_convolution_transpose
     (Int64.of_int groups)
     (if benchmark then 1 else 0)
     (if deterministic then 1 else 0)
-    (if allow_tf32 then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if allow_tf32 then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let cudnn_convolution_transpose_out
@@ -12749,9 +8850,7 @@ let cudnn_convolution_transpose_out
   ~deterministic
   ~allow_tf32
   =
-  let out__ = CArray.make t 1 in
   stubs_cudnn_convolution_transpose_out
-    (CArray.start out__)
     out
     self
     weight
@@ -12766,32 +8865,22 @@ let cudnn_convolution_transpose_out
     (Int64.of_int groups)
     (if benchmark then 1 else 0)
     (if deterministic then 1 else 0)
-    (if allow_tf32 then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if allow_tf32 then 1 else 0)
+  |> with_tensor_gc
 ;;
 
-let cudnn_grid_sampler self ~grid =
-  let out__ = CArray.make t 1 in
-  stubs_cudnn_grid_sampler (CArray.start out__) self grid;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let cudnn_grid_sampler self ~grid = stubs_cudnn_grid_sampler self grid |> with_tensor_gc
 
 let cudnn_grid_sampler_backward self ~grid ~grad_output =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_cudnn_grid_sampler_backward (CArray.start out__) self grid grad_output;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let cudnn_grid_sampler_backward_out ~out0 ~out1 self ~grid ~grad_output =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_cudnn_grid_sampler_backward_out
     (CArray.start out__)
     out0
@@ -12799,692 +8888,309 @@ let cudnn_grid_sampler_backward_out ~out0 ~out1 self ~grid ~grad_output =
     self
     grid
     grad_output;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let cudnn_grid_sampler_out ~out self ~grid =
-  let out__ = CArray.make t 1 in
-  stubs_cudnn_grid_sampler_out (CArray.start out__) out self grid;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_cudnn_grid_sampler_out out self grid |> with_tensor_gc
 ;;
 
 let cudnn_is_acceptable self = stubs_cudnn_is_acceptable self
 
 let cummax self ~dim =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_cummax (CArray.start out__) self (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let cummax_out ~values ~indices self ~dim =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_cummax_out (CArray.start out__) values indices self (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let cummaxmin_backward ~grad input ~indices ~dim =
-  let out__ = CArray.make t 1 in
-  stubs_cummaxmin_backward (CArray.start out__) grad input indices (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_cummaxmin_backward grad input indices (Int64.of_int dim) |> with_tensor_gc
 ;;
 
 let cummin self ~dim =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_cummin (CArray.start out__) self (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let cummin_out ~values ~indices self ~dim =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_cummin_out (CArray.start out__) values indices self (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let cumprod self ~dim ~dtype =
-  let out__ = CArray.make t 1 in
-  stubs_cumprod (CArray.start out__) self (Int64.of_int dim) (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_cumprod self (Int64.of_int dim) (Kind.packed_to_int dtype) |> with_tensor_gc
 ;;
 
 let cumprod_ self ~dim ~dtype =
-  let out__ = CArray.make t 1 in
-  stubs_cumprod_ (CArray.start out__) self (Int64.of_int dim) (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_cumprod_ self (Int64.of_int dim) (Kind.packed_to_int dtype) |> with_tensor_gc
 ;;
 
 let cumprod_backward ~grad input ~dim ~output =
-  let out__ = CArray.make t 1 in
-  stubs_cumprod_backward (CArray.start out__) grad input (Int64.of_int dim) output;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_cumprod_backward grad input (Int64.of_int dim) output |> with_tensor_gc
 ;;
 
 let cumprod_out ~out self ~dim ~dtype =
-  let out__ = CArray.make t 1 in
-  stubs_cumprod_out
-    (CArray.start out__)
-    out
-    self
-    (Int64.of_int dim)
-    (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_cumprod_out out self (Int64.of_int dim) (Kind.packed_to_int dtype)
+  |> with_tensor_gc
 ;;
 
 let cumsum self ~dim ~dtype =
-  let out__ = CArray.make t 1 in
-  stubs_cumsum (CArray.start out__) self (Int64.of_int dim) (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_cumsum self (Int64.of_int dim) (Kind.packed_to_int dtype) |> with_tensor_gc
 ;;
 
 let cumsum_ self ~dim ~dtype =
-  let out__ = CArray.make t 1 in
-  stubs_cumsum_ (CArray.start out__) self (Int64.of_int dim) (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_cumsum_ self (Int64.of_int dim) (Kind.packed_to_int dtype) |> with_tensor_gc
 ;;
 
 let cumsum_out ~out self ~dim ~dtype =
-  let out__ = CArray.make t 1 in
-  stubs_cumsum_out
-    (CArray.start out__)
-    out
-    self
-    (Int64.of_int dim)
-    (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_cumsum_out out self (Int64.of_int dim) (Kind.packed_to_int dtype)
+  |> with_tensor_gc
 ;;
 
 let cumulative_trapezoid ~y ~dim =
-  let out__ = CArray.make t 1 in
-  stubs_cumulative_trapezoid (CArray.start out__) y (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_cumulative_trapezoid y (Int64.of_int dim) |> with_tensor_gc
 ;;
 
 let cumulative_trapezoid_x ~y ~x ~dim =
-  let out__ = CArray.make t 1 in
-  stubs_cumulative_trapezoid_x (CArray.start out__) y x (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let data self =
-  let out__ = CArray.make t 1 in
-  stubs_data (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let deg2rad self =
-  let out__ = CArray.make t 1 in
-  stubs_deg2rad (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let deg2rad_ self =
-  let out__ = CArray.make t 1 in
-  stubs_deg2rad_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let deg2rad_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_deg2rad_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_cumulative_trapezoid_x y x (Int64.of_int dim) |> with_tensor_gc
 ;;
 
+let data self = stubs_data self |> with_tensor_gc
+let deg2rad self = stubs_deg2rad self |> with_tensor_gc
+let deg2rad_ self = stubs_deg2rad_ self |> with_tensor_gc
+let deg2rad_out ~out self = stubs_deg2rad_out out self |> with_tensor_gc
 let dense_dim self = stubs_dense_dim self
-
-let dequantize self =
-  let out__ = CArray.make t 1 in
-  stubs_dequantize (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let dequantize_self_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_dequantize_self_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let dequantize self = stubs_dequantize self |> with_tensor_gc
+let dequantize_self_out ~out self = stubs_dequantize_self_out out self |> with_tensor_gc
 
 let dequantize_tensors tensors =
   stubs_dequantize_tensors
-    (CArray.of_list t tensors |> CArray.start)
+    (CArray.of_list gc_tensor tensors |> CArray.start)
     (List.length tensors)
   |> to_tensor_list
 ;;
 
 let dequantize_tensors_out ~out tensors =
   stubs_dequantize_tensors_out
-    (CArray.of_list t out |> CArray.start)
+    (CArray.of_list gc_tensor out |> CArray.start)
     (List.length out)
-    (CArray.of_list t tensors |> CArray.start)
+    (CArray.of_list gc_tensor tensors |> CArray.start)
     (List.length tensors)
 ;;
 
-let det self =
-  let out__ = CArray.make t 1 in
-  stubs_det (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let detach self =
-  let out__ = CArray.make t 1 in
-  stubs_detach (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let detach_ self =
-  let out__ = CArray.make t 1 in
-  stubs_detach_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let detach_copy self =
-  let out__ = CArray.make t 1 in
-  stubs_detach_copy (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let detach_copy_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_detach_copy_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let diag self ~diagonal =
-  let out__ = CArray.make t 1 in
-  stubs_diag (CArray.start out__) self (Int64.of_int diagonal);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let det self = stubs_det self |> with_tensor_gc
+let detach self = stubs_detach self |> with_tensor_gc
+let detach_ self = stubs_detach_ self |> with_tensor_gc
+let detach_copy self = stubs_detach_copy self |> with_tensor_gc
+let detach_copy_out ~out self = stubs_detach_copy_out out self |> with_tensor_gc
+let diag self ~diagonal = stubs_diag self (Int64.of_int diagonal) |> with_tensor_gc
 
 let diag_embed self ~offset ~dim1 ~dim2 =
-  let out__ = CArray.make t 1 in
-  stubs_diag_embed
-    (CArray.start out__)
-    self
-    (Int64.of_int offset)
-    (Int64.of_int dim1)
-    (Int64.of_int dim2);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_diag_embed self (Int64.of_int offset) (Int64.of_int dim1) (Int64.of_int dim2)
+  |> with_tensor_gc
 ;;
 
 let diag_embed_out ~out self ~offset ~dim1 ~dim2 =
-  let out__ = CArray.make t 1 in
   stubs_diag_embed_out
-    (CArray.start out__)
     out
     self
     (Int64.of_int offset)
     (Int64.of_int dim1)
-    (Int64.of_int dim2);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int dim2)
+  |> with_tensor_gc
 ;;
 
 let diag_out ~out self ~diagonal =
-  let out__ = CArray.make t 1 in
-  stubs_diag_out (CArray.start out__) out self (Int64.of_int diagonal);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_diag_out out self (Int64.of_int diagonal) |> with_tensor_gc
 ;;
 
-let diagflat self ~offset =
-  let out__ = CArray.make t 1 in
-  stubs_diagflat (CArray.start out__) self (Int64.of_int offset);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let diagflat self ~offset = stubs_diagflat self (Int64.of_int offset) |> with_tensor_gc
 
 let diagonal self ~offset ~dim1 ~dim2 =
-  let out__ = CArray.make t 1 in
-  stubs_diagonal
-    (CArray.start out__)
-    self
-    (Int64.of_int offset)
-    (Int64.of_int dim1)
-    (Int64.of_int dim2);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_diagonal self (Int64.of_int offset) (Int64.of_int dim1) (Int64.of_int dim2)
+  |> with_tensor_gc
 ;;
 
 let diagonal_backward ~grad_output ~input_sizes ~offset ~dim1 ~dim2 =
-  let out__ = CArray.make t 1 in
   stubs_diagonal_backward
-    (CArray.start out__)
     grad_output
     (List.map Int64.of_int input_sizes |> CArray.of_list int64_t |> CArray.start)
     (List.length input_sizes)
     (Int64.of_int offset)
     (Int64.of_int dim1)
-    (Int64.of_int dim2);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int dim2)
+  |> with_tensor_gc
 ;;
 
 let diagonal_backward_out ~out ~grad_output ~input_sizes ~offset ~dim1 ~dim2 =
-  let out__ = CArray.make t 1 in
   stubs_diagonal_backward_out
-    (CArray.start out__)
     out
     grad_output
     (List.map Int64.of_int input_sizes |> CArray.of_list int64_t |> CArray.start)
     (List.length input_sizes)
     (Int64.of_int offset)
     (Int64.of_int dim1)
-    (Int64.of_int dim2);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int dim2)
+  |> with_tensor_gc
 ;;
 
 let diagonal_copy self ~offset ~dim1 ~dim2 =
-  let out__ = CArray.make t 1 in
-  stubs_diagonal_copy
-    (CArray.start out__)
-    self
-    (Int64.of_int offset)
-    (Int64.of_int dim1)
-    (Int64.of_int dim2);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_diagonal_copy self (Int64.of_int offset) (Int64.of_int dim1) (Int64.of_int dim2)
+  |> with_tensor_gc
 ;;
 
 let diagonal_copy_out ~out self ~offset ~dim1 ~dim2 =
-  let out__ = CArray.make t 1 in
   stubs_diagonal_copy_out
-    (CArray.start out__)
     out
     self
     (Int64.of_int offset)
     (Int64.of_int dim1)
-    (Int64.of_int dim2);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int dim2)
+  |> with_tensor_gc
 ;;
 
 let diagonal_scatter self ~src ~offset ~dim1 ~dim2 =
-  let out__ = CArray.make t 1 in
   stubs_diagonal_scatter
-    (CArray.start out__)
     self
     src
     (Int64.of_int offset)
     (Int64.of_int dim1)
-    (Int64.of_int dim2);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int dim2)
+  |> with_tensor_gc
 ;;
 
 let diagonal_scatter_out ~out self ~src ~offset ~dim1 ~dim2 =
-  let out__ = CArray.make t 1 in
   stubs_diagonal_scatter_out
-    (CArray.start out__)
     out
     self
     src
     (Int64.of_int offset)
     (Int64.of_int dim1)
-    (Int64.of_int dim2);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int dim2)
+  |> with_tensor_gc
 ;;
 
 let diff self ~n ~dim ~prepend ~append =
-  let out__ = CArray.make t 1 in
   stubs_diff
-    (CArray.start out__)
     self
     (Int64.of_int n)
     (Int64.of_int dim)
     (match prepend with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match append with
      | Some v -> v
-     | None -> null);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+  |> with_tensor_gc
 ;;
 
 let diff_out ~out self ~n ~dim ~prepend ~append =
-  let out__ = CArray.make t 1 in
   stubs_diff_out
-    (CArray.start out__)
     out
     self
     (Int64.of_int n)
     (Int64.of_int dim)
     (match prepend with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match append with
      | Some v -> v
-     | None -> null);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let digamma self =
-  let out__ = CArray.make t 1 in
-  stubs_digamma (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let digamma_ self =
-  let out__ = CArray.make t 1 in
-  stubs_digamma_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+  |> with_tensor_gc
 ;;
 
-let digamma_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_digamma_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let dist self other =
-  let out__ = CArray.make t 1 in
-  stubs_dist (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let dist_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_dist_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let div self other =
-  let out__ = CArray.make t 1 in
-  stubs_div (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let div_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_div_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let div_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_div_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let digamma self = stubs_digamma self |> with_tensor_gc
+let digamma_ self = stubs_digamma_ self |> with_tensor_gc
+let digamma_out ~out self = stubs_digamma_out out self |> with_tensor_gc
+let dist self other = stubs_dist self other |> with_tensor_gc
+let dist_out ~out self other = stubs_dist_out out self other |> with_tensor_gc
+let div self other = stubs_div self other |> with_tensor_gc
+let div_ self other = stubs_div_ self other |> with_tensor_gc
+let div_out ~out self other = stubs_div_out out self other |> with_tensor_gc
 
 let div_out_mode ~out self other ~rounding_mode =
-  let out__ = CArray.make t 1 in
-  stubs_div_out_mode (CArray.start out__) out self other rounding_mode;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let div_scalar self other =
-  let out__ = CArray.make t 1 in
-  stubs_div_scalar (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_div_out_mode out self other rounding_mode |> with_tensor_gc
 ;;
 
-let div_scalar_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_div_scalar_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let div_scalar self other = stubs_div_scalar self other |> with_tensor_gc
+let div_scalar_ self other = stubs_div_scalar_ self other |> with_tensor_gc
 
 let div_scalar_mode self other ~rounding_mode =
-  let out__ = CArray.make t 1 in
-  stubs_div_scalar_mode (CArray.start out__) self other rounding_mode;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_div_scalar_mode self other rounding_mode |> with_tensor_gc
 ;;
 
 let div_scalar_mode_ self other ~rounding_mode =
-  let out__ = CArray.make t 1 in
-  stubs_div_scalar_mode_ (CArray.start out__) self other rounding_mode;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_div_scalar_mode_ self other rounding_mode |> with_tensor_gc
 ;;
 
 let div_scalar_mode_out ~out self other ~rounding_mode =
-  let out__ = CArray.make t 1 in
-  stubs_div_scalar_mode_out (CArray.start out__) out self other rounding_mode;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_div_scalar_mode_out out self other rounding_mode |> with_tensor_gc
 ;;
 
-let div_scalar_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_div_scalar_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let div_scalar_out ~out self other = stubs_div_scalar_out out self other |> with_tensor_gc
 
 let div_tensor_mode self other ~rounding_mode =
-  let out__ = CArray.make t 1 in
-  stubs_div_tensor_mode (CArray.start out__) self other rounding_mode;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_div_tensor_mode self other rounding_mode |> with_tensor_gc
 ;;
 
 let div_tensor_mode_ self other ~rounding_mode =
-  let out__ = CArray.make t 1 in
-  stubs_div_tensor_mode_ (CArray.start out__) self other rounding_mode;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let divide self other =
-  let out__ = CArray.make t 1 in
-  stubs_divide (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let divide_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_divide_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_div_tensor_mode_ self other rounding_mode |> with_tensor_gc
 ;;
 
-let divide_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_divide_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let divide self other = stubs_divide self other |> with_tensor_gc
+let divide_ self other = stubs_divide_ self other |> with_tensor_gc
+let divide_out ~out self other = stubs_divide_out out self other |> with_tensor_gc
 
 let divide_out_mode ~out self other ~rounding_mode =
-  let out__ = CArray.make t 1 in
-  stubs_divide_out_mode (CArray.start out__) out self other rounding_mode;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_divide_out_mode out self other rounding_mode |> with_tensor_gc
 ;;
 
-let divide_scalar self other =
-  let out__ = CArray.make t 1 in
-  stubs_divide_scalar (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let divide_scalar_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_divide_scalar_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let divide_scalar self other = stubs_divide_scalar self other |> with_tensor_gc
+let divide_scalar_ self other = stubs_divide_scalar_ self other |> with_tensor_gc
 
 let divide_scalar_mode self other ~rounding_mode =
-  let out__ = CArray.make t 1 in
-  stubs_divide_scalar_mode (CArray.start out__) self other rounding_mode;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_divide_scalar_mode self other rounding_mode |> with_tensor_gc
 ;;
 
 let divide_scalar_mode_ self other ~rounding_mode =
-  let out__ = CArray.make t 1 in
-  stubs_divide_scalar_mode_ (CArray.start out__) self other rounding_mode;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_divide_scalar_mode_ self other rounding_mode |> with_tensor_gc
 ;;
 
 let divide_tensor_mode self other ~rounding_mode =
-  let out__ = CArray.make t 1 in
-  stubs_divide_tensor_mode (CArray.start out__) self other rounding_mode;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_divide_tensor_mode self other rounding_mode |> with_tensor_gc
 ;;
 
 let divide_tensor_mode_ self other ~rounding_mode =
-  let out__ = CArray.make t 1 in
-  stubs_divide_tensor_mode_ (CArray.start out__) self other rounding_mode;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let dot self tensor =
-  let out__ = CArray.make t 1 in
-  stubs_dot (CArray.start out__) self tensor;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_divide_tensor_mode_ self other rounding_mode |> with_tensor_gc
 ;;
 
-let dot_out ~out self tensor =
-  let out__ = CArray.make t 1 in
-  stubs_dot_out (CArray.start out__) out self tensor;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let dot self tensor = stubs_dot self tensor |> with_tensor_gc
+let dot_out ~out self tensor = stubs_dot_out out self tensor |> with_tensor_gc
 
 let dropout input ~p ~train =
-  let out__ = CArray.make t 1 in
-  stubs_dropout (CArray.start out__) input p (if train then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_dropout input p (if train then 1 else 0) |> with_tensor_gc
 ;;
 
 let dropout_ self ~p ~train =
-  let out__ = CArray.make t 1 in
-  stubs_dropout_ (CArray.start out__) self p (if train then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_dropout_ self p (if train then 1 else 0) |> with_tensor_gc
 ;;
 
 let dsplit self ~sections = stubs_dsplit self (Int64.of_int sections) |> to_tensor_list
@@ -13498,75 +9204,44 @@ let dsplit_array self ~indices =
 ;;
 
 let dstack tensors =
-  let out__ = CArray.make t 1 in
-  stubs_dstack
-    (CArray.start out__)
-    (CArray.of_list t tensors |> CArray.start)
-    (List.length tensors);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_dstack (CArray.of_list gc_tensor tensors |> CArray.start) (List.length tensors)
+  |> with_tensor_gc
 ;;
 
 let dstack_out ~out tensors =
-  let out__ = CArray.make t 1 in
   stubs_dstack_out
-    (CArray.start out__)
     out
-    (CArray.of_list t tensors |> CArray.start)
-    (List.length tensors);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (CArray.of_list gc_tensor tensors |> CArray.start)
+    (List.length tensors)
+  |> with_tensor_gc
 ;;
 
 let einsum ~equation tensors ~path =
-  let out__ = CArray.make t 1 in
   stubs_einsum
-    (CArray.start out__)
     equation
-    (CArray.of_list t tensors |> CArray.start)
+    (CArray.of_list gc_tensor tensors |> CArray.start)
     (List.length tensors)
     (match path with
      | None -> from_voidp int64_t null
      | Some v -> List.map Int64.of_int v |> CArray.of_list int64_t |> CArray.start)
     (match path with
      | None -> -1
-     | Some v -> List.length v);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let elu self =
-  let out__ = CArray.make t 1 in
-  stubs_elu (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | Some v -> List.length v)
+  |> with_tensor_gc
 ;;
 
-let elu_ self =
-  let out__ = CArray.make t 1 in
-  stubs_elu_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let elu self = stubs_elu self |> with_tensor_gc
+let elu_ self = stubs_elu_ self |> with_tensor_gc
 
 let elu_backward ~grad_output ~alpha ~scale ~input_scale ~is_result ~self_or_result =
-  let out__ = CArray.make t 1 in
   stubs_elu_backward
-    (CArray.start out__)
     grad_output
     alpha
     scale
     input_scale
     (if is_result then 1 else 0)
-    self_or_result;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    self_or_result
+  |> with_tensor_gc
 ;;
 
 let elu_backward_grad_input
@@ -13578,41 +9253,27 @@ let elu_backward_grad_input
   ~is_result
   ~self_or_result
   =
-  let out__ = CArray.make t 1 in
   stubs_elu_backward_grad_input
-    (CArray.start out__)
     grad_input
     grad_output
     alpha
     scale
     input_scale
     (if is_result then 1 else 0)
-    self_or_result;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    self_or_result
+  |> with_tensor_gc
 ;;
 
-let elu_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_elu_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let elu_out ~out self = stubs_elu_out out self |> with_tensor_gc
 
 let embedding ~weight ~indices ~padding_idx ~scale_grad_by_freq ~sparse =
-  let out__ = CArray.make t 1 in
   stubs_embedding
-    (CArray.start out__)
     weight
     indices
     (Int64.of_int padding_idx)
     (if scale_grad_by_freq then 1 else 0)
-    (if sparse then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if sparse then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let embedding_backward
@@ -13623,18 +9284,14 @@ let embedding_backward
   ~scale_grad_by_freq
   ~sparse
   =
-  let out__ = CArray.make t 1 in
   stubs_embedding_backward
-    (CArray.start out__)
     grad
     indices
     (Int64.of_int num_weights)
     (Int64.of_int padding_idx)
     (if scale_grad_by_freq then 1 else 0)
-    (if sparse then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if sparse then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let embedding_bag
@@ -13647,7 +9304,7 @@ let embedding_bag
   ~per_sample_weights
   ~include_last_offset
   =
-  let out__ = CArray.make t 4 in
+  let out__ = CArray.make raw_tensor 4 in
   stubs_embedding_bag
     (CArray.start out__)
     weight
@@ -13658,16 +9315,12 @@ let embedding_bag
     (if sparse then 1 else 0)
     (match per_sample_weights with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (if include_last_offset then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
-  let t3 = CArray.get out__ 3 in
-  Gc.finalise C.Tensor.free t3;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
+  let t3 = CArray.get out__ 3 |> with_tensor_gc in
   t0, t1, t2, t3
 ;;
 
@@ -13682,7 +9335,7 @@ let embedding_bag_padding_idx
   ~include_last_offset
   ~padding_idx
   =
-  let out__ = CArray.make t 4 in
+  let out__ = CArray.make raw_tensor 4 in
   stubs_embedding_bag_padding_idx
     (CArray.start out__)
     weight
@@ -13693,7 +9346,7 @@ let embedding_bag_padding_idx
     (if sparse then 1 else 0)
     (match per_sample_weights with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (if include_last_offset then 1 else 0)
     (match padding_idx with
      | None -> Int64.zero
@@ -13701,14 +9354,10 @@ let embedding_bag_padding_idx
     (match padding_idx with
      | Some _ -> 0
      | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
-  let t3 = CArray.get out__ 3 in
-  Gc.finalise C.Tensor.free t3;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
+  let t3 = CArray.get out__ 3 |> with_tensor_gc in
   t0, t1, t2, t3
 ;;
 
@@ -13719,17 +9368,13 @@ let embedding_dense_backward
   ~padding_idx
   ~scale_grad_by_freq
   =
-  let out__ = CArray.make t 1 in
   stubs_embedding_dense_backward
-    (CArray.start out__)
     grad_output
     indices
     (Int64.of_int num_weights)
     (Int64.of_int padding_idx)
-    (if scale_grad_by_freq then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if scale_grad_by_freq then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let embedding_dense_backward_out
@@ -13740,520 +9385,226 @@ let embedding_dense_backward_out
   ~padding_idx
   ~scale_grad_by_freq
   =
-  let out__ = CArray.make t 1 in
   stubs_embedding_dense_backward_out
-    (CArray.start out__)
     out
     grad_output
     indices
     (Int64.of_int num_weights)
     (Int64.of_int padding_idx)
-    (if scale_grad_by_freq then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if scale_grad_by_freq then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let embedding_out ~out ~weight ~indices ~padding_idx ~scale_grad_by_freq ~sparse =
-  let out__ = CArray.make t 1 in
   stubs_embedding_out
-    (CArray.start out__)
     out
     weight
     indices
     (Int64.of_int padding_idx)
     (if scale_grad_by_freq then 1 else 0)
-    (if sparse then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if sparse then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let embedding_renorm self ~indices ~max_norm ~norm_type =
-  let out__ = CArray.make t 1 in
-  stubs_embedding_renorm (CArray.start out__) self indices max_norm norm_type;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_embedding_renorm self indices max_norm norm_type |> with_tensor_gc
 ;;
 
 let embedding_renorm_ self ~indices ~max_norm ~norm_type =
-  let out__ = CArray.make t 1 in
-  stubs_embedding_renorm_ (CArray.start out__) self indices max_norm norm_type;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_embedding_renorm_ self indices max_norm norm_type |> with_tensor_gc
 ;;
 
 let embedding_renorm_out ~out self ~indices ~max_norm ~norm_type =
-  let out__ = CArray.make t 1 in
-  stubs_embedding_renorm_out (CArray.start out__) out self indices max_norm norm_type;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_embedding_renorm_out out self indices max_norm norm_type |> with_tensor_gc
 ;;
 
 let embedding_sparse_backward ~grad ~indices ~num_weights ~padding_idx ~scale_grad_by_freq
   =
-  let out__ = CArray.make t 1 in
   stubs_embedding_sparse_backward
-    (CArray.start out__)
     grad
     indices
     (Int64.of_int num_weights)
     (Int64.of_int padding_idx)
-    (if scale_grad_by_freq then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if scale_grad_by_freq then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let empty ~size ~options =
-  let out__ = CArray.make t 1 in
   stubs_empty
-    (CArray.start out__)
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let empty_like self =
-  let out__ = CArray.make t 1 in
-  stubs_empty_like (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
-let empty_like_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_empty_like_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let empty_like self = stubs_empty_like self |> with_tensor_gc
+let empty_like_out ~out self = stubs_empty_like_out out self |> with_tensor_gc
 
 let empty_out ~out ~size =
-  let out__ = CArray.make t 1 in
   stubs_empty_out
-    (CArray.start out__)
     out
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
-    (List.length size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length size)
+  |> with_tensor_gc
 ;;
 
 let empty_permuted ~size ~physical_layout ~options =
-  let out__ = CArray.make t 1 in
   stubs_empty_permuted
-    (CArray.start out__)
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     (List.map Int64.of_int physical_layout |> CArray.of_list int64_t |> CArray.start)
     (List.length physical_layout)
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let empty_permuted_out ~out ~size ~physical_layout =
-  let out__ = CArray.make t 1 in
   stubs_empty_permuted_out
-    (CArray.start out__)
     out
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     (List.map Int64.of_int physical_layout |> CArray.of_list int64_t |> CArray.start)
-    (List.length physical_layout);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length physical_layout)
+  |> with_tensor_gc
 ;;
 
 let empty_quantized ~size ~qtensor ~options =
-  let out__ = CArray.make t 1 in
   stubs_empty_quantized
-    (CArray.start out__)
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     qtensor
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let empty_quantized_out ~out ~size ~qtensor =
-  let out__ = CArray.make t 1 in
   stubs_empty_quantized_out
-    (CArray.start out__)
     out
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
-    qtensor;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    qtensor
+  |> with_tensor_gc
 ;;
 
 let empty_strided ~size ~stride ~options =
-  let out__ = CArray.make t 1 in
   stubs_empty_strided
-    (CArray.start out__)
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
     (List.length stride)
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let empty_strided_out ~out ~size ~stride =
-  let out__ = CArray.make t 1 in
   stubs_empty_strided_out
-    (CArray.start out__)
     out
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
-    (List.length stride);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let eq self other =
-  let out__ = CArray.make t 1 in
-  stubs_eq (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let eq_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_eq_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length stride)
+  |> with_tensor_gc
 ;;
 
-let eq_scalar_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_eq_scalar_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let eq self other = stubs_eq self other |> with_tensor_gc
+let eq_ self other = stubs_eq_ self other |> with_tensor_gc
+let eq_scalar_out ~out self other = stubs_eq_scalar_out out self other |> with_tensor_gc
+let eq_tensor self other = stubs_eq_tensor self other |> with_tensor_gc
+let eq_tensor_ self other = stubs_eq_tensor_ self other |> with_tensor_gc
+let eq_tensor_out ~out self other = stubs_eq_tensor_out out self other |> with_tensor_gc
+let equal self other = stubs_equal self other
+let erf self = stubs_erf self |> with_tensor_gc
+let erf_ self = stubs_erf_ self |> with_tensor_gc
+let erf_out ~out self = stubs_erf_out out self |> with_tensor_gc
+let erfc self = stubs_erfc self |> with_tensor_gc
+let erfc_ self = stubs_erfc_ self |> with_tensor_gc
+let erfc_out ~out self = stubs_erfc_out out self |> with_tensor_gc
+let erfinv self = stubs_erfinv self |> with_tensor_gc
+let erfinv_ self = stubs_erfinv_ self |> with_tensor_gc
+let erfinv_out ~out self = stubs_erfinv_out out self |> with_tensor_gc
+let exp self = stubs_exp self |> with_tensor_gc
+let exp2 self = stubs_exp2 self |> with_tensor_gc
+let exp2_ self = stubs_exp2_ self |> with_tensor_gc
+let exp2_out ~out self = stubs_exp2_out out self |> with_tensor_gc
+let exp_ self = stubs_exp_ self |> with_tensor_gc
+let exp_out ~out self = stubs_exp_out out self |> with_tensor_gc
 
-let eq_tensor self other =
-  let out__ = CArray.make t 1 in
-  stubs_eq_tensor (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+let expand self ~size ~implicit =
+  stubs_expand
+    self
+    (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
+    (List.length size)
+    (if implicit then 1 else 0)
+  |> with_tensor_gc
 ;;
 
-let eq_tensor_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_eq_tensor_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let eq_tensor_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_eq_tensor_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let equal self other = stubs_equal self other
-
-let erf self =
-  let out__ = CArray.make t 1 in
-  stubs_erf (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let erf_ self =
-  let out__ = CArray.make t 1 in
-  stubs_erf_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let erf_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_erf_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let erfc self =
-  let out__ = CArray.make t 1 in
-  stubs_erfc (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let erfc_ self =
-  let out__ = CArray.make t 1 in
-  stubs_erfc_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let erfc_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_erfc_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let erfinv self =
-  let out__ = CArray.make t 1 in
-  stubs_erfinv (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let erfinv_ self =
-  let out__ = CArray.make t 1 in
-  stubs_erfinv_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let erfinv_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_erfinv_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let exp self =
-  let out__ = CArray.make t 1 in
-  stubs_exp (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let exp2 self =
-  let out__ = CArray.make t 1 in
-  stubs_exp2 (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let exp2_ self =
-  let out__ = CArray.make t 1 in
-  stubs_exp2_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let exp2_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_exp2_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let exp_ self =
-  let out__ = CArray.make t 1 in
-  stubs_exp_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let exp_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_exp_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let expand self ~size ~implicit =
-  let out__ = CArray.make t 1 in
-  stubs_expand
-    (CArray.start out__)
-    self
-    (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
-    (List.length size)
-    (if implicit then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let expand_as self other =
-  let out__ = CArray.make t 1 in
-  stubs_expand_as (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let expand_as self other = stubs_expand_as self other |> with_tensor_gc
 
 let expand_copy self ~size ~implicit =
-  let out__ = CArray.make t 1 in
   stubs_expand_copy
-    (CArray.start out__)
     self
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
-    (if implicit then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if implicit then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let expand_copy_out ~out self ~size ~implicit =
-  let out__ = CArray.make t 1 in
   stubs_expand_copy_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
-    (if implicit then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if implicit then 1 else 0)
+  |> with_tensor_gc
 ;;
 
-let expm1 self =
-  let out__ = CArray.make t 1 in
-  stubs_expm1 (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let expm1_ self =
-  let out__ = CArray.make t 1 in
-  stubs_expm1_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let expm1_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_expm1_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let exponential self ~lambd =
-  let out__ = CArray.make t 1 in
-  stubs_exponential (CArray.start out__) self lambd;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let exponential_ self ~lambd =
-  let out__ = CArray.make t 1 in
-  stubs_exponential_ (CArray.start out__) self lambd;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let expm1 self = stubs_expm1 self |> with_tensor_gc
+let expm1_ self = stubs_expm1_ self |> with_tensor_gc
+let expm1_out ~out self = stubs_expm1_out out self |> with_tensor_gc
+let exponential self ~lambd = stubs_exponential self lambd |> with_tensor_gc
+let exponential_ self ~lambd = stubs_exponential_ self lambd |> with_tensor_gc
 
 let exponential_out ~out self ~lambd =
-  let out__ = CArray.make t 1 in
-  stubs_exponential_out (CArray.start out__) out self lambd;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_exponential_out out self lambd |> with_tensor_gc
 ;;
 
 let eye ~n ~options =
-  let out__ = CArray.make t 1 in
   stubs_eye
-    (CArray.start out__)
     (Int64.of_int n)
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let eye_m ~n ~m ~options =
-  let out__ = CArray.make t 1 in
   stubs_eye_m
-    (CArray.start out__)
     (Int64.of_int n)
     (Int64.of_int m)
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let eye_m_out ~out ~n ~m =
-  let out__ = CArray.make t 1 in
-  stubs_eye_m_out (CArray.start out__) out (Int64.of_int n) (Int64.of_int m);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_eye_m_out out (Int64.of_int n) (Int64.of_int m) |> with_tensor_gc
 ;;
 
-let eye_out ~out ~n =
-  let out__ = CArray.make t 1 in
-  stubs_eye_out (CArray.start out__) out (Int64.of_int n);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let eye_out ~out ~n = stubs_eye_out out (Int64.of_int n) |> with_tensor_gc
 
 let fake_quantize_per_channel_affine self ~scale ~zero_point ~axis ~quant_min ~quant_max =
-  let out__ = CArray.make t 1 in
   stubs_fake_quantize_per_channel_affine
-    (CArray.start out__)
     self
     scale
     zero_point
     (Int64.of_int axis)
     (Int64.of_int quant_min)
-    (Int64.of_int quant_max);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int quant_max)
+  |> with_tensor_gc
 ;;
 
 let fake_quantize_per_channel_affine_cachemask
@@ -14264,7 +9615,7 @@ let fake_quantize_per_channel_affine_cachemask
   ~quant_min
   ~quant_max
   =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_fake_quantize_per_channel_affine_cachemask
     (CArray.start out__)
     self
@@ -14273,19 +9624,13 @@ let fake_quantize_per_channel_affine_cachemask
     (Int64.of_int axis)
     (Int64.of_int quant_min)
     (Int64.of_int quant_max);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let fake_quantize_per_channel_affine_cachemask_backward ~grad ~mask =
-  let out__ = CArray.make t 1 in
-  stubs_fake_quantize_per_channel_affine_cachemask_backward (CArray.start out__) grad mask;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_fake_quantize_per_channel_affine_cachemask_backward grad mask |> with_tensor_gc
 ;;
 
 let fake_quantize_per_channel_affine_cachemask_out
@@ -14298,7 +9643,7 @@ let fake_quantize_per_channel_affine_cachemask_out
   ~quant_min
   ~quant_max
   =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_fake_quantize_per_channel_affine_cachemask_out
     (CArray.start out__)
     out0
@@ -14309,25 +9654,19 @@ let fake_quantize_per_channel_affine_cachemask_out
     (Int64.of_int axis)
     (Int64.of_int quant_min)
     (Int64.of_int quant_max);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let fake_quantize_per_tensor_affine self ~scale ~zero_point ~quant_min ~quant_max =
-  let out__ = CArray.make t 1 in
   stubs_fake_quantize_per_tensor_affine
-    (CArray.start out__)
     self
     scale
     (Int64.of_int zero_point)
     (Int64.of_int quant_min)
-    (Int64.of_int quant_max);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int quant_max)
+  |> with_tensor_gc
 ;;
 
 let fake_quantize_per_tensor_affine_cachemask
@@ -14337,7 +9676,7 @@ let fake_quantize_per_tensor_affine_cachemask
   ~quant_min
   ~quant_max
   =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_fake_quantize_per_tensor_affine_cachemask
     (CArray.start out__)
     self
@@ -14345,19 +9684,13 @@ let fake_quantize_per_tensor_affine_cachemask
     (Int64.of_int zero_point)
     (Int64.of_int quant_min)
     (Int64.of_int quant_max);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let fake_quantize_per_tensor_affine_cachemask_backward ~grad ~mask =
-  let out__ = CArray.make t 1 in
-  stubs_fake_quantize_per_tensor_affine_cachemask_backward (CArray.start out__) grad mask;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_fake_quantize_per_tensor_affine_cachemask_backward grad mask |> with_tensor_gc
 ;;
 
 let fake_quantize_per_tensor_affine_cachemask_out
@@ -14369,7 +9702,7 @@ let fake_quantize_per_tensor_affine_cachemask_out
   ~quant_min
   ~quant_max
   =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_fake_quantize_per_tensor_affine_cachemask_out
     (CArray.start out__)
     out0
@@ -14379,10 +9712,8 @@ let fake_quantize_per_tensor_affine_cachemask_out
     (Int64.of_int zero_point)
     (Int64.of_int quant_min)
     (Int64.of_int quant_max);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
@@ -14393,37 +9724,22 @@ let fake_quantize_per_tensor_affine_tensor_qparams
   ~quant_min
   ~quant_max
   =
-  let out__ = CArray.make t 1 in
   stubs_fake_quantize_per_tensor_affine_tensor_qparams
-    (CArray.start out__)
     self
     scale
     zero_point
     (Int64.of_int quant_min)
-    (Int64.of_int quant_max);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int quant_max)
+  |> with_tensor_gc
 ;;
 
 let fbgemm_linear_fp16_weight input ~packed_weight ~bias =
-  let out__ = CArray.make t 1 in
-  stubs_fbgemm_linear_fp16_weight (CArray.start out__) input packed_weight bias;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_fbgemm_linear_fp16_weight input packed_weight bias |> with_tensor_gc
 ;;
 
 let fbgemm_linear_fp16_weight_fp32_activation input ~packed_weight ~bias =
-  let out__ = CArray.make t 1 in
-  stubs_fbgemm_linear_fp16_weight_fp32_activation
-    (CArray.start out__)
-    input
-    packed_weight
-    bias;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_fbgemm_linear_fp16_weight_fp32_activation input packed_weight bias
+  |> with_tensor_gc
 ;;
 
 let fbgemm_linear_int8_weight
@@ -14435,19 +9751,15 @@ let fbgemm_linear_int8_weight
   ~weight_zero_point
   ~bias
   =
-  let out__ = CArray.make t 1 in
   stubs_fbgemm_linear_int8_weight
-    (CArray.start out__)
     input
     weight
     packed
     col_offsets
     weight_scale
     weight_zero_point
-    bias;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    bias
+  |> with_tensor_gc
 ;;
 
 let fbgemm_linear_int8_weight_fp32_activation
@@ -14459,85 +9771,48 @@ let fbgemm_linear_int8_weight_fp32_activation
   ~weight_zero_point
   ~bias
   =
-  let out__ = CArray.make t 1 in
   stubs_fbgemm_linear_int8_weight_fp32_activation
-    (CArray.start out__)
     input
     weight
     packed
     col_offsets
     weight_scale
     weight_zero_point
-    bias;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    bias
+  |> with_tensor_gc
 ;;
 
 let fbgemm_pack_gemm_matrix_fp16 input =
-  let out__ = CArray.make t 1 in
-  stubs_fbgemm_pack_gemm_matrix_fp16 (CArray.start out__) input;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_fbgemm_pack_gemm_matrix_fp16 input |> with_tensor_gc
 ;;
 
 let fbgemm_pack_quantized_matrix input =
-  let out__ = CArray.make t 1 in
-  stubs_fbgemm_pack_quantized_matrix (CArray.start out__) input;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_fbgemm_pack_quantized_matrix input |> with_tensor_gc
 ;;
 
 let fbgemm_pack_quantized_matrix_kn input ~k ~n =
-  let out__ = CArray.make t 1 in
-  stubs_fbgemm_pack_quantized_matrix_kn
-    (CArray.start out__)
-    input
-    (Int64.of_int k)
-    (Int64.of_int n);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_fbgemm_pack_quantized_matrix_kn input (Int64.of_int k) (Int64.of_int n)
+  |> with_tensor_gc
 ;;
 
 let feature_alpha_dropout input ~p ~train =
-  let out__ = CArray.make t 1 in
-  stubs_feature_alpha_dropout (CArray.start out__) input p (if train then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_feature_alpha_dropout input p (if train then 1 else 0) |> with_tensor_gc
 ;;
 
 let feature_alpha_dropout_ self ~p ~train =
-  let out__ = CArray.make t 1 in
-  stubs_feature_alpha_dropout_ (CArray.start out__) self p (if train then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_feature_alpha_dropout_ self p (if train then 1 else 0) |> with_tensor_gc
 ;;
 
 let feature_dropout input ~p ~train =
-  let out__ = CArray.make t 1 in
-  stubs_feature_dropout (CArray.start out__) input p (if train then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_feature_dropout input p (if train then 1 else 0) |> with_tensor_gc
 ;;
 
 let feature_dropout_ self ~p ~train =
-  let out__ = CArray.make t 1 in
-  stubs_feature_dropout_ (CArray.start out__) self p (if train then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_feature_dropout_ self p (if train then 1 else 0) |> with_tensor_gc
 ;;
 
 let fft_fft self ~n ~dim ~norm =
-  let out__ = CArray.make t 1 in
   stubs_fft_fft
-    (CArray.start out__)
     self
     (match n with
      | None -> Int64.zero
@@ -14546,16 +9821,12 @@ let fft_fft self ~n ~dim ~norm =
      | Some _ -> 0
      | None -> 1)
     (Int64.of_int dim)
-    norm;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    norm
+  |> with_tensor_gc
 ;;
 
 let fft_fft2 self ~s ~dim ~norm =
-  let out__ = CArray.make t 1 in
   stubs_fft_fft2
-    (CArray.start out__)
     self
     (match s with
      | None -> from_voidp int64_t null
@@ -14565,16 +9836,12 @@ let fft_fft2 self ~s ~dim ~norm =
      | Some v -> List.length v)
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
     (List.length dim)
-    norm;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    norm
+  |> with_tensor_gc
 ;;
 
 let fft_fft2_out ~out self ~s ~dim ~norm =
-  let out__ = CArray.make t 1 in
   stubs_fft_fft2_out
-    (CArray.start out__)
     out
     self
     (match s with
@@ -14585,16 +9852,12 @@ let fft_fft2_out ~out self ~s ~dim ~norm =
      | Some v -> List.length v)
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
     (List.length dim)
-    norm;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    norm
+  |> with_tensor_gc
 ;;
 
 let fft_fft_out ~out self ~n ~dim ~norm =
-  let out__ = CArray.make t 1 in
   stubs_fft_fft_out
-    (CArray.start out__)
     out
     self
     (match n with
@@ -14604,37 +9867,25 @@ let fft_fft_out ~out self ~n ~dim ~norm =
      | Some _ -> 0
      | None -> 1)
     (Int64.of_int dim)
-    norm;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    norm
+  |> with_tensor_gc
 ;;
 
 let fft_fftfreq ~n ~d ~options =
-  let out__ = CArray.make t 1 in
   stubs_fft_fftfreq
-    (CArray.start out__)
     (Int64.of_int n)
     d
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let fft_fftfreq_out ~out ~n ~d =
-  let out__ = CArray.make t 1 in
-  stubs_fft_fftfreq_out (CArray.start out__) out (Int64.of_int n) d;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_fft_fftfreq_out out (Int64.of_int n) d |> with_tensor_gc
 ;;
 
 let fft_fftn self ~s ~dim ~norm =
-  let out__ = CArray.make t 1 in
   stubs_fft_fftn
-    (CArray.start out__)
     self
     (match s with
      | None -> from_voidp int64_t null
@@ -14648,16 +9899,12 @@ let fft_fftn self ~s ~dim ~norm =
     (match dim with
      | None -> -1
      | Some v -> List.length v)
-    norm;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    norm
+  |> with_tensor_gc
 ;;
 
 let fft_fftn_out ~out self ~s ~dim ~norm =
-  let out__ = CArray.make t 1 in
   stubs_fft_fftn_out
-    (CArray.start out__)
     out
     self
     (match s with
@@ -14672,32 +9919,24 @@ let fft_fftn_out ~out self ~s ~dim ~norm =
     (match dim with
      | None -> -1
      | Some v -> List.length v)
-    norm;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    norm
+  |> with_tensor_gc
 ;;
 
 let fft_fftshift self ~dim =
-  let out__ = CArray.make t 1 in
   stubs_fft_fftshift
-    (CArray.start out__)
     self
     (match dim with
      | None -> from_voidp int64_t null
      | Some v -> List.map Int64.of_int v |> CArray.of_list int64_t |> CArray.start)
     (match dim with
      | None -> -1
-     | Some v -> List.length v);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | Some v -> List.length v)
+  |> with_tensor_gc
 ;;
 
 let fft_hfft self ~n ~dim ~norm =
-  let out__ = CArray.make t 1 in
   stubs_fft_hfft
-    (CArray.start out__)
     self
     (match n with
      | None -> Int64.zero
@@ -14706,16 +9945,12 @@ let fft_hfft self ~n ~dim ~norm =
      | Some _ -> 0
      | None -> 1)
     (Int64.of_int dim)
-    norm;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    norm
+  |> with_tensor_gc
 ;;
 
 let fft_hfft2 self ~s ~dim ~norm =
-  let out__ = CArray.make t 1 in
   stubs_fft_hfft2
-    (CArray.start out__)
     self
     (match s with
      | None -> from_voidp int64_t null
@@ -14725,16 +9960,12 @@ let fft_hfft2 self ~s ~dim ~norm =
      | Some v -> List.length v)
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
     (List.length dim)
-    norm;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    norm
+  |> with_tensor_gc
 ;;
 
 let fft_hfft2_out ~out self ~s ~dim ~norm =
-  let out__ = CArray.make t 1 in
   stubs_fft_hfft2_out
-    (CArray.start out__)
     out
     self
     (match s with
@@ -14745,16 +9976,12 @@ let fft_hfft2_out ~out self ~s ~dim ~norm =
      | Some v -> List.length v)
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
     (List.length dim)
-    norm;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    norm
+  |> with_tensor_gc
 ;;
 
 let fft_hfft_out ~out self ~n ~dim ~norm =
-  let out__ = CArray.make t 1 in
   stubs_fft_hfft_out
-    (CArray.start out__)
     out
     self
     (match n with
@@ -14764,16 +9991,12 @@ let fft_hfft_out ~out self ~n ~dim ~norm =
      | Some _ -> 0
      | None -> 1)
     (Int64.of_int dim)
-    norm;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    norm
+  |> with_tensor_gc
 ;;
 
 let fft_hfftn self ~s ~dim ~norm =
-  let out__ = CArray.make t 1 in
   stubs_fft_hfftn
-    (CArray.start out__)
     self
     (match s with
      | None -> from_voidp int64_t null
@@ -14787,16 +10010,12 @@ let fft_hfftn self ~s ~dim ~norm =
     (match dim with
      | None -> -1
      | Some v -> List.length v)
-    norm;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    norm
+  |> with_tensor_gc
 ;;
 
 let fft_hfftn_out ~out self ~s ~dim ~norm =
-  let out__ = CArray.make t 1 in
   stubs_fft_hfftn_out
-    (CArray.start out__)
     out
     self
     (match s with
@@ -14811,16 +10030,12 @@ let fft_hfftn_out ~out self ~s ~dim ~norm =
     (match dim with
      | None -> -1
      | Some v -> List.length v)
-    norm;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    norm
+  |> with_tensor_gc
 ;;
 
 let fft_ifft self ~n ~dim ~norm =
-  let out__ = CArray.make t 1 in
   stubs_fft_ifft
-    (CArray.start out__)
     self
     (match n with
      | None -> Int64.zero
@@ -14829,16 +10044,12 @@ let fft_ifft self ~n ~dim ~norm =
      | Some _ -> 0
      | None -> 1)
     (Int64.of_int dim)
-    norm;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    norm
+  |> with_tensor_gc
 ;;
 
 let fft_ifft2 self ~s ~dim ~norm =
-  let out__ = CArray.make t 1 in
   stubs_fft_ifft2
-    (CArray.start out__)
     self
     (match s with
      | None -> from_voidp int64_t null
@@ -14848,16 +10059,12 @@ let fft_ifft2 self ~s ~dim ~norm =
      | Some v -> List.length v)
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
     (List.length dim)
-    norm;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    norm
+  |> with_tensor_gc
 ;;
 
 let fft_ifft2_out ~out self ~s ~dim ~norm =
-  let out__ = CArray.make t 1 in
   stubs_fft_ifft2_out
-    (CArray.start out__)
     out
     self
     (match s with
@@ -14868,16 +10075,12 @@ let fft_ifft2_out ~out self ~s ~dim ~norm =
      | Some v -> List.length v)
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
     (List.length dim)
-    norm;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    norm
+  |> with_tensor_gc
 ;;
 
 let fft_ifft_out ~out self ~n ~dim ~norm =
-  let out__ = CArray.make t 1 in
   stubs_fft_ifft_out
-    (CArray.start out__)
     out
     self
     (match n with
@@ -14887,16 +10090,12 @@ let fft_ifft_out ~out self ~n ~dim ~norm =
      | Some _ -> 0
      | None -> 1)
     (Int64.of_int dim)
-    norm;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    norm
+  |> with_tensor_gc
 ;;
 
 let fft_ifftn self ~s ~dim ~norm =
-  let out__ = CArray.make t 1 in
   stubs_fft_ifftn
-    (CArray.start out__)
     self
     (match s with
      | None -> from_voidp int64_t null
@@ -14910,16 +10109,12 @@ let fft_ifftn self ~s ~dim ~norm =
     (match dim with
      | None -> -1
      | Some v -> List.length v)
-    norm;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    norm
+  |> with_tensor_gc
 ;;
 
 let fft_ifftn_out ~out self ~s ~dim ~norm =
-  let out__ = CArray.make t 1 in
   stubs_fft_ifftn_out
-    (CArray.start out__)
     out
     self
     (match s with
@@ -14934,32 +10129,24 @@ let fft_ifftn_out ~out self ~s ~dim ~norm =
     (match dim with
      | None -> -1
      | Some v -> List.length v)
-    norm;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    norm
+  |> with_tensor_gc
 ;;
 
 let fft_ifftshift self ~dim =
-  let out__ = CArray.make t 1 in
   stubs_fft_ifftshift
-    (CArray.start out__)
     self
     (match dim with
      | None -> from_voidp int64_t null
      | Some v -> List.map Int64.of_int v |> CArray.of_list int64_t |> CArray.start)
     (match dim with
      | None -> -1
-     | Some v -> List.length v);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | Some v -> List.length v)
+  |> with_tensor_gc
 ;;
 
 let fft_ihfft self ~n ~dim ~norm =
-  let out__ = CArray.make t 1 in
   stubs_fft_ihfft
-    (CArray.start out__)
     self
     (match n with
      | None -> Int64.zero
@@ -14968,16 +10155,12 @@ let fft_ihfft self ~n ~dim ~norm =
      | Some _ -> 0
      | None -> 1)
     (Int64.of_int dim)
-    norm;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    norm
+  |> with_tensor_gc
 ;;
 
 let fft_ihfft2 self ~s ~dim ~norm =
-  let out__ = CArray.make t 1 in
   stubs_fft_ihfft2
-    (CArray.start out__)
     self
     (match s with
      | None -> from_voidp int64_t null
@@ -14987,16 +10170,12 @@ let fft_ihfft2 self ~s ~dim ~norm =
      | Some v -> List.length v)
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
     (List.length dim)
-    norm;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    norm
+  |> with_tensor_gc
 ;;
 
 let fft_ihfft2_out ~out self ~s ~dim ~norm =
-  let out__ = CArray.make t 1 in
   stubs_fft_ihfft2_out
-    (CArray.start out__)
     out
     self
     (match s with
@@ -15007,16 +10186,12 @@ let fft_ihfft2_out ~out self ~s ~dim ~norm =
      | Some v -> List.length v)
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
     (List.length dim)
-    norm;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    norm
+  |> with_tensor_gc
 ;;
 
 let fft_ihfft_out ~out self ~n ~dim ~norm =
-  let out__ = CArray.make t 1 in
   stubs_fft_ihfft_out
-    (CArray.start out__)
     out
     self
     (match n with
@@ -15026,16 +10201,12 @@ let fft_ihfft_out ~out self ~n ~dim ~norm =
      | Some _ -> 0
      | None -> 1)
     (Int64.of_int dim)
-    norm;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    norm
+  |> with_tensor_gc
 ;;
 
 let fft_ihfftn self ~s ~dim ~norm =
-  let out__ = CArray.make t 1 in
   stubs_fft_ihfftn
-    (CArray.start out__)
     self
     (match s with
      | None -> from_voidp int64_t null
@@ -15049,16 +10220,12 @@ let fft_ihfftn self ~s ~dim ~norm =
     (match dim with
      | None -> -1
      | Some v -> List.length v)
-    norm;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    norm
+  |> with_tensor_gc
 ;;
 
 let fft_ihfftn_out ~out self ~s ~dim ~norm =
-  let out__ = CArray.make t 1 in
   stubs_fft_ihfftn_out
-    (CArray.start out__)
     out
     self
     (match s with
@@ -15073,16 +10240,12 @@ let fft_ihfftn_out ~out self ~s ~dim ~norm =
     (match dim with
      | None -> -1
      | Some v -> List.length v)
-    norm;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    norm
+  |> with_tensor_gc
 ;;
 
 let fft_irfft self ~n ~dim ~norm =
-  let out__ = CArray.make t 1 in
   stubs_fft_irfft
-    (CArray.start out__)
     self
     (match n with
      | None -> Int64.zero
@@ -15091,16 +10254,12 @@ let fft_irfft self ~n ~dim ~norm =
      | Some _ -> 0
      | None -> 1)
     (Int64.of_int dim)
-    norm;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    norm
+  |> with_tensor_gc
 ;;
 
 let fft_irfft2 self ~s ~dim ~norm =
-  let out__ = CArray.make t 1 in
   stubs_fft_irfft2
-    (CArray.start out__)
     self
     (match s with
      | None -> from_voidp int64_t null
@@ -15110,16 +10269,12 @@ let fft_irfft2 self ~s ~dim ~norm =
      | Some v -> List.length v)
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
     (List.length dim)
-    norm;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    norm
+  |> with_tensor_gc
 ;;
 
 let fft_irfft2_out ~out self ~s ~dim ~norm =
-  let out__ = CArray.make t 1 in
   stubs_fft_irfft2_out
-    (CArray.start out__)
     out
     self
     (match s with
@@ -15130,16 +10285,12 @@ let fft_irfft2_out ~out self ~s ~dim ~norm =
      | Some v -> List.length v)
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
     (List.length dim)
-    norm;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    norm
+  |> with_tensor_gc
 ;;
 
 let fft_irfft_out ~out self ~n ~dim ~norm =
-  let out__ = CArray.make t 1 in
   stubs_fft_irfft_out
-    (CArray.start out__)
     out
     self
     (match n with
@@ -15149,16 +10300,12 @@ let fft_irfft_out ~out self ~n ~dim ~norm =
      | Some _ -> 0
      | None -> 1)
     (Int64.of_int dim)
-    norm;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    norm
+  |> with_tensor_gc
 ;;
 
 let fft_irfftn self ~s ~dim ~norm =
-  let out__ = CArray.make t 1 in
   stubs_fft_irfftn
-    (CArray.start out__)
     self
     (match s with
      | None -> from_voidp int64_t null
@@ -15172,16 +10319,12 @@ let fft_irfftn self ~s ~dim ~norm =
     (match dim with
      | None -> -1
      | Some v -> List.length v)
-    norm;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    norm
+  |> with_tensor_gc
 ;;
 
 let fft_irfftn_out ~out self ~s ~dim ~norm =
-  let out__ = CArray.make t 1 in
   stubs_fft_irfftn_out
-    (CArray.start out__)
     out
     self
     (match s with
@@ -15196,16 +10339,12 @@ let fft_irfftn_out ~out self ~s ~dim ~norm =
     (match dim with
      | None -> -1
      | Some v -> List.length v)
-    norm;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    norm
+  |> with_tensor_gc
 ;;
 
 let fft_rfft self ~n ~dim ~norm =
-  let out__ = CArray.make t 1 in
   stubs_fft_rfft
-    (CArray.start out__)
     self
     (match n with
      | None -> Int64.zero
@@ -15214,16 +10353,12 @@ let fft_rfft self ~n ~dim ~norm =
      | Some _ -> 0
      | None -> 1)
     (Int64.of_int dim)
-    norm;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    norm
+  |> with_tensor_gc
 ;;
 
 let fft_rfft2 self ~s ~dim ~norm =
-  let out__ = CArray.make t 1 in
   stubs_fft_rfft2
-    (CArray.start out__)
     self
     (match s with
      | None -> from_voidp int64_t null
@@ -15233,16 +10368,12 @@ let fft_rfft2 self ~s ~dim ~norm =
      | Some v -> List.length v)
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
     (List.length dim)
-    norm;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    norm
+  |> with_tensor_gc
 ;;
 
 let fft_rfft2_out ~out self ~s ~dim ~norm =
-  let out__ = CArray.make t 1 in
   stubs_fft_rfft2_out
-    (CArray.start out__)
     out
     self
     (match s with
@@ -15253,16 +10384,12 @@ let fft_rfft2_out ~out self ~s ~dim ~norm =
      | Some v -> List.length v)
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
     (List.length dim)
-    norm;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    norm
+  |> with_tensor_gc
 ;;
 
 let fft_rfft_out ~out self ~n ~dim ~norm =
-  let out__ = CArray.make t 1 in
   stubs_fft_rfft_out
-    (CArray.start out__)
     out
     self
     (match n with
@@ -15272,37 +10399,25 @@ let fft_rfft_out ~out self ~n ~dim ~norm =
      | Some _ -> 0
      | None -> 1)
     (Int64.of_int dim)
-    norm;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    norm
+  |> with_tensor_gc
 ;;
 
 let fft_rfftfreq ~n ~d ~options =
-  let out__ = CArray.make t 1 in
   stubs_fft_rfftfreq
-    (CArray.start out__)
     (Int64.of_int n)
     d
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let fft_rfftfreq_out ~out ~n ~d =
-  let out__ = CArray.make t 1 in
-  stubs_fft_rfftfreq_out (CArray.start out__) out (Int64.of_int n) d;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_fft_rfftfreq_out out (Int64.of_int n) d |> with_tensor_gc
 ;;
 
 let fft_rfftn self ~s ~dim ~norm =
-  let out__ = CArray.make t 1 in
   stubs_fft_rfftn
-    (CArray.start out__)
     self
     (match s with
      | None -> from_voidp int64_t null
@@ -15316,16 +10431,12 @@ let fft_rfftn self ~s ~dim ~norm =
     (match dim with
      | None -> -1
      | Some v -> List.length v)
-    norm;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    norm
+  |> with_tensor_gc
 ;;
 
 let fft_rfftn_out ~out self ~s ~dim ~norm =
-  let out__ = CArray.make t 1 in
   stubs_fft_rfftn_out
-    (CArray.start out__)
     out
     self
     (match s with
@@ -15340,386 +10451,131 @@ let fft_rfftn_out ~out self ~s ~dim ~norm =
     (match dim with
      | None -> -1
      | Some v -> List.length v)
-    norm;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    norm
+  |> with_tensor_gc
 ;;
 
-let fill self ~value =
-  let out__ = CArray.make t 1 in
-  stubs_fill (CArray.start out__) self value;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let fill_ self ~value =
-  let out__ = CArray.make t 1 in
-  stubs_fill_ (CArray.start out__) self value;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let fill self ~value = stubs_fill self value |> with_tensor_gc
+let fill_ self ~value = stubs_fill_ self value |> with_tensor_gc
 
 let fill_diagonal_ self ~fill_value ~wrap =
-  let out__ = CArray.make t 1 in
-  stubs_fill_diagonal_ (CArray.start out__) self fill_value (if wrap then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_fill_diagonal_ self fill_value (if wrap then 1 else 0) |> with_tensor_gc
 ;;
 
 let fill_scalar_out ~out self ~value =
-  let out__ = CArray.make t 1 in
-  stubs_fill_scalar_out (CArray.start out__) out self value;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_fill_scalar_out out self value |> with_tensor_gc
 ;;
 
-let fill_tensor self ~value =
-  let out__ = CArray.make t 1 in
-  stubs_fill_tensor (CArray.start out__) self value;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let fill_tensor_ self ~value =
-  let out__ = CArray.make t 1 in
-  stubs_fill_tensor_ (CArray.start out__) self value;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let fill_tensor self ~value = stubs_fill_tensor self value |> with_tensor_gc
+let fill_tensor_ self ~value = stubs_fill_tensor_ self value |> with_tensor_gc
 
 let fill_tensor_out ~out self ~value =
-  let out__ = CArray.make t 1 in
-  stubs_fill_tensor_out (CArray.start out__) out self value;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_fill_tensor_out out self value |> with_tensor_gc
 ;;
 
-let fix self =
-  let out__ = CArray.make t 1 in
-  stubs_fix (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let fix_ self =
-  let out__ = CArray.make t 1 in
-  stubs_fix_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let fix_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_fix_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let fix self = stubs_fix self |> with_tensor_gc
+let fix_ self = stubs_fix_ self |> with_tensor_gc
+let fix_out ~out self = stubs_fix_out out self |> with_tensor_gc
 
 let flatten self ~start_dim ~end_dim =
-  let out__ = CArray.make t 1 in
-  stubs_flatten (CArray.start out__) self (Int64.of_int start_dim) (Int64.of_int end_dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_flatten self (Int64.of_int start_dim) (Int64.of_int end_dim) |> with_tensor_gc
 ;;
 
 let flatten_dense_tensors tensors =
-  let out__ = CArray.make t 1 in
   stubs_flatten_dense_tensors
-    (CArray.start out__)
-    (CArray.of_list t tensors |> CArray.start)
-    (List.length tensors);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (CArray.of_list gc_tensor tensors |> CArray.start)
+    (List.length tensors)
+  |> with_tensor_gc
 ;;
 
 let flip self ~dims =
-  let out__ = CArray.make t 1 in
   stubs_flip
-    (CArray.start out__)
     self
     (List.map Int64.of_int dims |> CArray.of_list int64_t |> CArray.start)
-    (List.length dims);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length dims)
+  |> with_tensor_gc
 ;;
 
 let flip_out ~out self ~dims =
-  let out__ = CArray.make t 1 in
   stubs_flip_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int dims |> CArray.of_list int64_t |> CArray.start)
-    (List.length dims);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length dims)
+  |> with_tensor_gc
 ;;
 
-let fliplr self =
-  let out__ = CArray.make t 1 in
-  stubs_fliplr (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let flipud self =
-  let out__ = CArray.make t 1 in
-  stubs_flipud (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let float_power self ~exponent =
-  let out__ = CArray.make t 1 in
-  stubs_float_power (CArray.start out__) self exponent;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let float_power_ self ~exponent =
-  let out__ = CArray.make t 1 in
-  stubs_float_power_ (CArray.start out__) self exponent;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let fliplr self = stubs_fliplr self |> with_tensor_gc
+let flipud self = stubs_flipud self |> with_tensor_gc
+let float_power self ~exponent = stubs_float_power self exponent |> with_tensor_gc
+let float_power_ self ~exponent = stubs_float_power_ self exponent |> with_tensor_gc
 
 let float_power_scalar self ~exponent =
-  let out__ = CArray.make t 1 in
-  stubs_float_power_scalar (CArray.start out__) self exponent;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_float_power_scalar self exponent |> with_tensor_gc
 ;;
 
 let float_power_scalar_out ~out self ~exponent =
-  let out__ = CArray.make t 1 in
-  stubs_float_power_scalar_out (CArray.start out__) out self exponent;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_float_power_scalar_out out self exponent |> with_tensor_gc
 ;;
 
 let float_power_tensor_ self ~exponent =
-  let out__ = CArray.make t 1 in
-  stubs_float_power_tensor_ (CArray.start out__) self exponent;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_float_power_tensor_ self exponent |> with_tensor_gc
 ;;
 
 let float_power_tensor_scalar self ~exponent =
-  let out__ = CArray.make t 1 in
-  stubs_float_power_tensor_scalar (CArray.start out__) self exponent;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_float_power_tensor_scalar self exponent |> with_tensor_gc
 ;;
 
 let float_power_tensor_scalar_out ~out self ~exponent =
-  let out__ = CArray.make t 1 in
-  stubs_float_power_tensor_scalar_out (CArray.start out__) out self exponent;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_float_power_tensor_scalar_out out self exponent |> with_tensor_gc
 ;;
 
 let float_power_tensor_tensor_out ~out self ~exponent =
-  let out__ = CArray.make t 1 in
-  stubs_float_power_tensor_tensor_out (CArray.start out__) out self exponent;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_float_power_tensor_tensor_out out self exponent |> with_tensor_gc
 ;;
 
-let floor self =
-  let out__ = CArray.make t 1 in
-  stubs_floor (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let floor_ self =
-  let out__ = CArray.make t 1 in
-  stubs_floor_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let floor_divide self other =
-  let out__ = CArray.make t 1 in
-  stubs_floor_divide (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let floor_divide_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_floor_divide_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let floor self = stubs_floor self |> with_tensor_gc
+let floor_ self = stubs_floor_ self |> with_tensor_gc
+let floor_divide self other = stubs_floor_divide self other |> with_tensor_gc
+let floor_divide_ self other = stubs_floor_divide_ self other |> with_tensor_gc
 
 let floor_divide_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_floor_divide_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_floor_divide_out out self other |> with_tensor_gc
 ;;
 
 let floor_divide_scalar self other =
-  let out__ = CArray.make t 1 in
-  stubs_floor_divide_scalar (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_floor_divide_scalar self other |> with_tensor_gc
 ;;
 
 let floor_divide_scalar_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_floor_divide_scalar_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let floor_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_floor_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let fmax self other =
-  let out__ = CArray.make t 1 in
-  stubs_fmax (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let fmax_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_fmax_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let fmin self other =
-  let out__ = CArray.make t 1 in
-  stubs_fmin (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_floor_divide_scalar_ self other |> with_tensor_gc
 ;;
 
-let fmin_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_fmin_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let fmod self other =
-  let out__ = CArray.make t 1 in
-  stubs_fmod (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let fmod_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_fmod_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let floor_out ~out self = stubs_floor_out out self |> with_tensor_gc
+let fmax self other = stubs_fmax self other |> with_tensor_gc
+let fmax_out ~out self other = stubs_fmax_out out self other |> with_tensor_gc
+let fmin self other = stubs_fmin self other |> with_tensor_gc
+let fmin_out ~out self other = stubs_fmin_out out self other |> with_tensor_gc
+let fmod self other = stubs_fmod self other |> with_tensor_gc
+let fmod_ self other = stubs_fmod_ self other |> with_tensor_gc
 
 let fmod_scalar_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_fmod_scalar_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let fmod_tensor self other =
-  let out__ = CArray.make t 1 in
-  stubs_fmod_tensor (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_fmod_scalar_out out self other |> with_tensor_gc
 ;;
 
-let fmod_tensor_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_fmod_tensor_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let fmod_tensor self other = stubs_fmod_tensor self other |> with_tensor_gc
+let fmod_tensor_ self other = stubs_fmod_tensor_ self other |> with_tensor_gc
 
 let fmod_tensor_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_fmod_tensor_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let frac self =
-  let out__ = CArray.make t 1 in
-  stubs_frac (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_fmod_tensor_out out self other |> with_tensor_gc
 ;;
 
-let frac_ self =
-  let out__ = CArray.make t 1 in
-  stubs_frac_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let frac_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_frac_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let frac self = stubs_frac self |> with_tensor_gc
+let frac_ self = stubs_frac_ self |> with_tensor_gc
+let frac_out ~out self = stubs_frac_out out self |> with_tensor_gc
 
 let fractional_max_pool2d self ~kernel_size ~output_size ~random_samples =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_fractional_max_pool2d
     (CArray.start out__)
     self
@@ -15728,27 +10584,21 @@ let fractional_max_pool2d self ~kernel_size ~output_size ~random_samples =
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
     (List.length output_size)
     random_samples;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let fractional_max_pool2d_backward ~grad_output self ~kernel_size ~output_size ~indices =
-  let out__ = CArray.make t 1 in
   stubs_fractional_max_pool2d_backward
-    (CArray.start out__)
     grad_output
     self
     (List.map Int64.of_int kernel_size |> CArray.of_list int64_t |> CArray.start)
     (List.length kernel_size)
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
     (List.length output_size)
-    indices;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    indices
+  |> with_tensor_gc
 ;;
 
 let fractional_max_pool2d_backward_grad_input
@@ -15759,9 +10609,7 @@ let fractional_max_pool2d_backward_grad_input
   ~output_size
   ~indices
   =
-  let out__ = CArray.make t 1 in
   stubs_fractional_max_pool2d_backward_grad_input
-    (CArray.start out__)
     grad_input
     grad_output
     self
@@ -15769,10 +10617,8 @@ let fractional_max_pool2d_backward_grad_input
     (List.length kernel_size)
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
     (List.length output_size)
-    indices;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    indices
+  |> with_tensor_gc
 ;;
 
 let fractional_max_pool2d_output
@@ -15783,7 +10629,7 @@ let fractional_max_pool2d_output
   ~output_size
   ~random_samples
   =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_fractional_max_pool2d_output
     (CArray.start out__)
     output
@@ -15794,15 +10640,13 @@ let fractional_max_pool2d_output
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
     (List.length output_size)
     random_samples;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let fractional_max_pool3d self ~kernel_size ~output_size ~random_samples =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_fractional_max_pool3d
     (CArray.start out__)
     self
@@ -15811,27 +10655,21 @@ let fractional_max_pool3d self ~kernel_size ~output_size ~random_samples =
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
     (List.length output_size)
     random_samples;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let fractional_max_pool3d_backward ~grad_output self ~kernel_size ~output_size ~indices =
-  let out__ = CArray.make t 1 in
   stubs_fractional_max_pool3d_backward
-    (CArray.start out__)
     grad_output
     self
     (List.map Int64.of_int kernel_size |> CArray.of_list int64_t |> CArray.start)
     (List.length kernel_size)
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
     (List.length output_size)
-    indices;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    indices
+  |> with_tensor_gc
 ;;
 
 let fractional_max_pool3d_backward_grad_input
@@ -15842,9 +10680,7 @@ let fractional_max_pool3d_backward_grad_input
   ~output_size
   ~indices
   =
-  let out__ = CArray.make t 1 in
   stubs_fractional_max_pool3d_backward_grad_input
-    (CArray.start out__)
     grad_input
     grad_output
     self
@@ -15852,10 +10688,8 @@ let fractional_max_pool3d_backward_grad_input
     (List.length kernel_size)
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
     (List.length output_size)
-    indices;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    indices
+  |> with_tensor_gc
 ;;
 
 let fractional_max_pool3d_output
@@ -15866,7 +10700,7 @@ let fractional_max_pool3d_output
   ~output_size
   ~random_samples
   =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_fractional_max_pool3d_output
     (CArray.start out__)
     output
@@ -15877,64 +10711,48 @@ let fractional_max_pool3d_output
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
     (List.length output_size)
     random_samples;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let frexp self =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_frexp (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let frexp_tensor_out ~mantissa ~exponent self =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_frexp_tensor_out (CArray.start out__) mantissa exponent self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let frobenius_norm self ~dim ~keepdim =
-  let out__ = CArray.make t 1 in
   stubs_frobenius_norm
-    (CArray.start out__)
     self
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
     (List.length dim)
-    (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if keepdim then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let frobenius_norm_out ~out self ~dim ~keepdim =
-  let out__ = CArray.make t 1 in
   stubs_frobenius_norm_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
     (List.length dim)
-    (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if keepdim then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let from_file ~filename ~shared ~size ~options =
-  let out__ = CArray.make t 1 in
   stubs_from_file
-    (CArray.start out__)
     filename
     (if shared then 1 else 0)
     (match size with
@@ -15944,16 +10762,12 @@ let from_file ~filename ~shared ~size ~options =
      | Some _ -> 0
      | None -> 1)
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let from_file_out ~out ~filename ~shared ~size =
-  let out__ = CArray.make t 1 in
   stubs_from_file_out
-    (CArray.start out__)
     out
     filename
     (if shared then 1 else 0)
@@ -15962,53 +10776,33 @@ let from_file_out ~out ~filename ~shared ~size =
      | Some v -> Int64.of_int v)
     (match size with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let full ~size ~fill_value ~options =
-  let out__ = CArray.make t 1 in
   stubs_full
-    (CArray.start out__)
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     fill_value
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
-let full_like self ~fill_value =
-  let out__ = CArray.make t 1 in
-  stubs_full_like (CArray.start out__) self fill_value;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let full_like self ~fill_value = stubs_full_like self fill_value |> with_tensor_gc
 
 let full_like_out ~out self ~fill_value =
-  let out__ = CArray.make t 1 in
-  stubs_full_like_out (CArray.start out__) out self fill_value;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_full_like_out out self fill_value |> with_tensor_gc
 ;;
 
 let full_out ~out ~size ~fill_value =
-  let out__ = CArray.make t 1 in
   stubs_full_out
-    (CArray.start out__)
     out
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
-    fill_value;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    fill_value
+  |> with_tensor_gc
 ;;
 
 let fused_moving_avg_obs_fake_quant
@@ -16026,9 +10820,7 @@ let fused_moving_avg_obs_fake_quant
   ~per_row_fake_quant
   ~symmetric_quant
   =
-  let out__ = CArray.make t 1 in
   stubs_fused_moving_avg_obs_fake_quant
-    (CArray.start out__)
     self
     observer_on
     fake_quant_on
@@ -16041,507 +10833,197 @@ let fused_moving_avg_obs_fake_quant
     (Int64.of_int quant_max)
     (Int64.of_int ch_axis)
     (if per_row_fake_quant then 1 else 0)
-    (if symmetric_quant then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if symmetric_quant then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let gather self ~dim ~index ~sparse_grad =
-  let out__ = CArray.make t 1 in
-  stubs_gather
-    (CArray.start out__)
-    self
-    (Int64.of_int dim)
-    index
-    (if sparse_grad then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_gather self (Int64.of_int dim) index (if sparse_grad then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let gather_backward ~grad self ~dim ~index ~sparse_grad =
-  let out__ = CArray.make t 1 in
-  stubs_gather_backward
-    (CArray.start out__)
-    grad
-    self
-    (Int64.of_int dim)
-    index
-    (if sparse_grad then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_gather_backward grad self (Int64.of_int dim) index (if sparse_grad then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let gather_out ~out self ~dim ~index ~sparse_grad =
-  let out__ = CArray.make t 1 in
-  stubs_gather_out
-    (CArray.start out__)
-    out
-    self
-    (Int64.of_int dim)
-    index
-    (if sparse_grad then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let gcd self other =
-  let out__ = CArray.make t 1 in
-  stubs_gcd (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let gcd_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_gcd_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let gcd_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_gcd_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let ge self other =
-  let out__ = CArray.make t 1 in
-  stubs_ge (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let ge_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_ge_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let ge_scalar_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_ge_scalar_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let ge_tensor self other =
-  let out__ = CArray.make t 1 in
-  stubs_ge_tensor (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let ge_tensor_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_ge_tensor_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let ge_tensor_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_ge_tensor_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let gelu self ~approximate =
-  let out__ = CArray.make t 1 in
-  stubs_gelu (CArray.start out__) self approximate;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let gelu_ self ~approximate =
-  let out__ = CArray.make t 1 in
-  stubs_gelu_ (CArray.start out__) self approximate;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+  stubs_gather_out out self (Int64.of_int dim) index (if sparse_grad then 1 else 0)
+  |> with_tensor_gc
+;;
+
+let gcd self other = stubs_gcd self other |> with_tensor_gc
+let gcd_ self other = stubs_gcd_ self other |> with_tensor_gc
+let gcd_out ~out self other = stubs_gcd_out out self other |> with_tensor_gc
+let ge self other = stubs_ge self other |> with_tensor_gc
+let ge_ self other = stubs_ge_ self other |> with_tensor_gc
+let ge_scalar_out ~out self other = stubs_ge_scalar_out out self other |> with_tensor_gc
+let ge_tensor self other = stubs_ge_tensor self other |> with_tensor_gc
+let ge_tensor_ self other = stubs_ge_tensor_ self other |> with_tensor_gc
+let ge_tensor_out ~out self other = stubs_ge_tensor_out out self other |> with_tensor_gc
+let gelu self ~approximate = stubs_gelu self approximate |> with_tensor_gc
+let gelu_ self ~approximate = stubs_gelu_ self approximate |> with_tensor_gc
 
 let gelu_backward ~grad_output self ~approximate =
-  let out__ = CArray.make t 1 in
-  stubs_gelu_backward (CArray.start out__) grad_output self approximate;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_gelu_backward grad_output self approximate |> with_tensor_gc
 ;;
 
 let gelu_backward_grad_input ~grad_input ~grad_output self ~approximate =
-  let out__ = CArray.make t 1 in
-  stubs_gelu_backward_grad_input
-    (CArray.start out__)
-    grad_input
-    grad_output
-    self
-    approximate;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_gelu_backward_grad_input grad_input grad_output self approximate |> with_tensor_gc
 ;;
 
 let gelu_out ~out self ~approximate =
-  let out__ = CArray.make t 1 in
-  stubs_gelu_out (CArray.start out__) out self approximate;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_gelu_out out self approximate |> with_tensor_gc
 ;;
 
-let geometric self ~p =
-  let out__ = CArray.make t 1 in
-  stubs_geometric (CArray.start out__) self p;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let geometric_ self ~p =
-  let out__ = CArray.make t 1 in
-  stubs_geometric_ (CArray.start out__) self p;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let geometric_out ~out self ~p =
-  let out__ = CArray.make t 1 in
-  stubs_geometric_out (CArray.start out__) out self p;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let geometric self ~p = stubs_geometric self p |> with_tensor_gc
+let geometric_ self ~p = stubs_geometric_ self p |> with_tensor_gc
+let geometric_out ~out self ~p = stubs_geometric_out out self p |> with_tensor_gc
 
 let geqrf self =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_geqrf (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let geqrf_a ~a ~tau self =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_geqrf_a (CArray.start out__) a tau self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
-let ger self ~vec2 =
-  let out__ = CArray.make t 1 in
-  stubs_ger (CArray.start out__) self vec2;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let ger_out ~out self ~vec2 =
-  let out__ = CArray.make t 1 in
-  stubs_ger_out (CArray.start out__) out self vec2;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let glu self ~dim =
-  let out__ = CArray.make t 1 in
-  stubs_glu (CArray.start out__) self (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let ger self ~vec2 = stubs_ger self vec2 |> with_tensor_gc
+let ger_out ~out self ~vec2 = stubs_ger_out out self vec2 |> with_tensor_gc
+let glu self ~dim = stubs_glu self (Int64.of_int dim) |> with_tensor_gc
 
 let glu_backward ~grad_output self ~dim =
-  let out__ = CArray.make t 1 in
-  stubs_glu_backward (CArray.start out__) grad_output self (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_glu_backward grad_output self (Int64.of_int dim) |> with_tensor_gc
 ;;
 
 let glu_backward_grad_input ~grad_input ~grad_output self ~dim =
-  let out__ = CArray.make t 1 in
-  stubs_glu_backward_grad_input
-    (CArray.start out__)
-    grad_input
-    grad_output
-    self
-    (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_glu_backward_grad_input grad_input grad_output self (Int64.of_int dim)
+  |> with_tensor_gc
 ;;
 
 let glu_backward_jvp ~grad_x ~grad_glu ~x ~dgrad_glu ~dx ~dim =
-  let out__ = CArray.make t 1 in
-  stubs_glu_backward_jvp
-    (CArray.start out__)
-    grad_x
-    grad_glu
-    x
-    dgrad_glu
-    dx
-    (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_glu_backward_jvp grad_x grad_glu x dgrad_glu dx (Int64.of_int dim)
+  |> with_tensor_gc
 ;;
 
 let glu_backward_jvp_out ~out ~grad_x ~grad_glu ~x ~dgrad_glu ~dx ~dim =
-  let out__ = CArray.make t 1 in
-  stubs_glu_backward_jvp_out
-    (CArray.start out__)
-    out
-    grad_x
-    grad_glu
-    x
-    dgrad_glu
-    dx
-    (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_glu_backward_jvp_out out grad_x grad_glu x dgrad_glu dx (Int64.of_int dim)
+  |> with_tensor_gc
 ;;
 
-let glu_jvp ~glu ~x ~dx ~dim =
-  let out__ = CArray.make t 1 in
-  stubs_glu_jvp (CArray.start out__) glu x dx (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let glu_jvp ~glu ~x ~dx ~dim = stubs_glu_jvp glu x dx (Int64.of_int dim) |> with_tensor_gc
 
 let glu_jvp_out ~out ~glu ~x ~dx ~dim =
-  let out__ = CArray.make t 1 in
-  stubs_glu_jvp_out (CArray.start out__) out glu x dx (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let glu_out ~out self ~dim =
-  let out__ = CArray.make t 1 in
-  stubs_glu_out (CArray.start out__) out self (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let grad self =
-  let out__ = CArray.make t 1 in
-  stubs_grad (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let greater self other =
-  let out__ = CArray.make t 1 in
-  stubs_greater (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let greater_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_greater_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let greater_equal self other =
-  let out__ = CArray.make t 1 in
-  stubs_greater_equal (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_glu_jvp_out out glu x dx (Int64.of_int dim) |> with_tensor_gc
 ;;
 
-let greater_equal_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_greater_equal_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let glu_out ~out self ~dim = stubs_glu_out out self (Int64.of_int dim) |> with_tensor_gc
+let grad self = stubs_grad self |> with_tensor_gc
+let greater self other = stubs_greater self other |> with_tensor_gc
+let greater_ self other = stubs_greater_ self other |> with_tensor_gc
+let greater_equal self other = stubs_greater_equal self other |> with_tensor_gc
+let greater_equal_ self other = stubs_greater_equal_ self other |> with_tensor_gc
 
 let greater_equal_scalar_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_greater_equal_scalar_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_greater_equal_scalar_out out self other |> with_tensor_gc
 ;;
 
 let greater_equal_tensor self other =
-  let out__ = CArray.make t 1 in
-  stubs_greater_equal_tensor (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_greater_equal_tensor self other |> with_tensor_gc
 ;;
 
 let greater_equal_tensor_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_greater_equal_tensor_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_greater_equal_tensor_ self other |> with_tensor_gc
 ;;
 
 let greater_equal_tensor_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_greater_equal_tensor_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_greater_equal_tensor_out out self other |> with_tensor_gc
 ;;
 
 let greater_scalar_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_greater_scalar_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let greater_tensor self other =
-  let out__ = CArray.make t 1 in
-  stubs_greater_tensor (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_greater_scalar_out out self other |> with_tensor_gc
 ;;
 
-let greater_tensor_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_greater_tensor_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let greater_tensor self other = stubs_greater_tensor self other |> with_tensor_gc
+let greater_tensor_ self other = stubs_greater_tensor_ self other |> with_tensor_gc
 
 let greater_tensor_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_greater_tensor_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_greater_tensor_out out self other |> with_tensor_gc
 ;;
 
 let grid_sampler input ~grid ~interpolation_mode ~padding_mode ~align_corners =
-  let out__ = CArray.make t 1 in
   stubs_grid_sampler
-    (CArray.start out__)
     input
     grid
     (Int64.of_int interpolation_mode)
     (Int64.of_int padding_mode)
-    (if align_corners then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if align_corners then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let grid_sampler_2d input ~grid ~interpolation_mode ~padding_mode ~align_corners =
-  let out__ = CArray.make t 1 in
   stubs_grid_sampler_2d
-    (CArray.start out__)
     input
     grid
     (Int64.of_int interpolation_mode)
     (Int64.of_int padding_mode)
-    (if align_corners then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if align_corners then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let grid_sampler_2d_out ~out input ~grid ~interpolation_mode ~padding_mode ~align_corners =
-  let out__ = CArray.make t 1 in
   stubs_grid_sampler_2d_out
-    (CArray.start out__)
     out
     input
     grid
     (Int64.of_int interpolation_mode)
     (Int64.of_int padding_mode)
-    (if align_corners then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if align_corners then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let grid_sampler_3d input ~grid ~interpolation_mode ~padding_mode ~align_corners =
-  let out__ = CArray.make t 1 in
   stubs_grid_sampler_3d
-    (CArray.start out__)
     input
     grid
     (Int64.of_int interpolation_mode)
     (Int64.of_int padding_mode)
-    (if align_corners then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if align_corners then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let grid_sampler_3d_out ~out input ~grid ~interpolation_mode ~padding_mode ~align_corners =
-  let out__ = CArray.make t 1 in
   stubs_grid_sampler_3d_out
-    (CArray.start out__)
     out
     input
     grid
     (Int64.of_int interpolation_mode)
     (Int64.of_int padding_mode)
-    (if align_corners then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if align_corners then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let group_norm input ~num_groups ~weight ~bias ~eps ~cudnn_enabled =
-  let out__ = CArray.make t 1 in
   stubs_group_norm
-    (CArray.start out__)
     input
     (Int64.of_int num_groups)
     (match weight with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     eps
-    (if cudnn_enabled then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if cudnn_enabled then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let gru
@@ -16555,12 +11037,12 @@ let gru
   ~bidirectional
   ~batch_first
   =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_gru
     (CArray.start out__)
     input
     hx
-    (CArray.of_list t params |> CArray.start)
+    (CArray.of_list gc_tensor params |> CArray.start)
     (List.length params)
     (if has_biases then 1 else 0)
     (Int64.of_int num_layers)
@@ -16568,30 +11050,24 @@ let gru
     (if train then 1 else 0)
     (if bidirectional then 1 else 0)
     (if batch_first then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let gru_cell input ~hx ~w_ih ~w_hh ~b_ih ~b_hh =
-  let out__ = CArray.make t 1 in
   stubs_gru_cell
-    (CArray.start out__)
     input
     hx
     w_ih
     w_hh
     (match b_ih with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match b_hh with
      | Some v -> v
-     | None -> null);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+  |> with_tensor_gc
 ;;
 
 let gru_data
@@ -16605,429 +11081,197 @@ let gru_data
   ~train
   ~bidirectional
   =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_gru_data
     (CArray.start out__)
     data
     batch_sizes
     hx
-    (CArray.of_list t params |> CArray.start)
+    (CArray.of_list gc_tensor params |> CArray.start)
     (List.length params)
     (if has_biases then 1 else 0)
     (Int64.of_int num_layers)
     dropout
     (if train then 1 else 0)
     (if bidirectional then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
-let gt self other =
-  let out__ = CArray.make t 1 in
-  stubs_gt (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let gt_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_gt_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let gt_scalar_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_gt_scalar_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let gt_tensor self other =
-  let out__ = CArray.make t 1 in
-  stubs_gt_tensor (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let gt_tensor_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_gt_tensor_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let gt_tensor_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_gt_tensor_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let gt self other = stubs_gt self other |> with_tensor_gc
+let gt_ self other = stubs_gt_ self other |> with_tensor_gc
+let gt_scalar_out ~out self other = stubs_gt_scalar_out out self other |> with_tensor_gc
+let gt_tensor self other = stubs_gt_tensor self other |> with_tensor_gc
+let gt_tensor_ self other = stubs_gt_tensor_ self other |> with_tensor_gc
+let gt_tensor_out ~out self other = stubs_gt_tensor_out out self other |> with_tensor_gc
 
 let hamming_window ~window_length ~options =
-  let out__ = CArray.make t 1 in
   stubs_hamming_window
-    (CArray.start out__)
     (Int64.of_int window_length)
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let hamming_window_out ~out ~window_length =
-  let out__ = CArray.make t 1 in
-  stubs_hamming_window_out (CArray.start out__) out (Int64.of_int window_length);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_hamming_window_out out (Int64.of_int window_length) |> with_tensor_gc
 ;;
 
 let hamming_window_periodic ~window_length ~periodic ~options =
-  let out__ = CArray.make t 1 in
   stubs_hamming_window_periodic
-    (CArray.start out__)
     (Int64.of_int window_length)
     (if periodic then 1 else 0)
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let hamming_window_periodic_alpha ~window_length ~periodic ~alpha ~options =
-  let out__ = CArray.make t 1 in
   stubs_hamming_window_periodic_alpha
-    (CArray.start out__)
     (Int64.of_int window_length)
     (if periodic then 1 else 0)
     alpha
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let hamming_window_periodic_alpha_beta ~window_length ~periodic ~alpha ~beta ~options =
-  let out__ = CArray.make t 1 in
   stubs_hamming_window_periodic_alpha_beta
-    (CArray.start out__)
     (Int64.of_int window_length)
     (if periodic then 1 else 0)
     alpha
     beta
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let hamming_window_periodic_alpha_beta_out ~out ~window_length ~periodic ~alpha ~beta =
-  let out__ = CArray.make t 1 in
   stubs_hamming_window_periodic_alpha_beta_out
-    (CArray.start out__)
     out
     (Int64.of_int window_length)
     (if periodic then 1 else 0)
     alpha
-    beta;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    beta
+  |> with_tensor_gc
 ;;
 
 let hamming_window_periodic_alpha_out ~out ~window_length ~periodic ~alpha =
-  let out__ = CArray.make t 1 in
   stubs_hamming_window_periodic_alpha_out
-    (CArray.start out__)
     out
     (Int64.of_int window_length)
     (if periodic then 1 else 0)
-    alpha;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    alpha
+  |> with_tensor_gc
 ;;
 
 let hamming_window_periodic_out ~out ~window_length ~periodic =
-  let out__ = CArray.make t 1 in
   stubs_hamming_window_periodic_out
-    (CArray.start out__)
     out
     (Int64.of_int window_length)
-    (if periodic then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if periodic then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let hann_window ~window_length ~options =
-  let out__ = CArray.make t 1 in
   stubs_hann_window
-    (CArray.start out__)
     (Int64.of_int window_length)
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let hann_window_out ~out ~window_length =
-  let out__ = CArray.make t 1 in
-  stubs_hann_window_out (CArray.start out__) out (Int64.of_int window_length);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_hann_window_out out (Int64.of_int window_length) |> with_tensor_gc
 ;;
 
 let hann_window_periodic ~window_length ~periodic ~options =
-  let out__ = CArray.make t 1 in
   stubs_hann_window_periodic
-    (CArray.start out__)
     (Int64.of_int window_length)
     (if periodic then 1 else 0)
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let hann_window_periodic_out ~out ~window_length ~periodic =
-  let out__ = CArray.make t 1 in
   stubs_hann_window_periodic_out
-    (CArray.start out__)
     out
     (Int64.of_int window_length)
-    (if periodic then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if periodic then 1 else 0)
+  |> with_tensor_gc
 ;;
 
-let hardshrink self =
-  let out__ = CArray.make t 1 in
-  stubs_hardshrink (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let hardshrink self = stubs_hardshrink self |> with_tensor_gc
 
 let hardshrink_backward ~grad_out self ~lambd =
-  let out__ = CArray.make t 1 in
-  stubs_hardshrink_backward (CArray.start out__) grad_out self lambd;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_hardshrink_backward grad_out self lambd |> with_tensor_gc
 ;;
 
 let hardshrink_backward_grad_input ~grad_input ~grad_out self ~lambd =
-  let out__ = CArray.make t 1 in
-  stubs_hardshrink_backward_grad_input (CArray.start out__) grad_input grad_out self lambd;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_hardshrink_backward_grad_input grad_input grad_out self lambd |> with_tensor_gc
 ;;
 
-let hardshrink_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_hardshrink_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let hardsigmoid self =
-  let out__ = CArray.make t 1 in
-  stubs_hardsigmoid (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let hardsigmoid_ self =
-  let out__ = CArray.make t 1 in
-  stubs_hardsigmoid_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let hardshrink_out ~out self = stubs_hardshrink_out out self |> with_tensor_gc
+let hardsigmoid self = stubs_hardsigmoid self |> with_tensor_gc
+let hardsigmoid_ self = stubs_hardsigmoid_ self |> with_tensor_gc
 
 let hardsigmoid_backward ~grad_output self =
-  let out__ = CArray.make t 1 in
-  stubs_hardsigmoid_backward (CArray.start out__) grad_output self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_hardsigmoid_backward grad_output self |> with_tensor_gc
 ;;
 
 let hardsigmoid_backward_grad_input ~grad_input ~grad_output self =
-  let out__ = CArray.make t 1 in
-  stubs_hardsigmoid_backward_grad_input (CArray.start out__) grad_input grad_output self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let hardsigmoid_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_hardsigmoid_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_hardsigmoid_backward_grad_input grad_input grad_output self |> with_tensor_gc
 ;;
 
-let hardswish self =
-  let out__ = CArray.make t 1 in
-  stubs_hardswish (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let hardswish_ self =
-  let out__ = CArray.make t 1 in
-  stubs_hardswish_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let hardsigmoid_out ~out self = stubs_hardsigmoid_out out self |> with_tensor_gc
+let hardswish self = stubs_hardswish self |> with_tensor_gc
+let hardswish_ self = stubs_hardswish_ self |> with_tensor_gc
 
 let hardswish_backward ~grad_output self =
-  let out__ = CArray.make t 1 in
-  stubs_hardswish_backward (CArray.start out__) grad_output self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_hardswish_backward grad_output self |> with_tensor_gc
 ;;
 
 let hardswish_backward_out ~out ~grad_output self =
-  let out__ = CArray.make t 1 in
-  stubs_hardswish_backward_out (CArray.start out__) out grad_output self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_hardswish_backward_out out grad_output self |> with_tensor_gc
 ;;
 
-let hardswish_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_hardswish_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let hardtanh self =
-  let out__ = CArray.make t 1 in
-  stubs_hardtanh (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let hardtanh_ self =
-  let out__ = CArray.make t 1 in
-  stubs_hardtanh_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let hardswish_out ~out self = stubs_hardswish_out out self |> with_tensor_gc
+let hardtanh self = stubs_hardtanh self |> with_tensor_gc
+let hardtanh_ self = stubs_hardtanh_ self |> with_tensor_gc
 
 let hardtanh_backward ~grad_output self ~min_val ~max_val =
-  let out__ = CArray.make t 1 in
-  stubs_hardtanh_backward (CArray.start out__) grad_output self min_val max_val;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_hardtanh_backward grad_output self min_val max_val |> with_tensor_gc
 ;;
 
 let hardtanh_backward_grad_input ~grad_input ~grad_output self ~min_val ~max_val =
-  let out__ = CArray.make t 1 in
-  stubs_hardtanh_backward_grad_input
-    (CArray.start out__)
-    grad_input
-    grad_output
-    self
-    min_val
-    max_val;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let hardtanh_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_hardtanh_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_hardtanh_backward_grad_input grad_input grad_output self min_val max_val
+  |> with_tensor_gc
 ;;
 
-let heaviside self ~values =
-  let out__ = CArray.make t 1 in
-  stubs_heaviside (CArray.start out__) self values;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let heaviside_ self ~values =
-  let out__ = CArray.make t 1 in
-  stubs_heaviside_ (CArray.start out__) self values;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let hardtanh_out ~out self = stubs_hardtanh_out out self |> with_tensor_gc
+let heaviside self ~values = stubs_heaviside self values |> with_tensor_gc
+let heaviside_ self ~values = stubs_heaviside_ self values |> with_tensor_gc
 
 let heaviside_out ~out self ~values =
-  let out__ = CArray.make t 1 in
-  stubs_heaviside_out (CArray.start out__) out self values;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_heaviside_out out self values |> with_tensor_gc
 ;;
 
 let hinge_embedding_loss self ~target ~margin ~reduction =
-  let out__ = CArray.make t 1 in
   stubs_hinge_embedding_loss
-    (CArray.start out__)
     self
     target
     margin
-    (Reduction.to_int reduction |> Int64.of_int);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Reduction.to_int reduction |> Int64.of_int)
+  |> with_tensor_gc
 ;;
 
-let histc self ~bins =
-  let out__ = CArray.make t 1 in
-  stubs_histc (CArray.start out__) self (Int64.of_int bins);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let histc self ~bins = stubs_histc self (Int64.of_int bins) |> with_tensor_gc
 
 let histc_out ~out self ~bins =
-  let out__ = CArray.make t 1 in
-  stubs_histc_out (CArray.start out__) out self (Int64.of_int bins);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_histc_out out self (Int64.of_int bins) |> with_tensor_gc
 ;;
 
 let hsplit self ~sections = stubs_hsplit self (Int64.of_int sections) |> to_tensor_list
@@ -17040,201 +11284,68 @@ let hsplit_array self ~indices =
   |> to_tensor_list
 ;;
 
-let hspmm ~mat1 ~mat2 =
-  let out__ = CArray.make t 1 in
-  stubs_hspmm (CArray.start out__) mat1 mat2;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let hspmm_out ~out ~mat1 ~mat2 =
-  let out__ = CArray.make t 1 in
-  stubs_hspmm_out (CArray.start out__) out mat1 mat2;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let hstack tensors =
-  let out__ = CArray.make t 1 in
-  stubs_hstack
-    (CArray.start out__)
-    (CArray.of_list t tensors |> CArray.start)
-    (List.length tensors);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let hstack_out ~out tensors =
-  let out__ = CArray.make t 1 in
-  stubs_hstack_out
-    (CArray.start out__)
-    out
-    (CArray.of_list t tensors |> CArray.start)
-    (List.length tensors);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let huber_loss self ~target ~reduction ~delta =
-  let out__ = CArray.make t 1 in
-  stubs_huber_loss
-    (CArray.start out__)
-    self
-    target
-    (Reduction.to_int reduction |> Int64.of_int)
-    delta;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let huber_loss_backward ~grad_output self ~target ~reduction ~delta =
-  let out__ = CArray.make t 1 in
-  stubs_huber_loss_backward
-    (CArray.start out__)
-    grad_output
-    self
-    target
-    (Reduction.to_int reduction |> Int64.of_int)
-    delta;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let huber_loss_backward_out ~grad_input ~grad_output self ~target ~reduction ~delta =
-  let out__ = CArray.make t 1 in
-  stubs_huber_loss_backward_out
-    (CArray.start out__)
-    grad_input
-    grad_output
-    self
-    target
-    (Reduction.to_int reduction |> Int64.of_int)
-    delta;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let huber_loss_out ~out self ~target ~reduction ~delta =
-  let out__ = CArray.make t 1 in
-  stubs_huber_loss_out
-    (CArray.start out__)
-    out
-    self
-    target
-    (Reduction.to_int reduction |> Int64.of_int)
-    delta;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let hypot self other =
-  let out__ = CArray.make t 1 in
-  stubs_hypot (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let hypot_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_hypot_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let hypot_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_hypot_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let i0 self =
-  let out__ = CArray.make t 1 in
-  stubs_i0 (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let i0_ self =
-  let out__ = CArray.make t 1 in
-  stubs_i0_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let i0_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_i0_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let hspmm ~mat1 ~mat2 = stubs_hspmm mat1 mat2 |> with_tensor_gc
+let hspmm_out ~out ~mat1 ~mat2 = stubs_hspmm_out out mat1 mat2 |> with_tensor_gc
 
-let igamma self other =
-  let out__ = CArray.make t 1 in
-  stubs_igamma (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+let hstack tensors =
+  stubs_hstack (CArray.of_list gc_tensor tensors |> CArray.start) (List.length tensors)
+  |> with_tensor_gc
 ;;
 
-let igamma_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_igamma_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+let hstack_out ~out tensors =
+  stubs_hstack_out
+    out
+    (CArray.of_list gc_tensor tensors |> CArray.start)
+    (List.length tensors)
+  |> with_tensor_gc
 ;;
 
-let igamma_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_igamma_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+let huber_loss self ~target ~reduction ~delta =
+  stubs_huber_loss self target (Reduction.to_int reduction |> Int64.of_int) delta
+  |> with_tensor_gc
 ;;
 
-let igammac self other =
-  let out__ = CArray.make t 1 in
-  stubs_igammac (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+let huber_loss_backward ~grad_output self ~target ~reduction ~delta =
+  stubs_huber_loss_backward
+    grad_output
+    self
+    target
+    (Reduction.to_int reduction |> Int64.of_int)
+    delta
+  |> with_tensor_gc
 ;;
 
-let igammac_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_igammac_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+let huber_loss_backward_out ~grad_input ~grad_output self ~target ~reduction ~delta =
+  stubs_huber_loss_backward_out
+    grad_input
+    grad_output
+    self
+    target
+    (Reduction.to_int reduction |> Int64.of_int)
+    delta
+  |> with_tensor_gc
 ;;
 
-let igammac_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_igammac_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let huber_loss_out ~out self ~target ~reduction ~delta =
+  stubs_huber_loss_out out self target (Reduction.to_int reduction |> Int64.of_int) delta
+  |> with_tensor_gc
+;;
+
+let hypot self other = stubs_hypot self other |> with_tensor_gc
+let hypot_ self other = stubs_hypot_ self other |> with_tensor_gc
+let hypot_out ~out self other = stubs_hypot_out out self other |> with_tensor_gc
+let i0 self = stubs_i0 self |> with_tensor_gc
+let i0_ self = stubs_i0_ self |> with_tensor_gc
+let i0_out ~out self = stubs_i0_out out self |> with_tensor_gc
+let igamma self other = stubs_igamma self other |> with_tensor_gc
+let igamma_ self other = stubs_igamma_ self other |> with_tensor_gc
+let igamma_out ~out self other = stubs_igamma_out out self other |> with_tensor_gc
+let igammac self other = stubs_igammac self other |> with_tensor_gc
+let igammac_ self other = stubs_igammac_ self other |> with_tensor_gc
+let igammac_out ~out self other = stubs_igammac_out out self other |> with_tensor_gc
 
 let im2col self ~kernel_size ~dilation ~padding ~stride =
-  let out__ = CArray.make t 1 in
   stubs_im2col
-    (CArray.start out__)
     self
     (List.map Int64.of_int kernel_size |> CArray.of_list int64_t |> CArray.start)
     (List.length kernel_size)
@@ -17243,16 +11354,12 @@ let im2col self ~kernel_size ~dilation ~padding ~stride =
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
     (List.length padding)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
-    (List.length stride);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length stride)
+  |> with_tensor_gc
 ;;
 
 let im2col_out ~out self ~kernel_size ~dilation ~padding ~stride =
-  let out__ = CArray.make t 1 in
   stubs_im2col_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int kernel_size |> CArray.of_list int64_t |> CArray.start)
@@ -17262,349 +11369,202 @@ let im2col_out ~out self ~kernel_size ~dilation ~padding ~stride =
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
     (List.length padding)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
-    (List.length stride);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length stride)
+  |> with_tensor_gc
 ;;
 
-let imag self =
-  let out__ = CArray.make t 1 in
-  stubs_imag (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let imag self = stubs_imag self |> with_tensor_gc
 
 let index self ~indices =
-  let out__ = CArray.make t 1 in
   stubs_index
-    (CArray.start out__)
     self
     (List.map
        (function
         | Some x -> x
-        | None -> null)
+        | None -> none_gc_tensor)
        indices
-     |> CArray.of_list t
+     |> CArray.of_list gc_tensor
      |> CArray.start)
-    (List.length indices);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length indices)
+  |> with_tensor_gc
 ;;
 
 let index_add self ~dim ~index ~source =
-  let out__ = CArray.make t 1 in
-  stubs_index_add (CArray.start out__) self (Int64.of_int dim) index source;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_index_add self (Int64.of_int dim) index source |> with_tensor_gc
 ;;
 
 let index_add_ self ~dim ~index ~source =
-  let out__ = CArray.make t 1 in
-  stubs_index_add_ (CArray.start out__) self (Int64.of_int dim) index source;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_index_add_ self (Int64.of_int dim) index source |> with_tensor_gc
 ;;
 
 let index_add_out ~out self ~dim ~index ~source =
-  let out__ = CArray.make t 1 in
-  stubs_index_add_out (CArray.start out__) out self (Int64.of_int dim) index source;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_index_add_out out self (Int64.of_int dim) index source |> with_tensor_gc
 ;;
 
 let index_copy self ~dim ~index ~source =
-  let out__ = CArray.make t 1 in
-  stubs_index_copy (CArray.start out__) self (Int64.of_int dim) index source;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_index_copy self (Int64.of_int dim) index source |> with_tensor_gc
 ;;
 
 let index_copy_ self ~dim ~index ~source =
-  let out__ = CArray.make t 1 in
-  stubs_index_copy_ (CArray.start out__) self (Int64.of_int dim) index source;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_index_copy_ self (Int64.of_int dim) index source |> with_tensor_gc
 ;;
 
 let index_copy_out ~out self ~dim ~index ~source =
-  let out__ = CArray.make t 1 in
-  stubs_index_copy_out (CArray.start out__) out self (Int64.of_int dim) index source;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_index_copy_out out self (Int64.of_int dim) index source |> with_tensor_gc
 ;;
 
 let index_fill self ~dim ~index ~value =
-  let out__ = CArray.make t 1 in
-  stubs_index_fill (CArray.start out__) self (Int64.of_int dim) index value;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_index_fill self (Int64.of_int dim) index value |> with_tensor_gc
 ;;
 
 let index_fill_ self ~dim ~index ~value =
-  let out__ = CArray.make t 1 in
-  stubs_index_fill_ (CArray.start out__) self (Int64.of_int dim) index value;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_index_fill_ self (Int64.of_int dim) index value |> with_tensor_gc
 ;;
 
 let index_fill_int_scalar_out ~out self ~dim ~index ~value =
-  let out__ = CArray.make t 1 in
-  stubs_index_fill_int_scalar_out
-    (CArray.start out__)
-    out
-    self
-    (Int64.of_int dim)
-    index
-    value;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_index_fill_int_scalar_out out self (Int64.of_int dim) index value
+  |> with_tensor_gc
 ;;
 
 let index_fill_int_tensor self ~dim ~index ~value =
-  let out__ = CArray.make t 1 in
-  stubs_index_fill_int_tensor (CArray.start out__) self (Int64.of_int dim) index value;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_index_fill_int_tensor self (Int64.of_int dim) index value |> with_tensor_gc
 ;;
 
 let index_fill_int_tensor_ self ~dim ~index ~value =
-  let out__ = CArray.make t 1 in
-  stubs_index_fill_int_tensor_ (CArray.start out__) self (Int64.of_int dim) index value;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_index_fill_int_tensor_ self (Int64.of_int dim) index value |> with_tensor_gc
 ;;
 
 let index_fill_int_tensor_out ~out self ~dim ~index ~value =
-  let out__ = CArray.make t 1 in
-  stubs_index_fill_int_tensor_out
-    (CArray.start out__)
-    out
-    self
-    (Int64.of_int dim)
-    index
-    value;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_index_fill_int_tensor_out out self (Int64.of_int dim) index value
+  |> with_tensor_gc
 ;;
 
 let index_put self ~indices ~values ~accumulate =
-  let out__ = CArray.make t 1 in
   stubs_index_put
-    (CArray.start out__)
     self
     (List.map
        (function
         | Some x -> x
-        | None -> null)
+        | None -> none_gc_tensor)
        indices
-     |> CArray.of_list t
+     |> CArray.of_list gc_tensor
      |> CArray.start)
     (List.length indices)
     values
-    (if accumulate then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if accumulate then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let index_put_ self ~indices ~values ~accumulate =
-  let out__ = CArray.make t 1 in
   stubs_index_put_
-    (CArray.start out__)
     self
     (List.map
        (function
         | Some x -> x
-        | None -> null)
+        | None -> none_gc_tensor)
        indices
-     |> CArray.of_list t
+     |> CArray.of_list gc_tensor
      |> CArray.start)
     (List.length indices)
     values
-    (if accumulate then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if accumulate then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let index_put_out ~out self ~indices ~values ~accumulate =
-  let out__ = CArray.make t 1 in
   stubs_index_put_out
-    (CArray.start out__)
     out
     self
     (List.map
        (function
         | Some x -> x
-        | None -> null)
+        | None -> none_gc_tensor)
        indices
-     |> CArray.of_list t
+     |> CArray.of_list gc_tensor
      |> CArray.start)
     (List.length indices)
     values
-    (if accumulate then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if accumulate then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let index_reduce self ~dim ~index ~source ~reduce ~include_self =
-  let out__ = CArray.make t 1 in
   stubs_index_reduce
-    (CArray.start out__)
     self
     (Int64.of_int dim)
     index
     source
     reduce
-    (if include_self then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if include_self then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let index_reduce_ self ~dim ~index ~source ~reduce ~include_self =
-  let out__ = CArray.make t 1 in
   stubs_index_reduce_
-    (CArray.start out__)
     self
     (Int64.of_int dim)
     index
     source
     reduce
-    (if include_self then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if include_self then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let index_reduce_out ~out self ~dim ~index ~source ~reduce ~include_self =
-  let out__ = CArray.make t 1 in
   stubs_index_reduce_out
-    (CArray.start out__)
     out
     self
     (Int64.of_int dim)
     index
     source
     reduce
-    (if include_self then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if include_self then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let index_select self ~dim ~index =
-  let out__ = CArray.make t 1 in
-  stubs_index_select (CArray.start out__) self (Int64.of_int dim) index;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_index_select self (Int64.of_int dim) index |> with_tensor_gc
 ;;
 
 let index_select_backward ~grad ~self_sizes ~dim ~index =
-  let out__ = CArray.make t 1 in
   stubs_index_select_backward
-    (CArray.start out__)
     grad
     (List.map Int64.of_int self_sizes |> CArray.of_list int64_t |> CArray.start)
     (List.length self_sizes)
     (Int64.of_int dim)
-    index;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    index
+  |> with_tensor_gc
 ;;
 
 let index_select_out ~out self ~dim ~index =
-  let out__ = CArray.make t 1 in
-  stubs_index_select_out (CArray.start out__) out self (Int64.of_int dim) index;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_index_select_out out self (Int64.of_int dim) index |> with_tensor_gc
 ;;
 
 let index_tensor_out ~out self ~indices =
-  let out__ = CArray.make t 1 in
   stubs_index_tensor_out
-    (CArray.start out__)
     out
     self
     (List.map
        (function
         | Some x -> x
-        | None -> null)
+        | None -> none_gc_tensor)
        indices
-     |> CArray.of_list t
+     |> CArray.of_list gc_tensor
      |> CArray.start)
-    (List.length indices);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let indices self =
-  let out__ = CArray.make t 1 in
-  stubs_indices (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let indices_copy self =
-  let out__ = CArray.make t 1 in
-  stubs_indices_copy (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length indices)
+  |> with_tensor_gc
 ;;
 
-let indices_copy_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_indices_copy_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let indices self = stubs_indices self |> with_tensor_gc
+let indices_copy self = stubs_indices_copy self |> with_tensor_gc
+let indices_copy_out ~out self = stubs_indices_copy_out out self |> with_tensor_gc
 
 let infinitely_differentiable_gelu_backward ~grad self =
-  let out__ = CArray.make t 1 in
-  stubs_infinitely_differentiable_gelu_backward (CArray.start out__) grad self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_infinitely_differentiable_gelu_backward grad self |> with_tensor_gc
 ;;
 
-let inner self other =
-  let out__ = CArray.make t 1 in
-  stubs_inner (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let inner_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_inner_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let inner self other = stubs_inner self other |> with_tensor_gc
+let inner_out ~out self other = stubs_inner_out out self other |> with_tensor_gc
 
 let instance_norm
   input
@@ -17617,63 +11577,31 @@ let instance_norm
   ~eps
   ~cudnn_enabled
   =
-  let out__ = CArray.make t 1 in
   stubs_instance_norm
-    (CArray.start out__)
     input
     (match weight with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match running_mean with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match running_var with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (if use_input_stats then 1 else 0)
     momentum
     eps
-    (if cudnn_enabled then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let int_repr self =
-  let out__ = CArray.make t 1 in
-  stubs_int_repr (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let int_repr_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_int_repr_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let inverse self =
-  let out__ = CArray.make t 1 in
-  stubs_inverse (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let inverse_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_inverse_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if cudnn_enabled then 1 else 0)
+  |> with_tensor_gc
 ;;
 
+let int_repr self = stubs_int_repr self |> with_tensor_gc
+let int_repr_out ~out self = stubs_int_repr_out out self |> with_tensor_gc
+let inverse self = stubs_inverse self |> with_tensor_gc
+let inverse_out ~out self = stubs_inverse_out out self |> with_tensor_gc
 let is_coalesced self = stubs_is_coalesced self
 let is_complex self = stubs_is_complex self
 let is_conj self = stubs_is_conj self
@@ -17690,173 +11618,77 @@ let is_signed self = stubs_is_signed self
 let is_vulkan_available = stubs_is_vulkan_available
 
 let isclose self other ~rtol ~atol ~equal_nan =
-  let out__ = CArray.make t 1 in
-  stubs_isclose (CArray.start out__) self other rtol atol (if equal_nan then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_isclose self other rtol atol (if equal_nan then 1 else 0) |> with_tensor_gc
 ;;
 
-let isfinite self =
-  let out__ = CArray.make t 1 in
-  stubs_isfinite (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let isfinite self = stubs_isfinite self |> with_tensor_gc
 
 let isin ~elements ~test_elements ~assume_unique ~invert =
-  let out__ = CArray.make t 1 in
   stubs_isin
-    (CArray.start out__)
     elements
     test_elements
     (if assume_unique then 1 else 0)
-    (if invert then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if invert then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let isin_scalar_tensor ~element ~test_elements ~assume_unique ~invert =
-  let out__ = CArray.make t 1 in
   stubs_isin_scalar_tensor
-    (CArray.start out__)
     element
     test_elements
     (if assume_unique then 1 else 0)
-    (if invert then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if invert then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let isin_scalar_tensor_out ~out ~element ~test_elements ~assume_unique ~invert =
-  let out__ = CArray.make t 1 in
   stubs_isin_scalar_tensor_out
-    (CArray.start out__)
     out
     element
     test_elements
     (if assume_unique then 1 else 0)
-    (if invert then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if invert then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let isin_tensor_scalar ~elements ~test_element ~assume_unique ~invert =
-  let out__ = CArray.make t 1 in
   stubs_isin_tensor_scalar
-    (CArray.start out__)
     elements
     test_element
     (if assume_unique then 1 else 0)
-    (if invert then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if invert then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let isin_tensor_scalar_out ~out ~elements ~test_element ~assume_unique ~invert =
-  let out__ = CArray.make t 1 in
   stubs_isin_tensor_scalar_out
-    (CArray.start out__)
     out
     elements
     test_element
     (if assume_unique then 1 else 0)
-    (if invert then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if invert then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let isin_tensor_tensor_out ~out ~elements ~test_elements ~assume_unique ~invert =
-  let out__ = CArray.make t 1 in
   stubs_isin_tensor_tensor_out
-    (CArray.start out__)
     out
     elements
     test_elements
     (if assume_unique then 1 else 0)
-    (if invert then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let isinf self =
-  let out__ = CArray.make t 1 in
-  stubs_isinf (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let isinf_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_isinf_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if invert then 1 else 0)
+  |> with_tensor_gc
 ;;
 
-let isnan self =
-  let out__ = CArray.make t 1 in
-  stubs_isnan (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let isnan_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_isnan_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let isneginf self =
-  let out__ = CArray.make t 1 in
-  stubs_isneginf (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let isneginf_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_isneginf_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let isposinf self =
-  let out__ = CArray.make t 1 in
-  stubs_isposinf (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let isposinf_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_isposinf_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let isreal self =
-  let out__ = CArray.make t 1 in
-  stubs_isreal (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let isinf self = stubs_isinf self |> with_tensor_gc
+let isinf_out ~out self = stubs_isinf_out out self |> with_tensor_gc
+let isnan self = stubs_isnan self |> with_tensor_gc
+let isnan_out ~out self = stubs_isnan_out out self |> with_tensor_gc
+let isneginf self = stubs_isneginf self |> with_tensor_gc
+let isneginf_out ~out self = stubs_isneginf_out out self |> with_tensor_gc
+let isposinf self = stubs_isposinf self |> with_tensor_gc
+let isposinf_out ~out self = stubs_isposinf_out out self |> with_tensor_gc
+let isreal self = stubs_isreal self |> with_tensor_gc
 
 let istft
   self
@@ -17870,9 +11702,7 @@ let istft
   ~length
   ~return_complex
   =
-  let out__ = CArray.make t 1 in
   stubs_istft
-    (CArray.start out__)
     self
     (Int64.of_int n_fft)
     (match hop_length with
@@ -17889,7 +11719,7 @@ let istft
      | None -> 1)
     (match window with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (if center then 1 else 0)
     (if normalized then 1 else 0)
     (if onesided then 1 else 0)
@@ -17899,130 +11729,85 @@ let istft
     (match length with
      | Some _ -> 0
      | None -> 1)
-    (if return_complex then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if return_complex then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let kaiser_window ~window_length ~options =
-  let out__ = CArray.make t 1 in
   stubs_kaiser_window
-    (CArray.start out__)
     (Int64.of_int window_length)
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let kaiser_window_beta ~window_length ~periodic ~beta ~options =
-  let out__ = CArray.make t 1 in
   stubs_kaiser_window_beta
-    (CArray.start out__)
     (Int64.of_int window_length)
     (if periodic then 1 else 0)
     beta
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let kaiser_window_beta_out ~out ~window_length ~periodic ~beta =
-  let out__ = CArray.make t 1 in
   stubs_kaiser_window_beta_out
-    (CArray.start out__)
     out
     (Int64.of_int window_length)
     (if periodic then 1 else 0)
-    beta;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    beta
+  |> with_tensor_gc
 ;;
 
 let kaiser_window_out ~out ~window_length =
-  let out__ = CArray.make t 1 in
-  stubs_kaiser_window_out (CArray.start out__) out (Int64.of_int window_length);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_kaiser_window_out out (Int64.of_int window_length) |> with_tensor_gc
 ;;
 
 let kaiser_window_periodic ~window_length ~periodic ~options =
-  let out__ = CArray.make t 1 in
   stubs_kaiser_window_periodic
-    (CArray.start out__)
     (Int64.of_int window_length)
     (if periodic then 1 else 0)
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let kaiser_window_periodic_out ~out ~window_length ~periodic =
-  let out__ = CArray.make t 1 in
   stubs_kaiser_window_periodic_out
-    (CArray.start out__)
     out
     (Int64.of_int window_length)
-    (if periodic then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if periodic then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let kl_div self ~target ~reduction ~log_target =
-  let out__ = CArray.make t 1 in
   stubs_kl_div
-    (CArray.start out__)
     self
     target
     (Reduction.to_int reduction |> Int64.of_int)
-    (if log_target then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let kron self other =
-  let out__ = CArray.make t 1 in
-  stubs_kron (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if log_target then 1 else 0)
+  |> with_tensor_gc
 ;;
 
-let kron_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_kron_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let kron self other = stubs_kron self other |> with_tensor_gc
+let kron_out ~out self other = stubs_kron_out out self other |> with_tensor_gc
 
 let kthvalue self ~k ~dim ~keepdim =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_kthvalue
     (CArray.start out__)
     self
     (Int64.of_int k)
     (Int64.of_int dim)
     (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let kthvalue_values ~values ~indices self ~k ~dim ~keepdim =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_kthvalue_values
     (CArray.start out__)
     values
@@ -18031,168 +11816,53 @@ let kthvalue_values ~values ~indices self ~k ~dim ~keepdim =
     (Int64.of_int k)
     (Int64.of_int dim)
     (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let l1_loss self ~target ~reduction =
-  let out__ = CArray.make t 1 in
-  stubs_l1_loss
-    (CArray.start out__)
-    self
-    target
-    (Reduction.to_int reduction |> Int64.of_int);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_l1_loss self target (Reduction.to_int reduction |> Int64.of_int) |> with_tensor_gc
 ;;
 
 let layer_norm input ~normalized_shape ~weight ~bias ~eps ~cudnn_enable =
-  let out__ = CArray.make t 1 in
   stubs_layer_norm
-    (CArray.start out__)
     input
     (List.map Int64.of_int normalized_shape |> CArray.of_list int64_t |> CArray.start)
     (List.length normalized_shape)
     (match weight with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     eps
-    (if cudnn_enable then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let lcm self other =
-  let out__ = CArray.make t 1 in
-  stubs_lcm (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let lcm_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_lcm_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let lcm_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_lcm_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let ldexp self other =
-  let out__ = CArray.make t 1 in
-  stubs_ldexp (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let ldexp_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_ldexp_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let ldexp_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_ldexp_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let le self other =
-  let out__ = CArray.make t 1 in
-  stubs_le (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let le_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_le_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let le_scalar_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_le_scalar_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let le_tensor self other =
-  let out__ = CArray.make t 1 in
-  stubs_le_tensor (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let le_tensor_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_le_tensor_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let le_tensor_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_le_tensor_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let leaky_relu self =
-  let out__ = CArray.make t 1 in
-  stubs_leaky_relu (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let leaky_relu_ self =
-  let out__ = CArray.make t 1 in
-  stubs_leaky_relu_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+    (if cudnn_enable then 1 else 0)
+  |> with_tensor_gc
+;;
+
+let lcm self other = stubs_lcm self other |> with_tensor_gc
+let lcm_ self other = stubs_lcm_ self other |> with_tensor_gc
+let lcm_out ~out self other = stubs_lcm_out out self other |> with_tensor_gc
+let ldexp self other = stubs_ldexp self other |> with_tensor_gc
+let ldexp_ self other = stubs_ldexp_ self other |> with_tensor_gc
+let ldexp_out ~out self other = stubs_ldexp_out out self other |> with_tensor_gc
+let le self other = stubs_le self other |> with_tensor_gc
+let le_ self other = stubs_le_ self other |> with_tensor_gc
+let le_scalar_out ~out self other = stubs_le_scalar_out out self other |> with_tensor_gc
+let le_tensor self other = stubs_le_tensor self other |> with_tensor_gc
+let le_tensor_ self other = stubs_le_tensor_ self other |> with_tensor_gc
+let le_tensor_out ~out self other = stubs_le_tensor_out out self other |> with_tensor_gc
+let leaky_relu self = stubs_leaky_relu self |> with_tensor_gc
+let leaky_relu_ self = stubs_leaky_relu_ self |> with_tensor_gc
 
 let leaky_relu_backward ~grad_output self ~negative_slope ~self_is_result =
-  let out__ = CArray.make t 1 in
   stubs_leaky_relu_backward
-    (CArray.start out__)
     grad_output
     self
     negative_slope
-    (if self_is_result then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if self_is_result then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let leaky_relu_backward_grad_input
@@ -18202,259 +11872,87 @@ let leaky_relu_backward_grad_input
   ~negative_slope
   ~self_is_result
   =
-  let out__ = CArray.make t 1 in
   stubs_leaky_relu_backward_grad_input
-    (CArray.start out__)
     grad_input
     grad_output
     self
     negative_slope
-    (if self_is_result then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let leaky_relu_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_leaky_relu_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let lerp self ~end_ ~weight =
-  let out__ = CArray.make t 1 in
-  stubs_lerp (CArray.start out__) self end_ weight;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if self_is_result then 1 else 0)
+  |> with_tensor_gc
 ;;
 
-let lerp_ self ~end_ ~weight =
-  let out__ = CArray.make t 1 in
-  stubs_lerp_ (CArray.start out__) self end_ weight;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let leaky_relu_out ~out self = stubs_leaky_relu_out out self |> with_tensor_gc
+let lerp self ~end_ ~weight = stubs_lerp self end_ weight |> with_tensor_gc
+let lerp_ self ~end_ ~weight = stubs_lerp_ self end_ weight |> with_tensor_gc
 
 let lerp_scalar_out ~out self ~end_ ~weight =
-  let out__ = CArray.make t 1 in
-  stubs_lerp_scalar_out (CArray.start out__) out self end_ weight;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_lerp_scalar_out out self end_ weight |> with_tensor_gc
 ;;
 
-let lerp_tensor self ~end_ ~weight =
-  let out__ = CArray.make t 1 in
-  stubs_lerp_tensor (CArray.start out__) self end_ weight;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let lerp_tensor self ~end_ ~weight = stubs_lerp_tensor self end_ weight |> with_tensor_gc
 
 let lerp_tensor_ self ~end_ ~weight =
-  let out__ = CArray.make t 1 in
-  stubs_lerp_tensor_ (CArray.start out__) self end_ weight;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_lerp_tensor_ self end_ weight |> with_tensor_gc
 ;;
 
 let lerp_tensor_out ~out self ~end_ ~weight =
-  let out__ = CArray.make t 1 in
-  stubs_lerp_tensor_out (CArray.start out__) out self end_ weight;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let less self other =
-  let out__ = CArray.make t 1 in
-  stubs_less (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let less_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_less_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let less_equal self other =
-  let out__ = CArray.make t 1 in
-  stubs_less_equal (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_lerp_tensor_out out self end_ weight |> with_tensor_gc
 ;;
 
-let less_equal_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_less_equal_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let less self other = stubs_less self other |> with_tensor_gc
+let less_ self other = stubs_less_ self other |> with_tensor_gc
+let less_equal self other = stubs_less_equal self other |> with_tensor_gc
+let less_equal_ self other = stubs_less_equal_ self other |> with_tensor_gc
 
 let less_equal_scalar_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_less_equal_scalar_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_less_equal_scalar_out out self other |> with_tensor_gc
 ;;
 
-let less_equal_tensor self other =
-  let out__ = CArray.make t 1 in
-  stubs_less_equal_tensor (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let less_equal_tensor_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_less_equal_tensor_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let less_equal_tensor self other = stubs_less_equal_tensor self other |> with_tensor_gc
+let less_equal_tensor_ self other = stubs_less_equal_tensor_ self other |> with_tensor_gc
 
 let less_equal_tensor_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_less_equal_tensor_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_less_equal_tensor_out out self other |> with_tensor_gc
 ;;
 
 let less_scalar_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_less_scalar_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let less_tensor self other =
-  let out__ = CArray.make t 1 in
-  stubs_less_tensor (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_less_scalar_out out self other |> with_tensor_gc
 ;;
 
-let less_tensor_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_less_tensor_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let less_tensor self other = stubs_less_tensor self other |> with_tensor_gc
+let less_tensor_ self other = stubs_less_tensor_ self other |> with_tensor_gc
 
 let less_tensor_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_less_tensor_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let lgamma self =
-  let out__ = CArray.make t 1 in
-  stubs_lgamma (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let lgamma_ self =
-  let out__ = CArray.make t 1 in
-  stubs_lgamma_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let lgamma_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_lgamma_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let lift self =
-  let out__ = CArray.make t 1 in
-  stubs_lift (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let lift_fresh self =
-  let out__ = CArray.make t 1 in
-  stubs_lift_fresh (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let lift_fresh_copy self =
-  let out__ = CArray.make t 1 in
-  stubs_lift_fresh_copy (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let lift_fresh_copy_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_lift_fresh_copy_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_less_tensor_out out self other |> with_tensor_gc
 ;;
 
-let lift_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_lift_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let lgamma self = stubs_lgamma self |> with_tensor_gc
+let lgamma_ self = stubs_lgamma_ self |> with_tensor_gc
+let lgamma_out ~out self = stubs_lgamma_out out self |> with_tensor_gc
+let lift self = stubs_lift self |> with_tensor_gc
+let lift_fresh self = stubs_lift_fresh self |> with_tensor_gc
+let lift_fresh_copy self = stubs_lift_fresh_copy self |> with_tensor_gc
+let lift_fresh_copy_out ~out self = stubs_lift_fresh_copy_out out self |> with_tensor_gc
+let lift_out ~out self = stubs_lift_out out self |> with_tensor_gc
 
 let linalg_cholesky self ~upper =
-  let out__ = CArray.make t 1 in
-  stubs_linalg_cholesky (CArray.start out__) self (if upper then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_linalg_cholesky self (if upper then 1 else 0) |> with_tensor_gc
 ;;
 
 let linalg_cholesky_ex self ~upper ~check_errors =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_linalg_cholesky_ex
     (CArray.start out__)
     self
     (if upper then 1 else 0)
     (if check_errors then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let linalg_cholesky_ex_l ~l ~info self ~upper ~check_errors =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_linalg_cholesky_ex_l
     (CArray.start out__)
     l
@@ -18462,255 +11960,135 @@ let linalg_cholesky_ex_l ~l ~info self ~upper ~check_errors =
     self
     (if upper then 1 else 0)
     (if check_errors then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let linalg_cholesky_out ~out self ~upper =
-  let out__ = CArray.make t 1 in
-  stubs_linalg_cholesky_out (CArray.start out__) out self (if upper then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let linalg_cond self ~p =
-  let out__ = CArray.make t 1 in
-  stubs_linalg_cond (CArray.start out__) self p;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let linalg_cond_out ~out self ~p =
-  let out__ = CArray.make t 1 in
-  stubs_linalg_cond_out (CArray.start out__) out self p;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_linalg_cholesky_out out self (if upper then 1 else 0) |> with_tensor_gc
 ;;
 
-let linalg_cond_p_str self ~p =
-  let out__ = CArray.make t 1 in
-  stubs_linalg_cond_p_str (CArray.start out__) self p;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let linalg_cond self ~p = stubs_linalg_cond self p |> with_tensor_gc
+let linalg_cond_out ~out self ~p = stubs_linalg_cond_out out self p |> with_tensor_gc
+let linalg_cond_p_str self ~p = stubs_linalg_cond_p_str self p |> with_tensor_gc
 
 let linalg_cond_p_str_out ~out self ~p =
-  let out__ = CArray.make t 1 in
-  stubs_linalg_cond_p_str_out (CArray.start out__) out self p;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_linalg_cond_p_str_out out self p |> with_tensor_gc
 ;;
 
 let linalg_cross self other ~dim =
-  let out__ = CArray.make t 1 in
-  stubs_linalg_cross (CArray.start out__) self other (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_linalg_cross self other (Int64.of_int dim) |> with_tensor_gc
 ;;
 
 let linalg_cross_out ~out self other ~dim =
-  let out__ = CArray.make t 1 in
-  stubs_linalg_cross_out (CArray.start out__) out self other (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let linalg_det ~a =
-  let out__ = CArray.make t 1 in
-  stubs_linalg_det (CArray.start out__) a;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_linalg_cross_out out self other (Int64.of_int dim) |> with_tensor_gc
 ;;
 
-let linalg_det_out ~out ~a =
-  let out__ = CArray.make t 1 in
-  stubs_linalg_det_out (CArray.start out__) out a;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let linalg_det ~a = stubs_linalg_det a |> with_tensor_gc
+let linalg_det_out ~out ~a = stubs_linalg_det_out out a |> with_tensor_gc
 
 let linalg_diagonal ~a ~offset ~dim1 ~dim2 =
-  let out__ = CArray.make t 1 in
-  stubs_linalg_diagonal
-    (CArray.start out__)
-    a
-    (Int64.of_int offset)
-    (Int64.of_int dim1)
-    (Int64.of_int dim2);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_linalg_diagonal a (Int64.of_int offset) (Int64.of_int dim1) (Int64.of_int dim2)
+  |> with_tensor_gc
 ;;
 
 let linalg_eig self =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_linalg_eig (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let linalg_eig_out ~eigenvalues ~eigenvectors self =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_linalg_eig_out (CArray.start out__) eigenvalues eigenvectors self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let linalg_eigh self ~uplo =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_linalg_eigh (CArray.start out__) self uplo;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let linalg_eigh_eigvals ~eigvals ~eigvecs self ~uplo =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_linalg_eigh_eigvals (CArray.start out__) eigvals eigvecs self uplo;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
-let linalg_eigvals self =
-  let out__ = CArray.make t 1 in
-  stubs_linalg_eigvals (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let linalg_eigvals_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_linalg_eigvals_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let linalg_eigvalsh self ~uplo =
-  let out__ = CArray.make t 1 in
-  stubs_linalg_eigvalsh (CArray.start out__) self uplo;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let linalg_eigvals self = stubs_linalg_eigvals self |> with_tensor_gc
+let linalg_eigvals_out ~out self = stubs_linalg_eigvals_out out self |> with_tensor_gc
+let linalg_eigvalsh self ~uplo = stubs_linalg_eigvalsh self uplo |> with_tensor_gc
 
 let linalg_eigvalsh_out ~out self ~uplo =
-  let out__ = CArray.make t 1 in
-  stubs_linalg_eigvalsh_out (CArray.start out__) out self uplo;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_linalg_eigvalsh_out out self uplo |> with_tensor_gc
 ;;
 
 let linalg_householder_product input ~tau =
-  let out__ = CArray.make t 1 in
-  stubs_linalg_householder_product (CArray.start out__) input tau;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_linalg_householder_product input tau |> with_tensor_gc
 ;;
 
 let linalg_householder_product_out ~out input ~tau =
-  let out__ = CArray.make t 1 in
-  stubs_linalg_householder_product_out (CArray.start out__) out input tau;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_linalg_householder_product_out out input tau |> with_tensor_gc
 ;;
 
-let linalg_inv ~a =
-  let out__ = CArray.make t 1 in
-  stubs_linalg_inv (CArray.start out__) a;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let linalg_inv ~a = stubs_linalg_inv a |> with_tensor_gc
 
 let linalg_inv_ex ~a ~check_errors =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_linalg_inv_ex (CArray.start out__) a (if check_errors then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let linalg_inv_ex_inverse ~inverse ~info ~a ~check_errors =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_linalg_inv_ex_inverse
     (CArray.start out__)
     inverse
     info
     a
     (if check_errors then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
-let linalg_inv_out ~out ~a =
-  let out__ = CArray.make t 1 in
-  stubs_linalg_inv_out (CArray.start out__) out a;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let linalg_inv_out ~out ~a = stubs_linalg_inv_out out a |> with_tensor_gc
 
 let linalg_ldl_factor self ~hermitian =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_linalg_ldl_factor (CArray.start out__) self (if hermitian then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let linalg_ldl_factor_ex self ~hermitian ~check_errors =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs_linalg_ldl_factor_ex
     (CArray.start out__)
     self
     (if hermitian then 1 else 0)
     (if check_errors then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
 let linalg_ldl_factor_ex_out ~ld ~pivots ~info self ~hermitian ~check_errors =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs_linalg_ldl_factor_ex_out
     (CArray.start out__)
     ld
@@ -18719,54 +12097,36 @@ let linalg_ldl_factor_ex_out ~ld ~pivots ~info self ~hermitian ~check_errors =
     self
     (if hermitian then 1 else 0)
     (if check_errors then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
 let linalg_ldl_factor_out ~ld ~pivots self ~hermitian =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_linalg_ldl_factor_out
     (CArray.start out__)
     ld
     pivots
     self
     (if hermitian then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let linalg_ldl_solve ~ld ~pivots ~b ~hermitian =
-  let out__ = CArray.make t 1 in
-  stubs_linalg_ldl_solve (CArray.start out__) ld pivots b (if hermitian then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_linalg_ldl_solve ld pivots b (if hermitian then 1 else 0) |> with_tensor_gc
 ;;
 
 let linalg_ldl_solve_out ~out ~ld ~pivots ~b ~hermitian =
-  let out__ = CArray.make t 1 in
-  stubs_linalg_ldl_solve_out
-    (CArray.start out__)
-    out
-    ld
-    pivots
-    b
-    (if hermitian then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_linalg_ldl_solve_out out ld pivots b (if hermitian then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let linalg_lstsq self ~b ~rcond ~driver =
-  let out__ = CArray.make t 4 in
+  let out__ = CArray.make raw_tensor 4 in
   stubs_linalg_lstsq
     (CArray.start out__)
     self
@@ -18776,19 +12136,15 @@ let linalg_lstsq self ~b ~rcond ~driver =
      | Some _ -> 0
      | None -> 1)
     driver;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
-  let t3 = CArray.get out__ 3 in
-  Gc.finalise C.Tensor.free t3;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
+  let t3 = CArray.get out__ 3 |> with_tensor_gc in
   t0, t1, t2, t3
 ;;
 
 let linalg_lstsq_out ~solution ~residuals ~rank ~singular_values self ~b ~rcond ~driver =
-  let out__ = CArray.make t 4 in
+  let out__ = CArray.make raw_tensor 4 in
   stubs_linalg_lstsq_out
     (CArray.start out__)
     solution
@@ -18802,57 +12158,45 @@ let linalg_lstsq_out ~solution ~residuals ~rank ~singular_values self ~b ~rcond
      | Some _ -> 0
      | None -> 1)
     driver;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
-  let t3 = CArray.get out__ 3 in
-  Gc.finalise C.Tensor.free t3;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
+  let t3 = CArray.get out__ 3 |> with_tensor_gc in
   t0, t1, t2, t3
 ;;
 
 let linalg_lu ~a ~pivot =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs_linalg_lu (CArray.start out__) a (if pivot then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
 let linalg_lu_factor ~a ~pivot =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_linalg_lu_factor (CArray.start out__) a (if pivot then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let linalg_lu_factor_ex ~a ~pivot ~check_errors =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs_linalg_lu_factor_ex
     (CArray.start out__)
     a
     (if pivot then 1 else 0)
     (if check_errors then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
 let linalg_lu_factor_ex_out ~lu ~pivots ~info ~a ~pivot ~check_errors =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs_linalg_lu_factor_ex_out
     (CArray.start out__)
     lu
@@ -18861,126 +12205,71 @@ let linalg_lu_factor_ex_out ~lu ~pivots ~info ~a ~pivot ~check_errors =
     a
     (if pivot then 1 else 0)
     (if check_errors then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
 let linalg_lu_factor_out ~lu ~pivots ~a ~pivot =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_linalg_lu_factor_out (CArray.start out__) lu pivots a (if pivot then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let linalg_lu_out ~p ~l ~u ~a ~pivot =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs_linalg_lu_out (CArray.start out__) p l u a (if pivot then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
 let linalg_lu_solve ~lu ~pivots ~b ~left ~adjoint =
-  let out__ = CArray.make t 1 in
-  stubs_linalg_lu_solve
-    (CArray.start out__)
-    lu
-    pivots
-    b
-    (if left then 1 else 0)
-    (if adjoint then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_linalg_lu_solve lu pivots b (if left then 1 else 0) (if adjoint then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let linalg_lu_solve_out ~out ~lu ~pivots ~b ~left ~adjoint =
-  let out__ = CArray.make t 1 in
   stubs_linalg_lu_solve_out
-    (CArray.start out__)
     out
     lu
     pivots
     b
     (if left then 1 else 0)
-    (if adjoint then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if adjoint then 1 else 0)
+  |> with_tensor_gc
 ;;
 
-let linalg_matmul self other =
-  let out__ = CArray.make t 1 in
-  stubs_linalg_matmul (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let linalg_matmul self other = stubs_linalg_matmul self other |> with_tensor_gc
 
 let linalg_matmul_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_linalg_matmul_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_linalg_matmul_out out self other |> with_tensor_gc
 ;;
 
-let linalg_matrix_exp self =
-  let out__ = CArray.make t 1 in
-  stubs_linalg_matrix_exp (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let linalg_matrix_exp self = stubs_linalg_matrix_exp self |> with_tensor_gc
 
 let linalg_matrix_exp_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_linalg_matrix_exp_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_linalg_matrix_exp_out out self |> with_tensor_gc
 ;;
 
 let linalg_matrix_power self ~n =
-  let out__ = CArray.make t 1 in
-  stubs_linalg_matrix_power (CArray.start out__) self (Int64.of_int n);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_linalg_matrix_power self (Int64.of_int n) |> with_tensor_gc
 ;;
 
 let linalg_matrix_power_out ~out self ~n =
-  let out__ = CArray.make t 1 in
-  stubs_linalg_matrix_power_out (CArray.start out__) out self (Int64.of_int n);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_linalg_matrix_power_out out self (Int64.of_int n) |> with_tensor_gc
 ;;
 
 let linalg_matrix_rank self ~tol ~hermitian =
-  let out__ = CArray.make t 1 in
-  stubs_linalg_matrix_rank (CArray.start out__) self tol (if hermitian then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_linalg_matrix_rank self tol (if hermitian then 1 else 0) |> with_tensor_gc
 ;;
 
 let linalg_matrix_rank_atol_rtol_float self ~atol ~rtol ~hermitian =
-  let out__ = CArray.make t 1 in
   stubs_linalg_matrix_rank_atol_rtol_float
-    (CArray.start out__)
     self
     (Option.value atol ~default:0.0)
     (match atol with
@@ -18990,16 +12279,12 @@ let linalg_matrix_rank_atol_rtol_float self ~atol ~rtol ~hermitian =
     (match rtol with
      | Some _ -> 0
      | None -> 1)
-    (if hermitian then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if hermitian then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let linalg_matrix_rank_atol_rtol_float_out ~out self ~atol ~rtol ~hermitian =
-  let out__ = CArray.make t 1 in
   stubs_linalg_matrix_rank_atol_rtol_float_out
-    (CArray.start out__)
     out
     self
     (Option.value atol ~default:0.0)
@@ -19010,120 +12295,72 @@ let linalg_matrix_rank_atol_rtol_float_out ~out self ~atol ~rtol ~hermitian =
     (match rtol with
      | Some _ -> 0
      | None -> 1)
-    (if hermitian then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if hermitian then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let linalg_matrix_rank_atol_rtol_tensor input ~atol ~rtol ~hermitian =
-  let out__ = CArray.make t 1 in
   stubs_linalg_matrix_rank_atol_rtol_tensor
-    (CArray.start out__)
     input
     (match atol with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match rtol with
      | Some v -> v
-     | None -> null)
-    (if hermitian then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+    (if hermitian then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let linalg_matrix_rank_atol_rtol_tensor_out ~out input ~atol ~rtol ~hermitian =
-  let out__ = CArray.make t 1 in
   stubs_linalg_matrix_rank_atol_rtol_tensor_out
-    (CArray.start out__)
     out
     input
     (match atol with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match rtol with
      | Some v -> v
-     | None -> null)
-    (if hermitian then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+    (if hermitian then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let linalg_matrix_rank_out ~out self ~tol ~hermitian =
-  let out__ = CArray.make t 1 in
-  stubs_linalg_matrix_rank_out
-    (CArray.start out__)
-    out
-    self
-    tol
-    (if hermitian then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_linalg_matrix_rank_out out self tol (if hermitian then 1 else 0) |> with_tensor_gc
 ;;
 
 let linalg_matrix_rank_out_tol_tensor ~out input ~tol ~hermitian =
-  let out__ = CArray.make t 1 in
-  stubs_linalg_matrix_rank_out_tol_tensor
-    (CArray.start out__)
-    out
-    input
-    tol
-    (if hermitian then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_linalg_matrix_rank_out_tol_tensor out input tol (if hermitian then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let linalg_matrix_rank_tol_tensor input ~tol ~hermitian =
-  let out__ = CArray.make t 1 in
-  stubs_linalg_matrix_rank_tol_tensor
-    (CArray.start out__)
-    input
-    tol
-    (if hermitian then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_linalg_matrix_rank_tol_tensor input tol (if hermitian then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let linalg_multi_dot tensors =
-  let out__ = CArray.make t 1 in
   stubs_linalg_multi_dot
-    (CArray.start out__)
-    (CArray.of_list t tensors |> CArray.start)
-    (List.length tensors);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (CArray.of_list gc_tensor tensors |> CArray.start)
+    (List.length tensors)
+  |> with_tensor_gc
 ;;
 
 let linalg_multi_dot_out ~out tensors =
-  let out__ = CArray.make t 1 in
   stubs_linalg_multi_dot_out
-    (CArray.start out__)
     out
-    (CArray.of_list t tensors |> CArray.start)
-    (List.length tensors);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (CArray.of_list gc_tensor tensors |> CArray.start)
+    (List.length tensors)
+  |> with_tensor_gc
 ;;
 
 let linalg_pinv self ~rcond ~hermitian =
-  let out__ = CArray.make t 1 in
-  stubs_linalg_pinv (CArray.start out__) self rcond (if hermitian then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_linalg_pinv self rcond (if hermitian then 1 else 0) |> with_tensor_gc
 ;;
 
 let linalg_pinv_atol_rtol_float self ~atol ~rtol ~hermitian =
-  let out__ = CArray.make t 1 in
   stubs_linalg_pinv_atol_rtol_float
-    (CArray.start out__)
     self
     (Option.value atol ~default:0.0)
     (match atol with
@@ -19133,16 +12370,12 @@ let linalg_pinv_atol_rtol_float self ~atol ~rtol ~hermitian =
     (match rtol with
      | Some _ -> 0
      | None -> 1)
-    (if hermitian then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if hermitian then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let linalg_pinv_atol_rtol_float_out ~out self ~atol ~rtol ~hermitian =
-  let out__ = CArray.make t 1 in
   stubs_linalg_pinv_atol_rtol_float_out
-    (CArray.start out__)
     out
     self
     (Option.value atol ~default:0.0)
@@ -19153,145 +12386,101 @@ let linalg_pinv_atol_rtol_float_out ~out self ~atol ~rtol ~hermitian =
     (match rtol with
      | Some _ -> 0
      | None -> 1)
-    (if hermitian then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if hermitian then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let linalg_pinv_atol_rtol_tensor self ~atol ~rtol ~hermitian =
-  let out__ = CArray.make t 1 in
   stubs_linalg_pinv_atol_rtol_tensor
-    (CArray.start out__)
     self
     (match atol with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match rtol with
      | Some v -> v
-     | None -> null)
-    (if hermitian then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+    (if hermitian then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let linalg_pinv_atol_rtol_tensor_out ~out self ~atol ~rtol ~hermitian =
-  let out__ = CArray.make t 1 in
   stubs_linalg_pinv_atol_rtol_tensor_out
-    (CArray.start out__)
     out
     self
     (match atol with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match rtol with
      | Some v -> v
-     | None -> null)
-    (if hermitian then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+    (if hermitian then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let linalg_pinv_out ~out self ~rcond ~hermitian =
-  let out__ = CArray.make t 1 in
-  stubs_linalg_pinv_out (CArray.start out__) out self rcond (if hermitian then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let linalg_pinv_out_rcond_tensor ~out self ~rcond ~hermitian =
-  let out__ = CArray.make t 1 in
-  stubs_linalg_pinv_out_rcond_tensor
-    (CArray.start out__)
-    out
-    self
-    rcond
-    (if hermitian then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_linalg_pinv_out out self rcond (if hermitian then 1 else 0) |> with_tensor_gc
+;;
+
+let linalg_pinv_out_rcond_tensor ~out self ~rcond ~hermitian =
+  stubs_linalg_pinv_out_rcond_tensor out self rcond (if hermitian then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let linalg_pinv_rcond_tensor self ~rcond ~hermitian =
-  let out__ = CArray.make t 1 in
-  stubs_linalg_pinv_rcond_tensor
-    (CArray.start out__)
-    self
-    rcond
-    (if hermitian then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_linalg_pinv_rcond_tensor self rcond (if hermitian then 1 else 0) |> with_tensor_gc
 ;;
 
 let linalg_qr ~a ~mode =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_linalg_qr (CArray.start out__) a mode;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let linalg_qr_out ~q ~r ~a ~mode =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_linalg_qr_out (CArray.start out__) q r a mode;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let linalg_slogdet ~a =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_linalg_slogdet (CArray.start out__) a;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let linalg_slogdet_out ~sign ~logabsdet ~a =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_linalg_slogdet_out (CArray.start out__) sign logabsdet a;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let linalg_solve ~a ~b ~left =
-  let out__ = CArray.make t 1 in
-  stubs_linalg_solve (CArray.start out__) a b (if left then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_linalg_solve a b (if left then 1 else 0) |> with_tensor_gc
 ;;
 
 let linalg_solve_ex ~a ~b ~left ~check_errors =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_linalg_solve_ex
     (CArray.start out__)
     a
     b
     (if left then 1 else 0)
     (if check_errors then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let linalg_solve_ex_out result ~info ~a ~b ~left ~check_errors =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_linalg_solve_ex_out
     (CArray.start out__)
     result
@@ -19300,110 +12489,70 @@ let linalg_solve_ex_out result ~info ~a ~b ~left ~check_errors =
     b
     (if left then 1 else 0)
     (if check_errors then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let linalg_solve_out ~out ~a ~b ~left =
-  let out__ = CArray.make t 1 in
-  stubs_linalg_solve_out (CArray.start out__) out a b (if left then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_linalg_solve_out out a b (if left then 1 else 0) |> with_tensor_gc
 ;;
 
 let linalg_solve_triangular self ~b ~upper ~left ~unitriangular =
-  let out__ = CArray.make t 1 in
   stubs_linalg_solve_triangular
-    (CArray.start out__)
     self
     b
     (if upper then 1 else 0)
     (if left then 1 else 0)
-    (if unitriangular then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if unitriangular then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let linalg_solve_triangular_out ~out self ~b ~upper ~left ~unitriangular =
-  let out__ = CArray.make t 1 in
   stubs_linalg_solve_triangular_out
-    (CArray.start out__)
     out
     self
     b
     (if upper then 1 else 0)
     (if left then 1 else 0)
-    (if unitriangular then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if unitriangular then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let linalg_svd ~a ~full_matrices ~driver =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs_linalg_svd (CArray.start out__) a (if full_matrices then 1 else 0) driver;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
 let linalg_svd_u ~u ~s ~vh ~a ~full_matrices ~driver =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs_linalg_svd_u (CArray.start out__) u s vh a (if full_matrices then 1 else 0) driver;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
-let linalg_svdvals ~a ~driver =
-  let out__ = CArray.make t 1 in
-  stubs_linalg_svdvals (CArray.start out__) a driver;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let linalg_svdvals ~a ~driver = stubs_linalg_svdvals a driver |> with_tensor_gc
 
 let linalg_svdvals_out ~out ~a ~driver =
-  let out__ = CArray.make t 1 in
-  stubs_linalg_svdvals_out (CArray.start out__) out a driver;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_linalg_svdvals_out out a driver |> with_tensor_gc
 ;;
 
 let linalg_tensorinv self ~ind =
-  let out__ = CArray.make t 1 in
-  stubs_linalg_tensorinv (CArray.start out__) self (Int64.of_int ind);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_linalg_tensorinv self (Int64.of_int ind) |> with_tensor_gc
 ;;
 
 let linalg_tensorinv_out ~out self ~ind =
-  let out__ = CArray.make t 1 in
-  stubs_linalg_tensorinv_out (CArray.start out__) out self (Int64.of_int ind);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_linalg_tensorinv_out out self (Int64.of_int ind) |> with_tensor_gc
 ;;
 
 let linalg_tensorsolve self other ~dims =
-  let out__ = CArray.make t 1 in
   stubs_linalg_tensorsolve
-    (CArray.start out__)
     self
     other
     (match dims with
@@ -19411,16 +12560,12 @@ let linalg_tensorsolve self other ~dims =
      | Some v -> List.map Int64.of_int v |> CArray.of_list int64_t |> CArray.start)
     (match dims with
      | None -> -1
-     | Some v -> List.length v);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | Some v -> List.length v)
+  |> with_tensor_gc
 ;;
 
 let linalg_tensorsolve_out ~out self other ~dims =
-  let out__ = CArray.make t 1 in
   stubs_linalg_tensorsolve_out
-    (CArray.start out__)
     out
     self
     other
@@ -19429,551 +12574,223 @@ let linalg_tensorsolve_out ~out self other ~dims =
      | Some v -> List.map Int64.of_int v |> CArray.of_list int64_t |> CArray.start)
     (match dims with
      | None -> -1
-     | Some v -> List.length v);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | Some v -> List.length v)
+  |> with_tensor_gc
 ;;
 
 let linalg_vander ~x ~n =
-  let out__ = CArray.make t 1 in
   stubs_linalg_vander
-    (CArray.start out__)
     x
     (match n with
      | None -> Int64.zero
      | Some v -> Int64.of_int v)
     (match n with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let linalg_vecdot ~x ~y ~dim =
-  let out__ = CArray.make t 1 in
-  stubs_linalg_vecdot (CArray.start out__) x y (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_linalg_vecdot x y (Int64.of_int dim) |> with_tensor_gc
 ;;
 
 let linalg_vecdot_out ~out ~x ~y ~dim =
-  let out__ = CArray.make t 1 in
-  stubs_linalg_vecdot_out (CArray.start out__) out x y (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_linalg_vecdot_out out x y (Int64.of_int dim) |> with_tensor_gc
 ;;
 
 let linear input ~weight ~bias =
-  let out__ = CArray.make t 1 in
   stubs_linear
-    (CArray.start out__)
     input
     weight
     (match bias with
      | Some v -> v
-     | None -> null);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+  |> with_tensor_gc
 ;;
 
 let linear_out ~out input ~weight ~bias =
-  let out__ = CArray.make t 1 in
   stubs_linear_out
-    (CArray.start out__)
     out
     input
     weight
     (match bias with
      | Some v -> v
-     | None -> null);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+  |> with_tensor_gc
 ;;
 
 let linspace ~start ~end_ ~steps ~options =
-  let out__ = CArray.make t 1 in
   stubs_linspace
-    (CArray.start out__)
     start
     end_
     (Int64.of_int steps)
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let linspace_out ~out ~start ~end_ ~steps =
-  let out__ = CArray.make t 1 in
-  stubs_linspace_out (CArray.start out__) out start end_ (Int64.of_int steps);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let log self =
-  let out__ = CArray.make t 1 in
-  stubs_log (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let log10 self =
-  let out__ = CArray.make t 1 in
-  stubs_log10 (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let log10_ self =
-  let out__ = CArray.make t 1 in
-  stubs_log10_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let log10_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_log10_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let log1p self =
-  let out__ = CArray.make t 1 in
-  stubs_log1p (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let log1p_ self =
-  let out__ = CArray.make t 1 in
-  stubs_log1p_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let log1p_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_log1p_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let log2 self =
-  let out__ = CArray.make t 1 in
-  stubs_log2 (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let log2_ self =
-  let out__ = CArray.make t 1 in
-  stubs_log2_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let log2_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_log2_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let log_ self =
-  let out__ = CArray.make t 1 in
-  stubs_log_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let log_normal self ~mean ~std =
-  let out__ = CArray.make t 1 in
-  stubs_log_normal (CArray.start out__) self mean std;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let log_normal_ self ~mean ~std =
-  let out__ = CArray.make t 1 in
-  stubs_log_normal_ (CArray.start out__) self mean std;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+  stubs_linspace_out out start end_ (Int64.of_int steps) |> with_tensor_gc
+;;
+
+let log self = stubs_log self |> with_tensor_gc
+let log10 self = stubs_log10 self |> with_tensor_gc
+let log10_ self = stubs_log10_ self |> with_tensor_gc
+let log10_out ~out self = stubs_log10_out out self |> with_tensor_gc
+let log1p self = stubs_log1p self |> with_tensor_gc
+let log1p_ self = stubs_log1p_ self |> with_tensor_gc
+let log1p_out ~out self = stubs_log1p_out out self |> with_tensor_gc
+let log2 self = stubs_log2 self |> with_tensor_gc
+let log2_ self = stubs_log2_ self |> with_tensor_gc
+let log2_out ~out self = stubs_log2_out out self |> with_tensor_gc
+let log_ self = stubs_log_ self |> with_tensor_gc
+let log_normal self ~mean ~std = stubs_log_normal self mean std |> with_tensor_gc
+let log_normal_ self ~mean ~std = stubs_log_normal_ self mean std |> with_tensor_gc
 
 let log_normal_out ~out self ~mean ~std =
-  let out__ = CArray.make t 1 in
-  stubs_log_normal_out (CArray.start out__) out self mean std;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let log_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_log_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_log_normal_out out self mean std |> with_tensor_gc
 ;;
 
-let log_sigmoid self =
-  let out__ = CArray.make t 1 in
-  stubs_log_sigmoid (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let log_out ~out self = stubs_log_out out self |> with_tensor_gc
+let log_sigmoid self = stubs_log_sigmoid self |> with_tensor_gc
 
 let log_sigmoid_backward ~grad_output self ~buffer =
-  let out__ = CArray.make t 1 in
-  stubs_log_sigmoid_backward (CArray.start out__) grad_output self buffer;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_log_sigmoid_backward grad_output self buffer |> with_tensor_gc
 ;;
 
 let log_sigmoid_backward_grad_input ~grad_input ~grad_output self ~buffer =
-  let out__ = CArray.make t 1 in
-  stubs_log_sigmoid_backward_grad_input
-    (CArray.start out__)
-    grad_input
-    grad_output
-    self
-    buffer;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_log_sigmoid_backward_grad_input grad_input grad_output self buffer
+  |> with_tensor_gc
 ;;
 
-let log_sigmoid_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_log_sigmoid_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let log_sigmoid_out ~out self = stubs_log_sigmoid_out out self |> with_tensor_gc
 
 let log_softmax self ~dim ~dtype =
-  let out__ = CArray.make t 1 in
-  stubs_log_softmax
-    (CArray.start out__)
-    self
-    (Int64.of_int dim)
-    (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_log_softmax self (Int64.of_int dim) (Kind.packed_to_int dtype) |> with_tensor_gc
 ;;
 
 let log_softmax_int_out ~out self ~dim ~dtype =
-  let out__ = CArray.make t 1 in
-  stubs_log_softmax_int_out
-    (CArray.start out__)
-    out
-    self
-    (Int64.of_int dim)
-    (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let logaddexp self other =
-  let out__ = CArray.make t 1 in
-  stubs_logaddexp (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let logaddexp2 self other =
-  let out__ = CArray.make t 1 in
-  stubs_logaddexp2 (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let logaddexp2_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_logaddexp2_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_log_softmax_int_out out self (Int64.of_int dim) (Kind.packed_to_int dtype)
+  |> with_tensor_gc
 ;;
 
-let logaddexp_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_logaddexp_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let logcumsumexp self ~dim =
-  let out__ = CArray.make t 1 in
-  stubs_logcumsumexp (CArray.start out__) self (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let logaddexp self other = stubs_logaddexp self other |> with_tensor_gc
+let logaddexp2 self other = stubs_logaddexp2 self other |> with_tensor_gc
+let logaddexp2_out ~out self other = stubs_logaddexp2_out out self other |> with_tensor_gc
+let logaddexp_out ~out self other = stubs_logaddexp_out out self other |> with_tensor_gc
+let logcumsumexp self ~dim = stubs_logcumsumexp self (Int64.of_int dim) |> with_tensor_gc
 
 let logcumsumexp_out ~out self ~dim =
-  let out__ = CArray.make t 1 in
-  stubs_logcumsumexp_out (CArray.start out__) out self (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let logdet self =
-  let out__ = CArray.make t 1 in
-  stubs_logdet (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let logical_and self other =
-  let out__ = CArray.make t 1 in
-  stubs_logical_and (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_logcumsumexp_out out self (Int64.of_int dim) |> with_tensor_gc
 ;;
 
-let logical_and_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_logical_and_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let logdet self = stubs_logdet self |> with_tensor_gc
+let logical_and self other = stubs_logical_and self other |> with_tensor_gc
+let logical_and_ self other = stubs_logical_and_ self other |> with_tensor_gc
 
 let logical_and_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_logical_and_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_logical_and_out out self other |> with_tensor_gc
 ;;
 
-let logical_not self =
-  let out__ = CArray.make t 1 in
-  stubs_logical_not (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let logical_not_ self =
-  let out__ = CArray.make t 1 in
-  stubs_logical_not_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let logical_not_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_logical_not_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let logical_or self other =
-  let out__ = CArray.make t 1 in
-  stubs_logical_or (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let logical_or_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_logical_or_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let logical_or_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_logical_or_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let logical_xor self other =
-  let out__ = CArray.make t 1 in
-  stubs_logical_xor (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let logical_xor_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_logical_xor_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let logical_not self = stubs_logical_not self |> with_tensor_gc
+let logical_not_ self = stubs_logical_not_ self |> with_tensor_gc
+let logical_not_out ~out self = stubs_logical_not_out out self |> with_tensor_gc
+let logical_or self other = stubs_logical_or self other |> with_tensor_gc
+let logical_or_ self other = stubs_logical_or_ self other |> with_tensor_gc
+let logical_or_out ~out self other = stubs_logical_or_out out self other |> with_tensor_gc
+let logical_xor self other = stubs_logical_xor self other |> with_tensor_gc
+let logical_xor_ self other = stubs_logical_xor_ self other |> with_tensor_gc
 
 let logical_xor_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_logical_xor_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_logical_xor_out out self other |> with_tensor_gc
 ;;
 
 let logit self ~eps =
-  let out__ = CArray.make t 1 in
   stubs_logit
-    (CArray.start out__)
     self
     (Option.value eps ~default:0.0)
     (match eps with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let logit_ self ~eps =
-  let out__ = CArray.make t 1 in
   stubs_logit_
-    (CArray.start out__)
     self
     (Option.value eps ~default:0.0)
     (match eps with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let logit_backward ~grad_output self ~eps =
-  let out__ = CArray.make t 1 in
   stubs_logit_backward
-    (CArray.start out__)
     grad_output
     self
     (Option.value eps ~default:0.0)
     (match eps with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let logit_backward_grad_input ~grad_input ~grad_output self ~eps =
-  let out__ = CArray.make t 1 in
   stubs_logit_backward_grad_input
-    (CArray.start out__)
     grad_input
     grad_output
     self
     (Option.value eps ~default:0.0)
     (match eps with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let logit_out ~out self ~eps =
-  let out__ = CArray.make t 1 in
   stubs_logit_out
-    (CArray.start out__)
     out
     self
     (Option.value eps ~default:0.0)
     (match eps with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let logspace ~start ~end_ ~steps ~base ~options =
-  let out__ = CArray.make t 1 in
   stubs_logspace
-    (CArray.start out__)
     start
     end_
     (Int64.of_int steps)
     base
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let logspace_out ~out ~start ~end_ ~steps ~base =
-  let out__ = CArray.make t 1 in
-  stubs_logspace_out (CArray.start out__) out start end_ (Int64.of_int steps) base;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_logspace_out out start end_ (Int64.of_int steps) base |> with_tensor_gc
 ;;
 
 let logsumexp self ~dim ~keepdim =
-  let out__ = CArray.make t 1 in
   stubs_logsumexp
-    (CArray.start out__)
     self
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
     (List.length dim)
-    (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if keepdim then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let logsumexp_out ~out self ~dim ~keepdim =
-  let out__ = CArray.make t 1 in
   stubs_logsumexp_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
     (List.length dim)
-    (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if keepdim then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let lstm
@@ -19987,13 +12804,13 @@ let lstm
   ~bidirectional
   ~batch_first
   =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs_lstm
     (CArray.start out__)
     input
-    (CArray.of_list t hx |> CArray.start)
+    (CArray.of_list gc_tensor hx |> CArray.start)
     (List.length hx)
-    (CArray.of_list t params |> CArray.start)
+    (CArray.of_list gc_tensor params |> CArray.start)
     (List.length params)
     (if has_biases then 1 else 0)
     (Int64.of_int num_layers)
@@ -20001,34 +12818,29 @@ let lstm
     (if train then 1 else 0)
     (if bidirectional then 1 else 0)
     (if batch_first then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
 let lstm_cell input ~hx ~w_ih ~w_hh ~b_ih ~b_hh =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_lstm_cell
     (CArray.start out__)
     input
-    (CArray.of_list t hx |> CArray.start)
+    (CArray.of_list gc_tensor hx |> CArray.start)
     (List.length hx)
     w_ih
     w_hh
     (match b_ih with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match b_hh with
      | Some v -> v
-     | None -> null);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+     | None -> none_gc_tensor);
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
@@ -20043,26 +12855,23 @@ let lstm_data
   ~train
   ~bidirectional
   =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs_lstm_data
     (CArray.start out__)
     data
     batch_sizes
-    (CArray.of_list t hx |> CArray.start)
+    (CArray.of_list gc_tensor hx |> CArray.start)
     (List.length hx)
-    (CArray.of_list t params |> CArray.start)
+    (CArray.of_list gc_tensor params |> CArray.start)
     (List.length params)
     (if has_biases then 1 else 0)
     (Int64.of_int num_layers)
     dropout
     (if train then 1 else 0)
     (if bidirectional then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
@@ -20088,26 +12897,26 @@ let lstm_mps_backward
   =
   stubs_lstm_mps_backward
     out0
-    (CArray.of_list t out1 |> CArray.start)
+    (CArray.of_list gc_tensor out1 |> CArray.start)
     (List.length out1)
-    (CArray.of_list t out2 |> CArray.start)
+    (CArray.of_list gc_tensor out2 |> CArray.start)
     (List.length out2)
     (match grad_y with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match grad_hy with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match grad_cy with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     z_state
     cell_state_fwd
     input
     layersoutputs
-    (CArray.of_list t hx |> CArray.start)
+    (CArray.of_list gc_tensor hx |> CArray.start)
     (List.length hx)
-    (CArray.of_list t params |> CArray.start)
+    (CArray.of_list gc_tensor params |> CArray.start)
     (List.length params)
     (if has_biases then 1 else 0)
     (Int64.of_int num_layers)
@@ -20117,89 +12926,37 @@ let lstm_mps_backward
     (if batch_first then 1 else 0)
 ;;
 
-let lt self other =
-  let out__ = CArray.make t 1 in
-  stubs_lt (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let lt_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_lt_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let lt_scalar_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_lt_scalar_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let lt_tensor self other =
-  let out__ = CArray.make t 1 in
-  stubs_lt_tensor (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let lt_tensor_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_lt_tensor_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let lt_tensor_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_lt_tensor_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let lt self other = stubs_lt self other |> with_tensor_gc
+let lt_ self other = stubs_lt_ self other |> with_tensor_gc
+let lt_scalar_out ~out self other = stubs_lt_scalar_out out self other |> with_tensor_gc
+let lt_tensor self other = stubs_lt_tensor self other |> with_tensor_gc
+let lt_tensor_ self other = stubs_lt_tensor_ self other |> with_tensor_gc
+let lt_tensor_out ~out self other = stubs_lt_tensor_out out self other |> with_tensor_gc
 
 let lu_solve self ~lu_data ~lu_pivots =
-  let out__ = CArray.make t 1 in
-  stubs_lu_solve (CArray.start out__) self lu_data lu_pivots;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_lu_solve self lu_data lu_pivots |> with_tensor_gc
 ;;
 
 let lu_solve_out ~out self ~lu_data ~lu_pivots =
-  let out__ = CArray.make t 1 in
-  stubs_lu_solve_out (CArray.start out__) out self lu_data lu_pivots;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_lu_solve_out out self lu_data lu_pivots |> with_tensor_gc
 ;;
 
 let lu_unpack ~lu_data ~lu_pivots ~unpack_data ~unpack_pivots =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs_lu_unpack
     (CArray.start out__)
     lu_data
     lu_pivots
     (if unpack_data then 1 else 0)
     (if unpack_pivots then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
 let lu_unpack_out ~p ~l ~u ~lu_data ~lu_pivots ~unpack_data ~unpack_pivots =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs_lu_unpack_out
     (CArray.start out__)
     p
@@ -20209,201 +12966,86 @@ let lu_unpack_out ~p ~l ~u ~lu_data ~lu_pivots ~unpack_data ~unpack_pivots =
     lu_pivots
     (if unpack_data then 1 else 0)
     (if unpack_pivots then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
 let margin_ranking_loss ~input1 ~input2 ~target ~margin ~reduction =
-  let out__ = CArray.make t 1 in
   stubs_margin_ranking_loss
-    (CArray.start out__)
     input1
     input2
     target
     margin
-    (Reduction.to_int reduction |> Int64.of_int);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let masked_fill self ~mask ~value =
-  let out__ = CArray.make t 1 in
-  stubs_masked_fill (CArray.start out__) self mask value;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Reduction.to_int reduction |> Int64.of_int)
+  |> with_tensor_gc
 ;;
 
-let masked_fill_ self ~mask ~value =
-  let out__ = CArray.make t 1 in
-  stubs_masked_fill_ (CArray.start out__) self mask value;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let masked_fill self ~mask ~value = stubs_masked_fill self mask value |> with_tensor_gc
+let masked_fill_ self ~mask ~value = stubs_masked_fill_ self mask value |> with_tensor_gc
 
 let masked_fill_scalar_out ~out self ~mask ~value =
-  let out__ = CArray.make t 1 in
-  stubs_masked_fill_scalar_out (CArray.start out__) out self mask value;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_masked_fill_scalar_out out self mask value |> with_tensor_gc
 ;;
 
 let masked_fill_tensor self ~mask ~value =
-  let out__ = CArray.make t 1 in
-  stubs_masked_fill_tensor (CArray.start out__) self mask value;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_masked_fill_tensor self mask value |> with_tensor_gc
 ;;
 
 let masked_fill_tensor_ self ~mask ~value =
-  let out__ = CArray.make t 1 in
-  stubs_masked_fill_tensor_ (CArray.start out__) self mask value;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_masked_fill_tensor_ self mask value |> with_tensor_gc
 ;;
 
 let masked_fill_tensor_out ~out self ~mask ~value =
-  let out__ = CArray.make t 1 in
-  stubs_masked_fill_tensor_out (CArray.start out__) out self mask value;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_masked_fill_tensor_out out self mask value |> with_tensor_gc
 ;;
 
 let masked_scatter self ~mask ~source =
-  let out__ = CArray.make t 1 in
-  stubs_masked_scatter (CArray.start out__) self mask source;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_masked_scatter self mask source |> with_tensor_gc
 ;;
 
 let masked_scatter_ self ~mask ~source =
-  let out__ = CArray.make t 1 in
-  stubs_masked_scatter_ (CArray.start out__) self mask source;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_masked_scatter_ self mask source |> with_tensor_gc
 ;;
 
 let masked_scatter_out ~out self ~mask ~source =
-  let out__ = CArray.make t 1 in
-  stubs_masked_scatter_out (CArray.start out__) out self mask source;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_masked_scatter_out out self mask source |> with_tensor_gc
 ;;
 
-let masked_select self ~mask =
-  let out__ = CArray.make t 1 in
-  stubs_masked_select (CArray.start out__) self mask;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let masked_select self ~mask = stubs_masked_select self mask |> with_tensor_gc
 
 let masked_select_backward ~grad input ~mask =
-  let out__ = CArray.make t 1 in
-  stubs_masked_select_backward (CArray.start out__) grad input mask;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_masked_select_backward grad input mask |> with_tensor_gc
 ;;
 
 let masked_select_out ~out self ~mask =
-  let out__ = CArray.make t 1 in
-  stubs_masked_select_out (CArray.start out__) out self mask;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let matmul self other =
-  let out__ = CArray.make t 1 in
-  stubs_matmul (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let matmul_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_matmul_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let matrix_exp self =
-  let out__ = CArray.make t 1 in
-  stubs_matrix_exp (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let matrix_exp_backward self ~grad =
-  let out__ = CArray.make t 1 in
-  stubs_matrix_exp_backward (CArray.start out__) self grad;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let matrix_h self =
-  let out__ = CArray.make t 1 in
-  stubs_matrix_h (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_masked_select_out out self mask |> with_tensor_gc
 ;;
 
-let matrix_power self ~n =
-  let out__ = CArray.make t 1 in
-  stubs_matrix_power (CArray.start out__) self (Int64.of_int n);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let matmul self other = stubs_matmul self other |> with_tensor_gc
+let matmul_out ~out self other = stubs_matmul_out out self other |> with_tensor_gc
+let matrix_exp self = stubs_matrix_exp self |> with_tensor_gc
+let matrix_exp_backward self ~grad = stubs_matrix_exp_backward self grad |> with_tensor_gc
+let matrix_h self = stubs_matrix_h self |> with_tensor_gc
+let matrix_power self ~n = stubs_matrix_power self (Int64.of_int n) |> with_tensor_gc
 
 let matrix_power_out ~out self ~n =
-  let out__ = CArray.make t 1 in
-  stubs_matrix_power_out (CArray.start out__) out self (Int64.of_int n);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_matrix_power_out out self (Int64.of_int n) |> with_tensor_gc
 ;;
 
-let max self =
-  let out__ = CArray.make t 1 in
-  stubs_max (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let max self = stubs_max self |> with_tensor_gc
 
 let max_dim self ~dim ~keepdim =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_max_dim (CArray.start out__) self (Int64.of_int dim) (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let max_dim_max ~max ~max_values self ~dim ~keepdim =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_max_dim_max
     (CArray.start out__)
     max
@@ -20411,33 +13053,16 @@ let max_dim_max ~max ~max_values self ~dim ~keepdim =
     self
     (Int64.of_int dim)
     (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
-let max_other self other =
-  let out__ = CArray.make t 1 in
-  stubs_max_other (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let max_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_max_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let max_other self other = stubs_max_other self other |> with_tensor_gc
+let max_out ~out self other = stubs_max_out out self other |> with_tensor_gc
 
 let max_pool1d self ~kernel_size ~stride ~padding ~dilation ~ceil_mode =
-  let out__ = CArray.make t 1 in
   stubs_max_pool1d
-    (CArray.start out__)
     self
     (List.map Int64.of_int kernel_size |> CArray.of_list int64_t |> CArray.start)
     (List.length kernel_size)
@@ -20447,14 +13072,12 @@ let max_pool1d self ~kernel_size ~stride ~padding ~dilation ~ceil_mode =
     (List.length padding)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
-    (if ceil_mode then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if ceil_mode then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let max_pool1d_with_indices self ~kernel_size ~stride ~padding ~dilation ~ceil_mode =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_max_pool1d_with_indices
     (CArray.start out__)
     self
@@ -20467,17 +13090,13 @@ let max_pool1d_with_indices self ~kernel_size ~stride ~padding ~dilation ~ceil_m
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
     (if ceil_mode then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let max_pool2d self ~kernel_size ~stride ~padding ~dilation ~ceil_mode =
-  let out__ = CArray.make t 1 in
   stubs_max_pool2d
-    (CArray.start out__)
     self
     (List.map Int64.of_int kernel_size |> CArray.of_list int64_t |> CArray.start)
     (List.length kernel_size)
@@ -20487,10 +13106,8 @@ let max_pool2d self ~kernel_size ~stride ~padding ~dilation ~ceil_mode =
     (List.length padding)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
-    (if ceil_mode then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if ceil_mode then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let max_pool2d_backward
@@ -20502,9 +13119,7 @@ let max_pool2d_backward
   ~dilation
   ~ceil_mode
   =
-  let out__ = CArray.make t 1 in
   stubs_max_pool2d_backward
-    (CArray.start out__)
     grad_output
     self
     (List.map Int64.of_int kernel_size |> CArray.of_list int64_t |> CArray.start)
@@ -20515,10 +13130,8 @@ let max_pool2d_backward
     (List.length padding)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
-    (if ceil_mode then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if ceil_mode then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let max_pool2d_backward_out
@@ -20531,9 +13144,7 @@ let max_pool2d_backward_out
   ~dilation
   ~ceil_mode
   =
-  let out__ = CArray.make t 1 in
   stubs_max_pool2d_backward_out
-    (CArray.start out__)
     out
     grad_output
     self
@@ -20545,14 +13156,12 @@ let max_pool2d_backward_out
     (List.length padding)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
-    (if ceil_mode then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if ceil_mode then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let max_pool2d_with_indices self ~kernel_size ~stride ~padding ~dilation ~ceil_mode =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_max_pool2d_with_indices
     (CArray.start out__)
     self
@@ -20565,10 +13174,8 @@ let max_pool2d_with_indices self ~kernel_size ~stride ~padding ~dilation ~ceil_m
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
     (if ceil_mode then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
@@ -20582,9 +13189,7 @@ let max_pool2d_with_indices_backward
   ~ceil_mode
   ~indices
   =
-  let out__ = CArray.make t 1 in
   stubs_max_pool2d_with_indices_backward
-    (CArray.start out__)
     grad_output
     self
     (List.map Int64.of_int kernel_size |> CArray.of_list int64_t |> CArray.start)
@@ -20596,10 +13201,8 @@ let max_pool2d_with_indices_backward
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
     (if ceil_mode then 1 else 0)
-    indices;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    indices
+  |> with_tensor_gc
 ;;
 
 let max_pool2d_with_indices_backward_grad_input
@@ -20613,9 +13216,7 @@ let max_pool2d_with_indices_backward_grad_input
   ~ceil_mode
   ~indices
   =
-  let out__ = CArray.make t 1 in
   stubs_max_pool2d_with_indices_backward_grad_input
-    (CArray.start out__)
     grad_input
     grad_output
     self
@@ -20628,10 +13229,8 @@ let max_pool2d_with_indices_backward_grad_input
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
     (if ceil_mode then 1 else 0)
-    indices;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    indices
+  |> with_tensor_gc
 ;;
 
 let max_pool2d_with_indices_out
@@ -20644,7 +13243,7 @@ let max_pool2d_with_indices_out
   ~dilation
   ~ceil_mode
   =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_max_pool2d_with_indices_out
     (CArray.start out__)
     out
@@ -20659,17 +13258,13 @@ let max_pool2d_with_indices_out
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
     (if ceil_mode then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let max_pool3d self ~kernel_size ~stride ~padding ~dilation ~ceil_mode =
-  let out__ = CArray.make t 1 in
   stubs_max_pool3d
-    (CArray.start out__)
     self
     (List.map Int64.of_int kernel_size |> CArray.of_list int64_t |> CArray.start)
     (List.length kernel_size)
@@ -20679,14 +13274,12 @@ let max_pool3d self ~kernel_size ~stride ~padding ~dilation ~ceil_mode =
     (List.length padding)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
-    (if ceil_mode then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if ceil_mode then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let max_pool3d_with_indices self ~kernel_size ~stride ~padding ~dilation ~ceil_mode =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_max_pool3d_with_indices
     (CArray.start out__)
     self
@@ -20699,10 +13292,8 @@ let max_pool3d_with_indices self ~kernel_size ~stride ~padding ~dilation ~ceil_m
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
     (if ceil_mode then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
@@ -20716,9 +13307,7 @@ let max_pool3d_with_indices_backward
   ~ceil_mode
   ~indices
   =
-  let out__ = CArray.make t 1 in
   stubs_max_pool3d_with_indices_backward
-    (CArray.start out__)
     grad_output
     self
     (List.map Int64.of_int kernel_size |> CArray.of_list int64_t |> CArray.start)
@@ -20730,10 +13319,8 @@ let max_pool3d_with_indices_backward
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
     (if ceil_mode then 1 else 0)
-    indices;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    indices
+  |> with_tensor_gc
 ;;
 
 let max_pool3d_with_indices_backward_grad_input
@@ -20747,9 +13334,7 @@ let max_pool3d_with_indices_backward_grad_input
   ~ceil_mode
   ~indices
   =
-  let out__ = CArray.make t 1 in
   stubs_max_pool3d_with_indices_backward_grad_input
-    (CArray.start out__)
     grad_input
     grad_output
     self
@@ -20762,10 +13347,8 @@ let max_pool3d_with_indices_backward_grad_input
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
     (if ceil_mode then 1 else 0)
-    indices;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    indices
+  |> with_tensor_gc
 ;;
 
 let max_pool3d_with_indices_out
@@ -20778,7 +13361,7 @@ let max_pool3d_with_indices_out
   ~dilation
   ~ceil_mode
   =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_max_pool3d_with_indices_out
     (CArray.start out__)
     out
@@ -20793,52 +13376,34 @@ let max_pool3d_with_indices_out
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
     (if ceil_mode then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
-let max_unary_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_max_unary_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let max_unary_out ~out self = stubs_max_unary_out out self |> with_tensor_gc
 
 let max_unpool2d self ~indices ~output_size =
-  let out__ = CArray.make t 1 in
   stubs_max_unpool2d
-    (CArray.start out__)
     self
     indices
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
-    (List.length output_size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length output_size)
+  |> with_tensor_gc
 ;;
 
 let max_unpool2d_out ~out self ~indices ~output_size =
-  let out__ = CArray.make t 1 in
   stubs_max_unpool2d_out
-    (CArray.start out__)
     out
     self
     indices
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
-    (List.length output_size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length output_size)
+  |> with_tensor_gc
 ;;
 
 let max_unpool3d self ~indices ~output_size ~stride ~padding =
-  let out__ = CArray.make t 1 in
   stubs_max_unpool3d
-    (CArray.start out__)
     self
     indices
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
@@ -20846,16 +13411,12 @@ let max_unpool3d self ~indices ~output_size ~stride ~padding =
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
     (List.length stride)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
-    (List.length padding);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length padding)
+  |> with_tensor_gc
 ;;
 
 let max_unpool3d_out ~out self ~indices ~output_size ~stride ~padding =
-  let out__ = CArray.make t 1 in
   stubs_max_unpool3d_out
-    (CArray.start out__)
     out
     self
     indices
@@ -20864,40 +13425,16 @@ let max_unpool3d_out ~out self ~indices ~output_size ~stride ~padding =
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
     (List.length stride)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
-    (List.length padding);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let maximum self other =
-  let out__ = CArray.make t 1 in
-  stubs_maximum (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let maximum_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_maximum_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length padding)
+  |> with_tensor_gc
 ;;
 
-let mean self ~dtype =
-  let out__ = CArray.make t 1 in
-  stubs_mean (CArray.start out__) self (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let maximum self other = stubs_maximum self other |> with_tensor_gc
+let maximum_out ~out self other = stubs_maximum_out out self other |> with_tensor_gc
+let mean self ~dtype = stubs_mean self (Kind.packed_to_int dtype) |> with_tensor_gc
 
 let mean_dim self ~dim ~keepdim ~dtype =
-  let out__ = CArray.make t 1 in
   stubs_mean_dim
-    (CArray.start out__)
     self
     (match dim with
      | None -> from_voidp int64_t null
@@ -20906,16 +13443,12 @@ let mean_dim self ~dim ~keepdim ~dtype =
      | None -> -1
      | Some v -> List.length v)
     (if keepdim then 1 else 0)
-    (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Kind.packed_to_int dtype)
+  |> with_tensor_gc
 ;;
 
 let mean_out ~out self ~dim ~keepdim ~dtype =
-  let out__ = CArray.make t 1 in
   stubs_mean_out
-    (CArray.start out__)
     out
     self
     (match dim with
@@ -20925,32 +13458,22 @@ let mean_out ~out self ~dim ~keepdim ~dtype =
      | None -> -1
      | Some v -> List.length v)
     (if keepdim then 1 else 0)
-    (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Kind.packed_to_int dtype)
+  |> with_tensor_gc
 ;;
 
-let median self =
-  let out__ = CArray.make t 1 in
-  stubs_median (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let median self = stubs_median self |> with_tensor_gc
 
 let median_dim self ~dim ~keepdim =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_median_dim (CArray.start out__) self (Int64.of_int dim) (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let median_dim_values ~values ~indices self ~dim ~keepdim =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_median_dim_values
     (CArray.start out__)
     values
@@ -20958,62 +13481,39 @@ let median_dim_values ~values ~indices self ~dim ~keepdim =
     self
     (Int64.of_int dim)
     (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
-let median_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_median_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let median_out ~out self = stubs_median_out out self |> with_tensor_gc
 
 let meshgrid tensors =
-  stubs_meshgrid (CArray.of_list t tensors |> CArray.start) (List.length tensors)
+  stubs_meshgrid (CArray.of_list gc_tensor tensors |> CArray.start) (List.length tensors)
   |> to_tensor_list
 ;;
 
 let meshgrid_indexing tensors ~indexing =
   stubs_meshgrid_indexing
-    (CArray.of_list t tensors |> CArray.start)
+    (CArray.of_list gc_tensor tensors |> CArray.start)
     (List.length tensors)
     indexing
   |> to_tensor_list
 ;;
 
-let mh self =
-  let out__ = CArray.make t 1 in
-  stubs_mh (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let min self =
-  let out__ = CArray.make t 1 in
-  stubs_min (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let mh self = stubs_mh self |> with_tensor_gc
+let min self = stubs_min self |> with_tensor_gc
 
 let min_dim self ~dim ~keepdim =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_min_dim (CArray.start out__) self (Int64.of_int dim) (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let min_dim_min ~min ~min_indices self ~dim ~keepdim =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_min_dim_min
     (CArray.start out__)
     min
@@ -21021,52 +13521,16 @@ let min_dim_min ~min ~min_indices self ~dim ~keepdim =
     self
     (Int64.of_int dim)
     (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
-let min_other self other =
-  let out__ = CArray.make t 1 in
-  stubs_min_other (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let min_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_min_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let min_unary_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_min_unary_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let minimum self other =
-  let out__ = CArray.make t 1 in
-  stubs_minimum (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let minimum_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_minimum_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let min_other self other = stubs_min_other self other |> with_tensor_gc
+let min_out ~out self other = stubs_min_out out self other |> with_tensor_gc
+let min_unary_out ~out self = stubs_min_unary_out out self |> with_tensor_gc
+let minimum self other = stubs_minimum self other |> with_tensor_gc
+let minimum_out ~out self other = stubs_minimum_out out self other |> with_tensor_gc
 
 let miopen_batch_norm
   input
@@ -21078,29 +13542,26 @@ let miopen_batch_norm
   ~exponential_average_factor
   ~epsilon
   =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs_miopen_batch_norm
     (CArray.start out__)
     input
     weight
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match running_mean with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match running_var with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (if training then 1 else 0)
     exponential_average_factor
     epsilon;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
@@ -21114,7 +13575,7 @@ let miopen_batch_norm_backward
   ~save_var
   ~epsilon
   =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs_miopen_batch_norm_backward
     (CArray.start out__)
     input
@@ -21122,23 +13583,20 @@ let miopen_batch_norm_backward
     weight
     (match running_mean with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match running_var with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match save_mean with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match save_var with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     epsilon;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
@@ -21155,7 +13613,7 @@ let miopen_batch_norm_backward_out
   ~save_var
   ~epsilon
   =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs_miopen_batch_norm_backward_out
     (CArray.start out__)
     out0
@@ -21166,23 +13624,20 @@ let miopen_batch_norm_backward_out
     weight
     (match running_mean with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match running_var with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match save_mean with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match save_var with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     epsilon;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
@@ -21199,7 +13654,7 @@ let miopen_batch_norm_out
   ~exponential_average_factor
   ~epsilon
   =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs_miopen_batch_norm_out
     (CArray.start out__)
     out0
@@ -21209,22 +13664,19 @@ let miopen_batch_norm_out
     weight
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match running_mean with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match running_var with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (if training then 1 else 0)
     exponential_average_factor
     epsilon;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
@@ -21239,14 +13691,12 @@ let miopen_convolution
   ~benchmark
   ~deterministic
   =
-  let out__ = CArray.make t 1 in
   stubs_miopen_convolution
-    (CArray.start out__)
     self
     weight
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
     (List.length padding)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
@@ -21255,10 +13705,8 @@ let miopen_convolution
     (List.length dilation)
     (Int64.of_int groups)
     (if benchmark then 1 else 0)
-    (if deterministic then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if deterministic then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let miopen_convolution_add_relu
@@ -21272,26 +13720,22 @@ let miopen_convolution_add_relu
   ~dilation
   ~groups
   =
-  let out__ = CArray.make t 1 in
   stubs_miopen_convolution_add_relu
-    (CArray.start out__)
     self
     weight
     z
     alpha
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
     (List.length stride)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
     (List.length padding)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
-    (Int64.of_int groups);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int groups)
+  |> with_tensor_gc
 ;;
 
 let miopen_convolution_out
@@ -21306,15 +13750,13 @@ let miopen_convolution_out
   ~benchmark
   ~deterministic
   =
-  let out__ = CArray.make t 1 in
   stubs_miopen_convolution_out
-    (CArray.start out__)
     out
     self
     weight
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
     (List.length padding)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
@@ -21323,31 +13765,25 @@ let miopen_convolution_out
     (List.length dilation)
     (Int64.of_int groups)
     (if benchmark then 1 else 0)
-    (if deterministic then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if deterministic then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let miopen_convolution_relu self ~weight ~bias ~stride ~padding ~dilation ~groups =
-  let out__ = CArray.make t 1 in
   stubs_miopen_convolution_relu
-    (CArray.start out__)
     self
     weight
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
     (List.length stride)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
     (List.length padding)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
-    (Int64.of_int groups);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int groups)
+  |> with_tensor_gc
 ;;
 
 let miopen_convolution_transpose
@@ -21362,14 +13798,12 @@ let miopen_convolution_transpose
   ~benchmark
   ~deterministic
   =
-  let out__ = CArray.make t 1 in
   stubs_miopen_convolution_transpose
-    (CArray.start out__)
     self
     weight
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
     (List.length padding)
     (List.map Int64.of_int output_padding |> CArray.of_list int64_t |> CArray.start)
@@ -21380,10 +13814,8 @@ let miopen_convolution_transpose
     (List.length dilation)
     (Int64.of_int groups)
     (if benchmark then 1 else 0)
-    (if deterministic then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if deterministic then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let miopen_convolution_transpose_out
@@ -21399,15 +13831,13 @@ let miopen_convolution_transpose_out
   ~benchmark
   ~deterministic
   =
-  let out__ = CArray.make t 1 in
   stubs_miopen_convolution_transpose_out
-    (CArray.start out__)
     out
     self
     weight
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
     (List.length padding)
     (List.map Int64.of_int output_padding |> CArray.of_list int64_t |> CArray.start)
@@ -21418,10 +13848,8 @@ let miopen_convolution_transpose_out
     (List.length dilation)
     (Int64.of_int groups)
     (if benchmark then 1 else 0)
-    (if deterministic then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if deterministic then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let miopen_depthwise_convolution
@@ -21435,14 +13863,12 @@ let miopen_depthwise_convolution
   ~benchmark
   ~deterministic
   =
-  let out__ = CArray.make t 1 in
   stubs_miopen_depthwise_convolution
-    (CArray.start out__)
     self
     weight
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
     (List.length padding)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
@@ -21451,10 +13877,8 @@ let miopen_depthwise_convolution
     (List.length dilation)
     (Int64.of_int groups)
     (if benchmark then 1 else 0)
-    (if deterministic then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if deterministic then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let miopen_depthwise_convolution_out
@@ -21469,15 +13893,13 @@ let miopen_depthwise_convolution_out
   ~benchmark
   ~deterministic
   =
-  let out__ = CArray.make t 1 in
   stubs_miopen_depthwise_convolution_out
-    (CArray.start out__)
     out
     self
     weight
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
     (List.length padding)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
@@ -21486,10 +13908,8 @@ let miopen_depthwise_convolution_out
     (List.length dilation)
     (Int64.of_int groups)
     (if benchmark then 1 else 0)
-    (if deterministic then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if deterministic then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let miopen_rnn
@@ -21508,17 +13928,17 @@ let miopen_rnn
   ~batch_sizes
   ~dropout_state
   =
-  let out__ = CArray.make t 5 in
+  let out__ = CArray.make raw_tensor 5 in
   stubs_miopen_rnn
     (CArray.start out__)
     input
-    (CArray.of_list t weight |> CArray.start)
+    (CArray.of_list gc_tensor weight |> CArray.start)
     (List.length weight)
     (Int64.of_int weight_stride0)
     hx
     (match cx with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (Int64.of_int mode)
     (Int64.of_int hidden_size)
     (Int64.of_int num_layers)
@@ -21530,17 +13950,12 @@ let miopen_rnn
     (List.length batch_sizes)
     (match dropout_state with
      | Some v -> v
-     | None -> null);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
-  let t3 = CArray.get out__ 3 in
-  Gc.finalise C.Tensor.free t3;
-  let t4 = CArray.get out__ 4 in
-  Gc.finalise C.Tensor.free t4;
+     | None -> none_gc_tensor);
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
+  let t3 = CArray.get out__ 3 |> with_tensor_gc in
+  let t4 = CArray.get out__ 4 |> with_tensor_gc in
   t0, t1, t2, t3, t4
 ;;
 
@@ -21565,7 +13980,7 @@ let miopen_rnn_out
   ~batch_sizes
   ~dropout_state
   =
-  let out__ = CArray.make t 5 in
+  let out__ = CArray.make raw_tensor 5 in
   stubs_miopen_rnn_out
     (CArray.start out__)
     out0
@@ -21574,13 +13989,13 @@ let miopen_rnn_out
     out3
     out4
     input
-    (CArray.of_list t weight |> CArray.start)
+    (CArray.of_list gc_tensor weight |> CArray.start)
     (List.length weight)
     (Int64.of_int weight_stride0)
     hx
     (match cx with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (Int64.of_int mode)
     (Int64.of_int hidden_size)
     (Int64.of_int num_layers)
@@ -21592,189 +14007,123 @@ let miopen_rnn_out
     (List.length batch_sizes)
     (match dropout_state with
      | Some v -> v
-     | None -> null);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
-  let t3 = CArray.get out__ 3 in
-  Gc.finalise C.Tensor.free t3;
-  let t4 = CArray.get out__ 4 in
-  Gc.finalise C.Tensor.free t4;
+     | None -> none_gc_tensor);
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
+  let t3 = CArray.get out__ 3 |> with_tensor_gc in
+  let t4 = CArray.get out__ 4 |> with_tensor_gc in
   t0, t1, t2, t3, t4
 ;;
 
-let mish self =
-  let out__ = CArray.make t 1 in
-  stubs_mish (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let mish_ self =
-  let out__ = CArray.make t 1 in
-  stubs_mish_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let mish self = stubs_mish self |> with_tensor_gc
+let mish_ self = stubs_mish_ self |> with_tensor_gc
 
 let mish_backward ~grad_output self =
-  let out__ = CArray.make t 1 in
-  stubs_mish_backward (CArray.start out__) grad_output self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_mish_backward grad_output self |> with_tensor_gc
 ;;
 
-let mish_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_mish_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let mish_out ~out self = stubs_mish_out out self |> with_tensor_gc
 
 let mkldnn_adaptive_avg_pool2d self ~output_size =
-  let out__ = CArray.make t 1 in
   stubs_mkldnn_adaptive_avg_pool2d
-    (CArray.start out__)
     self
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
-    (List.length output_size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length output_size)
+  |> with_tensor_gc
 ;;
 
 let mkldnn_adaptive_avg_pool2d_backward ~grad_output self =
-  let out__ = CArray.make t 1 in
-  stubs_mkldnn_adaptive_avg_pool2d_backward (CArray.start out__) grad_output self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_mkldnn_adaptive_avg_pool2d_backward grad_output self |> with_tensor_gc
 ;;
 
 let mkldnn_adaptive_avg_pool2d_backward_out ~out ~grad_output self =
-  let out__ = CArray.make t 1 in
-  stubs_mkldnn_adaptive_avg_pool2d_backward_out (CArray.start out__) out grad_output self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_mkldnn_adaptive_avg_pool2d_backward_out out grad_output self |> with_tensor_gc
 ;;
 
 let mkldnn_adaptive_avg_pool2d_out ~out self ~output_size =
-  let out__ = CArray.make t 1 in
   stubs_mkldnn_adaptive_avg_pool2d_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
-    (List.length output_size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length output_size)
+  |> with_tensor_gc
 ;;
 
 let mkldnn_convolution self ~weight ~bias ~padding ~stride ~dilation ~groups =
-  let out__ = CArray.make t 1 in
   stubs_mkldnn_convolution
-    (CArray.start out__)
     self
     weight
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
     (List.length padding)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
     (List.length stride)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
-    (Int64.of_int groups);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int groups)
+  |> with_tensor_gc
 ;;
 
 let mkldnn_convolution_out ~out self ~weight ~bias ~padding ~stride ~dilation ~groups =
-  let out__ = CArray.make t 1 in
   stubs_mkldnn_convolution_out
-    (CArray.start out__)
     out
     self
     weight
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
     (List.length padding)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
     (List.length stride)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
-    (Int64.of_int groups);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int groups)
+  |> with_tensor_gc
 ;;
 
 let mkldnn_linear self ~weight ~bias =
-  let out__ = CArray.make t 1 in
   stubs_mkldnn_linear
-    (CArray.start out__)
     self
     weight
     (match bias with
      | Some v -> v
-     | None -> null);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+  |> with_tensor_gc
 ;;
 
 let mkldnn_linear_backward_input ~input_size ~grad_output ~weight =
-  let out__ = CArray.make t 1 in
   stubs_mkldnn_linear_backward_input
-    (CArray.start out__)
     (List.map Int64.of_int input_size |> CArray.of_list int64_t |> CArray.start)
     (List.length input_size)
     grad_output
-    weight;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    weight
+  |> with_tensor_gc
 ;;
 
 let mkldnn_linear_backward_input_out ~out ~input_size ~grad_output ~weight =
-  let out__ = CArray.make t 1 in
   stubs_mkldnn_linear_backward_input_out
-    (CArray.start out__)
     out
     (List.map Int64.of_int input_size |> CArray.of_list int64_t |> CArray.start)
     (List.length input_size)
     grad_output
-    weight;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    weight
+  |> with_tensor_gc
 ;;
 
 let mkldnn_linear_backward_weights ~grad_output input ~weight ~bias_defined =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_mkldnn_linear_backward_weights
     (CArray.start out__)
     grad_output
     input
     weight
     (if bias_defined then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
@@ -21786,7 +14135,7 @@ let mkldnn_linear_backward_weights_out
   ~weight
   ~bias_defined
   =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_mkldnn_linear_backward_weights_out
     (CArray.start out__)
     out0
@@ -21795,32 +14144,24 @@ let mkldnn_linear_backward_weights_out
     input
     weight
     (if bias_defined then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let mkldnn_linear_out ~out self ~weight ~bias =
-  let out__ = CArray.make t 1 in
   stubs_mkldnn_linear_out
-    (CArray.start out__)
     out
     self
     weight
     (match bias with
      | Some v -> v
-     | None -> null);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+  |> with_tensor_gc
 ;;
 
 let mkldnn_max_pool2d self ~kernel_size ~stride ~padding ~dilation ~ceil_mode =
-  let out__ = CArray.make t 1 in
   stubs_mkldnn_max_pool2d
-    (CArray.start out__)
     self
     (List.map Int64.of_int kernel_size |> CArray.of_list int64_t |> CArray.start)
     (List.length kernel_size)
@@ -21830,10 +14171,8 @@ let mkldnn_max_pool2d self ~kernel_size ~stride ~padding ~dilation ~ceil_mode =
     (List.length padding)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
-    (if ceil_mode then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if ceil_mode then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let mkldnn_max_pool2d_backward
@@ -21846,9 +14185,7 @@ let mkldnn_max_pool2d_backward
   ~dilation
   ~ceil_mode
   =
-  let out__ = CArray.make t 1 in
   stubs_mkldnn_max_pool2d_backward
-    (CArray.start out__)
     grad_output
     output
     input
@@ -21860,10 +14197,8 @@ let mkldnn_max_pool2d_backward
     (List.length padding)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
-    (if ceil_mode then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if ceil_mode then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let mkldnn_max_pool2d_backward_out
@@ -21877,9 +14212,7 @@ let mkldnn_max_pool2d_backward_out
   ~dilation
   ~ceil_mode
   =
-  let out__ = CArray.make t 1 in
   stubs_mkldnn_max_pool2d_backward_out
-    (CArray.start out__)
     out
     grad_output
     output
@@ -21892,16 +14225,12 @@ let mkldnn_max_pool2d_backward_out
     (List.length padding)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
-    (if ceil_mode then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if ceil_mode then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let mkldnn_max_pool2d_out ~out self ~kernel_size ~stride ~padding ~dilation ~ceil_mode =
-  let out__ = CArray.make t 1 in
   stubs_mkldnn_max_pool2d_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int kernel_size |> CArray.of_list int64_t |> CArray.start)
@@ -21912,16 +14241,12 @@ let mkldnn_max_pool2d_out ~out self ~kernel_size ~stride ~padding ~dilation ~cei
     (List.length padding)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
-    (if ceil_mode then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if ceil_mode then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let mkldnn_max_pool3d self ~kernel_size ~stride ~padding ~dilation ~ceil_mode =
-  let out__ = CArray.make t 1 in
   stubs_mkldnn_max_pool3d
-    (CArray.start out__)
     self
     (List.map Int64.of_int kernel_size |> CArray.of_list int64_t |> CArray.start)
     (List.length kernel_size)
@@ -21931,10 +14256,8 @@ let mkldnn_max_pool3d self ~kernel_size ~stride ~padding ~dilation ~ceil_mode =
     (List.length padding)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
-    (if ceil_mode then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if ceil_mode then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let mkldnn_max_pool3d_backward
@@ -21947,9 +14270,7 @@ let mkldnn_max_pool3d_backward
   ~dilation
   ~ceil_mode
   =
-  let out__ = CArray.make t 1 in
   stubs_mkldnn_max_pool3d_backward
-    (CArray.start out__)
     grad_output
     output
     input
@@ -21961,10 +14282,8 @@ let mkldnn_max_pool3d_backward
     (List.length padding)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
-    (if ceil_mode then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if ceil_mode then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let mkldnn_max_pool3d_backward_out
@@ -21978,9 +14297,7 @@ let mkldnn_max_pool3d_backward_out
   ~dilation
   ~ceil_mode
   =
-  let out__ = CArray.make t 1 in
   stubs_mkldnn_max_pool3d_backward_out
-    (CArray.start out__)
     out
     grad_output
     output
@@ -21993,16 +14310,12 @@ let mkldnn_max_pool3d_backward_out
     (List.length padding)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
-    (if ceil_mode then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if ceil_mode then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let mkldnn_max_pool3d_out ~out self ~kernel_size ~stride ~padding ~dilation ~ceil_mode =
-  let out__ = CArray.make t 1 in
   stubs_mkldnn_max_pool3d_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int kernel_size |> CArray.of_list int64_t |> CArray.start)
@@ -22013,16 +14326,12 @@ let mkldnn_max_pool3d_out ~out self ~kernel_size ~stride ~padding ~dilation ~cei
     (List.length padding)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
-    (if ceil_mode then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if ceil_mode then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let mkldnn_reorder_conv2d_weight self ~padding ~stride ~dilation ~groups ~input_size =
-  let out__ = CArray.make t 1 in
   stubs_mkldnn_reorder_conv2d_weight
-    (CArray.start out__)
     self
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
     (List.length padding)
@@ -22036,10 +14345,8 @@ let mkldnn_reorder_conv2d_weight self ~padding ~stride ~dilation ~groups ~input_
      | Some v -> List.map Int64.of_int v |> CArray.of_list int64_t |> CArray.start)
     (match input_size with
      | None -> -1
-     | Some v -> List.length v);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | Some v -> List.length v)
+  |> with_tensor_gc
 ;;
 
 let mkldnn_reorder_conv2d_weight_out
@@ -22051,9 +14358,7 @@ let mkldnn_reorder_conv2d_weight_out
   ~groups
   ~input_size
   =
-  let out__ = CArray.make t 1 in
   stubs_mkldnn_reorder_conv2d_weight_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
@@ -22068,16 +14373,12 @@ let mkldnn_reorder_conv2d_weight_out
      | Some v -> List.map Int64.of_int v |> CArray.of_list int64_t |> CArray.start)
     (match input_size with
      | None -> -1
-     | Some v -> List.length v);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | Some v -> List.length v)
+  |> with_tensor_gc
 ;;
 
 let mkldnn_reorder_conv3d_weight self ~padding ~stride ~dilation ~groups =
-  let out__ = CArray.make t 1 in
   stubs_mkldnn_reorder_conv3d_weight
-    (CArray.start out__)
     self
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
     (List.length padding)
@@ -22085,16 +14386,12 @@ let mkldnn_reorder_conv3d_weight self ~padding ~stride ~dilation ~groups =
     (List.length stride)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
-    (Int64.of_int groups);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int groups)
+  |> with_tensor_gc
 ;;
 
 let mkldnn_reorder_conv3d_weight_out ~out self ~padding ~stride ~dilation ~groups =
-  let out__ = CArray.make t 1 in
   stubs_mkldnn_reorder_conv3d_weight_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
@@ -22103,10 +14400,8 @@ let mkldnn_reorder_conv3d_weight_out ~out self ~padding ~stride ~dilation ~group
     (List.length stride)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
-    (Int64.of_int groups);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int groups)
+  |> with_tensor_gc
 ;;
 
 let mkldnn_rnn_layer
@@ -22127,7 +14422,7 @@ let mkldnn_rnn_layer
   ~batch_first
   ~train
   =
-  let out__ = CArray.make t 4 in
+  let out__ = CArray.make raw_tensor 4 in
   stubs_mkldnn_rnn_layer
     (CArray.start out__)
     input
@@ -22147,14 +14442,10 @@ let mkldnn_rnn_layer
     (if bidirectional then 1 else 0)
     (if batch_first then 1 else 0)
     (if train then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
-  let t3 = CArray.get out__ 3 in
-  Gc.finalise C.Tensor.free t3;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
+  let t3 = CArray.get out__ 3 |> with_tensor_gc in
   t0, t1, t2, t3
 ;;
 
@@ -22183,7 +14474,7 @@ let mkldnn_rnn_layer_backward
   ~batch_first
   ~workspace
   =
-  let out__ = CArray.make t 7 in
+  let out__ = CArray.make raw_tensor 7 in
   stubs_mkldnn_rnn_layer_backward
     (CArray.start out__)
     input
@@ -22198,13 +14489,13 @@ let mkldnn_rnn_layer_backward
     cy_
     (match grad_output with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match grad_hy with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match grad_cy with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (if reverse then 1 else 0)
     (Int64.of_int mode)
     (Int64.of_int hidden_size)
@@ -22216,20 +14507,13 @@ let mkldnn_rnn_layer_backward
     (List.length batch_sizes)
     (if batch_first then 1 else 0)
     workspace;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
-  let t3 = CArray.get out__ 3 in
-  Gc.finalise C.Tensor.free t3;
-  let t4 = CArray.get out__ 4 in
-  Gc.finalise C.Tensor.free t4;
-  let t5 = CArray.get out__ 5 in
-  Gc.finalise C.Tensor.free t5;
-  let t6 = CArray.get out__ 6 in
-  Gc.finalise C.Tensor.free t6;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
+  let t3 = CArray.get out__ 3 |> with_tensor_gc in
+  let t4 = CArray.get out__ 4 |> with_tensor_gc in
+  let t5 = CArray.get out__ 5 |> with_tensor_gc in
+  let t6 = CArray.get out__ 6 |> with_tensor_gc in
   t0, t1, t2, t3, t4, t5, t6
 ;;
 
@@ -22265,7 +14549,7 @@ let mkldnn_rnn_layer_backward_out
   ~batch_first
   ~workspace
   =
-  let out__ = CArray.make t 7 in
+  let out__ = CArray.make raw_tensor 7 in
   stubs_mkldnn_rnn_layer_backward_out
     (CArray.start out__)
     out0
@@ -22287,13 +14571,13 @@ let mkldnn_rnn_layer_backward_out
     cy_
     (match grad_output with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match grad_hy with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match grad_cy with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (if reverse then 1 else 0)
     (Int64.of_int mode)
     (Int64.of_int hidden_size)
@@ -22305,20 +14589,13 @@ let mkldnn_rnn_layer_backward_out
     (List.length batch_sizes)
     (if batch_first then 1 else 0)
     workspace;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
-  let t3 = CArray.get out__ 3 in
-  Gc.finalise C.Tensor.free t3;
-  let t4 = CArray.get out__ 4 in
-  Gc.finalise C.Tensor.free t4;
-  let t5 = CArray.get out__ 5 in
-  Gc.finalise C.Tensor.free t5;
-  let t6 = CArray.get out__ 6 in
-  Gc.finalise C.Tensor.free t6;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
+  let t3 = CArray.get out__ 3 |> with_tensor_gc in
+  let t4 = CArray.get out__ 4 |> with_tensor_gc in
+  let t5 = CArray.get out__ 5 |> with_tensor_gc in
+  let t6 = CArray.get out__ 6 |> with_tensor_gc in
   t0, t1, t2, t3, t4, t5, t6
 ;;
 
@@ -22344,7 +14621,7 @@ let mkldnn_rnn_layer_out
   ~batch_first
   ~train
   =
-  let out__ = CArray.make t 4 in
+  let out__ = CArray.make raw_tensor 4 in
   stubs_mkldnn_rnn_layer_out
     (CArray.start out__)
     out0
@@ -22368,45 +14645,26 @@ let mkldnn_rnn_layer_out
     (if bidirectional then 1 else 0)
     (if batch_first then 1 else 0)
     (if train then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
-  let t3 = CArray.get out__ 3 in
-  Gc.finalise C.Tensor.free t3;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
+  let t3 = CArray.get out__ 3 |> with_tensor_gc in
   t0, t1, t2, t3
 ;;
 
-let mm self ~mat2 =
-  let out__ = CArray.make t 1 in
-  stubs_mm (CArray.start out__) self mat2;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let mm_out ~out self ~mat2 =
-  let out__ = CArray.make t 1 in
-  stubs_mm_out (CArray.start out__) out self mat2;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let mm self ~mat2 = stubs_mm self mat2 |> with_tensor_gc
+let mm_out ~out self ~mat2 = stubs_mm_out out self mat2 |> with_tensor_gc
 
 let mode self ~dim ~keepdim =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_mode (CArray.start out__) self (Int64.of_int dim) (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let mode_values ~values ~indices self ~dim ~keepdim =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_mode_values
     (CArray.start out__)
     values
@@ -22414,193 +14672,82 @@ let mode_values ~values ~indices self ~dim ~keepdim =
     self
     (Int64.of_int dim)
     (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let moveaxis self ~source ~destination =
-  let out__ = CArray.make t 1 in
   stubs_moveaxis
-    (CArray.start out__)
     self
     (List.map Int64.of_int source |> CArray.of_list int64_t |> CArray.start)
     (List.length source)
     (List.map Int64.of_int destination |> CArray.of_list int64_t |> CArray.start)
-    (List.length destination);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length destination)
+  |> with_tensor_gc
 ;;
 
 let moveaxis_int self ~source ~destination =
-  let out__ = CArray.make t 1 in
-  stubs_moveaxis_int
-    (CArray.start out__)
-    self
-    (Int64.of_int source)
-    (Int64.of_int destination);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_moveaxis_int self (Int64.of_int source) (Int64.of_int destination)
+  |> with_tensor_gc
 ;;
 
 let movedim self ~source ~destination =
-  let out__ = CArray.make t 1 in
   stubs_movedim
-    (CArray.start out__)
     self
     (List.map Int64.of_int source |> CArray.of_list int64_t |> CArray.start)
     (List.length source)
     (List.map Int64.of_int destination |> CArray.of_list int64_t |> CArray.start)
-    (List.length destination);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length destination)
+  |> with_tensor_gc
 ;;
 
 let movedim_int self ~source ~destination =
-  let out__ = CArray.make t 1 in
-  stubs_movedim_int
-    (CArray.start out__)
-    self
-    (Int64.of_int source)
-    (Int64.of_int destination);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_movedim_int self (Int64.of_int source) (Int64.of_int destination)
+  |> with_tensor_gc
 ;;
 
 let mse_loss self ~target ~reduction =
-  let out__ = CArray.make t 1 in
-  stubs_mse_loss
-    (CArray.start out__)
-    self
-    target
-    (Reduction.to_int reduction |> Int64.of_int);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_mse_loss self target (Reduction.to_int reduction |> Int64.of_int)
+  |> with_tensor_gc
 ;;
 
 let mse_loss_backward ~grad_output self ~target ~reduction =
-  let out__ = CArray.make t 1 in
   stubs_mse_loss_backward
-    (CArray.start out__)
     grad_output
     self
     target
-    (Reduction.to_int reduction |> Int64.of_int);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Reduction.to_int reduction |> Int64.of_int)
+  |> with_tensor_gc
 ;;
 
 let mse_loss_backward_grad_input ~grad_input ~grad_output self ~target ~reduction =
-  let out__ = CArray.make t 1 in
   stubs_mse_loss_backward_grad_input
-    (CArray.start out__)
     grad_input
     grad_output
     self
     target
-    (Reduction.to_int reduction |> Int64.of_int);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Reduction.to_int reduction |> Int64.of_int)
+  |> with_tensor_gc
 ;;
 
 let mse_loss_out ~out self ~target ~reduction =
-  let out__ = CArray.make t 1 in
-  stubs_mse_loss_out
-    (CArray.start out__)
-    out
-    self
-    target
-    (Reduction.to_int reduction |> Int64.of_int);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let msort self =
-  let out__ = CArray.make t 1 in
-  stubs_msort (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let msort_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_msort_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let mt self =
-  let out__ = CArray.make t 1 in
-  stubs_mt (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let mul self other =
-  let out__ = CArray.make t 1 in
-  stubs_mul (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let mul_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_mul_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let mul_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_mul_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let mul_scalar self other =
-  let out__ = CArray.make t 1 in
-  stubs_mul_scalar (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let mul_scalar_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_mul_scalar_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_mse_loss_out out self target (Reduction.to_int reduction |> Int64.of_int)
+  |> with_tensor_gc
 ;;
 
-let mul_scalar_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_mul_scalar_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let msort self = stubs_msort self |> with_tensor_gc
+let msort_out ~out self = stubs_msort_out out self |> with_tensor_gc
+let mt self = stubs_mt self |> with_tensor_gc
+let mul self other = stubs_mul self other |> with_tensor_gc
+let mul_ self other = stubs_mul_ self other |> with_tensor_gc
+let mul_out ~out self other = stubs_mul_out out self other |> with_tensor_gc
+let mul_scalar self other = stubs_mul_scalar self other |> with_tensor_gc
+let mul_scalar_ self other = stubs_mul_scalar_ self other |> with_tensor_gc
+let mul_scalar_out ~out self other = stubs_mul_scalar_out out self other |> with_tensor_gc
 
 let multi_margin_loss_backward ~grad_output self ~target ~p ~margin ~weight ~reduction =
-  let out__ = CArray.make t 1 in
   stubs_multi_margin_loss_backward
-    (CArray.start out__)
     grad_output
     self
     target
@@ -22608,11 +14755,9 @@ let multi_margin_loss_backward ~grad_output self ~target ~p ~margin ~weight ~red
     margin
     (match weight with
      | Some v -> v
-     | None -> null)
-    (Reduction.to_int reduction |> Int64.of_int);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+    (Reduction.to_int reduction |> Int64.of_int)
+  |> with_tensor_gc
 ;;
 
 let multi_margin_loss_backward_grad_input
@@ -22625,9 +14770,7 @@ let multi_margin_loss_backward_grad_input
   ~weight
   ~reduction
   =
-  let out__ = CArray.make t 1 in
   stubs_multi_margin_loss_backward_grad_input
-    (CArray.start out__)
     grad_input
     grad_output
     self
@@ -22636,37 +14779,24 @@ let multi_margin_loss_backward_grad_input
     margin
     (match weight with
      | Some v -> v
-     | None -> null)
-    (Reduction.to_int reduction |> Int64.of_int);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+    (Reduction.to_int reduction |> Int64.of_int)
+  |> with_tensor_gc
 ;;
 
 let multilabel_margin_loss self ~target ~reduction =
-  let out__ = CArray.make t 1 in
-  stubs_multilabel_margin_loss
-    (CArray.start out__)
-    self
-    target
-    (Reduction.to_int reduction |> Int64.of_int);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_multilabel_margin_loss self target (Reduction.to_int reduction |> Int64.of_int)
+  |> with_tensor_gc
 ;;
 
 let multilabel_margin_loss_backward ~grad_output self ~target ~reduction ~is_target =
-  let out__ = CArray.make t 1 in
   stubs_multilabel_margin_loss_backward
-    (CArray.start out__)
     grad_output
     self
     target
     (Reduction.to_int reduction |> Int64.of_int)
-    is_target;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    is_target
+  |> with_tensor_gc
 ;;
 
 let multilabel_margin_loss_backward_grad_input
@@ -22677,142 +14807,51 @@ let multilabel_margin_loss_backward_grad_input
   ~reduction
   ~is_target
   =
-  let out__ = CArray.make t 1 in
   stubs_multilabel_margin_loss_backward_grad_input
-    (CArray.start out__)
     grad_input
     grad_output
     self
     target
     (Reduction.to_int reduction |> Int64.of_int)
-    is_target;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    is_target
+  |> with_tensor_gc
 ;;
 
 let multilabel_margin_loss_out ~out self ~target ~reduction =
-  let out__ = CArray.make t 1 in
   stubs_multilabel_margin_loss_out
-    (CArray.start out__)
     out
     self
     target
-    (Reduction.to_int reduction |> Int64.of_int);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Reduction.to_int reduction |> Int64.of_int)
+  |> with_tensor_gc
 ;;
 
 let multinomial self ~num_samples ~replacement =
-  let out__ = CArray.make t 1 in
-  stubs_multinomial
-    (CArray.start out__)
-    self
-    (Int64.of_int num_samples)
-    (if replacement then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_multinomial self (Int64.of_int num_samples) (if replacement then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let multinomial_out ~out self ~num_samples ~replacement =
-  let out__ = CArray.make t 1 in
-  stubs_multinomial_out
-    (CArray.start out__)
-    out
-    self
-    (Int64.of_int num_samples)
-    (if replacement then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let multiply self other =
-  let out__ = CArray.make t 1 in
-  stubs_multiply (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let multiply_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_multiply_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_multinomial_out out self (Int64.of_int num_samples) (if replacement then 1 else 0)
+  |> with_tensor_gc
 ;;
 
-let multiply_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_multiply_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let multiply_scalar self other =
-  let out__ = CArray.make t 1 in
-  stubs_multiply_scalar (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let multiply_scalar_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_multiply_scalar_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let mv self ~vec =
-  let out__ = CArray.make t 1 in
-  stubs_mv (CArray.start out__) self vec;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let mv_out ~out self ~vec =
-  let out__ = CArray.make t 1 in
-  stubs_mv_out (CArray.start out__) out self vec;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let mvlgamma self ~p =
-  let out__ = CArray.make t 1 in
-  stubs_mvlgamma (CArray.start out__) self (Int64.of_int p);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let mvlgamma_ self ~p =
-  let out__ = CArray.make t 1 in
-  stubs_mvlgamma_ (CArray.start out__) self (Int64.of_int p);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let multiply self other = stubs_multiply self other |> with_tensor_gc
+let multiply_ self other = stubs_multiply_ self other |> with_tensor_gc
+let multiply_out ~out self other = stubs_multiply_out out self other |> with_tensor_gc
+let multiply_scalar self other = stubs_multiply_scalar self other |> with_tensor_gc
+let multiply_scalar_ self other = stubs_multiply_scalar_ self other |> with_tensor_gc
+let mv self ~vec = stubs_mv self vec |> with_tensor_gc
+let mv_out ~out self ~vec = stubs_mv_out out self vec |> with_tensor_gc
+let mvlgamma self ~p = stubs_mvlgamma self (Int64.of_int p) |> with_tensor_gc
+let mvlgamma_ self ~p = stubs_mvlgamma_ self (Int64.of_int p) |> with_tensor_gc
 
 let mvlgamma_out ~out self ~p =
-  let out__ = CArray.make t 1 in
-  stubs_mvlgamma_out (CArray.start out__) out self (Int64.of_int p);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_mvlgamma_out out self (Int64.of_int p) |> with_tensor_gc
 ;;
 
 let nan_to_num self ~nan ~posinf ~neginf =
-  let out__ = CArray.make t 1 in
   stubs_nan_to_num
-    (CArray.start out__)
     self
     (Option.value nan ~default:0.0)
     (match nan with
@@ -22825,16 +14864,12 @@ let nan_to_num self ~nan ~posinf ~neginf =
     (Option.value neginf ~default:0.0)
     (match neginf with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let nan_to_num_ self ~nan ~posinf ~neginf =
-  let out__ = CArray.make t 1 in
   stubs_nan_to_num_
-    (CArray.start out__)
     self
     (Option.value nan ~default:0.0)
     (match nan with
@@ -22847,16 +14882,12 @@ let nan_to_num_ self ~nan ~posinf ~neginf =
     (Option.value neginf ~default:0.0)
     (match neginf with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let nan_to_num_out ~out self ~nan ~posinf ~neginf =
-  let out__ = CArray.make t 1 in
   stubs_nan_to_num_out
-    (CArray.start out__)
     out
     self
     (Option.value nan ~default:0.0)
@@ -22870,16 +14901,12 @@ let nan_to_num_out ~out self ~nan ~posinf ~neginf =
     (Option.value neginf ~default:0.0)
     (match neginf with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let nanmean self ~dim ~keepdim ~dtype =
-  let out__ = CArray.make t 1 in
   stubs_nanmean
-    (CArray.start out__)
     self
     (match dim with
      | None -> from_voidp int64_t null
@@ -22888,16 +14915,12 @@ let nanmean self ~dim ~keepdim ~dtype =
      | None -> -1
      | Some v -> List.length v)
     (if keepdim then 1 else 0)
-    (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Kind.packed_to_int dtype)
+  |> with_tensor_gc
 ;;
 
 let nanmean_out ~out self ~dim ~keepdim ~dtype =
-  let out__ = CArray.make t 1 in
   stubs_nanmean_out
-    (CArray.start out__)
     out
     self
     (match dim with
@@ -22907,36 +14930,26 @@ let nanmean_out ~out self ~dim ~keepdim ~dtype =
      | None -> -1
      | Some v -> List.length v)
     (if keepdim then 1 else 0)
-    (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Kind.packed_to_int dtype)
+  |> with_tensor_gc
 ;;
 
-let nanmedian self =
-  let out__ = CArray.make t 1 in
-  stubs_nanmedian (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let nanmedian self = stubs_nanmedian self |> with_tensor_gc
 
 let nanmedian_dim self ~dim ~keepdim =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_nanmedian_dim
     (CArray.start out__)
     self
     (Int64.of_int dim)
     (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let nanmedian_dim_values ~values ~indices self ~dim ~keepdim =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_nanmedian_dim_values
     (CArray.start out__)
     values
@@ -22944,25 +14957,15 @@ let nanmedian_dim_values ~values ~indices self ~dim ~keepdim =
     self
     (Int64.of_int dim)
     (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
-let nanmedian_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_nanmedian_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let nanmedian_out ~out self = stubs_nanmedian_out out self |> with_tensor_gc
 
 let nanquantile self ~q ~dim ~keepdim ~interpolation =
-  let out__ = CArray.make t 1 in
   stubs_nanquantile
-    (CArray.start out__)
     self
     q
     (match dim with
@@ -22972,16 +14975,12 @@ let nanquantile self ~q ~dim ~keepdim ~interpolation =
      | Some _ -> 0
      | None -> 1)
     (if keepdim then 1 else 0)
-    interpolation;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    interpolation
+  |> with_tensor_gc
 ;;
 
 let nanquantile_out ~out self ~q ~dim ~keepdim ~interpolation =
-  let out__ = CArray.make t 1 in
   stubs_nanquantile_out
-    (CArray.start out__)
     out
     self
     q
@@ -22992,16 +14991,12 @@ let nanquantile_out ~out self ~q ~dim ~keepdim ~interpolation =
      | Some _ -> 0
      | None -> 1)
     (if keepdim then 1 else 0)
-    interpolation;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    interpolation
+  |> with_tensor_gc
 ;;
 
 let nanquantile_scalar self ~q ~dim ~keepdim ~interpolation =
-  let out__ = CArray.make t 1 in
   stubs_nanquantile_scalar
-    (CArray.start out__)
     self
     q
     (match dim with
@@ -23011,16 +15006,12 @@ let nanquantile_scalar self ~q ~dim ~keepdim ~interpolation =
      | Some _ -> 0
      | None -> 1)
     (if keepdim then 1 else 0)
-    interpolation;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    interpolation
+  |> with_tensor_gc
 ;;
 
 let nanquantile_scalar_out ~out self ~q ~dim ~keepdim ~interpolation =
-  let out__ = CArray.make t 1 in
   stubs_nanquantile_scalar_out
-    (CArray.start out__)
     out
     self
     q
@@ -23031,16 +15022,12 @@ let nanquantile_scalar_out ~out self ~q ~dim ~keepdim ~interpolation =
      | Some _ -> 0
      | None -> 1)
     (if keepdim then 1 else 0)
-    interpolation;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    interpolation
+  |> with_tensor_gc
 ;;
 
 let nansum self ~dim ~keepdim ~dtype =
-  let out__ = CArray.make t 1 in
   stubs_nansum
-    (CArray.start out__)
     self
     (match dim with
      | None -> from_voidp int64_t null
@@ -23049,16 +15036,12 @@ let nansum self ~dim ~keepdim ~dtype =
      | None -> -1
      | Some v -> List.length v)
     (if keepdim then 1 else 0)
-    (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Kind.packed_to_int dtype)
+  |> with_tensor_gc
 ;;
 
 let nansum_out ~out self ~dim ~keepdim ~dtype =
-  let out__ = CArray.make t 1 in
   stubs_nansum_out
-    (CArray.start out__)
     out
     self
     (match dim with
@@ -23068,63 +15051,33 @@ let nansum_out ~out self ~dim ~keepdim ~dtype =
      | None -> -1
      | Some v -> List.length v)
     (if keepdim then 1 else 0)
-    (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Kind.packed_to_int dtype)
+  |> with_tensor_gc
 ;;
 
 let narrow self ~dim ~start ~length =
-  let out__ = CArray.make t 1 in
-  stubs_narrow
-    (CArray.start out__)
-    self
-    (Int64.of_int dim)
-    (Int64.of_int start)
-    (Int64.of_int length);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_narrow self (Int64.of_int dim) (Int64.of_int start) (Int64.of_int length)
+  |> with_tensor_gc
 ;;
 
 let narrow_copy self ~dim ~start ~length =
-  let out__ = CArray.make t 1 in
-  stubs_narrow_copy
-    (CArray.start out__)
-    self
-    (Int64.of_int dim)
-    (Int64.of_int start)
-    (Int64.of_int length);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_narrow_copy self (Int64.of_int dim) (Int64.of_int start) (Int64.of_int length)
+  |> with_tensor_gc
 ;;
 
 let narrow_copy_out ~out self ~dim ~start ~length =
-  let out__ = CArray.make t 1 in
   stubs_narrow_copy_out
-    (CArray.start out__)
     out
     self
     (Int64.of_int dim)
     (Int64.of_int start)
-    (Int64.of_int length);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int length)
+  |> with_tensor_gc
 ;;
 
 let narrow_tensor self ~dim ~start ~length =
-  let out__ = CArray.make t 1 in
-  stubs_narrow_tensor
-    (CArray.start out__)
-    self
-    (Int64.of_int dim)
-    start
-    (Int64.of_int length);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_narrow_tensor self (Int64.of_int dim) start (Int64.of_int length)
+  |> with_tensor_gc
 ;;
 
 let native_batch_norm
@@ -23137,31 +15090,28 @@ let native_batch_norm
   ~momentum
   ~eps
   =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs_native_batch_norm
     (CArray.start out__)
     input
     (match weight with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match running_mean with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match running_var with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (if training then 1 else 0)
     momentum
     eps;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
@@ -23178,7 +15128,7 @@ let native_batch_norm_out
   ~momentum
   ~eps
   =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs_native_batch_norm_out
     (CArray.start out__)
     out
@@ -23187,99 +15137,77 @@ let native_batch_norm_out
     input
     (match weight with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match running_mean with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match running_var with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (if training then 1 else 0)
     momentum
     eps;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
 let native_channel_shuffle self ~groups =
-  let out__ = CArray.make t 1 in
-  stubs_native_channel_shuffle (CArray.start out__) self (Int64.of_int groups);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_native_channel_shuffle self (Int64.of_int groups) |> with_tensor_gc
 ;;
 
 let native_dropout input ~p ~train =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_native_dropout (CArray.start out__) input p (if train then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let native_dropout_backward ~grad_output ~mask ~scale =
-  let out__ = CArray.make t 1 in
-  stubs_native_dropout_backward (CArray.start out__) grad_output mask scale;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_native_dropout_backward grad_output mask scale |> with_tensor_gc
 ;;
 
 let native_dropout_backward_out ~out ~grad_output ~mask ~scale =
-  let out__ = CArray.make t 1 in
-  stubs_native_dropout_backward_out (CArray.start out__) out grad_output mask scale;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_native_dropout_backward_out out grad_output mask scale |> with_tensor_gc
 ;;
 
 let native_dropout_out ~out0 ~out1 input ~p ~train =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_native_dropout_out (CArray.start out__) out0 out1 input p (if train then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let native_group_norm input ~weight ~bias ~n ~c ~hxw ~group ~eps =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs_native_group_norm
     (CArray.start out__)
     input
     (match weight with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (Int64.of_int n)
     (Int64.of_int c)
     (Int64.of_int hxw)
     (Int64.of_int group)
     eps;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
 let native_group_norm_out ~out0 ~out1 ~out2 input ~weight ~bias ~n ~c ~hxw ~group ~eps =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs_native_group_norm_out
     (CArray.start out__)
     out0
@@ -23288,26 +15216,23 @@ let native_group_norm_out ~out0 ~out1 ~out2 input ~weight ~bias ~n ~c ~hxw ~grou
     input
     (match weight with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (Int64.of_int n)
     (Int64.of_int c)
     (Int64.of_int hxw)
     (Int64.of_int group)
     eps;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
 let native_layer_norm input ~normalized_shape ~weight ~bias ~eps =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs_native_layer_norm
     (CArray.start out__)
     input
@@ -23315,22 +15240,19 @@ let native_layer_norm input ~normalized_shape ~weight ~bias ~eps =
     (List.length normalized_shape)
     (match weight with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     eps;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
 let native_layer_norm_out ~out0 ~out1 ~out2 input ~normalized_shape ~weight ~bias ~eps =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs_native_layer_norm_out
     (CArray.start out__)
     out0
@@ -23341,167 +15263,58 @@ let native_layer_norm_out ~out0 ~out1 ~out2 input ~normalized_shape ~weight ~bia
     (List.length normalized_shape)
     (match weight with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     eps;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
-let native_norm self =
-  let out__ = CArray.make t 1 in
-  stubs_native_norm (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let native_norm_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_native_norm_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let native_norm self = stubs_native_norm self |> with_tensor_gc
+let native_norm_out ~out self = stubs_native_norm_out out self |> with_tensor_gc
 
 let native_norm_scalaropt_dim_dtype self ~p ~dim ~keepdim ~dtype =
-  let out__ = CArray.make t 1 in
   stubs_native_norm_scalaropt_dim_dtype
-    (CArray.start out__)
     self
     p
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
     (List.length dim)
     (if keepdim then 1 else 0)
-    (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Kind.packed_to_int dtype)
+  |> with_tensor_gc
 ;;
 
 let native_norm_scalaropt_dim_dtype_out ~out self ~p ~dim ~keepdim ~dtype =
-  let out__ = CArray.make t 1 in
   stubs_native_norm_scalaropt_dim_dtype_out
-    (CArray.start out__)
     out
     self
     p
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
     (List.length dim)
     (if keepdim then 1 else 0)
-    (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let ne self other =
-  let out__ = CArray.make t 1 in
-  stubs_ne (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let ne_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_ne_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let ne_scalar_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_ne_scalar_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let ne_tensor self other =
-  let out__ = CArray.make t 1 in
-  stubs_ne_tensor (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let ne_tensor_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_ne_tensor_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let ne_tensor_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_ne_tensor_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let neg self =
-  let out__ = CArray.make t 1 in
-  stubs_neg (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let neg_ self =
-  let out__ = CArray.make t 1 in
-  stubs_neg_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let neg_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_neg_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let negative self =
-  let out__ = CArray.make t 1 in
-  stubs_negative (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let negative_ self =
-  let out__ = CArray.make t 1 in
-  stubs_negative_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let negative_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_negative_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+    (Kind.packed_to_int dtype)
+  |> with_tensor_gc
+;;
+
+let ne self other = stubs_ne self other |> with_tensor_gc
+let ne_ self other = stubs_ne_ self other |> with_tensor_gc
+let ne_scalar_out ~out self other = stubs_ne_scalar_out out self other |> with_tensor_gc
+let ne_tensor self other = stubs_ne_tensor self other |> with_tensor_gc
+let ne_tensor_ self other = stubs_ne_tensor_ self other |> with_tensor_gc
+let ne_tensor_out ~out self other = stubs_ne_tensor_out out self other |> with_tensor_gc
+let neg self = stubs_neg self |> with_tensor_gc
+let neg_ self = stubs_neg_ self |> with_tensor_gc
+let neg_out ~out self = stubs_neg_out out self |> with_tensor_gc
+let negative self = stubs_negative self |> with_tensor_gc
+let negative_ self = stubs_negative_ self |> with_tensor_gc
+let negative_out ~out self = stubs_negative_out out self |> with_tensor_gc
 
 let nested_to_padded_tensor self ~padding ~output_size =
-  let out__ = CArray.make t 1 in
   stubs_nested_to_padded_tensor
-    (CArray.start out__)
     self
     padding
     (match output_size with
@@ -23509,207 +15322,137 @@ let nested_to_padded_tensor self ~padding ~output_size =
      | Some v -> List.map Int64.of_int v |> CArray.of_list int64_t |> CArray.start)
     (match output_size with
      | None -> -1
-     | Some v -> List.length v);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | Some v -> List.length v)
+  |> with_tensor_gc
 ;;
 
 let new_empty self ~size ~options =
-  let out__ = CArray.make t 1 in
   stubs_new_empty
-    (CArray.start out__)
     self
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let new_empty_out ~out self ~size =
-  let out__ = CArray.make t 1 in
   stubs_new_empty_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
-    (List.length size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length size)
+  |> with_tensor_gc
 ;;
 
 let new_empty_strided self ~size ~stride ~options =
-  let out__ = CArray.make t 1 in
   stubs_new_empty_strided
-    (CArray.start out__)
     self
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
     (List.length stride)
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let new_empty_strided_out ~out self ~size ~stride =
-  let out__ = CArray.make t 1 in
   stubs_new_empty_strided_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
-    (List.length stride);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length stride)
+  |> with_tensor_gc
 ;;
 
 let new_full self ~size ~fill_value ~options =
-  let out__ = CArray.make t 1 in
   stubs_new_full
-    (CArray.start out__)
     self
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     fill_value
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let new_full_out ~out self ~size ~fill_value =
-  let out__ = CArray.make t 1 in
   stubs_new_full_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
-    fill_value;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    fill_value
+  |> with_tensor_gc
 ;;
 
 let new_ones self ~size ~options =
-  let out__ = CArray.make t 1 in
   stubs_new_ones
-    (CArray.start out__)
     self
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let new_ones_out ~out self ~size =
-  let out__ = CArray.make t 1 in
   stubs_new_ones_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
-    (List.length size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length size)
+  |> with_tensor_gc
 ;;
 
 let new_zeros self ~size ~options =
-  let out__ = CArray.make t 1 in
   stubs_new_zeros
-    (CArray.start out__)
     self
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let new_zeros_out ~out self ~size =
-  let out__ = CArray.make t 1 in
   stubs_new_zeros_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
-    (List.length size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let nextafter self other =
-  let out__ = CArray.make t 1 in
-  stubs_nextafter (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let nextafter_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_nextafter_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length size)
+  |> with_tensor_gc
 ;;
 
-let nextafter_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_nextafter_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let nextafter self other = stubs_nextafter self other |> with_tensor_gc
+let nextafter_ self other = stubs_nextafter_ self other |> with_tensor_gc
+let nextafter_out ~out self other = stubs_nextafter_out out self other |> with_tensor_gc
 
 let nll_loss self ~target ~weight ~reduction ~ignore_index =
-  let out__ = CArray.make t 1 in
   stubs_nll_loss
-    (CArray.start out__)
     self
     target
     (match weight with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (Reduction.to_int reduction |> Int64.of_int)
-    (Int64.of_int ignore_index);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int ignore_index)
+  |> with_tensor_gc
 ;;
 
 let nll_loss2d self ~target ~weight ~reduction ~ignore_index =
-  let out__ = CArray.make t 1 in
   stubs_nll_loss2d
-    (CArray.start out__)
     self
     target
     (match weight with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (Reduction.to_int reduction |> Int64.of_int)
-    (Int64.of_int ignore_index);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int ignore_index)
+  |> with_tensor_gc
 ;;
 
 let nll_loss2d_backward
@@ -23721,21 +15464,17 @@ let nll_loss2d_backward
   ~ignore_index
   ~total_weight
   =
-  let out__ = CArray.make t 1 in
   stubs_nll_loss2d_backward
-    (CArray.start out__)
     grad_output
     self
     target
     (match weight with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (Reduction.to_int reduction |> Int64.of_int)
     (Int64.of_int ignore_index)
-    total_weight;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    total_weight
+  |> with_tensor_gc
 ;;
 
 let nll_loss2d_backward_grad_input
@@ -23748,39 +15487,31 @@ let nll_loss2d_backward_grad_input
   ~ignore_index
   ~total_weight
   =
-  let out__ = CArray.make t 1 in
   stubs_nll_loss2d_backward_grad_input
-    (CArray.start out__)
     grad_input
     grad_output
     self
     target
     (match weight with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (Reduction.to_int reduction |> Int64.of_int)
     (Int64.of_int ignore_index)
-    total_weight;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    total_weight
+  |> with_tensor_gc
 ;;
 
 let nll_loss2d_out ~out self ~target ~weight ~reduction ~ignore_index =
-  let out__ = CArray.make t 1 in
   stubs_nll_loss2d_out
-    (CArray.start out__)
     out
     self
     target
     (match weight with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (Reduction.to_int reduction |> Int64.of_int)
-    (Int64.of_int ignore_index);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int ignore_index)
+  |> with_tensor_gc
 ;;
 
 let nll_loss_backward
@@ -23792,21 +15523,17 @@ let nll_loss_backward
   ~ignore_index
   ~total_weight
   =
-  let out__ = CArray.make t 1 in
   stubs_nll_loss_backward
-    (CArray.start out__)
     grad_output
     self
     target
     (match weight with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (Reduction.to_int reduction |> Int64.of_int)
     (Int64.of_int ignore_index)
-    total_weight;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    total_weight
+  |> with_tensor_gc
 ;;
 
 let nll_loss_backward_grad_input
@@ -23819,436 +15546,217 @@ let nll_loss_backward_grad_input
   ~ignore_index
   ~total_weight
   =
-  let out__ = CArray.make t 1 in
   stubs_nll_loss_backward_grad_input
-    (CArray.start out__)
     grad_input
     grad_output
     self
     target
     (match weight with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (Reduction.to_int reduction |> Int64.of_int)
     (Int64.of_int ignore_index)
-    total_weight;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    total_weight
+  |> with_tensor_gc
 ;;
 
 let nll_loss_nd self ~target ~weight ~reduction ~ignore_index =
-  let out__ = CArray.make t 1 in
   stubs_nll_loss_nd
-    (CArray.start out__)
     self
     target
     (match weight with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (Reduction.to_int reduction |> Int64.of_int)
-    (Int64.of_int ignore_index);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int ignore_index)
+  |> with_tensor_gc
 ;;
 
 let nll_loss_out ~out self ~target ~weight ~reduction ~ignore_index =
-  let out__ = CArray.make t 1 in
   stubs_nll_loss_out
-    (CArray.start out__)
     out
     self
     target
     (match weight with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (Reduction.to_int reduction |> Int64.of_int)
-    (Int64.of_int ignore_index);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let nonzero self =
-  let out__ = CArray.make t 1 in
-  stubs_nonzero (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int ignore_index)
+  |> with_tensor_gc
 ;;
 
+let nonzero self = stubs_nonzero self |> with_tensor_gc
 let nonzero_numpy self = stubs_nonzero_numpy self |> to_tensor_list
-
-let nonzero_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_nonzero_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let nonzero_out ~out self = stubs_nonzero_out out self |> with_tensor_gc
 
 let nonzero_static self ~size ~fill_value =
-  let out__ = CArray.make t 1 in
-  stubs_nonzero_static
-    (CArray.start out__)
-    self
-    (Int64.of_int size)
-    (Int64.of_int fill_value);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_nonzero_static self (Int64.of_int size) (Int64.of_int fill_value)
+  |> with_tensor_gc
 ;;
 
 let nonzero_static_out ~out self ~size ~fill_value =
-  let out__ = CArray.make t 1 in
-  stubs_nonzero_static_out
-    (CArray.start out__)
-    out
-    self
-    (Int64.of_int size)
-    (Int64.of_int fill_value);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_nonzero_static_out out self (Int64.of_int size) (Int64.of_int fill_value)
+  |> with_tensor_gc
 ;;
 
-let norm self =
-  let out__ = CArray.make t 1 in
-  stubs_norm (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let norm self = stubs_norm self |> with_tensor_gc
 
 let norm_dtype_out ~out self ~p ~dim ~keepdim ~dtype =
-  let out__ = CArray.make t 1 in
   stubs_norm_dtype_out
-    (CArray.start out__)
     out
     self
     p
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
     (List.length dim)
     (if keepdim then 1 else 0)
-    (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Kind.packed_to_int dtype)
+  |> with_tensor_gc
 ;;
 
 let norm_except_dim ~v ~pow ~dim =
-  let out__ = CArray.make t 1 in
-  stubs_norm_except_dim (CArray.start out__) v (Int64.of_int pow) (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_norm_except_dim v (Int64.of_int pow) (Int64.of_int dim) |> with_tensor_gc
 ;;
 
 let norm_out ~out self ~p ~dim ~keepdim =
-  let out__ = CArray.make t 1 in
   stubs_norm_out
-    (CArray.start out__)
     out
     self
     p
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
     (List.length dim)
-    (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if keepdim then 1 else 0)
+  |> with_tensor_gc
 ;;
 
-let norm_scalar_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_norm_scalar_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let norm_scalar_out ~out self = stubs_norm_scalar_out out self |> with_tensor_gc
 
 let norm_scalaropt_dim self ~p ~dim ~keepdim =
-  let out__ = CArray.make t 1 in
   stubs_norm_scalaropt_dim
-    (CArray.start out__)
     self
     p
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
     (List.length dim)
-    (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if keepdim then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let norm_scalaropt_dim_dtype self ~p ~dim ~keepdim ~dtype =
-  let out__ = CArray.make t 1 in
   stubs_norm_scalaropt_dim_dtype
-    (CArray.start out__)
     self
     p
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
     (List.length dim)
     (if keepdim then 1 else 0)
-    (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Kind.packed_to_int dtype)
+  |> with_tensor_gc
 ;;
 
 let norm_scalaropt_dtype self ~p ~dtype =
-  let out__ = CArray.make t 1 in
-  stubs_norm_scalaropt_dtype (CArray.start out__) self p (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_norm_scalaropt_dtype self p (Kind.packed_to_int dtype) |> with_tensor_gc
 ;;
 
 let norm_scalaropt_dtype_out ~out self ~p ~dtype =
-  let out__ = CArray.make t 1 in
-  stubs_norm_scalaropt_dtype_out
-    (CArray.start out__)
-    out
-    self
-    p
-    (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_norm_scalaropt_dtype_out out self p (Kind.packed_to_int dtype) |> with_tensor_gc
 ;;
 
-let normal_ self ~mean ~std =
-  let out__ = CArray.make t 1 in
-  stubs_normal_ (CArray.start out__) self mean std;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let normal_ self ~mean ~std = stubs_normal_ self mean std |> with_tensor_gc
 
 let normal_functional self ~mean ~std =
-  let out__ = CArray.make t 1 in
-  stubs_normal_functional (CArray.start out__) self mean std;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let not_equal self other =
-  let out__ = CArray.make t 1 in
-  stubs_not_equal (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_normal_functional self mean std |> with_tensor_gc
 ;;
 
-let not_equal_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_not_equal_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let not_equal self other = stubs_not_equal self other |> with_tensor_gc
+let not_equal_ self other = stubs_not_equal_ self other |> with_tensor_gc
 
 let not_equal_scalar_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_not_equal_scalar_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let not_equal_tensor self other =
-  let out__ = CArray.make t 1 in
-  stubs_not_equal_tensor (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_not_equal_scalar_out out self other |> with_tensor_gc
 ;;
 
-let not_equal_tensor_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_not_equal_tensor_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let not_equal_tensor self other = stubs_not_equal_tensor self other |> with_tensor_gc
+let not_equal_tensor_ self other = stubs_not_equal_tensor_ self other |> with_tensor_gc
 
 let not_equal_tensor_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_not_equal_tensor_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_not_equal_tensor_out out self other |> with_tensor_gc
 ;;
 
 let nuclear_norm self ~keepdim =
-  let out__ = CArray.make t 1 in
-  stubs_nuclear_norm (CArray.start out__) self (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_nuclear_norm self (if keepdim then 1 else 0) |> with_tensor_gc
 ;;
 
 let nuclear_norm_dim self ~dim ~keepdim =
-  let out__ = CArray.make t 1 in
   stubs_nuclear_norm_dim
-    (CArray.start out__)
     self
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
     (List.length dim)
-    (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if keepdim then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let nuclear_norm_dim_out ~out self ~dim ~keepdim =
-  let out__ = CArray.make t 1 in
   stubs_nuclear_norm_dim_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
     (List.length dim)
-    (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if keepdim then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let nuclear_norm_out ~out self ~keepdim =
-  let out__ = CArray.make t 1 in
-  stubs_nuclear_norm_out (CArray.start out__) out self (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_nuclear_norm_out out self (if keepdim then 1 else 0) |> with_tensor_gc
 ;;
 
-let numpy_t self =
-  let out__ = CArray.make t 1 in
-  stubs_numpy_t (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let numpy_t self = stubs_numpy_t self |> with_tensor_gc
 
 let one_hot self ~num_classes =
-  let out__ = CArray.make t 1 in
-  stubs_one_hot (CArray.start out__) self (Int64.of_int num_classes);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_one_hot self (Int64.of_int num_classes) |> with_tensor_gc
 ;;
 
 let ones ~size ~options =
-  let out__ = CArray.make t 1 in
   stubs_ones
-    (CArray.start out__)
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let ones_like self =
-  let out__ = CArray.make t 1 in
-  stubs_ones_like (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
-let ones_like_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_ones_like_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let ones_like self = stubs_ones_like self |> with_tensor_gc
+let ones_like_out ~out self = stubs_ones_like_out out self |> with_tensor_gc
 
 let ones_out ~out ~size =
-  let out__ = CArray.make t 1 in
   stubs_ones_out
-    (CArray.start out__)
     out
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
-    (List.length size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let orgqr self ~input2 =
-  let out__ = CArray.make t 1 in
-  stubs_orgqr (CArray.start out__) self input2;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length size)
+  |> with_tensor_gc
 ;;
 
-let orgqr_out ~out self ~input2 =
-  let out__ = CArray.make t 1 in
-  stubs_orgqr_out (CArray.start out__) out self input2;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let orgqr self ~input2 = stubs_orgqr self input2 |> with_tensor_gc
+let orgqr_out ~out self ~input2 = stubs_orgqr_out out self input2 |> with_tensor_gc
 
 let ormqr self ~input2 ~input3 ~left ~transpose =
-  let out__ = CArray.make t 1 in
-  stubs_ormqr
-    (CArray.start out__)
-    self
-    input2
-    input3
-    (if left then 1 else 0)
-    (if transpose then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_ormqr self input2 input3 (if left then 1 else 0) (if transpose then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let ormqr_out ~out self ~input2 ~input3 ~left ~transpose =
-  let out__ = CArray.make t 1 in
   stubs_ormqr_out
-    (CArray.start out__)
     out
     self
     input2
     input3
     (if left then 1 else 0)
-    (if transpose then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let outer self ~vec2 =
-  let out__ = CArray.make t 1 in
-  stubs_outer (CArray.start out__) self vec2;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let outer_out ~out self ~vec2 =
-  let out__ = CArray.make t 1 in
-  stubs_outer_out (CArray.start out__) out self vec2;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if transpose then 1 else 0)
+  |> with_tensor_gc
 ;;
 
+let outer self ~vec2 = stubs_outer self vec2 |> with_tensor_gc
+let outer_out ~out self ~vec2 = stubs_outer_out out self vec2 |> with_tensor_gc
 let output_nr self = stubs_output_nr self
 
 let pad self ~pad ~mode ~value =
-  let out__ = CArray.make t 1 in
   stubs_pad
-    (CArray.start out__)
     self
     (List.map Int64.of_int pad |> CArray.of_list int64_t |> CArray.start)
     (List.length pad)
@@ -24256,405 +15764,192 @@ let pad self ~pad ~mode ~value =
     (Option.value value ~default:0.0)
     (match value with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let pad_sequence ~sequences ~batch_first ~padding_value =
-  let out__ = CArray.make t 1 in
   stubs_pad_sequence
-    (CArray.start out__)
-    (CArray.of_list t sequences |> CArray.start)
+    (CArray.of_list gc_tensor sequences |> CArray.start)
     (List.length sequences)
     (if batch_first then 1 else 0)
-    padding_value;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    padding_value
+  |> with_tensor_gc
 ;;
 
 let pairwise_distance ~x1 ~x2 ~p ~eps ~keepdim =
-  let out__ = CArray.make t 1 in
-  stubs_pairwise_distance (CArray.start out__) x1 x2 p eps (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_pairwise_distance x1 x2 p eps (if keepdim then 1 else 0) |> with_tensor_gc
 ;;
 
-let pdist self ~p =
-  let out__ = CArray.make t 1 in
-  stubs_pdist (CArray.start out__) self p;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let pdist self ~p = stubs_pdist self p |> with_tensor_gc
 
 let permute self ~dims =
-  let out__ = CArray.make t 1 in
   stubs_permute
-    (CArray.start out__)
     self
     (List.map Int64.of_int dims |> CArray.of_list int64_t |> CArray.start)
-    (List.length dims);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length dims)
+  |> with_tensor_gc
 ;;
 
 let permute_copy self ~dims =
-  let out__ = CArray.make t 1 in
   stubs_permute_copy
-    (CArray.start out__)
     self
     (List.map Int64.of_int dims |> CArray.of_list int64_t |> CArray.start)
-    (List.length dims);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length dims)
+  |> with_tensor_gc
 ;;
 
 let permute_copy_out ~out self ~dims =
-  let out__ = CArray.make t 1 in
   stubs_permute_copy_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int dims |> CArray.of_list int64_t |> CArray.start)
-    (List.length dims);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length dims)
+  |> with_tensor_gc
 ;;
 
 let pin_memory self ~device =
-  let out__ = CArray.make t 1 in
-  stubs_pin_memory (CArray.start out__) self (Device.to_int device);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_pin_memory self (Device.to_int device) |> with_tensor_gc
 ;;
 
-let pinverse self ~rcond =
-  let out__ = CArray.make t 1 in
-  stubs_pinverse (CArray.start out__) self rcond;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let pinverse self ~rcond = stubs_pinverse self rcond |> with_tensor_gc
 
 let pixel_shuffle self ~upscale_factor =
-  let out__ = CArray.make t 1 in
-  stubs_pixel_shuffle (CArray.start out__) self (Int64.of_int upscale_factor);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_pixel_shuffle self (Int64.of_int upscale_factor) |> with_tensor_gc
 ;;
 
 let pixel_shuffle_out ~out self ~upscale_factor =
-  let out__ = CArray.make t 1 in
-  stubs_pixel_shuffle_out (CArray.start out__) out self (Int64.of_int upscale_factor);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_pixel_shuffle_out out self (Int64.of_int upscale_factor) |> with_tensor_gc
 ;;
 
 let pixel_unshuffle self ~downscale_factor =
-  let out__ = CArray.make t 1 in
-  stubs_pixel_unshuffle (CArray.start out__) self (Int64.of_int downscale_factor);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_pixel_unshuffle self (Int64.of_int downscale_factor) |> with_tensor_gc
 ;;
 
 let pixel_unshuffle_out ~out self ~downscale_factor =
-  let out__ = CArray.make t 1 in
-  stubs_pixel_unshuffle_out (CArray.start out__) out self (Int64.of_int downscale_factor);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_pixel_unshuffle_out out self (Int64.of_int downscale_factor) |> with_tensor_gc
 ;;
 
-let poisson self =
-  let out__ = CArray.make t 1 in
-  stubs_poisson (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let poisson self = stubs_poisson self |> with_tensor_gc
 
 let poisson_nll_loss input ~target ~log_input ~full ~eps ~reduction =
-  let out__ = CArray.make t 1 in
   stubs_poisson_nll_loss
-    (CArray.start out__)
     input
     target
     (if log_input then 1 else 0)
     (if full then 1 else 0)
     eps
-    (Reduction.to_int reduction |> Int64.of_int);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let poisson_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_poisson_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let polar ~abs ~angle =
-  let out__ = CArray.make t 1 in
-  stubs_polar (CArray.start out__) abs angle;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let polar_out ~out ~abs ~angle =
-  let out__ = CArray.make t 1 in
-  stubs_polar_out (CArray.start out__) out abs angle;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let polygamma ~n self =
-  let out__ = CArray.make t 1 in
-  stubs_polygamma (CArray.start out__) (Int64.of_int n) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Reduction.to_int reduction |> Int64.of_int)
+  |> with_tensor_gc
 ;;
 
-let polygamma_ self ~n =
-  let out__ = CArray.make t 1 in
-  stubs_polygamma_ (CArray.start out__) self (Int64.of_int n);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let poisson_out ~out self = stubs_poisson_out out self |> with_tensor_gc
+let polar ~abs ~angle = stubs_polar abs angle |> with_tensor_gc
+let polar_out ~out ~abs ~angle = stubs_polar_out out abs angle |> with_tensor_gc
+let polygamma ~n self = stubs_polygamma (Int64.of_int n) self |> with_tensor_gc
+let polygamma_ self ~n = stubs_polygamma_ self (Int64.of_int n) |> with_tensor_gc
 
 let polygamma_out ~out ~n self =
-  let out__ = CArray.make t 1 in
-  stubs_polygamma_out (CArray.start out__) out (Int64.of_int n) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let positive self =
-  let out__ = CArray.make t 1 in
-  stubs_positive (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let pow self ~exponent =
-  let out__ = CArray.make t 1 in
-  stubs_pow (CArray.start out__) self exponent;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let pow_ self ~exponent =
-  let out__ = CArray.make t 1 in
-  stubs_pow_ (CArray.start out__) self exponent;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_polygamma_out out (Int64.of_int n) self |> with_tensor_gc
 ;;
 
-let pow_scalar self ~exponent =
-  let out__ = CArray.make t 1 in
-  stubs_pow_scalar (CArray.start out__) self exponent;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let positive self = stubs_positive self |> with_tensor_gc
+let pow self ~exponent = stubs_pow self exponent |> with_tensor_gc
+let pow_ self ~exponent = stubs_pow_ self exponent |> with_tensor_gc
+let pow_scalar self ~exponent = stubs_pow_scalar self exponent |> with_tensor_gc
 
 let pow_scalar_out ~out self ~exponent =
-  let out__ = CArray.make t 1 in
-  stubs_pow_scalar_out (CArray.start out__) out self exponent;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_pow_scalar_out out self exponent |> with_tensor_gc
 ;;
 
-let pow_tensor_ self ~exponent =
-  let out__ = CArray.make t 1 in
-  stubs_pow_tensor_ (CArray.start out__) self exponent;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let pow_tensor_ self ~exponent = stubs_pow_tensor_ self exponent |> with_tensor_gc
 
 let pow_tensor_scalar self ~exponent =
-  let out__ = CArray.make t 1 in
-  stubs_pow_tensor_scalar (CArray.start out__) self exponent;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_pow_tensor_scalar self exponent |> with_tensor_gc
 ;;
 
 let pow_tensor_scalar_out ~out self ~exponent =
-  let out__ = CArray.make t 1 in
-  stubs_pow_tensor_scalar_out (CArray.start out__) out self exponent;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_pow_tensor_scalar_out out self exponent |> with_tensor_gc
 ;;
 
 let pow_tensor_tensor_out ~out self ~exponent =
-  let out__ = CArray.make t 1 in
-  stubs_pow_tensor_tensor_out (CArray.start out__) out self exponent;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let prelu self ~weight =
-  let out__ = CArray.make t 1 in
-  stubs_prelu (CArray.start out__) self weight;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_pow_tensor_tensor_out out self exponent |> with_tensor_gc
 ;;
 
-let prod self ~dtype =
-  let out__ = CArray.make t 1 in
-  stubs_prod (CArray.start out__) self (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let prelu self ~weight = stubs_prelu self weight |> with_tensor_gc
+let prod self ~dtype = stubs_prod self (Kind.packed_to_int dtype) |> with_tensor_gc
 
 let prod_dim_int self ~dim ~keepdim ~dtype =
-  let out__ = CArray.make t 1 in
   stubs_prod_dim_int
-    (CArray.start out__)
     self
     (Int64.of_int dim)
     (if keepdim then 1 else 0)
-    (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Kind.packed_to_int dtype)
+  |> with_tensor_gc
 ;;
 
 let prod_int_out ~out self ~dim ~keepdim ~dtype =
-  let out__ = CArray.make t 1 in
   stubs_prod_int_out
-    (CArray.start out__)
     out
     self
     (Int64.of_int dim)
     (if keepdim then 1 else 0)
-    (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Kind.packed_to_int dtype)
+  |> with_tensor_gc
 ;;
 
 let prod_out ~out self ~dtype =
-  let out__ = CArray.make t 1 in
-  stubs_prod_out (CArray.start out__) out self (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_prod_out out self (Kind.packed_to_int dtype) |> with_tensor_gc
 ;;
 
 let put self ~index ~source ~accumulate =
-  let out__ = CArray.make t 1 in
-  stubs_put (CArray.start out__) self index source (if accumulate then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_put self index source (if accumulate then 1 else 0) |> with_tensor_gc
 ;;
 
 let put_ self ~index ~source ~accumulate =
-  let out__ = CArray.make t 1 in
-  stubs_put_ (CArray.start out__) self index source (if accumulate then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_put_ self index source (if accumulate then 1 else 0) |> with_tensor_gc
 ;;
 
 let put_out ~out self ~index ~source ~accumulate =
-  let out__ = CArray.make t 1 in
-  stubs_put_out (CArray.start out__) out self index source (if accumulate then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_put_out out self index source (if accumulate then 1 else 0) |> with_tensor_gc
 ;;
 
 let q_per_channel_axis self = stubs_q_per_channel_axis self
-
-let q_per_channel_scales self =
-  let out__ = CArray.make t 1 in
-  stubs_q_per_channel_scales (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let q_per_channel_scales self = stubs_q_per_channel_scales self |> with_tensor_gc
 
 let q_per_channel_scales_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_q_per_channel_scales_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_q_per_channel_scales_out out self |> with_tensor_gc
 ;;
 
 let q_per_channel_zero_points self =
-  let out__ = CArray.make t 1 in
-  stubs_q_per_channel_zero_points (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_q_per_channel_zero_points self |> with_tensor_gc
 ;;
 
 let q_per_channel_zero_points_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_q_per_channel_zero_points_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_q_per_channel_zero_points_out out self |> with_tensor_gc
 ;;
 
 let q_scale self = stubs_q_scale self
 let q_zero_point self = stubs_q_zero_point self
 
 let qr self ~some =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_qr (CArray.start out__) self (if some then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let qr_q ~q ~r self ~some =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_qr_q (CArray.start out__) q r self (if some then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let quantile self ~q ~dim ~keepdim ~interpolation =
-  let out__ = CArray.make t 1 in
   stubs_quantile
-    (CArray.start out__)
     self
     q
     (match dim with
@@ -24664,16 +15959,12 @@ let quantile self ~q ~dim ~keepdim ~interpolation =
      | Some _ -> 0
      | None -> 1)
     (if keepdim then 1 else 0)
-    interpolation;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    interpolation
+  |> with_tensor_gc
 ;;
 
 let quantile_out ~out self ~q ~dim ~keepdim ~interpolation =
-  let out__ = CArray.make t 1 in
   stubs_quantile_out
-    (CArray.start out__)
     out
     self
     q
@@ -24684,16 +15975,12 @@ let quantile_out ~out self ~q ~dim ~keepdim ~interpolation =
      | Some _ -> 0
      | None -> 1)
     (if keepdim then 1 else 0)
-    interpolation;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    interpolation
+  |> with_tensor_gc
 ;;
 
 let quantile_scalar self ~q ~dim ~keepdim ~interpolation =
-  let out__ = CArray.make t 1 in
   stubs_quantile_scalar
-    (CArray.start out__)
     self
     q
     (match dim with
@@ -24703,16 +15990,12 @@ let quantile_scalar self ~q ~dim ~keepdim ~interpolation =
      | Some _ -> 0
      | None -> 1)
     (if keepdim then 1 else 0)
-    interpolation;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    interpolation
+  |> with_tensor_gc
 ;;
 
 let quantile_scalar_out ~out self ~q ~dim ~keepdim ~interpolation =
-  let out__ = CArray.make t 1 in
   stubs_quantile_scalar_out
-    (CArray.start out__)
     out
     self
     q
@@ -24723,123 +16006,89 @@ let quantile_scalar_out ~out self ~q ~dim ~keepdim ~interpolation =
      | Some _ -> 0
      | None -> 1)
     (if keepdim then 1 else 0)
-    interpolation;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    interpolation
+  |> with_tensor_gc
 ;;
 
 let quantize_per_channel self ~scales ~zero_points ~axis ~dtype =
-  let out__ = CArray.make t 1 in
   stubs_quantize_per_channel
-    (CArray.start out__)
     self
     scales
     zero_points
     (Int64.of_int axis)
-    (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Kind.packed_to_int dtype)
+  |> with_tensor_gc
 ;;
 
 let quantize_per_channel_out ~out self ~scales ~zero_points ~axis ~dtype =
-  let out__ = CArray.make t 1 in
   stubs_quantize_per_channel_out
-    (CArray.start out__)
     out
     self
     scales
     zero_points
     (Int64.of_int axis)
-    (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Kind.packed_to_int dtype)
+  |> with_tensor_gc
 ;;
 
 let quantize_per_tensor self ~scale ~zero_point ~dtype =
-  let out__ = CArray.make t 1 in
   stubs_quantize_per_tensor
-    (CArray.start out__)
     self
     scale
     (Int64.of_int zero_point)
-    (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Kind.packed_to_int dtype)
+  |> with_tensor_gc
 ;;
 
 let quantize_per_tensor_dynamic self ~dtype ~reduce_range =
-  let out__ = CArray.make t 1 in
   stubs_quantize_per_tensor_dynamic
-    (CArray.start out__)
     self
     (Kind.packed_to_int dtype)
-    (if reduce_range then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if reduce_range then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let quantize_per_tensor_dynamic_out ~out self ~dtype ~reduce_range =
-  let out__ = CArray.make t 1 in
   stubs_quantize_per_tensor_dynamic_out
-    (CArray.start out__)
     out
     self
     (Kind.packed_to_int dtype)
-    (if reduce_range then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if reduce_range then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let quantize_per_tensor_out ~out self ~scale ~zero_point ~dtype =
-  let out__ = CArray.make t 1 in
   stubs_quantize_per_tensor_out
-    (CArray.start out__)
     out
     self
     scale
     (Int64.of_int zero_point)
-    (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Kind.packed_to_int dtype)
+  |> with_tensor_gc
 ;;
 
 let quantize_per_tensor_tensor_qparams self ~scale ~zero_point ~dtype =
-  let out__ = CArray.make t 1 in
   stubs_quantize_per_tensor_tensor_qparams
-    (CArray.start out__)
     self
     scale
     zero_point
-    (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Kind.packed_to_int dtype)
+  |> with_tensor_gc
 ;;
 
 let quantize_per_tensor_tensor_qparams_out ~out self ~scale ~zero_point ~dtype =
-  let out__ = CArray.make t 1 in
   stubs_quantize_per_tensor_tensor_qparams_out
-    (CArray.start out__)
     out
     self
     scale
     zero_point
-    (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Kind.packed_to_int dtype)
+  |> with_tensor_gc
 ;;
 
 let quantize_per_tensor_tensors tensors ~scales ~zero_points ~dtype =
   stubs_quantize_per_tensor_tensors
-    (CArray.of_list t tensors |> CArray.start)
+    (CArray.of_list gc_tensor tensors |> CArray.start)
     (List.length tensors)
     scales
     zero_points
@@ -24849,9 +16098,9 @@ let quantize_per_tensor_tensors tensors ~scales ~zero_points ~dtype =
 
 let quantize_per_tensor_tensors_out ~out tensors ~scales ~zero_points ~dtype =
   stubs_quantize_per_tensor_tensors_out
-    (CArray.of_list t out |> CArray.start)
+    (CArray.of_list gc_tensor out |> CArray.start)
     (List.length out)
-    (CArray.of_list t tensors |> CArray.start)
+    (CArray.of_list gc_tensor tensors |> CArray.start)
     (List.length tensors)
     scales
     zero_points
@@ -24868,24 +16117,20 @@ let quantized_batch_norm
   ~output_scale
   ~output_zero_point
   =
-  let out__ = CArray.make t 1 in
   stubs_quantized_batch_norm
-    (CArray.start out__)
     input
     (match weight with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     mean
     var
     eps
     output_scale
-    (Int64.of_int output_zero_point);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int output_zero_point)
+  |> with_tensor_gc
 ;;
 
 let quantized_batch_norm_out
@@ -24899,25 +16144,21 @@ let quantized_batch_norm_out
   ~output_scale
   ~output_zero_point
   =
-  let out__ = CArray.make t 1 in
   stubs_quantized_batch_norm_out
-    (CArray.start out__)
     out
     input
     (match weight with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     mean
     var
     eps
     output_scale
-    (Int64.of_int output_zero_point);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int output_zero_point)
+  |> with_tensor_gc
 ;;
 
 let quantized_gru_cell
@@ -24936,9 +16177,7 @@ let quantized_gru_cell
   ~zero_point_ih
   ~zero_point_hh
   =
-  let out__ = CArray.make t 1 in
   stubs_quantized_gru_cell
-    (CArray.start out__)
     input
     hx
     w_ih
@@ -24952,10 +16191,8 @@ let quantized_gru_cell
     scale_ih
     scale_hh
     zero_point_ih
-    zero_point_hh;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    zero_point_hh
+  |> with_tensor_gc
 ;;
 
 let quantized_lstm_cell
@@ -24974,11 +16211,11 @@ let quantized_lstm_cell
   ~zero_point_ih
   ~zero_point_hh
   =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_quantized_lstm_cell
     (CArray.start out__)
     input
-    (CArray.of_list t hx |> CArray.start)
+    (CArray.of_list gc_tensor hx |> CArray.start)
     (List.length hx)
     w_ih
     w_hh
@@ -24992,17 +16229,13 @@ let quantized_lstm_cell
     scale_hh
     zero_point_ih
     zero_point_hh;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let quantized_max_pool1d self ~kernel_size ~stride ~padding ~dilation ~ceil_mode =
-  let out__ = CArray.make t 1 in
   stubs_quantized_max_pool1d
-    (CArray.start out__)
     self
     (List.map Int64.of_int kernel_size |> CArray.of_list int64_t |> CArray.start)
     (List.length kernel_size)
@@ -25012,16 +16245,12 @@ let quantized_max_pool1d self ~kernel_size ~stride ~padding ~dilation ~ceil_mode
     (List.length padding)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
-    (if ceil_mode then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if ceil_mode then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let quantized_max_pool1d_out ~out self ~kernel_size ~stride ~padding ~dilation ~ceil_mode =
-  let out__ = CArray.make t 1 in
   stubs_quantized_max_pool1d_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int kernel_size |> CArray.of_list int64_t |> CArray.start)
@@ -25032,16 +16261,12 @@ let quantized_max_pool1d_out ~out self ~kernel_size ~stride ~padding ~dilation ~
     (List.length padding)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
-    (if ceil_mode then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if ceil_mode then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let quantized_max_pool2d self ~kernel_size ~stride ~padding ~dilation ~ceil_mode =
-  let out__ = CArray.make t 1 in
   stubs_quantized_max_pool2d
-    (CArray.start out__)
     self
     (List.map Int64.of_int kernel_size |> CArray.of_list int64_t |> CArray.start)
     (List.length kernel_size)
@@ -25051,16 +16276,12 @@ let quantized_max_pool2d self ~kernel_size ~stride ~padding ~dilation ~ceil_mode
     (List.length padding)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
-    (if ceil_mode then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if ceil_mode then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let quantized_max_pool2d_out ~out self ~kernel_size ~stride ~padding ~dilation ~ceil_mode =
-  let out__ = CArray.make t 1 in
   stubs_quantized_max_pool2d_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int kernel_size |> CArray.of_list int64_t |> CArray.start)
@@ -25071,16 +16292,12 @@ let quantized_max_pool2d_out ~out self ~kernel_size ~stride ~padding ~dilation ~
     (List.length padding)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
-    (if ceil_mode then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if ceil_mode then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let quantized_max_pool3d self ~kernel_size ~stride ~padding ~dilation ~ceil_mode =
-  let out__ = CArray.make t 1 in
   stubs_quantized_max_pool3d
-    (CArray.start out__)
     self
     (List.map Int64.of_int kernel_size |> CArray.of_list int64_t |> CArray.start)
     (List.length kernel_size)
@@ -25090,16 +16307,12 @@ let quantized_max_pool3d self ~kernel_size ~stride ~padding ~dilation ~ceil_mode
     (List.length padding)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
-    (if ceil_mode then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if ceil_mode then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let quantized_max_pool3d_out ~out self ~kernel_size ~stride ~padding ~dilation ~ceil_mode =
-  let out__ = CArray.make t 1 in
   stubs_quantized_max_pool3d_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int kernel_size |> CArray.of_list int64_t |> CArray.start)
@@ -25110,10 +16323,8 @@ let quantized_max_pool3d_out ~out self ~kernel_size ~stride ~padding ~dilation ~
     (List.length padding)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
     (List.length dilation)
-    (if ceil_mode then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if ceil_mode then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let quantized_rnn_relu_cell
@@ -25132,9 +16343,7 @@ let quantized_rnn_relu_cell
   ~zero_point_ih
   ~zero_point_hh
   =
-  let out__ = CArray.make t 1 in
   stubs_quantized_rnn_relu_cell
-    (CArray.start out__)
     input
     hx
     w_ih
@@ -25148,10 +16357,8 @@ let quantized_rnn_relu_cell
     scale_ih
     scale_hh
     zero_point_ih
-    zero_point_hh;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    zero_point_hh
+  |> with_tensor_gc
 ;;
 
 let quantized_rnn_tanh_cell
@@ -25170,9 +16377,7 @@ let quantized_rnn_tanh_cell
   ~zero_point_ih
   ~zero_point_hh
   =
-  let out__ = CArray.make t 1 in
   stubs_quantized_rnn_tanh_cell
-    (CArray.start out__)
     input
     hx
     w_ih
@@ -25186,235 +16391,117 @@ let quantized_rnn_tanh_cell
     scale_ih
     scale_hh
     zero_point_ih
-    zero_point_hh;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let rad2deg self =
-  let out__ = CArray.make t 1 in
-  stubs_rad2deg (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let rad2deg_ self =
-  let out__ = CArray.make t 1 in
-  stubs_rad2deg_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    zero_point_hh
+  |> with_tensor_gc
 ;;
 
-let rad2deg_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_rad2deg_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let rad2deg self = stubs_rad2deg self |> with_tensor_gc
+let rad2deg_ self = stubs_rad2deg_ self |> with_tensor_gc
+let rad2deg_out ~out self = stubs_rad2deg_out out self |> with_tensor_gc
 
 let rand ~size ~options =
-  let out__ = CArray.make t 1 in
   stubs_rand
-    (CArray.start out__)
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let rand_like self =
-  let out__ = CArray.make t 1 in
-  stubs_rand_like (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
-let rand_like_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_rand_like_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let rand_like self = stubs_rand_like self |> with_tensor_gc
+let rand_like_out ~out self = stubs_rand_like_out out self |> with_tensor_gc
 
 let rand_out ~out ~size =
-  let out__ = CArray.make t 1 in
   stubs_rand_out
-    (CArray.start out__)
     out
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
-    (List.length size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length size)
+  |> with_tensor_gc
 ;;
 
 let randint ~high ~size ~options =
-  let out__ = CArray.make t 1 in
   stubs_randint
-    (CArray.start out__)
     (Int64.of_int high)
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let randint_like self ~high =
-  let out__ = CArray.make t 1 in
-  stubs_randint_like (CArray.start out__) self (Int64.of_int high);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_randint_like self (Int64.of_int high) |> with_tensor_gc
 ;;
 
 let randint_like_low_dtype self ~low ~high =
-  let out__ = CArray.make t 1 in
-  stubs_randint_like_low_dtype
-    (CArray.start out__)
-    self
-    (Int64.of_int low)
-    (Int64.of_int high);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_randint_like_low_dtype self (Int64.of_int low) (Int64.of_int high)
+  |> with_tensor_gc
 ;;
 
 let randint_like_low_dtype_out ~out self ~low ~high =
-  let out__ = CArray.make t 1 in
-  stubs_randint_like_low_dtype_out
-    (CArray.start out__)
-    out
-    self
-    (Int64.of_int low)
-    (Int64.of_int high);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_randint_like_low_dtype_out out self (Int64.of_int low) (Int64.of_int high)
+  |> with_tensor_gc
 ;;
 
 let randint_like_out ~out self ~high =
-  let out__ = CArray.make t 1 in
-  stubs_randint_like_out (CArray.start out__) out self (Int64.of_int high);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_randint_like_out out self (Int64.of_int high) |> with_tensor_gc
 ;;
 
 let randint_low ~low ~high ~size ~options =
-  let out__ = CArray.make t 1 in
   stubs_randint_low
-    (CArray.start out__)
     (Int64.of_int low)
     (Int64.of_int high)
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let randint_low_out ~out ~low ~high ~size =
-  let out__ = CArray.make t 1 in
   stubs_randint_low_out
-    (CArray.start out__)
     out
     (Int64.of_int low)
     (Int64.of_int high)
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
-    (List.length size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length size)
+  |> with_tensor_gc
 ;;
 
 let randint_out ~out ~high ~size =
-  let out__ = CArray.make t 1 in
   stubs_randint_out
-    (CArray.start out__)
     out
     (Int64.of_int high)
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
-    (List.length size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length size)
+  |> with_tensor_gc
 ;;
 
 let randn ~size ~options =
-  let out__ = CArray.make t 1 in
   stubs_randn
-    (CArray.start out__)
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let randn_like self =
-  let out__ = CArray.make t 1 in
-  stubs_randn_like (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
-let randn_like_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_randn_like_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let randn_like self = stubs_randn_like self |> with_tensor_gc
+let randn_like_out ~out self = stubs_randn_like_out out self |> with_tensor_gc
 
 let randn_out ~out ~size =
-  let out__ = CArray.make t 1 in
   stubs_randn_out
-    (CArray.start out__)
     out
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
-    (List.length size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let random self =
-  let out__ = CArray.make t 1 in
-  stubs_random (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length size)
+  |> with_tensor_gc
 ;;
 
-let random_ self =
-  let out__ = CArray.make t 1 in
-  stubs_random_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let random self = stubs_random self |> with_tensor_gc
+let random_ self = stubs_random_ self |> with_tensor_gc
 
 let random_from self ~from ~to_ =
-  let out__ = CArray.make t 1 in
   stubs_random_from
-    (CArray.start out__)
     self
     (Int64.of_int from)
     (match to_ with
@@ -25422,16 +16509,12 @@ let random_from self ~from ~to_ =
      | Some v -> Int64.of_int v)
     (match to_ with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let random_from_ self ~from ~to_ =
-  let out__ = CArray.make t 1 in
   stubs_random_from_
-    (CArray.start out__)
     self
     (Int64.of_int from)
     (match to_ with
@@ -25439,16 +16522,12 @@ let random_from_ self ~from ~to_ =
      | Some v -> Int64.of_int v)
     (match to_ with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let random_from_out ~out self ~from ~to_ =
-  let out__ = CArray.make t 1 in
   stubs_random_from_out
-    (CArray.start out__)
     out
     self
     (Int64.of_int from)
@@ -25457,462 +16536,220 @@ let random_from_out ~out self ~from ~to_ =
      | Some v -> Int64.of_int v)
     (match to_ with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let random_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_random_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let random_to self ~to_ =
-  let out__ = CArray.make t 1 in
-  stubs_random_to (CArray.start out__) self (Int64.of_int to_);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
-let random_to_ self ~to_ =
-  let out__ = CArray.make t 1 in
-  stubs_random_to_ (CArray.start out__) self (Int64.of_int to_);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let random_out ~out self = stubs_random_out out self |> with_tensor_gc
+let random_to self ~to_ = stubs_random_to self (Int64.of_int to_) |> with_tensor_gc
+let random_to_ self ~to_ = stubs_random_to_ self (Int64.of_int to_) |> with_tensor_gc
 
 let random_to_out ~out self ~to_ =
-  let out__ = CArray.make t 1 in
-  stubs_random_to_out (CArray.start out__) out self (Int64.of_int to_);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_random_to_out out self (Int64.of_int to_) |> with_tensor_gc
 ;;
 
 let randperm ~n ~options =
-  let out__ = CArray.make t 1 in
   stubs_randperm
-    (CArray.start out__)
     (Int64.of_int n)
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
-let randperm_out ~out ~n =
-  let out__ = CArray.make t 1 in
-  stubs_randperm_out (CArray.start out__) out (Int64.of_int n);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let randperm_out ~out ~n = stubs_randperm_out out (Int64.of_int n) |> with_tensor_gc
 
 let range ~start ~end_ ~options =
-  let out__ = CArray.make t 1 in
-  stubs_range
-    (CArray.start out__)
-    start
-    end_
-    (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let range_out ~out ~start ~end_ =
-  let out__ = CArray.make t 1 in
-  stubs_range_out (CArray.start out__) out start end_;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_range start end_ (Kind.packed_to_int (fst options)) (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
-let range_out_ ~out ~start ~end_ =
-  let out__ = CArray.make t 1 in
-  stubs_range_out_ (CArray.start out__) out start end_;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let range_out ~out ~start ~end_ = stubs_range_out out start end_ |> with_tensor_gc
+let range_out_ ~out ~start ~end_ = stubs_range_out_ out start end_ |> with_tensor_gc
 
 let range_step ~start ~end_ ~options =
-  let out__ = CArray.make t 1 in
   stubs_range_step
-    (CArray.start out__)
     start
     end_
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let ravel self =
-  let out__ = CArray.make t 1 in
-  stubs_ravel (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let real self =
-  let out__ = CArray.make t 1 in
-  stubs_real (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let reciprocal self =
-  let out__ = CArray.make t 1 in
-  stubs_reciprocal (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let reciprocal_ self =
-  let out__ = CArray.make t 1 in
-  stubs_reciprocal_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
-let reciprocal_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_reciprocal_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let ravel self = stubs_ravel self |> with_tensor_gc
+let real self = stubs_real self |> with_tensor_gc
+let reciprocal self = stubs_reciprocal self |> with_tensor_gc
+let reciprocal_ self = stubs_reciprocal_ self |> with_tensor_gc
+let reciprocal_out ~out self = stubs_reciprocal_out out self |> with_tensor_gc
 
 let reflection_pad1d self ~padding =
-  let out__ = CArray.make t 1 in
   stubs_reflection_pad1d
-    (CArray.start out__)
     self
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
-    (List.length padding);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length padding)
+  |> with_tensor_gc
 ;;
 
 let reflection_pad1d_backward ~grad_output self ~padding =
-  let out__ = CArray.make t 1 in
   stubs_reflection_pad1d_backward
-    (CArray.start out__)
     grad_output
     self
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
-    (List.length padding);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length padding)
+  |> with_tensor_gc
 ;;
 
 let reflection_pad1d_backward_grad_input ~grad_input ~grad_output self ~padding =
-  let out__ = CArray.make t 1 in
   stubs_reflection_pad1d_backward_grad_input
-    (CArray.start out__)
     grad_input
     grad_output
     self
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
-    (List.length padding);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length padding)
+  |> with_tensor_gc
 ;;
 
 let reflection_pad1d_out ~out self ~padding =
-  let out__ = CArray.make t 1 in
   stubs_reflection_pad1d_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
-    (List.length padding);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length padding)
+  |> with_tensor_gc
 ;;
 
 let reflection_pad2d self ~padding =
-  let out__ = CArray.make t 1 in
   stubs_reflection_pad2d
-    (CArray.start out__)
     self
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
-    (List.length padding);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length padding)
+  |> with_tensor_gc
 ;;
 
 let reflection_pad2d_backward ~grad_output self ~padding =
-  let out__ = CArray.make t 1 in
   stubs_reflection_pad2d_backward
-    (CArray.start out__)
     grad_output
     self
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
-    (List.length padding);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length padding)
+  |> with_tensor_gc
 ;;
 
 let reflection_pad2d_backward_grad_input ~grad_input ~grad_output self ~padding =
-  let out__ = CArray.make t 1 in
   stubs_reflection_pad2d_backward_grad_input
-    (CArray.start out__)
     grad_input
     grad_output
     self
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
-    (List.length padding);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length padding)
+  |> with_tensor_gc
 ;;
 
 let reflection_pad2d_out ~out self ~padding =
-  let out__ = CArray.make t 1 in
   stubs_reflection_pad2d_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
-    (List.length padding);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length padding)
+  |> with_tensor_gc
 ;;
 
 let reflection_pad3d self ~padding =
-  let out__ = CArray.make t 1 in
   stubs_reflection_pad3d
-    (CArray.start out__)
     self
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
-    (List.length padding);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length padding)
+  |> with_tensor_gc
 ;;
 
 let reflection_pad3d_backward ~grad_output self ~padding =
-  let out__ = CArray.make t 1 in
   stubs_reflection_pad3d_backward
-    (CArray.start out__)
     grad_output
     self
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
-    (List.length padding);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length padding)
+  |> with_tensor_gc
 ;;
 
 let reflection_pad3d_backward_grad_input ~grad_input ~grad_output self ~padding =
-  let out__ = CArray.make t 1 in
   stubs_reflection_pad3d_backward_grad_input
-    (CArray.start out__)
     grad_input
     grad_output
     self
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
-    (List.length padding);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length padding)
+  |> with_tensor_gc
 ;;
 
 let reflection_pad3d_out ~out self ~padding =
-  let out__ = CArray.make t 1 in
   stubs_reflection_pad3d_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
-    (List.length padding);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let relu self =
-  let out__ = CArray.make t 1 in
-  stubs_relu (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let relu6 self =
-  let out__ = CArray.make t 1 in
-  stubs_relu6 (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let relu6_ self =
-  let out__ = CArray.make t 1 in
-  stubs_relu6_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let relu_ self =
-  let out__ = CArray.make t 1 in
-  stubs_relu_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let relu_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_relu_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let remainder self other =
-  let out__ = CArray.make t 1 in
-  stubs_remainder (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length padding)
+  |> with_tensor_gc
 ;;
 
-let remainder_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_remainder_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let relu self = stubs_relu self |> with_tensor_gc
+let relu6 self = stubs_relu6 self |> with_tensor_gc
+let relu6_ self = stubs_relu6_ self |> with_tensor_gc
+let relu_ self = stubs_relu_ self |> with_tensor_gc
+let relu_out ~out self = stubs_relu_out out self |> with_tensor_gc
+let remainder self other = stubs_remainder self other |> with_tensor_gc
+let remainder_ self other = stubs_remainder_ self other |> with_tensor_gc
 
 let remainder_scalar_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_remainder_scalar_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_remainder_scalar_out out self other |> with_tensor_gc
 ;;
 
 let remainder_scalar_tensor self other =
-  let out__ = CArray.make t 1 in
-  stubs_remainder_scalar_tensor (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_remainder_scalar_tensor self other |> with_tensor_gc
 ;;
 
 let remainder_scalar_tensor_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_remainder_scalar_tensor_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let remainder_tensor self other =
-  let out__ = CArray.make t 1 in
-  stubs_remainder_tensor (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_remainder_scalar_tensor_out out self other |> with_tensor_gc
 ;;
 
-let remainder_tensor_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_remainder_tensor_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let remainder_tensor self other = stubs_remainder_tensor self other |> with_tensor_gc
+let remainder_tensor_ self other = stubs_remainder_tensor_ self other |> with_tensor_gc
 
 let remainder_tensor_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_remainder_tensor_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_remainder_tensor_out out self other |> with_tensor_gc
 ;;
 
 let renorm self ~p ~dim ~maxnorm =
-  let out__ = CArray.make t 1 in
-  stubs_renorm (CArray.start out__) self p (Int64.of_int dim) maxnorm;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_renorm self p (Int64.of_int dim) maxnorm |> with_tensor_gc
 ;;
 
 let renorm_ self ~p ~dim ~maxnorm =
-  let out__ = CArray.make t 1 in
-  stubs_renorm_ (CArray.start out__) self p (Int64.of_int dim) maxnorm;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_renorm_ self p (Int64.of_int dim) maxnorm |> with_tensor_gc
 ;;
 
 let renorm_out ~out self ~p ~dim ~maxnorm =
-  let out__ = CArray.make t 1 in
-  stubs_renorm_out (CArray.start out__) out self p (Int64.of_int dim) maxnorm;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_renorm_out out self p (Int64.of_int dim) maxnorm |> with_tensor_gc
 ;;
 
 let repeat self ~repeats =
-  let out__ = CArray.make t 1 in
   stubs_repeat
-    (CArray.start out__)
     self
     (List.map Int64.of_int repeats |> CArray.of_list int64_t |> CArray.start)
-    (List.length repeats);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length repeats)
+  |> with_tensor_gc
 ;;
 
 let repeat_interleave ~repeats ~output_size =
-  let out__ = CArray.make t 1 in
   stubs_repeat_interleave
-    (CArray.start out__)
     repeats
     (match output_size with
      | None -> Int64.zero
      | Some v -> Int64.of_int v)
     (match output_size with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let repeat_interleave_self_int self ~repeats ~dim ~output_size =
-  let out__ = CArray.make t 1 in
   stubs_repeat_interleave_self_int
-    (CArray.start out__)
     self
     (Int64.of_int repeats)
     (match dim with
@@ -25926,16 +16763,12 @@ let repeat_interleave_self_int self ~repeats ~dim ~output_size =
      | Some v -> Int64.of_int v)
     (match output_size with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let repeat_interleave_self_tensor self ~repeats ~dim ~output_size =
-  let out__ = CArray.make t 1 in
   stubs_repeat_interleave_self_tensor
-    (CArray.start out__)
     self
     repeats
     (match dim with
@@ -25949,16 +16782,12 @@ let repeat_interleave_self_tensor self ~repeats ~dim ~output_size =
      | Some v -> Int64.of_int v)
     (match output_size with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let repeat_interleave_tensor_out ~out ~repeats ~output_size =
-  let out__ = CArray.make t 1 in
   stubs_repeat_interleave_tensor_out
-    (CArray.start out__)
     out
     repeats
     (match output_size with
@@ -25966,310 +16795,187 @@ let repeat_interleave_tensor_out ~out ~repeats ~output_size =
      | Some v -> Int64.of_int v)
     (match output_size with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let repeat_out ~out self ~repeats =
-  let out__ = CArray.make t 1 in
   stubs_repeat_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int repeats |> CArray.of_list int64_t |> CArray.start)
-    (List.length repeats);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length repeats)
+  |> with_tensor_gc
 ;;
 
 let replication_pad1d self ~padding =
-  let out__ = CArray.make t 1 in
   stubs_replication_pad1d
-    (CArray.start out__)
     self
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
-    (List.length padding);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length padding)
+  |> with_tensor_gc
 ;;
 
 let replication_pad1d_backward ~grad_output self ~padding =
-  let out__ = CArray.make t 1 in
   stubs_replication_pad1d_backward
-    (CArray.start out__)
     grad_output
     self
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
-    (List.length padding);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length padding)
+  |> with_tensor_gc
 ;;
 
 let replication_pad1d_backward_grad_input ~grad_input ~grad_output self ~padding =
-  let out__ = CArray.make t 1 in
   stubs_replication_pad1d_backward_grad_input
-    (CArray.start out__)
     grad_input
     grad_output
     self
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
-    (List.length padding);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length padding)
+  |> with_tensor_gc
 ;;
 
 let replication_pad1d_out ~out self ~padding =
-  let out__ = CArray.make t 1 in
   stubs_replication_pad1d_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
-    (List.length padding);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length padding)
+  |> with_tensor_gc
 ;;
 
 let replication_pad2d self ~padding =
-  let out__ = CArray.make t 1 in
   stubs_replication_pad2d
-    (CArray.start out__)
     self
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
-    (List.length padding);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length padding)
+  |> with_tensor_gc
 ;;
 
 let replication_pad2d_backward ~grad_output self ~padding =
-  let out__ = CArray.make t 1 in
   stubs_replication_pad2d_backward
-    (CArray.start out__)
     grad_output
     self
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
-    (List.length padding);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length padding)
+  |> with_tensor_gc
 ;;
 
 let replication_pad2d_backward_grad_input ~grad_input ~grad_output self ~padding =
-  let out__ = CArray.make t 1 in
   stubs_replication_pad2d_backward_grad_input
-    (CArray.start out__)
     grad_input
     grad_output
     self
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
-    (List.length padding);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length padding)
+  |> with_tensor_gc
 ;;
 
 let replication_pad2d_out ~out self ~padding =
-  let out__ = CArray.make t 1 in
   stubs_replication_pad2d_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
-    (List.length padding);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length padding)
+  |> with_tensor_gc
 ;;
 
 let replication_pad3d self ~padding =
-  let out__ = CArray.make t 1 in
   stubs_replication_pad3d
-    (CArray.start out__)
     self
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
-    (List.length padding);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length padding)
+  |> with_tensor_gc
 ;;
 
 let replication_pad3d_backward ~grad_output self ~padding =
-  let out__ = CArray.make t 1 in
   stubs_replication_pad3d_backward
-    (CArray.start out__)
     grad_output
     self
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
-    (List.length padding);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length padding)
+  |> with_tensor_gc
 ;;
 
 let replication_pad3d_backward_grad_input ~grad_input ~grad_output self ~padding =
-  let out__ = CArray.make t 1 in
   stubs_replication_pad3d_backward_grad_input
-    (CArray.start out__)
     grad_input
     grad_output
     self
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
-    (List.length padding);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length padding)
+  |> with_tensor_gc
 ;;
 
 let replication_pad3d_out ~out self ~padding =
-  let out__ = CArray.make t 1 in
   stubs_replication_pad3d_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
-    (List.length padding);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length padding)
+  |> with_tensor_gc
 ;;
 
 let requires_grad_ self ~requires_grad =
-  let out__ = CArray.make t 1 in
-  stubs_requires_grad_ (CArray.start out__) self (if requires_grad then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_requires_grad_ self (if requires_grad then 1 else 0) |> with_tensor_gc
 ;;
 
 let reshape self ~shape =
-  let out__ = CArray.make t 1 in
   stubs_reshape
-    (CArray.start out__)
     self
     (List.map Int64.of_int shape |> CArray.of_list int64_t |> CArray.start)
-    (List.length shape);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length shape)
+  |> with_tensor_gc
 ;;
 
-let reshape_as self other =
-  let out__ = CArray.make t 1 in
-  stubs_reshape_as (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let reshape_as self other = stubs_reshape_as self other |> with_tensor_gc
 
 let resize self ~size =
-  let out__ = CArray.make t 1 in
   stubs_resize
-    (CArray.start out__)
     self
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
-    (List.length size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length size)
+  |> with_tensor_gc
 ;;
 
 let resize_ self ~size =
-  let out__ = CArray.make t 1 in
   stubs_resize_
-    (CArray.start out__)
     self
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
-    (List.length size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let resize_as self ~the_template =
-  let out__ = CArray.make t 1 in
-  stubs_resize_as (CArray.start out__) self the_template;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length size)
+  |> with_tensor_gc
 ;;
 
-let resize_as_ self ~the_template =
-  let out__ = CArray.make t 1 in
-  stubs_resize_as_ (CArray.start out__) self the_template;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let resize_as self ~the_template = stubs_resize_as self the_template |> with_tensor_gc
+let resize_as_ self ~the_template = stubs_resize_as_ self the_template |> with_tensor_gc
 
 let resize_as_out ~out self ~the_template =
-  let out__ = CArray.make t 1 in
-  stubs_resize_as_out (CArray.start out__) out self the_template;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_resize_as_out out self the_template |> with_tensor_gc
 ;;
 
 let resize_as_sparse self ~the_template =
-  let out__ = CArray.make t 1 in
-  stubs_resize_as_sparse (CArray.start out__) self the_template;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_resize_as_sparse self the_template |> with_tensor_gc
 ;;
 
 let resize_as_sparse_ self ~the_template =
-  let out__ = CArray.make t 1 in
-  stubs_resize_as_sparse_ (CArray.start out__) self the_template;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_resize_as_sparse_ self the_template |> with_tensor_gc
 ;;
 
 let resize_as_sparse_out ~out self ~the_template =
-  let out__ = CArray.make t 1 in
-  stubs_resize_as_sparse_out (CArray.start out__) out self the_template;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_resize_as_sparse_out out self the_template |> with_tensor_gc
 ;;
 
 let resize_out ~out self ~size =
-  let out__ = CArray.make t 1 in
   stubs_resize_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
-    (List.length size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let resolve_conj self =
-  let out__ = CArray.make t 1 in
-  stubs_resolve_conj (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let resolve_neg self =
-  let out__ = CArray.make t 1 in
-  stubs_resolve_neg (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length size)
+  |> with_tensor_gc
 ;;
 
+let resolve_conj self = stubs_resolve_conj self |> with_tensor_gc
+let resolve_neg self = stubs_resolve_neg self |> with_tensor_gc
 let retains_grad self = stubs_retains_grad self
 
 let rnn_relu
@@ -26283,12 +16989,12 @@ let rnn_relu
   ~bidirectional
   ~batch_first
   =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_rnn_relu
     (CArray.start out__)
     input
     hx
-    (CArray.of_list t params |> CArray.start)
+    (CArray.of_list gc_tensor params |> CArray.start)
     (List.length params)
     (if has_biases then 1 else 0)
     (Int64.of_int num_layers)
@@ -26296,30 +17002,24 @@ let rnn_relu
     (if train then 1 else 0)
     (if bidirectional then 1 else 0)
     (if batch_first then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let rnn_relu_cell input ~hx ~w_ih ~w_hh ~b_ih ~b_hh =
-  let out__ = CArray.make t 1 in
   stubs_rnn_relu_cell
-    (CArray.start out__)
     input
     hx
     w_ih
     w_hh
     (match b_ih with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match b_hh with
      | Some v -> v
-     | None -> null);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+  |> with_tensor_gc
 ;;
 
 let rnn_relu_data
@@ -26333,23 +17033,21 @@ let rnn_relu_data
   ~train
   ~bidirectional
   =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_rnn_relu_data
     (CArray.start out__)
     data
     batch_sizes
     hx
-    (CArray.of_list t params |> CArray.start)
+    (CArray.of_list gc_tensor params |> CArray.start)
     (List.length params)
     (if has_biases then 1 else 0)
     (Int64.of_int num_layers)
     dropout
     (if train then 1 else 0)
     (if bidirectional then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
@@ -26364,12 +17062,12 @@ let rnn_tanh
   ~bidirectional
   ~batch_first
   =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_rnn_tanh
     (CArray.start out__)
     input
     hx
-    (CArray.of_list t params |> CArray.start)
+    (CArray.of_list gc_tensor params |> CArray.start)
     (List.length params)
     (if has_biases then 1 else 0)
     (Int64.of_int num_layers)
@@ -26377,30 +17075,24 @@ let rnn_tanh
     (if train then 1 else 0)
     (if bidirectional then 1 else 0)
     (if batch_first then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let rnn_tanh_cell input ~hx ~w_ih ~w_hh ~b_ih ~b_hh =
-  let out__ = CArray.make t 1 in
   stubs_rnn_tanh_cell
-    (CArray.start out__)
     input
     hx
     w_ih
     w_hh
     (match b_ih with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match b_hh with
      | Some v -> v
-     | None -> null);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+  |> with_tensor_gc
 ;;
 
 let rnn_tanh_data
@@ -26414,207 +17106,109 @@ let rnn_tanh_data
   ~train
   ~bidirectional
   =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_rnn_tanh_data
     (CArray.start out__)
     data
     batch_sizes
     hx
-    (CArray.of_list t params |> CArray.start)
+    (CArray.of_list gc_tensor params |> CArray.start)
     (List.length params)
     (if has_biases then 1 else 0)
     (Int64.of_int num_layers)
     dropout
     (if train then 1 else 0)
     (if bidirectional then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let roll self ~shifts ~dims =
-  let out__ = CArray.make t 1 in
   stubs_roll
-    (CArray.start out__)
     self
     (List.map Int64.of_int shifts |> CArray.of_list int64_t |> CArray.start)
     (List.length shifts)
     (List.map Int64.of_int dims |> CArray.of_list int64_t |> CArray.start)
-    (List.length dims);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length dims)
+  |> with_tensor_gc
 ;;
 
 let roll_out ~out self ~shifts ~dims =
-  let out__ = CArray.make t 1 in
   stubs_roll_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int shifts |> CArray.of_list int64_t |> CArray.start)
     (List.length shifts)
     (List.map Int64.of_int dims |> CArray.of_list int64_t |> CArray.start)
-    (List.length dims);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length dims)
+  |> with_tensor_gc
 ;;
 
 let rot90 self ~k ~dims =
-  let out__ = CArray.make t 1 in
   stubs_rot90
-    (CArray.start out__)
     self
     (Int64.of_int k)
     (List.map Int64.of_int dims |> CArray.of_list int64_t |> CArray.start)
-    (List.length dims);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length dims)
+  |> with_tensor_gc
 ;;
 
 let rot90_out ~out self ~k ~dims =
-  let out__ = CArray.make t 1 in
   stubs_rot90_out
-    (CArray.start out__)
     out
     self
     (Int64.of_int k)
     (List.map Int64.of_int dims |> CArray.of_list int64_t |> CArray.start)
-    (List.length dims);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let round self =
-  let out__ = CArray.make t 1 in
-  stubs_round (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length dims)
+  |> with_tensor_gc
 ;;
 
-let round_ self =
-  let out__ = CArray.make t 1 in
-  stubs_round_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let round self = stubs_round self |> with_tensor_gc
+let round_ self = stubs_round_ self |> with_tensor_gc
 
 let round_decimals self ~decimals =
-  let out__ = CArray.make t 1 in
-  stubs_round_decimals (CArray.start out__) self (Int64.of_int decimals);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_round_decimals self (Int64.of_int decimals) |> with_tensor_gc
 ;;
 
 let round_decimals_ self ~decimals =
-  let out__ = CArray.make t 1 in
-  stubs_round_decimals_ (CArray.start out__) self (Int64.of_int decimals);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_round_decimals_ self (Int64.of_int decimals) |> with_tensor_gc
 ;;
 
 let round_decimals_out ~out self ~decimals =
-  let out__ = CArray.make t 1 in
-  stubs_round_decimals_out (CArray.start out__) out self (Int64.of_int decimals);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let round_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_round_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let row_indices self =
-  let out__ = CArray.make t 1 in
-  stubs_row_indices (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let row_indices_copy self =
-  let out__ = CArray.make t 1 in
-  stubs_row_indices_copy (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_round_decimals_out out self (Int64.of_int decimals) |> with_tensor_gc
 ;;
 
-let row_indices_copy_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_row_indices_copy_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let round_out ~out self = stubs_round_out out self |> with_tensor_gc
+let row_indices self = stubs_row_indices self |> with_tensor_gc
+let row_indices_copy self = stubs_row_indices_copy self |> with_tensor_gc
+let row_indices_copy_out ~out self = stubs_row_indices_copy_out out self |> with_tensor_gc
 
 let row_stack tensors =
-  let out__ = CArray.make t 1 in
-  stubs_row_stack
-    (CArray.start out__)
-    (CArray.of_list t tensors |> CArray.start)
-    (List.length tensors);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_row_stack (CArray.of_list gc_tensor tensors |> CArray.start) (List.length tensors)
+  |> with_tensor_gc
 ;;
 
 let row_stack_out ~out tensors =
-  let out__ = CArray.make t 1 in
   stubs_row_stack_out
-    (CArray.start out__)
     out
-    (CArray.of_list t tensors |> CArray.start)
-    (List.length tensors);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (CArray.of_list gc_tensor tensors |> CArray.start)
+    (List.length tensors)
+  |> with_tensor_gc
 ;;
 
-let rrelu self ~training =
-  let out__ = CArray.make t 1 in
-  stubs_rrelu (CArray.start out__) self (if training then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let rrelu self ~training = stubs_rrelu self (if training then 1 else 0) |> with_tensor_gc
 
 let rrelu_ self ~training =
-  let out__ = CArray.make t 1 in
-  stubs_rrelu_ (CArray.start out__) self (if training then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_rrelu_ self (if training then 1 else 0) |> with_tensor_gc
 ;;
 
 let rrelu_with_noise self ~noise ~training =
-  let out__ = CArray.make t 1 in
-  stubs_rrelu_with_noise (CArray.start out__) self noise (if training then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_rrelu_with_noise self noise (if training then 1 else 0) |> with_tensor_gc
 ;;
 
 let rrelu_with_noise_ self ~noise ~training =
-  let out__ = CArray.make t 1 in
-  stubs_rrelu_with_noise_ (CArray.start out__) self noise (if training then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_rrelu_with_noise_ self noise (if training then 1 else 0) |> with_tensor_gc
 ;;
 
 let rrelu_with_noise_backward
@@ -26626,19 +17220,15 @@ let rrelu_with_noise_backward
   ~training
   ~self_is_result
   =
-  let out__ = CArray.make t 1 in
   stubs_rrelu_with_noise_backward
-    (CArray.start out__)
     grad_output
     self
     noise
     lower
     upper
     (if training then 1 else 0)
-    (if self_is_result then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if self_is_result then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let rrelu_with_noise_backward_out
@@ -26651,9 +17241,7 @@ let rrelu_with_noise_backward_out
   ~training
   ~self_is_result
   =
-  let out__ = CArray.make t 1 in
   stubs_rrelu_with_noise_backward_out
-    (CArray.start out__)
     out
     grad_output
     self
@@ -26661,100 +17249,34 @@ let rrelu_with_noise_backward_out
     lower
     upper
     (if training then 1 else 0)
-    (if self_is_result then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if self_is_result then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let rrelu_with_noise_out ~out self ~noise ~training =
-  let out__ = CArray.make t 1 in
-  stubs_rrelu_with_noise_out
-    (CArray.start out__)
-    out
-    self
-    noise
-    (if training then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let rsqrt self =
-  let out__ = CArray.make t 1 in
-  stubs_rsqrt (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let rsqrt_ self =
-  let out__ = CArray.make t 1 in
-  stubs_rsqrt_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let rsqrt_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_rsqrt_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let rsub self other =
-  let out__ = CArray.make t 1 in
-  stubs_rsub (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_rrelu_with_noise_out out self noise (if training then 1 else 0) |> with_tensor_gc
 ;;
 
-let rsub_scalar self other =
-  let out__ = CArray.make t 1 in
-  stubs_rsub_scalar (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let rsqrt self = stubs_rsqrt self |> with_tensor_gc
+let rsqrt_ self = stubs_rsqrt_ self |> with_tensor_gc
+let rsqrt_out ~out self = stubs_rsqrt_out out self |> with_tensor_gc
+let rsub self other = stubs_rsub self other |> with_tensor_gc
+let rsub_scalar self other = stubs_rsub_scalar self other |> with_tensor_gc
 
 let rsub_scalar_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_rsub_scalar_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_rsub_scalar_out out self other |> with_tensor_gc
 ;;
 
 let rsub_tensor_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_rsub_tensor_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_rsub_tensor_out out self other |> with_tensor_gc
 ;;
 
 let scalar_tensor ~s ~options =
-  let out__ = CArray.make t 1 in
-  stubs_scalar_tensor
-    (CArray.start out__)
-    s
-    (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_scalar_tensor s (Kind.packed_to_int (fst options)) (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
-let scalar_tensor_out ~out ~s =
-  let out__ = CArray.make t 1 in
-  stubs_scalar_tensor_out (CArray.start out__) out s;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let scalar_tensor_out ~out ~s = stubs_scalar_tensor_out out s |> with_tensor_gc
 
 let scaled_dot_product_attention
   ~query
@@ -26765,176 +17287,85 @@ let scaled_dot_product_attention
   ~is_causal
   ~scale
   =
-  let out__ = CArray.make t 1 in
   stubs_scaled_dot_product_attention
-    (CArray.start out__)
     query
     key
     value
     (match attn_mask with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     dropout_p
     (if is_causal then 1 else 0)
     (Option.value scale ~default:0.0)
     (match scale with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let scatter self ~dim ~index ~src =
-  let out__ = CArray.make t 1 in
-  stubs_scatter (CArray.start out__) self (Int64.of_int dim) index src;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_scatter self (Int64.of_int dim) index src |> with_tensor_gc
 ;;
 
 let scatter_ self ~dim ~index ~src =
-  let out__ = CArray.make t 1 in
-  stubs_scatter_ (CArray.start out__) self (Int64.of_int dim) index src;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_scatter_ self (Int64.of_int dim) index src |> with_tensor_gc
 ;;
 
 let scatter_add self ~dim ~index ~src =
-  let out__ = CArray.make t 1 in
-  stubs_scatter_add (CArray.start out__) self (Int64.of_int dim) index src;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_scatter_add self (Int64.of_int dim) index src |> with_tensor_gc
 ;;
 
 let scatter_add_ self ~dim ~index ~src =
-  let out__ = CArray.make t 1 in
-  stubs_scatter_add_ (CArray.start out__) self (Int64.of_int dim) index src;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_scatter_add_ self (Int64.of_int dim) index src |> with_tensor_gc
 ;;
 
 let scatter_add_out ~out self ~dim ~index ~src =
-  let out__ = CArray.make t 1 in
-  stubs_scatter_add_out (CArray.start out__) out self (Int64.of_int dim) index src;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_scatter_add_out out self (Int64.of_int dim) index src |> with_tensor_gc
 ;;
 
 let scatter_reduce self ~dim ~index ~src ~reduce =
-  let out__ = CArray.make t 1 in
-  stubs_scatter_reduce (CArray.start out__) self (Int64.of_int dim) index src reduce;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_scatter_reduce self (Int64.of_int dim) index src reduce |> with_tensor_gc
 ;;
 
 let scatter_reduce_ self ~dim ~index ~src ~reduce =
-  let out__ = CArray.make t 1 in
-  stubs_scatter_reduce_ (CArray.start out__) self (Int64.of_int dim) index src reduce;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_scatter_reduce_ self (Int64.of_int dim) index src reduce |> with_tensor_gc
 ;;
 
 let scatter_reduce_out ~out self ~dim ~index ~src ~reduce =
-  let out__ = CArray.make t 1 in
-  stubs_scatter_reduce_out
-    (CArray.start out__)
-    out
-    self
-    (Int64.of_int dim)
-    index
-    src
-    reduce;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_scatter_reduce_out out self (Int64.of_int dim) index src reduce |> with_tensor_gc
 ;;
 
 let scatter_src_out ~out self ~dim ~index ~src =
-  let out__ = CArray.make t 1 in
-  stubs_scatter_src_out (CArray.start out__) out self (Int64.of_int dim) index src;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_scatter_src_out out self (Int64.of_int dim) index src |> with_tensor_gc
 ;;
 
 let scatter_value self ~dim ~index ~value =
-  let out__ = CArray.make t 1 in
-  stubs_scatter_value (CArray.start out__) self (Int64.of_int dim) index value;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_scatter_value self (Int64.of_int dim) index value |> with_tensor_gc
 ;;
 
 let scatter_value_ self ~dim ~index ~value =
-  let out__ = CArray.make t 1 in
-  stubs_scatter_value_ (CArray.start out__) self (Int64.of_int dim) index value;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_scatter_value_ self (Int64.of_int dim) index value |> with_tensor_gc
 ;;
 
 let scatter_value_out ~out self ~dim ~index ~value =
-  let out__ = CArray.make t 1 in
-  stubs_scatter_value_out (CArray.start out__) out self (Int64.of_int dim) index value;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_scatter_value_out out self (Int64.of_int dim) index value |> with_tensor_gc
 ;;
 
 let scatter_value_reduce self ~dim ~index ~value ~reduce =
-  let out__ = CArray.make t 1 in
-  stubs_scatter_value_reduce
-    (CArray.start out__)
-    self
-    (Int64.of_int dim)
-    index
-    value
-    reduce;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_scatter_value_reduce self (Int64.of_int dim) index value reduce |> with_tensor_gc
 ;;
 
 let scatter_value_reduce_ self ~dim ~index ~value ~reduce =
-  let out__ = CArray.make t 1 in
-  stubs_scatter_value_reduce_
-    (CArray.start out__)
-    self
-    (Int64.of_int dim)
-    index
-    value
-    reduce;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_scatter_value_reduce_ self (Int64.of_int dim) index value reduce |> with_tensor_gc
 ;;
 
 let scatter_value_reduce_out ~out self ~dim ~index ~value ~reduce =
-  let out__ = CArray.make t 1 in
-  stubs_scatter_value_reduce_out
-    (CArray.start out__)
-    out
-    self
-    (Int64.of_int dim)
-    index
-    value
-    reduce;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_scatter_value_reduce_out out self (Int64.of_int dim) index value reduce
+  |> with_tensor_gc
 ;;
 
 let searchsorted ~sorted_sequence self ~out_int32 ~right ~side ~sorter =
-  let out__ = CArray.make t 1 in
   stubs_searchsorted
-    (CArray.start out__)
     sorted_sequence
     self
     (if out_int32 then 1 else 0)
@@ -26942,16 +17373,12 @@ let searchsorted ~sorted_sequence self ~out_int32 ~right ~side ~sorter =
     side
     (match sorter with
      | Some v -> v
-     | None -> null);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+  |> with_tensor_gc
 ;;
 
 let searchsorted_scalar ~sorted_sequence self ~out_int32 ~right ~side ~sorter =
-  let out__ = CArray.make t 1 in
   stubs_searchsorted_scalar
-    (CArray.start out__)
     sorted_sequence
     self
     (if out_int32 then 1 else 0)
@@ -26959,16 +17386,12 @@ let searchsorted_scalar ~sorted_sequence self ~out_int32 ~right ~side ~sorter =
     side
     (match sorter with
      | Some v -> v
-     | None -> null);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+  |> with_tensor_gc
 ;;
 
 let searchsorted_scalar_out ~out ~sorted_sequence self ~out_int32 ~right ~side ~sorter =
-  let out__ = CArray.make t 1 in
   stubs_searchsorted_scalar_out
-    (CArray.start out__)
     out
     sorted_sequence
     self
@@ -26977,16 +17400,12 @@ let searchsorted_scalar_out ~out ~sorted_sequence self ~out_int32 ~right ~side ~
     side
     (match sorter with
      | Some v -> v
-     | None -> null);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+  |> with_tensor_gc
 ;;
 
 let searchsorted_tensor_out ~out ~sorted_sequence self ~out_int32 ~right ~side ~sorter =
-  let out__ = CArray.make t 1 in
   stubs_searchsorted_tensor_out
-    (CArray.start out__)
     out
     sorted_sequence
     self
@@ -26995,33 +17414,27 @@ let searchsorted_tensor_out ~out ~sorted_sequence self ~out_int32 ~right ~side ~
     side
     (match sorter with
      | Some v -> v
-     | None -> null);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> none_gc_tensor)
+  |> with_tensor_gc
 ;;
 
 let segment_reduce ~data ~reduce ~lengths ~indices ~offsets ~axis ~unsafe ~initial =
-  let out__ = CArray.make t 1 in
   stubs_segment_reduce
-    (CArray.start out__)
     data
     reduce
     (match lengths with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match indices with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match offsets with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (Int64.of_int axis)
     (if unsafe then 1 else 0)
-    initial;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    initial
+  |> with_tensor_gc
 ;;
 
 let segment_reduce_out
@@ -27035,424 +17448,145 @@ let segment_reduce_out
   ~unsafe
   ~initial
   =
-  let out__ = CArray.make t 1 in
   stubs_segment_reduce_out
-    (CArray.start out__)
     out
     data
     reduce
     (match lengths with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match indices with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (match offsets with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (Int64.of_int axis)
     (if unsafe then 1 else 0)
-    initial;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    initial
+  |> with_tensor_gc
 ;;
 
 let select self ~dim ~index =
-  let out__ = CArray.make t 1 in
-  stubs_select (CArray.start out__) self (Int64.of_int dim) (Int64.of_int index);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_select self (Int64.of_int dim) (Int64.of_int index) |> with_tensor_gc
 ;;
 
 let select_backward ~grad_output ~input_sizes ~dim ~index =
-  let out__ = CArray.make t 1 in
   stubs_select_backward
-    (CArray.start out__)
     grad_output
     (List.map Int64.of_int input_sizes |> CArray.of_list int64_t |> CArray.start)
     (List.length input_sizes)
     (Int64.of_int dim)
-    (Int64.of_int index);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int index)
+  |> with_tensor_gc
 ;;
 
 let select_backward_out ~out ~grad_output ~input_sizes ~dim ~index =
-  let out__ = CArray.make t 1 in
   stubs_select_backward_out
-    (CArray.start out__)
     out
     grad_output
     (List.map Int64.of_int input_sizes |> CArray.of_list int64_t |> CArray.start)
     (List.length input_sizes)
     (Int64.of_int dim)
-    (Int64.of_int index);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int index)
+  |> with_tensor_gc
 ;;
 
 let select_copy self ~dim ~index =
-  let out__ = CArray.make t 1 in
-  stubs_select_copy (CArray.start out__) self (Int64.of_int dim) (Int64.of_int index);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_select_copy self (Int64.of_int dim) (Int64.of_int index) |> with_tensor_gc
 ;;
 
 let select_copy_int_out ~out self ~dim ~index =
-  let out__ = CArray.make t 1 in
-  stubs_select_copy_int_out
-    (CArray.start out__)
-    out
-    self
-    (Int64.of_int dim)
-    (Int64.of_int index);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_select_copy_int_out out self (Int64.of_int dim) (Int64.of_int index)
+  |> with_tensor_gc
 ;;
 
 let select_scatter self ~src ~dim ~index =
-  let out__ = CArray.make t 1 in
-  stubs_select_scatter
-    (CArray.start out__)
-    self
-    src
-    (Int64.of_int dim)
-    (Int64.of_int index);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_select_scatter self src (Int64.of_int dim) (Int64.of_int index) |> with_tensor_gc
 ;;
 
 let select_scatter_out ~out self ~src ~dim ~index =
-  let out__ = CArray.make t 1 in
-  stubs_select_scatter_out
-    (CArray.start out__)
-    out
-    self
-    src
-    (Int64.of_int dim)
-    (Int64.of_int index);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let selu self =
-  let out__ = CArray.make t 1 in
-  stubs_selu (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_select_scatter_out out self src (Int64.of_int dim) (Int64.of_int index)
+  |> with_tensor_gc
 ;;
 
-let selu_ self =
-  let out__ = CArray.make t 1 in
-  stubs_selu_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let set self =
-  let out__ = CArray.make t 1 in
-  stubs_set (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let set_ self =
-  let out__ = CArray.make t 1 in
-  stubs_set_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let set_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_set_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let selu self = stubs_selu self |> with_tensor_gc
+let selu_ self = stubs_selu_ self |> with_tensor_gc
+let set self = stubs_set self |> with_tensor_gc
+let set_ self = stubs_set_ self |> with_tensor_gc
+let set_out ~out self = stubs_set_out out self |> with_tensor_gc
 
 let set_requires_grad self ~r =
-  let out__ = CArray.make t 1 in
-  stubs_set_requires_grad (CArray.start out__) self (if r then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_set_requires_grad self (if r then 1 else 0) |> with_tensor_gc
 ;;
 
-let set_source_tensor self ~source =
-  let out__ = CArray.make t 1 in
-  stubs_set_source_tensor (CArray.start out__) self source;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let set_source_tensor self ~source = stubs_set_source_tensor self source |> with_tensor_gc
 
 let set_source_tensor_ self ~source =
-  let out__ = CArray.make t 1 in
-  stubs_set_source_tensor_ (CArray.start out__) self source;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_set_source_tensor_ self source |> with_tensor_gc
 ;;
 
 let set_source_tensor_out ~out self ~source =
-  let out__ = CArray.make t 1 in
-  stubs_set_source_tensor_out (CArray.start out__) out self source;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_set_source_tensor_out out self source |> with_tensor_gc
 ;;
 
 let set_source_tensor_storage_offset_ self ~source ~storage_offset ~size ~stride =
-  let out__ = CArray.make t 1 in
   stubs_set_source_tensor_storage_offset_
-    (CArray.start out__)
     self
     source
     (Int64.of_int storage_offset)
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
-    (List.length stride);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let sgn self =
-  let out__ = CArray.make t 1 in
-  stubs_sgn (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let sgn_ self =
-  let out__ = CArray.make t 1 in
-  stubs_sgn_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let sgn_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_sgn_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let sigmoid self =
-  let out__ = CArray.make t 1 in
-  stubs_sigmoid (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length stride)
+  |> with_tensor_gc
 ;;
 
-let sigmoid_ self =
-  let out__ = CArray.make t 1 in
-  stubs_sigmoid_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let sgn self = stubs_sgn self |> with_tensor_gc
+let sgn_ self = stubs_sgn_ self |> with_tensor_gc
+let sgn_out ~out self = stubs_sgn_out out self |> with_tensor_gc
+let sigmoid self = stubs_sigmoid self |> with_tensor_gc
+let sigmoid_ self = stubs_sigmoid_ self |> with_tensor_gc
 
 let sigmoid_backward ~grad_output ~output =
-  let out__ = CArray.make t 1 in
-  stubs_sigmoid_backward (CArray.start out__) grad_output output;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_sigmoid_backward grad_output output |> with_tensor_gc
 ;;
 
 let sigmoid_backward_grad_input ~grad_input ~grad_output ~output =
-  let out__ = CArray.make t 1 in
-  stubs_sigmoid_backward_grad_input (CArray.start out__) grad_input grad_output output;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let sigmoid_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_sigmoid_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let sign self =
-  let out__ = CArray.make t 1 in
-  stubs_sign (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let sign_ self =
-  let out__ = CArray.make t 1 in
-  stubs_sign_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let sign_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_sign_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let signbit self =
-  let out__ = CArray.make t 1 in
-  stubs_signbit (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let signbit_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_signbit_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let silu self =
-  let out__ = CArray.make t 1 in
-  stubs_silu (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_sigmoid_backward_grad_input grad_input grad_output output |> with_tensor_gc
 ;;
 
-let silu_ self =
-  let out__ = CArray.make t 1 in
-  stubs_silu_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let sigmoid_out ~out self = stubs_sigmoid_out out self |> with_tensor_gc
+let sign self = stubs_sign self |> with_tensor_gc
+let sign_ self = stubs_sign_ self |> with_tensor_gc
+let sign_out ~out self = stubs_sign_out out self |> with_tensor_gc
+let signbit self = stubs_signbit self |> with_tensor_gc
+let signbit_out ~out self = stubs_signbit_out out self |> with_tensor_gc
+let silu self = stubs_silu self |> with_tensor_gc
+let silu_ self = stubs_silu_ self |> with_tensor_gc
 
 let silu_backward ~grad_output self =
-  let out__ = CArray.make t 1 in
-  stubs_silu_backward (CArray.start out__) grad_output self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_silu_backward grad_output self |> with_tensor_gc
 ;;
 
 let silu_backward_grad_input ~grad_input ~grad_output self =
-  let out__ = CArray.make t 1 in
-  stubs_silu_backward_grad_input (CArray.start out__) grad_input grad_output self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let silu_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_silu_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let sin self =
-  let out__ = CArray.make t 1 in
-  stubs_sin (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let sin_ self =
-  let out__ = CArray.make t 1 in
-  stubs_sin_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let sin_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_sin_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let sinc self =
-  let out__ = CArray.make t 1 in
-  stubs_sinc (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let sinc_ self =
-  let out__ = CArray.make t 1 in
-  stubs_sinc_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let sinc_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_sinc_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let sinh self =
-  let out__ = CArray.make t 1 in
-  stubs_sinh (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let sinh_ self =
-  let out__ = CArray.make t 1 in
-  stubs_sinh_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let sinh_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_sinh_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
+  stubs_silu_backward_grad_input grad_input grad_output self |> with_tensor_gc
+;;
+
+let silu_out ~out self = stubs_silu_out out self |> with_tensor_gc
+let sin self = stubs_sin self |> with_tensor_gc
+let sin_ self = stubs_sin_ self |> with_tensor_gc
+let sin_out ~out self = stubs_sin_out out self |> with_tensor_gc
+let sinc self = stubs_sinc self |> with_tensor_gc
+let sinc_ self = stubs_sinc_ self |> with_tensor_gc
+let sinc_out ~out self = stubs_sinc_out out self |> with_tensor_gc
+let sinh self = stubs_sinh self |> with_tensor_gc
+let sinh_ self = stubs_sinh_ self |> with_tensor_gc
+let sinh_out ~out self = stubs_sinh_out out self |> with_tensor_gc
 let size self ~dim = stubs_size self (Int64.of_int dim)
 
 let slice self ~dim ~start ~end_ ~step =
-  let out__ = CArray.make t 1 in
   stubs_slice
-    (CArray.start out__)
     self
     (Int64.of_int dim)
     (match start with
@@ -27467,32 +17601,24 @@ let slice self ~dim ~start ~end_ ~step =
     (match end_ with
      | Some _ -> 0
      | None -> 1)
-    (Int64.of_int step);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int step)
+  |> with_tensor_gc
 ;;
 
 let slice_backward ~grad_output ~input_sizes ~dim ~start ~end_ ~step =
-  let out__ = CArray.make t 1 in
   stubs_slice_backward
-    (CArray.start out__)
     grad_output
     (List.map Int64.of_int input_sizes |> CArray.of_list int64_t |> CArray.start)
     (List.length input_sizes)
     (Int64.of_int dim)
     (Int64.of_int start)
     (Int64.of_int end_)
-    (Int64.of_int step);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int step)
+  |> with_tensor_gc
 ;;
 
 let slice_backward_out ~out ~grad_output ~input_sizes ~dim ~start ~end_ ~step =
-  let out__ = CArray.make t 1 in
   stubs_slice_backward_out
-    (CArray.start out__)
     out
     grad_output
     (List.map Int64.of_int input_sizes |> CArray.of_list int64_t |> CArray.start)
@@ -27500,16 +17626,12 @@ let slice_backward_out ~out ~grad_output ~input_sizes ~dim ~start ~end_ ~step =
     (Int64.of_int dim)
     (Int64.of_int start)
     (Int64.of_int end_)
-    (Int64.of_int step);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int step)
+  |> with_tensor_gc
 ;;
 
 let slice_copy self ~dim ~start ~end_ ~step =
-  let out__ = CArray.make t 1 in
   stubs_slice_copy
-    (CArray.start out__)
     self
     (Int64.of_int dim)
     (match start with
@@ -27524,16 +17646,12 @@ let slice_copy self ~dim ~start ~end_ ~step =
     (match end_ with
      | Some _ -> 0
      | None -> 1)
-    (Int64.of_int step);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int step)
+  |> with_tensor_gc
 ;;
 
 let slice_copy_tensor_out ~out self ~dim ~start ~end_ ~step =
-  let out__ = CArray.make t 1 in
   stubs_slice_copy_tensor_out
-    (CArray.start out__)
     out
     self
     (Int64.of_int dim)
@@ -27549,16 +17667,12 @@ let slice_copy_tensor_out ~out self ~dim ~start ~end_ ~step =
     (match end_ with
      | Some _ -> 0
      | None -> 1)
-    (Int64.of_int step);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int step)
+  |> with_tensor_gc
 ;;
 
 let slice_scatter self ~src ~dim ~start ~end_ ~step =
-  let out__ = CArray.make t 1 in
   stubs_slice_scatter
-    (CArray.start out__)
     self
     src
     (Int64.of_int dim)
@@ -27574,16 +17688,12 @@ let slice_scatter self ~src ~dim ~start ~end_ ~step =
     (match end_ with
      | Some _ -> 0
      | None -> 1)
-    (Int64.of_int step);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int step)
+  |> with_tensor_gc
 ;;
 
 let slice_scatter_out ~out self ~src ~dim ~start ~end_ ~step =
-  let out__ = CArray.make t 1 in
   stubs_slice_scatter_out
-    (CArray.start out__)
     out
     self
     src
@@ -27600,56 +17710,44 @@ let slice_scatter_out ~out self ~src ~dim ~start ~end_ ~step =
     (match end_ with
      | Some _ -> 0
      | None -> 1)
-    (Int64.of_int step);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int step)
+  |> with_tensor_gc
 ;;
 
 let slogdet self =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_slogdet (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let slogdet_out ~sign ~logabsdet self =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_slogdet_out (CArray.start out__) sign logabsdet self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let slow_conv3d self ~weight ~kernel_size ~bias ~stride ~padding =
-  let out__ = CArray.make t 1 in
   stubs_slow_conv3d
-    (CArray.start out__)
     self
     weight
     (List.map Int64.of_int kernel_size |> CArray.of_list int64_t |> CArray.start)
     (List.length kernel_size)
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
     (List.length stride)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
-    (List.length padding);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length padding)
+  |> with_tensor_gc
 ;;
 
 let slow_conv3d_out ~out self ~weight ~kernel_size ~bias ~stride ~padding =
-  let out__ = CArray.make t 1 in
   stubs_slow_conv3d_out
-    (CArray.start out__)
     out
     self
     weight
@@ -27657,36 +17755,30 @@ let slow_conv3d_out ~out self ~weight ~kernel_size ~bias ~stride ~padding =
     (List.length kernel_size)
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
     (List.length stride)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
-    (List.length padding);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length padding)
+  |> with_tensor_gc
 ;;
 
 let slow_conv_dilated2d self ~weight ~kernel_size ~bias ~stride ~padding ~dilation =
-  let out__ = CArray.make t 1 in
   stubs_slow_conv_dilated2d
-    (CArray.start out__)
     self
     weight
     (List.map Int64.of_int kernel_size |> CArray.of_list int64_t |> CArray.start)
     (List.length kernel_size)
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
     (List.length stride)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
     (List.length padding)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
-    (List.length dilation);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length dilation)
+  |> with_tensor_gc
 ;;
 
 let slow_conv_dilated2d_out
@@ -27699,9 +17791,7 @@ let slow_conv_dilated2d_out
   ~padding
   ~dilation
   =
-  let out__ = CArray.make t 1 in
   stubs_slow_conv_dilated2d_out
-    (CArray.start out__)
     out
     self
     weight
@@ -27709,38 +17799,32 @@ let slow_conv_dilated2d_out
     (List.length kernel_size)
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
     (List.length stride)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
     (List.length padding)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
-    (List.length dilation);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length dilation)
+  |> with_tensor_gc
 ;;
 
 let slow_conv_dilated3d self ~weight ~kernel_size ~bias ~stride ~padding ~dilation =
-  let out__ = CArray.make t 1 in
   stubs_slow_conv_dilated3d
-    (CArray.start out__)
     self
     weight
     (List.map Int64.of_int kernel_size |> CArray.of_list int64_t |> CArray.start)
     (List.length kernel_size)
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
     (List.length stride)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
     (List.length padding)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
-    (List.length dilation);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length dilation)
+  |> with_tensor_gc
 ;;
 
 let slow_conv_dilated3d_out
@@ -27753,9 +17837,7 @@ let slow_conv_dilated3d_out
   ~padding
   ~dilation
   =
-  let out__ = CArray.make t 1 in
   stubs_slow_conv_dilated3d_out
-    (CArray.start out__)
     out
     self
     weight
@@ -27763,16 +17845,14 @@ let slow_conv_dilated3d_out
     (List.length kernel_size)
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
     (List.length stride)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
     (List.length padding)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
-    (List.length dilation);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length dilation)
+  |> with_tensor_gc
 ;;
 
 let slow_conv_transpose2d
@@ -27785,16 +17865,14 @@ let slow_conv_transpose2d
   ~output_padding
   ~dilation
   =
-  let out__ = CArray.make t 1 in
   stubs_slow_conv_transpose2d
-    (CArray.start out__)
     self
     weight
     (List.map Int64.of_int kernel_size |> CArray.of_list int64_t |> CArray.start)
     (List.length kernel_size)
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
     (List.length stride)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
@@ -27802,10 +17880,8 @@ let slow_conv_transpose2d
     (List.map Int64.of_int output_padding |> CArray.of_list int64_t |> CArray.start)
     (List.length output_padding)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
-    (List.length dilation);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length dilation)
+  |> with_tensor_gc
 ;;
 
 let slow_conv_transpose2d_out
@@ -27819,9 +17895,7 @@ let slow_conv_transpose2d_out
   ~output_padding
   ~dilation
   =
-  let out__ = CArray.make t 1 in
   stubs_slow_conv_transpose2d_out
-    (CArray.start out__)
     out
     self
     weight
@@ -27829,7 +17903,7 @@ let slow_conv_transpose2d_out
     (List.length kernel_size)
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
     (List.length stride)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
@@ -27837,10 +17911,8 @@ let slow_conv_transpose2d_out
     (List.map Int64.of_int output_padding |> CArray.of_list int64_t |> CArray.start)
     (List.length output_padding)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
-    (List.length dilation);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length dilation)
+  |> with_tensor_gc
 ;;
 
 let slow_conv_transpose3d
@@ -27853,16 +17925,14 @@ let slow_conv_transpose3d
   ~output_padding
   ~dilation
   =
-  let out__ = CArray.make t 1 in
   stubs_slow_conv_transpose3d
-    (CArray.start out__)
     self
     weight
     (List.map Int64.of_int kernel_size |> CArray.of_list int64_t |> CArray.start)
     (List.length kernel_size)
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
     (List.length stride)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
@@ -27870,10 +17940,8 @@ let slow_conv_transpose3d
     (List.map Int64.of_int output_padding |> CArray.of_list int64_t |> CArray.start)
     (List.length output_padding)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
-    (List.length dilation);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length dilation)
+  |> with_tensor_gc
 ;;
 
 let slow_conv_transpose3d_out
@@ -27887,9 +17955,7 @@ let slow_conv_transpose3d_out
   ~output_padding
   ~dilation
   =
-  let out__ = CArray.make t 1 in
   stubs_slow_conv_transpose3d_out
-    (CArray.start out__)
     out
     self
     weight
@@ -27897,7 +17963,7 @@ let slow_conv_transpose3d_out
     (List.length kernel_size)
     (match bias with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (List.map Int64.of_int stride |> CArray.of_list int64_t |> CArray.start)
     (List.length stride)
     (List.map Int64.of_int padding |> CArray.of_list int64_t |> CArray.start)
@@ -27905,45 +17971,25 @@ let slow_conv_transpose3d_out
     (List.map Int64.of_int output_padding |> CArray.of_list int64_t |> CArray.start)
     (List.length output_padding)
     (List.map Int64.of_int dilation |> CArray.of_list int64_t |> CArray.start)
-    (List.length dilation);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length dilation)
+  |> with_tensor_gc
 ;;
 
-let smm self ~mat2 =
-  let out__ = CArray.make t 1 in
-  stubs_smm (CArray.start out__) self mat2;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let smm self ~mat2 = stubs_smm self mat2 |> with_tensor_gc
 
 let smooth_l1_loss self ~target ~reduction ~beta =
-  let out__ = CArray.make t 1 in
-  stubs_smooth_l1_loss
-    (CArray.start out__)
-    self
-    target
-    (Reduction.to_int reduction |> Int64.of_int)
-    beta;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_smooth_l1_loss self target (Reduction.to_int reduction |> Int64.of_int) beta
+  |> with_tensor_gc
 ;;
 
 let smooth_l1_loss_backward ~grad_output self ~target ~reduction ~beta =
-  let out__ = CArray.make t 1 in
   stubs_smooth_l1_loss_backward
-    (CArray.start out__)
     grad_output
     self
     target
     (Reduction.to_int reduction |> Int64.of_int)
-    beta;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    beta
+  |> with_tensor_gc
 ;;
 
 let smooth_l1_loss_backward_grad_input
@@ -27954,209 +18000,111 @@ let smooth_l1_loss_backward_grad_input
   ~reduction
   ~beta
   =
-  let out__ = CArray.make t 1 in
   stubs_smooth_l1_loss_backward_grad_input
-    (CArray.start out__)
     grad_input
     grad_output
     self
     target
     (Reduction.to_int reduction |> Int64.of_int)
-    beta;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    beta
+  |> with_tensor_gc
 ;;
 
 let smooth_l1_loss_out ~out self ~target ~reduction ~beta =
-  let out__ = CArray.make t 1 in
   stubs_smooth_l1_loss_out
-    (CArray.start out__)
     out
     self
     target
     (Reduction.to_int reduction |> Int64.of_int)
-    beta;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    beta
+  |> with_tensor_gc
 ;;
 
 let soft_margin_loss self ~target ~reduction =
-  let out__ = CArray.make t 1 in
-  stubs_soft_margin_loss
-    (CArray.start out__)
-    self
-    target
-    (Reduction.to_int reduction |> Int64.of_int);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_soft_margin_loss self target (Reduction.to_int reduction |> Int64.of_int)
+  |> with_tensor_gc
 ;;
 
 let soft_margin_loss_backward ~grad_output self ~target ~reduction =
-  let out__ = CArray.make t 1 in
   stubs_soft_margin_loss_backward
-    (CArray.start out__)
     grad_output
     self
     target
-    (Reduction.to_int reduction |> Int64.of_int);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Reduction.to_int reduction |> Int64.of_int)
+  |> with_tensor_gc
 ;;
 
 let soft_margin_loss_backward_grad_input ~grad_input ~grad_output self ~target ~reduction =
-  let out__ = CArray.make t 1 in
   stubs_soft_margin_loss_backward_grad_input
-    (CArray.start out__)
     grad_input
     grad_output
     self
     target
-    (Reduction.to_int reduction |> Int64.of_int);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Reduction.to_int reduction |> Int64.of_int)
+  |> with_tensor_gc
 ;;
 
 let soft_margin_loss_out ~out self ~target ~reduction =
-  let out__ = CArray.make t 1 in
-  stubs_soft_margin_loss_out
-    (CArray.start out__)
-    out
-    self
-    target
-    (Reduction.to_int reduction |> Int64.of_int);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_soft_margin_loss_out out self target (Reduction.to_int reduction |> Int64.of_int)
+  |> with_tensor_gc
 ;;
 
 let softmax self ~dim ~dtype =
-  let out__ = CArray.make t 1 in
-  stubs_softmax (CArray.start out__) self (Int64.of_int dim) (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let softmax_int_out ~out self ~dim ~dtype =
-  let out__ = CArray.make t 1 in
-  stubs_softmax_int_out
-    (CArray.start out__)
-    out
-    self
-    (Int64.of_int dim)
-    (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let softplus self =
-  let out__ = CArray.make t 1 in
-  stubs_softplus (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let softplus_backward ~grad_output self ~beta ~threshold =
-  let out__ = CArray.make t 1 in
-  stubs_softplus_backward (CArray.start out__) grad_output self beta threshold;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let softplus_backward_grad_input ~grad_input ~grad_output self ~beta ~threshold =
-  let out__ = CArray.make t 1 in
-  stubs_softplus_backward_grad_input
-    (CArray.start out__)
-    grad_input
-    grad_output
-    self
-    beta
-    threshold;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_softmax self (Int64.of_int dim) (Kind.packed_to_int dtype) |> with_tensor_gc
 ;;
 
-let softplus_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_softplus_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+let softmax_int_out ~out self ~dim ~dtype =
+  stubs_softmax_int_out out self (Int64.of_int dim) (Kind.packed_to_int dtype)
+  |> with_tensor_gc
+;;
+
+let softplus self = stubs_softplus self |> with_tensor_gc
+
+let softplus_backward ~grad_output self ~beta ~threshold =
+  stubs_softplus_backward grad_output self beta threshold |> with_tensor_gc
 ;;
 
-let softshrink self =
-  let out__ = CArray.make t 1 in
-  stubs_softshrink (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+let softplus_backward_grad_input ~grad_input ~grad_output self ~beta ~threshold =
+  stubs_softplus_backward_grad_input grad_input grad_output self beta threshold
+  |> with_tensor_gc
 ;;
 
+let softplus_out ~out self = stubs_softplus_out out self |> with_tensor_gc
+let softshrink self = stubs_softshrink self |> with_tensor_gc
+
 let softshrink_backward ~grad_output self ~lambd =
-  let out__ = CArray.make t 1 in
-  stubs_softshrink_backward (CArray.start out__) grad_output self lambd;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_softshrink_backward grad_output self lambd |> with_tensor_gc
 ;;
 
 let softshrink_backward_grad_input ~grad_input ~grad_output self ~lambd =
-  let out__ = CArray.make t 1 in
-  stubs_softshrink_backward_grad_input
-    (CArray.start out__)
-    grad_input
-    grad_output
-    self
-    lambd;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_softshrink_backward_grad_input grad_input grad_output self lambd |> with_tensor_gc
 ;;
 
-let softshrink_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_softshrink_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let softshrink_out ~out self = stubs_softshrink_out out self |> with_tensor_gc
 
 let sort self ~dim ~descending =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_sort (CArray.start out__) self (Int64.of_int dim) (if descending then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let sort_stable self ~stable ~dim ~descending =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_sort_stable
     (CArray.start out__)
     self
     (if stable then 1 else 0)
     (Int64.of_int dim)
     (if descending then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let sort_values ~values ~indices self ~dim ~descending =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_sort_values
     (CArray.start out__)
     values
@@ -28164,15 +18112,13 @@ let sort_values ~values ~indices self ~dim ~descending =
     self
     (Int64.of_int dim)
     (if descending then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let sort_values_stable ~values ~indices self ~stable ~dim ~descending =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_sort_values_stable
     (CArray.start out__)
     values
@@ -28181,25 +18127,19 @@ let sort_values_stable ~values ~indices self ~stable ~dim ~descending =
     (if stable then 1 else 0)
     (Int64.of_int dim)
     (if descending then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let sparse_bsc_tensor ~ccol_indices ~row_indices ~values ~options =
-  let out__ = CArray.make t 1 in
   stubs_sparse_bsc_tensor
-    (CArray.start out__)
     ccol_indices
     row_indices
     values
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let sparse_bsc_tensor_ccol_row_value_size
@@ -28209,33 +18149,25 @@ let sparse_bsc_tensor_ccol_row_value_size
   ~size
   ~options
   =
-  let out__ = CArray.make t 1 in
   stubs_sparse_bsc_tensor_ccol_row_value_size
-    (CArray.start out__)
     ccol_indices
     row_indices
     values
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let sparse_bsr_tensor ~crow_indices ~col_indices ~values ~options =
-  let out__ = CArray.make t 1 in
   stubs_sparse_bsr_tensor
-    (CArray.start out__)
     crow_indices
     col_indices
     values
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let sparse_bsr_tensor_crow_col_value_size
@@ -28245,33 +18177,25 @@ let sparse_bsr_tensor_crow_col_value_size
   ~size
   ~options
   =
-  let out__ = CArray.make t 1 in
   stubs_sparse_bsr_tensor_crow_col_value_size
-    (CArray.start out__)
     crow_indices
     col_indices
     values
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let sparse_compressed_tensor ~compressed_indices ~plain_indices ~values ~options =
-  let out__ = CArray.make t 1 in
   stubs_sparse_compressed_tensor
-    (CArray.start out__)
     compressed_indices
     plain_indices
     values
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let sparse_compressed_tensor_comp_plain_value_size
@@ -28281,88 +18205,64 @@ let sparse_compressed_tensor_comp_plain_value_size
   ~size
   ~options
   =
-  let out__ = CArray.make t 1 in
   stubs_sparse_compressed_tensor_comp_plain_value_size
-    (CArray.start out__)
     compressed_indices
     plain_indices
     values
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let sparse_coo_tensor ~size ~options =
-  let out__ = CArray.make t 1 in
   stubs_sparse_coo_tensor
-    (CArray.start out__)
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let sparse_coo_tensor_indices ~indices ~values ~options ~is_coalesced =
-  let out__ = CArray.make t 1 in
   stubs_sparse_coo_tensor_indices
-    (CArray.start out__)
     indices
     values
     (Kind.packed_to_int (fst options))
     (Device.to_int (snd options))
-    (if is_coalesced then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if is_coalesced then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let sparse_coo_tensor_indices_size ~indices ~values ~size ~options ~is_coalesced =
-  let out__ = CArray.make t 1 in
   stubs_sparse_coo_tensor_indices_size
-    (CArray.start out__)
     indices
     values
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     (Kind.packed_to_int (fst options))
     (Device.to_int (snd options))
-    (if is_coalesced then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if is_coalesced then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let sparse_coo_tensor_size_out ~out ~size =
-  let out__ = CArray.make t 1 in
   stubs_sparse_coo_tensor_size_out
-    (CArray.start out__)
     out
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
-    (List.length size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length size)
+  |> with_tensor_gc
 ;;
 
 let sparse_csc_tensor ~ccol_indices ~row_indices ~values ~options =
-  let out__ = CArray.make t 1 in
   stubs_sparse_csc_tensor
-    (CArray.start out__)
     ccol_indices
     row_indices
     values
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let sparse_csc_tensor_ccol_row_value_size
@@ -28372,33 +18272,25 @@ let sparse_csc_tensor_ccol_row_value_size
   ~size
   ~options
   =
-  let out__ = CArray.make t 1 in
   stubs_sparse_csc_tensor_ccol_row_value_size
-    (CArray.start out__)
     ccol_indices
     row_indices
     values
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let sparse_csr_tensor ~crow_indices ~col_indices ~values ~options =
-  let out__ = CArray.make t 1 in
   stubs_sparse_csr_tensor
-    (CArray.start out__)
     crow_indices
     col_indices
     values
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let sparse_csr_tensor_crow_col_value_size
@@ -28408,1531 +18300,658 @@ let sparse_csr_tensor_crow_col_value_size
   ~size
   ~options
   =
-  let out__ = CArray.make t 1 in
   stubs_sparse_csr_tensor_crow_col_value_size
-    (CArray.start out__)
     crow_indices
     col_indices
     values
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let sparse_dim self = stubs_sparse_dim self
-
-let sparse_mask self ~mask =
-  let out__ = CArray.make t 1 in
-  stubs_sparse_mask (CArray.start out__) self mask;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let sparse_mask self ~mask = stubs_sparse_mask self mask |> with_tensor_gc
 
 let sparse_mask_out ~out self ~mask =
-  let out__ = CArray.make t 1 in
-  stubs_sparse_mask_out (CArray.start out__) out self mask;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_sparse_mask_out out self mask |> with_tensor_gc
 ;;
 
 let sparse_resize self ~size ~sparse_dim ~dense_dim =
-  let out__ = CArray.make t 1 in
   stubs_sparse_resize
-    (CArray.start out__)
     self
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     (Int64.of_int sparse_dim)
-    (Int64.of_int dense_dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int dense_dim)
+  |> with_tensor_gc
 ;;
 
 let sparse_resize_ self ~size ~sparse_dim ~dense_dim =
-  let out__ = CArray.make t 1 in
   stubs_sparse_resize_
-    (CArray.start out__)
     self
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     (Int64.of_int sparse_dim)
-    (Int64.of_int dense_dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int dense_dim)
+  |> with_tensor_gc
 ;;
 
 let sparse_resize_and_clear self ~size ~sparse_dim ~dense_dim =
-  let out__ = CArray.make t 1 in
   stubs_sparse_resize_and_clear
-    (CArray.start out__)
     self
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     (Int64.of_int sparse_dim)
-    (Int64.of_int dense_dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int dense_dim)
+  |> with_tensor_gc
 ;;
 
 let sparse_resize_and_clear_ self ~size ~sparse_dim ~dense_dim =
-  let out__ = CArray.make t 1 in
   stubs_sparse_resize_and_clear_
-    (CArray.start out__)
     self
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     (Int64.of_int sparse_dim)
-    (Int64.of_int dense_dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int dense_dim)
+  |> with_tensor_gc
 ;;
 
 let sparse_resize_and_clear_out ~out self ~size ~sparse_dim ~dense_dim =
-  let out__ = CArray.make t 1 in
   stubs_sparse_resize_and_clear_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     (Int64.of_int sparse_dim)
-    (Int64.of_int dense_dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int dense_dim)
+  |> with_tensor_gc
 ;;
 
 let sparse_resize_out ~out self ~size ~sparse_dim ~dense_dim =
-  let out__ = CArray.make t 1 in
   stubs_sparse_resize_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     (Int64.of_int sparse_dim)
-    (Int64.of_int dense_dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int dense_dim)
+  |> with_tensor_gc
 ;;
 
 let sparse_sampled_addmm self ~mat1 ~mat2 =
-  let out__ = CArray.make t 1 in
-  stubs_sparse_sampled_addmm (CArray.start out__) self mat1 mat2;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_sparse_sampled_addmm self mat1 mat2 |> with_tensor_gc
 ;;
 
 let sparse_sampled_addmm_out ~out self ~mat1 ~mat2 =
-  let out__ = CArray.make t 1 in
-  stubs_sparse_sampled_addmm_out (CArray.start out__) out self mat1 mat2;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let special_airy_ai ~x =
-  let out__ = CArray.make t 1 in
-  stubs_special_airy_ai (CArray.start out__) x;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let special_airy_ai_out ~out ~x =
-  let out__ = CArray.make t 1 in
-  stubs_special_airy_ai_out (CArray.start out__) out x;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_sparse_sampled_addmm_out out self mat1 mat2 |> with_tensor_gc
 ;;
 
-let special_bessel_j0 self =
-  let out__ = CArray.make t 1 in
-  stubs_special_bessel_j0 (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let special_airy_ai ~x = stubs_special_airy_ai x |> with_tensor_gc
+let special_airy_ai_out ~out ~x = stubs_special_airy_ai_out out x |> with_tensor_gc
+let special_bessel_j0 self = stubs_special_bessel_j0 self |> with_tensor_gc
 
 let special_bessel_j0_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_special_bessel_j0_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_bessel_j0_out out self |> with_tensor_gc
 ;;
 
-let special_bessel_j1 self =
-  let out__ = CArray.make t 1 in
-  stubs_special_bessel_j1 (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let special_bessel_j1 self = stubs_special_bessel_j1 self |> with_tensor_gc
 
 let special_bessel_j1_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_special_bessel_j1_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_bessel_j1_out out self |> with_tensor_gc
 ;;
 
-let special_bessel_y0 self =
-  let out__ = CArray.make t 1 in
-  stubs_special_bessel_y0 (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let special_bessel_y0 self = stubs_special_bessel_y0 self |> with_tensor_gc
 
 let special_bessel_y0_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_special_bessel_y0_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_bessel_y0_out out self |> with_tensor_gc
 ;;
 
-let special_bessel_y1 self =
-  let out__ = CArray.make t 1 in
-  stubs_special_bessel_y1 (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let special_bessel_y1 self = stubs_special_bessel_y1 self |> with_tensor_gc
 
 let special_bessel_y1_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_special_bessel_y1_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_bessel_y1_out out self |> with_tensor_gc
 ;;
 
 let special_chebyshev_polynomial_t ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_chebyshev_polynomial_t (CArray.start out__) x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_chebyshev_polynomial_t x n |> with_tensor_gc
 ;;
 
 let special_chebyshev_polynomial_t_n_scalar ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_chebyshev_polynomial_t_n_scalar (CArray.start out__) x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_chebyshev_polynomial_t_n_scalar x n |> with_tensor_gc
 ;;
 
 let special_chebyshev_polynomial_t_n_scalar_out ~out ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_chebyshev_polynomial_t_n_scalar_out (CArray.start out__) out x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_chebyshev_polynomial_t_n_scalar_out out x n |> with_tensor_gc
 ;;
 
 let special_chebyshev_polynomial_t_out ~out ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_chebyshev_polynomial_t_out (CArray.start out__) out x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_chebyshev_polynomial_t_out out x n |> with_tensor_gc
 ;;
 
 let special_chebyshev_polynomial_t_x_scalar ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_chebyshev_polynomial_t_x_scalar (CArray.start out__) x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_chebyshev_polynomial_t_x_scalar x n |> with_tensor_gc
 ;;
 
 let special_chebyshev_polynomial_t_x_scalar_out ~out ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_chebyshev_polynomial_t_x_scalar_out (CArray.start out__) out x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_chebyshev_polynomial_t_x_scalar_out out x n |> with_tensor_gc
 ;;
 
 let special_chebyshev_polynomial_u ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_chebyshev_polynomial_u (CArray.start out__) x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_chebyshev_polynomial_u x n |> with_tensor_gc
 ;;
 
 let special_chebyshev_polynomial_u_n_scalar ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_chebyshev_polynomial_u_n_scalar (CArray.start out__) x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_chebyshev_polynomial_u_n_scalar x n |> with_tensor_gc
 ;;
 
 let special_chebyshev_polynomial_u_n_scalar_out ~out ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_chebyshev_polynomial_u_n_scalar_out (CArray.start out__) out x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_chebyshev_polynomial_u_n_scalar_out out x n |> with_tensor_gc
 ;;
 
 let special_chebyshev_polynomial_u_out ~out ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_chebyshev_polynomial_u_out (CArray.start out__) out x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_chebyshev_polynomial_u_out out x n |> with_tensor_gc
 ;;
 
 let special_chebyshev_polynomial_u_x_scalar ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_chebyshev_polynomial_u_x_scalar (CArray.start out__) x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_chebyshev_polynomial_u_x_scalar x n |> with_tensor_gc
 ;;
 
 let special_chebyshev_polynomial_u_x_scalar_out ~out ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_chebyshev_polynomial_u_x_scalar_out (CArray.start out__) out x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_chebyshev_polynomial_u_x_scalar_out out x n |> with_tensor_gc
 ;;
 
 let special_chebyshev_polynomial_v ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_chebyshev_polynomial_v (CArray.start out__) x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_chebyshev_polynomial_v x n |> with_tensor_gc
 ;;
 
 let special_chebyshev_polynomial_v_n_scalar ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_chebyshev_polynomial_v_n_scalar (CArray.start out__) x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_chebyshev_polynomial_v_n_scalar x n |> with_tensor_gc
 ;;
 
 let special_chebyshev_polynomial_v_n_scalar_out ~out ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_chebyshev_polynomial_v_n_scalar_out (CArray.start out__) out x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_chebyshev_polynomial_v_n_scalar_out out x n |> with_tensor_gc
 ;;
 
 let special_chebyshev_polynomial_v_out ~out ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_chebyshev_polynomial_v_out (CArray.start out__) out x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_chebyshev_polynomial_v_out out x n |> with_tensor_gc
 ;;
 
 let special_chebyshev_polynomial_v_x_scalar ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_chebyshev_polynomial_v_x_scalar (CArray.start out__) x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_chebyshev_polynomial_v_x_scalar x n |> with_tensor_gc
 ;;
 
 let special_chebyshev_polynomial_v_x_scalar_out ~out ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_chebyshev_polynomial_v_x_scalar_out (CArray.start out__) out x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_chebyshev_polynomial_v_x_scalar_out out x n |> with_tensor_gc
 ;;
 
 let special_chebyshev_polynomial_w ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_chebyshev_polynomial_w (CArray.start out__) x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_chebyshev_polynomial_w x n |> with_tensor_gc
 ;;
 
 let special_chebyshev_polynomial_w_n_scalar ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_chebyshev_polynomial_w_n_scalar (CArray.start out__) x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_chebyshev_polynomial_w_n_scalar x n |> with_tensor_gc
 ;;
 
 let special_chebyshev_polynomial_w_n_scalar_out ~out ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_chebyshev_polynomial_w_n_scalar_out (CArray.start out__) out x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_chebyshev_polynomial_w_n_scalar_out out x n |> with_tensor_gc
 ;;
 
 let special_chebyshev_polynomial_w_out ~out ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_chebyshev_polynomial_w_out (CArray.start out__) out x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_chebyshev_polynomial_w_out out x n |> with_tensor_gc
 ;;
 
 let special_chebyshev_polynomial_w_x_scalar ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_chebyshev_polynomial_w_x_scalar (CArray.start out__) x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_chebyshev_polynomial_w_x_scalar x n |> with_tensor_gc
 ;;
 
 let special_chebyshev_polynomial_w_x_scalar_out ~out ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_chebyshev_polynomial_w_x_scalar_out (CArray.start out__) out x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let special_digamma self =
-  let out__ = CArray.make t 1 in
-  stubs_special_digamma (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let special_digamma_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_special_digamma_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let special_entr self =
-  let out__ = CArray.make t 1 in
-  stubs_special_entr (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let special_entr_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_special_entr_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let special_erf self =
-  let out__ = CArray.make t 1 in
-  stubs_special_erf (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let special_erf_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_special_erf_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let special_erfc self =
-  let out__ = CArray.make t 1 in
-  stubs_special_erfc (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let special_erfc_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_special_erfc_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let special_erfcx self =
-  let out__ = CArray.make t 1 in
-  stubs_special_erfcx (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let special_erfcx_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_special_erfcx_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let special_erfinv self =
-  let out__ = CArray.make t 1 in
-  stubs_special_erfinv (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let special_erfinv_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_special_erfinv_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let special_exp2 self =
-  let out__ = CArray.make t 1 in
-  stubs_special_exp2 (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let special_exp2_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_special_exp2_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let special_expit self =
-  let out__ = CArray.make t 1 in
-  stubs_special_expit (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let special_expit_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_special_expit_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let special_expm1 self =
-  let out__ = CArray.make t 1 in
-  stubs_special_expm1 (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let special_expm1_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_special_expm1_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let special_gammainc self other =
-  let out__ = CArray.make t 1 in
-  stubs_special_gammainc (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+  stubs_special_chebyshev_polynomial_w_x_scalar_out out x n |> with_tensor_gc
+;;
+
+let special_digamma self = stubs_special_digamma self |> with_tensor_gc
+let special_digamma_out ~out self = stubs_special_digamma_out out self |> with_tensor_gc
+let special_entr self = stubs_special_entr self |> with_tensor_gc
+let special_entr_out ~out self = stubs_special_entr_out out self |> with_tensor_gc
+let special_erf self = stubs_special_erf self |> with_tensor_gc
+let special_erf_out ~out self = stubs_special_erf_out out self |> with_tensor_gc
+let special_erfc self = stubs_special_erfc self |> with_tensor_gc
+let special_erfc_out ~out self = stubs_special_erfc_out out self |> with_tensor_gc
+let special_erfcx self = stubs_special_erfcx self |> with_tensor_gc
+let special_erfcx_out ~out self = stubs_special_erfcx_out out self |> with_tensor_gc
+let special_erfinv self = stubs_special_erfinv self |> with_tensor_gc
+let special_erfinv_out ~out self = stubs_special_erfinv_out out self |> with_tensor_gc
+let special_exp2 self = stubs_special_exp2 self |> with_tensor_gc
+let special_exp2_out ~out self = stubs_special_exp2_out out self |> with_tensor_gc
+let special_expit self = stubs_special_expit self |> with_tensor_gc
+let special_expit_out ~out self = stubs_special_expit_out out self |> with_tensor_gc
+let special_expm1 self = stubs_special_expm1 self |> with_tensor_gc
+let special_expm1_out ~out self = stubs_special_expm1_out out self |> with_tensor_gc
+let special_gammainc self other = stubs_special_gammainc self other |> with_tensor_gc
 
 let special_gammainc_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_special_gammainc_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_gammainc_out out self other |> with_tensor_gc
 ;;
 
-let special_gammaincc self other =
-  let out__ = CArray.make t 1 in
-  stubs_special_gammaincc (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let special_gammaincc self other = stubs_special_gammaincc self other |> with_tensor_gc
 
 let special_gammaincc_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_special_gammaincc_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_gammaincc_out out self other |> with_tensor_gc
 ;;
 
-let special_gammaln self =
-  let out__ = CArray.make t 1 in
-  stubs_special_gammaln (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let special_gammaln_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_special_gammaln_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let special_gammaln self = stubs_special_gammaln self |> with_tensor_gc
+let special_gammaln_out ~out self = stubs_special_gammaln_out out self |> with_tensor_gc
 
 let special_hermite_polynomial_h ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_hermite_polynomial_h (CArray.start out__) x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_hermite_polynomial_h x n |> with_tensor_gc
 ;;
 
 let special_hermite_polynomial_h_n_scalar ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_hermite_polynomial_h_n_scalar (CArray.start out__) x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_hermite_polynomial_h_n_scalar x n |> with_tensor_gc
 ;;
 
 let special_hermite_polynomial_h_n_scalar_out ~out ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_hermite_polynomial_h_n_scalar_out (CArray.start out__) out x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_hermite_polynomial_h_n_scalar_out out x n |> with_tensor_gc
 ;;
 
 let special_hermite_polynomial_h_out ~out ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_hermite_polynomial_h_out (CArray.start out__) out x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_hermite_polynomial_h_out out x n |> with_tensor_gc
 ;;
 
 let special_hermite_polynomial_h_x_scalar ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_hermite_polynomial_h_x_scalar (CArray.start out__) x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_hermite_polynomial_h_x_scalar x n |> with_tensor_gc
 ;;
 
 let special_hermite_polynomial_h_x_scalar_out ~out ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_hermite_polynomial_h_x_scalar_out (CArray.start out__) out x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_hermite_polynomial_h_x_scalar_out out x n |> with_tensor_gc
 ;;
 
 let special_hermite_polynomial_he ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_hermite_polynomial_he (CArray.start out__) x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_hermite_polynomial_he x n |> with_tensor_gc
 ;;
 
 let special_hermite_polynomial_he_n_scalar ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_hermite_polynomial_he_n_scalar (CArray.start out__) x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_hermite_polynomial_he_n_scalar x n |> with_tensor_gc
 ;;
 
 let special_hermite_polynomial_he_n_scalar_out ~out ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_hermite_polynomial_he_n_scalar_out (CArray.start out__) out x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_hermite_polynomial_he_n_scalar_out out x n |> with_tensor_gc
 ;;
 
 let special_hermite_polynomial_he_out ~out ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_hermite_polynomial_he_out (CArray.start out__) out x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_hermite_polynomial_he_out out x n |> with_tensor_gc
 ;;
 
 let special_hermite_polynomial_he_x_scalar ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_hermite_polynomial_he_x_scalar (CArray.start out__) x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_hermite_polynomial_he_x_scalar x n |> with_tensor_gc
 ;;
 
 let special_hermite_polynomial_he_x_scalar_out ~out ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_hermite_polynomial_he_x_scalar_out (CArray.start out__) out x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let special_i0 self =
-  let out__ = CArray.make t 1 in
-  stubs_special_i0 (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let special_i0_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_special_i0_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let special_i0e self =
-  let out__ = CArray.make t 1 in
-  stubs_special_i0e (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let special_i0e_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_special_i0e_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let special_i1 self =
-  let out__ = CArray.make t 1 in
-  stubs_special_i1 (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let special_i1_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_special_i1_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let special_i1e self =
-  let out__ = CArray.make t 1 in
-  stubs_special_i1e (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_hermite_polynomial_he_x_scalar_out out x n |> with_tensor_gc
 ;;
 
-let special_i1e_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_special_i1e_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let special_i0 self = stubs_special_i0 self |> with_tensor_gc
+let special_i0_out ~out self = stubs_special_i0_out out self |> with_tensor_gc
+let special_i0e self = stubs_special_i0e self |> with_tensor_gc
+let special_i0e_out ~out self = stubs_special_i0e_out out self |> with_tensor_gc
+let special_i1 self = stubs_special_i1 self |> with_tensor_gc
+let special_i1_out ~out self = stubs_special_i1_out out self |> with_tensor_gc
+let special_i1e self = stubs_special_i1e self |> with_tensor_gc
+let special_i1e_out ~out self = stubs_special_i1e_out out self |> with_tensor_gc
 
 let special_laguerre_polynomial_l ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_laguerre_polynomial_l (CArray.start out__) x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_laguerre_polynomial_l x n |> with_tensor_gc
 ;;
 
 let special_laguerre_polynomial_l_n_scalar ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_laguerre_polynomial_l_n_scalar (CArray.start out__) x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_laguerre_polynomial_l_n_scalar x n |> with_tensor_gc
 ;;
 
 let special_laguerre_polynomial_l_n_scalar_out ~out ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_laguerre_polynomial_l_n_scalar_out (CArray.start out__) out x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_laguerre_polynomial_l_n_scalar_out out x n |> with_tensor_gc
 ;;
 
 let special_laguerre_polynomial_l_out ~out ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_laguerre_polynomial_l_out (CArray.start out__) out x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_laguerre_polynomial_l_out out x n |> with_tensor_gc
 ;;
 
 let special_laguerre_polynomial_l_x_scalar ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_laguerre_polynomial_l_x_scalar (CArray.start out__) x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_laguerre_polynomial_l_x_scalar x n |> with_tensor_gc
 ;;
 
 let special_laguerre_polynomial_l_x_scalar_out ~out ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_laguerre_polynomial_l_x_scalar_out (CArray.start out__) out x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_laguerre_polynomial_l_x_scalar_out out x n |> with_tensor_gc
 ;;
 
 let special_legendre_polynomial_p ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_legendre_polynomial_p (CArray.start out__) x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_legendre_polynomial_p x n |> with_tensor_gc
 ;;
 
 let special_legendre_polynomial_p_n_scalar ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_legendre_polynomial_p_n_scalar (CArray.start out__) x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_legendre_polynomial_p_n_scalar x n |> with_tensor_gc
 ;;
 
 let special_legendre_polynomial_p_n_scalar_out ~out ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_legendre_polynomial_p_n_scalar_out (CArray.start out__) out x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_legendre_polynomial_p_n_scalar_out out x n |> with_tensor_gc
 ;;
 
 let special_legendre_polynomial_p_out ~out ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_legendre_polynomial_p_out (CArray.start out__) out x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_legendre_polynomial_p_out out x n |> with_tensor_gc
 ;;
 
 let special_legendre_polynomial_p_x_scalar ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_legendre_polynomial_p_x_scalar (CArray.start out__) x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_legendre_polynomial_p_x_scalar x n |> with_tensor_gc
 ;;
 
 let special_legendre_polynomial_p_x_scalar_out ~out ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_legendre_polynomial_p_x_scalar_out (CArray.start out__) out x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let special_log1p self =
-  let out__ = CArray.make t 1 in
-  stubs_special_log1p (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_legendre_polynomial_p_x_scalar_out out x n |> with_tensor_gc
 ;;
 
-let special_log1p_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_special_log1p_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let special_log_ndtr self =
-  let out__ = CArray.make t 1 in
-  stubs_special_log_ndtr (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let special_log_ndtr_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_special_log_ndtr_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let special_log1p self = stubs_special_log1p self |> with_tensor_gc
+let special_log1p_out ~out self = stubs_special_log1p_out out self |> with_tensor_gc
+let special_log_ndtr self = stubs_special_log_ndtr self |> with_tensor_gc
+let special_log_ndtr_out ~out self = stubs_special_log_ndtr_out out self |> with_tensor_gc
 
 let special_log_softmax self ~dim ~dtype =
-  let out__ = CArray.make t 1 in
-  stubs_special_log_softmax
-    (CArray.start out__)
-    self
-    (Int64.of_int dim)
-    (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_log_softmax self (Int64.of_int dim) (Kind.packed_to_int dtype)
+  |> with_tensor_gc
 ;;
 
 let special_logit self ~eps =
-  let out__ = CArray.make t 1 in
   stubs_special_logit
-    (CArray.start out__)
     self
     (Option.value eps ~default:0.0)
     (match eps with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let special_logit_out ~out self ~eps =
-  let out__ = CArray.make t 1 in
   stubs_special_logit_out
-    (CArray.start out__)
     out
     self
     (Option.value eps ~default:0.0)
     (match eps with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let special_logsumexp self ~dim ~keepdim =
-  let out__ = CArray.make t 1 in
   stubs_special_logsumexp
-    (CArray.start out__)
     self
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
     (List.length dim)
-    (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if keepdim then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let special_logsumexp_out ~out self ~dim ~keepdim =
-  let out__ = CArray.make t 1 in
   stubs_special_logsumexp_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
     (List.length dim)
-    (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if keepdim then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let special_modified_bessel_i0 self =
-  let out__ = CArray.make t 1 in
-  stubs_special_modified_bessel_i0 (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_modified_bessel_i0 self |> with_tensor_gc
 ;;
 
 let special_modified_bessel_i0_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_special_modified_bessel_i0_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_modified_bessel_i0_out out self |> with_tensor_gc
 ;;
 
 let special_modified_bessel_i1 self =
-  let out__ = CArray.make t 1 in
-  stubs_special_modified_bessel_i1 (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_modified_bessel_i1 self |> with_tensor_gc
 ;;
 
 let special_modified_bessel_i1_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_special_modified_bessel_i1_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_modified_bessel_i1_out out self |> with_tensor_gc
 ;;
 
 let special_modified_bessel_k0 self =
-  let out__ = CArray.make t 1 in
-  stubs_special_modified_bessel_k0 (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_modified_bessel_k0 self |> with_tensor_gc
 ;;
 
 let special_modified_bessel_k0_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_special_modified_bessel_k0_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_modified_bessel_k0_out out self |> with_tensor_gc
 ;;
 
 let special_modified_bessel_k1 self =
-  let out__ = CArray.make t 1 in
-  stubs_special_modified_bessel_k1 (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_modified_bessel_k1 self |> with_tensor_gc
 ;;
 
 let special_modified_bessel_k1_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_special_modified_bessel_k1_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_modified_bessel_k1_out out self |> with_tensor_gc
 ;;
 
 let special_multigammaln self ~p =
-  let out__ = CArray.make t 1 in
-  stubs_special_multigammaln (CArray.start out__) self (Int64.of_int p);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_multigammaln self (Int64.of_int p) |> with_tensor_gc
 ;;
 
 let special_multigammaln_out ~out self ~p =
-  let out__ = CArray.make t 1 in
-  stubs_special_multigammaln_out (CArray.start out__) out self (Int64.of_int p);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let special_ndtr self =
-  let out__ = CArray.make t 1 in
-  stubs_special_ndtr (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let special_ndtr_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_special_ndtr_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let special_ndtri self =
-  let out__ = CArray.make t 1 in
-  stubs_special_ndtri (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_multigammaln_out out self (Int64.of_int p) |> with_tensor_gc
 ;;
 
-let special_ndtri_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_special_ndtri_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let special_ndtr self = stubs_special_ndtr self |> with_tensor_gc
+let special_ndtr_out ~out self = stubs_special_ndtr_out out self |> with_tensor_gc
+let special_ndtri self = stubs_special_ndtri self |> with_tensor_gc
+let special_ndtri_out ~out self = stubs_special_ndtri_out out self |> with_tensor_gc
 
 let special_polygamma ~n self =
-  let out__ = CArray.make t 1 in
-  stubs_special_polygamma (CArray.start out__) (Int64.of_int n) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_polygamma (Int64.of_int n) self |> with_tensor_gc
 ;;
 
 let special_polygamma_out ~out ~n self =
-  let out__ = CArray.make t 1 in
-  stubs_special_polygamma_out (CArray.start out__) out (Int64.of_int n) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_polygamma_out out (Int64.of_int n) self |> with_tensor_gc
 ;;
 
-let special_psi self =
-  let out__ = CArray.make t 1 in
-  stubs_special_psi (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let special_psi_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_special_psi_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let special_psi self = stubs_special_psi self |> with_tensor_gc
+let special_psi_out ~out self = stubs_special_psi_out out self |> with_tensor_gc
 
 let special_round self ~decimals =
-  let out__ = CArray.make t 1 in
-  stubs_special_round (CArray.start out__) self (Int64.of_int decimals);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_round self (Int64.of_int decimals) |> with_tensor_gc
 ;;
 
 let special_round_out ~out self ~decimals =
-  let out__ = CArray.make t 1 in
-  stubs_special_round_out (CArray.start out__) out self (Int64.of_int decimals);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_round_out out self (Int64.of_int decimals) |> with_tensor_gc
 ;;
 
 let special_scaled_modified_bessel_k0 ~x =
-  let out__ = CArray.make t 1 in
-  stubs_special_scaled_modified_bessel_k0 (CArray.start out__) x;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_scaled_modified_bessel_k0 x |> with_tensor_gc
 ;;
 
 let special_scaled_modified_bessel_k0_out ~out ~x =
-  let out__ = CArray.make t 1 in
-  stubs_special_scaled_modified_bessel_k0_out (CArray.start out__) out x;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_scaled_modified_bessel_k0_out out x |> with_tensor_gc
 ;;
 
 let special_scaled_modified_bessel_k1 ~x =
-  let out__ = CArray.make t 1 in
-  stubs_special_scaled_modified_bessel_k1 (CArray.start out__) x;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_scaled_modified_bessel_k1 x |> with_tensor_gc
 ;;
 
 let special_scaled_modified_bessel_k1_out ~out ~x =
-  let out__ = CArray.make t 1 in
-  stubs_special_scaled_modified_bessel_k1_out (CArray.start out__) out x;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_scaled_modified_bessel_k1_out out x |> with_tensor_gc
 ;;
 
 let special_shifted_chebyshev_polynomial_t ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_shifted_chebyshev_polynomial_t (CArray.start out__) x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_shifted_chebyshev_polynomial_t x n |> with_tensor_gc
 ;;
 
 let special_shifted_chebyshev_polynomial_t_n_scalar ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_shifted_chebyshev_polynomial_t_n_scalar (CArray.start out__) x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_shifted_chebyshev_polynomial_t_n_scalar x n |> with_tensor_gc
 ;;
 
 let special_shifted_chebyshev_polynomial_t_n_scalar_out ~out ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_shifted_chebyshev_polynomial_t_n_scalar_out (CArray.start out__) out x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_shifted_chebyshev_polynomial_t_n_scalar_out out x n |> with_tensor_gc
 ;;
 
 let special_shifted_chebyshev_polynomial_t_out ~out ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_shifted_chebyshev_polynomial_t_out (CArray.start out__) out x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_shifted_chebyshev_polynomial_t_out out x n |> with_tensor_gc
 ;;
 
 let special_shifted_chebyshev_polynomial_t_x_scalar ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_shifted_chebyshev_polynomial_t_x_scalar (CArray.start out__) x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_shifted_chebyshev_polynomial_t_x_scalar x n |> with_tensor_gc
 ;;
 
 let special_shifted_chebyshev_polynomial_t_x_scalar_out ~out ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_shifted_chebyshev_polynomial_t_x_scalar_out (CArray.start out__) out x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_shifted_chebyshev_polynomial_t_x_scalar_out out x n |> with_tensor_gc
 ;;
 
 let special_shifted_chebyshev_polynomial_u ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_shifted_chebyshev_polynomial_u (CArray.start out__) x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_shifted_chebyshev_polynomial_u x n |> with_tensor_gc
 ;;
 
 let special_shifted_chebyshev_polynomial_u_n_scalar ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_shifted_chebyshev_polynomial_u_n_scalar (CArray.start out__) x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_shifted_chebyshev_polynomial_u_n_scalar x n |> with_tensor_gc
 ;;
 
 let special_shifted_chebyshev_polynomial_u_n_scalar_out ~out ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_shifted_chebyshev_polynomial_u_n_scalar_out (CArray.start out__) out x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_shifted_chebyshev_polynomial_u_n_scalar_out out x n |> with_tensor_gc
 ;;
 
 let special_shifted_chebyshev_polynomial_u_out ~out ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_shifted_chebyshev_polynomial_u_out (CArray.start out__) out x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_shifted_chebyshev_polynomial_u_out out x n |> with_tensor_gc
 ;;
 
 let special_shifted_chebyshev_polynomial_u_x_scalar ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_shifted_chebyshev_polynomial_u_x_scalar (CArray.start out__) x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_shifted_chebyshev_polynomial_u_x_scalar x n |> with_tensor_gc
 ;;
 
 let special_shifted_chebyshev_polynomial_u_x_scalar_out ~out ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_shifted_chebyshev_polynomial_u_x_scalar_out (CArray.start out__) out x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_shifted_chebyshev_polynomial_u_x_scalar_out out x n |> with_tensor_gc
 ;;
 
 let special_shifted_chebyshev_polynomial_v ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_shifted_chebyshev_polynomial_v (CArray.start out__) x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_shifted_chebyshev_polynomial_v x n |> with_tensor_gc
 ;;
 
 let special_shifted_chebyshev_polynomial_v_n_scalar ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_shifted_chebyshev_polynomial_v_n_scalar (CArray.start out__) x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_shifted_chebyshev_polynomial_v_n_scalar x n |> with_tensor_gc
 ;;
 
 let special_shifted_chebyshev_polynomial_v_n_scalar_out ~out ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_shifted_chebyshev_polynomial_v_n_scalar_out (CArray.start out__) out x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_shifted_chebyshev_polynomial_v_n_scalar_out out x n |> with_tensor_gc
 ;;
 
 let special_shifted_chebyshev_polynomial_v_out ~out ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_shifted_chebyshev_polynomial_v_out (CArray.start out__) out x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_shifted_chebyshev_polynomial_v_out out x n |> with_tensor_gc
 ;;
 
 let special_shifted_chebyshev_polynomial_v_x_scalar ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_shifted_chebyshev_polynomial_v_x_scalar (CArray.start out__) x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_shifted_chebyshev_polynomial_v_x_scalar x n |> with_tensor_gc
 ;;
 
 let special_shifted_chebyshev_polynomial_v_x_scalar_out ~out ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_shifted_chebyshev_polynomial_v_x_scalar_out (CArray.start out__) out x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_shifted_chebyshev_polynomial_v_x_scalar_out out x n |> with_tensor_gc
 ;;
 
 let special_shifted_chebyshev_polynomial_w ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_shifted_chebyshev_polynomial_w (CArray.start out__) x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_shifted_chebyshev_polynomial_w x n |> with_tensor_gc
 ;;
 
 let special_shifted_chebyshev_polynomial_w_n_scalar ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_shifted_chebyshev_polynomial_w_n_scalar (CArray.start out__) x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_shifted_chebyshev_polynomial_w_n_scalar x n |> with_tensor_gc
 ;;
 
 let special_shifted_chebyshev_polynomial_w_n_scalar_out ~out ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_shifted_chebyshev_polynomial_w_n_scalar_out (CArray.start out__) out x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_shifted_chebyshev_polynomial_w_n_scalar_out out x n |> with_tensor_gc
 ;;
 
 let special_shifted_chebyshev_polynomial_w_out ~out ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_shifted_chebyshev_polynomial_w_out (CArray.start out__) out x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_shifted_chebyshev_polynomial_w_out out x n |> with_tensor_gc
 ;;
 
 let special_shifted_chebyshev_polynomial_w_x_scalar ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_shifted_chebyshev_polynomial_w_x_scalar (CArray.start out__) x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_shifted_chebyshev_polynomial_w_x_scalar x n |> with_tensor_gc
 ;;
 
 let special_shifted_chebyshev_polynomial_w_x_scalar_out ~out ~x ~n =
-  let out__ = CArray.make t 1 in
-  stubs_special_shifted_chebyshev_polynomial_w_x_scalar_out (CArray.start out__) out x n;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let special_sinc self =
-  let out__ = CArray.make t 1 in
-  stubs_special_sinc (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_shifted_chebyshev_polynomial_w_x_scalar_out out x n |> with_tensor_gc
 ;;
 
-let special_sinc_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_special_sinc_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let special_sinc self = stubs_special_sinc self |> with_tensor_gc
+let special_sinc_out ~out self = stubs_special_sinc_out out self |> with_tensor_gc
 
 let special_softmax self ~dim ~dtype =
-  let out__ = CArray.make t 1 in
-  stubs_special_softmax
-    (CArray.start out__)
-    self
-    (Int64.of_int dim)
-    (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_softmax self (Int64.of_int dim) (Kind.packed_to_int dtype)
+  |> with_tensor_gc
 ;;
 
-let special_spherical_bessel_j0 ~x =
-  let out__ = CArray.make t 1 in
-  stubs_special_spherical_bessel_j0 (CArray.start out__) x;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let special_spherical_bessel_j0 ~x = stubs_special_spherical_bessel_j0 x |> with_tensor_gc
 
 let special_spherical_bessel_j0_out ~out ~x =
-  let out__ = CArray.make t 1 in
-  stubs_special_spherical_bessel_j0_out (CArray.start out__) out x;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_spherical_bessel_j0_out out x |> with_tensor_gc
 ;;
 
-let special_xlog1py self other =
-  let out__ = CArray.make t 1 in
-  stubs_special_xlog1py (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let special_xlog1py self other = stubs_special_xlog1py self other |> with_tensor_gc
 
 let special_xlog1py_other_scalar self other =
-  let out__ = CArray.make t 1 in
-  stubs_special_xlog1py_other_scalar (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_xlog1py_other_scalar self other |> with_tensor_gc
 ;;
 
 let special_xlog1py_other_scalar_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_special_xlog1py_other_scalar_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_xlog1py_other_scalar_out out self other |> with_tensor_gc
 ;;
 
 let special_xlog1py_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_special_xlog1py_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_xlog1py_out out self other |> with_tensor_gc
 ;;
 
 let special_xlog1py_self_scalar self other =
-  let out__ = CArray.make t 1 in
-  stubs_special_xlog1py_self_scalar (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_xlog1py_self_scalar self other |> with_tensor_gc
 ;;
 
 let special_xlog1py_self_scalar_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_special_xlog1py_self_scalar_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_xlog1py_self_scalar_out out self other |> with_tensor_gc
 ;;
 
-let special_xlogy self other =
-  let out__ = CArray.make t 1 in
-  stubs_special_xlogy (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let special_xlogy self other = stubs_special_xlogy self other |> with_tensor_gc
 
 let special_xlogy_other_scalar self other =
-  let out__ = CArray.make t 1 in
-  stubs_special_xlogy_other_scalar (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_xlogy_other_scalar self other |> with_tensor_gc
 ;;
 
 let special_xlogy_other_scalar_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_special_xlogy_other_scalar_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_xlogy_other_scalar_out out self other |> with_tensor_gc
 ;;
 
 let special_xlogy_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_special_xlogy_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_xlogy_out out self other |> with_tensor_gc
 ;;
 
 let special_xlogy_self_scalar self other =
-  let out__ = CArray.make t 1 in
-  stubs_special_xlogy_self_scalar (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_xlogy_self_scalar self other |> with_tensor_gc
 ;;
 
 let special_xlogy_self_scalar_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_special_xlogy_self_scalar_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_xlogy_self_scalar_out out self other |> with_tensor_gc
 ;;
 
-let special_zeta self other =
-  let out__ = CArray.make t 1 in
-  stubs_special_zeta (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let special_zeta self other = stubs_special_zeta self other |> with_tensor_gc
 
 let special_zeta_other_scalar self other =
-  let out__ = CArray.make t 1 in
-  stubs_special_zeta_other_scalar (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_zeta_other_scalar self other |> with_tensor_gc
 ;;
 
 let special_zeta_other_scalar_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_special_zeta_other_scalar_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_zeta_other_scalar_out out self other |> with_tensor_gc
 ;;
 
 let special_zeta_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_special_zeta_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_zeta_out out self other |> with_tensor_gc
 ;;
 
 let special_zeta_self_scalar self other =
-  let out__ = CArray.make t 1 in
-  stubs_special_zeta_self_scalar (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_zeta_self_scalar self other |> with_tensor_gc
 ;;
 
 let special_zeta_self_scalar_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_special_zeta_self_scalar_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_special_zeta_self_scalar_out out self other |> with_tensor_gc
 ;;
 
 let split self ~split_size ~dim =
@@ -29945,7 +18964,7 @@ let split_copy self ~split_size ~dim =
 
 let split_copy_tensor_out ~out self ~split_size ~dim =
   stubs_split_copy_tensor_out
-    (CArray.of_list t out |> CArray.start)
+    (CArray.of_list gc_tensor out |> CArray.start)
     (List.length out)
     self
     (Int64.of_int split_size)
@@ -29981,7 +19000,7 @@ let split_with_sizes_copy self ~split_sizes ~dim =
 
 let split_with_sizes_copy_out ~out self ~split_sizes ~dim =
   stubs_split_with_sizes_copy_out
-    (CArray.of_list t out |> CArray.start)
+    (CArray.of_list gc_tensor out |> CArray.start)
     (List.length out)
     self
     (List.map Int64.of_int split_sizes |> CArray.of_list int64_t |> CArray.start)
@@ -29989,220 +19008,88 @@ let split_with_sizes_copy_out ~out self ~split_sizes ~dim =
     (Int64.of_int dim)
 ;;
 
-let sqrt self =
-  let out__ = CArray.make t 1 in
-  stubs_sqrt (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let sqrt_ self =
-  let out__ = CArray.make t 1 in
-  stubs_sqrt_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let sqrt_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_sqrt_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let square self =
-  let out__ = CArray.make t 1 in
-  stubs_square (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let square_ self =
-  let out__ = CArray.make t 1 in
-  stubs_square_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let square_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_square_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let squeeze self =
-  let out__ = CArray.make t 1 in
-  stubs_squeeze (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let squeeze_ self =
-  let out__ = CArray.make t 1 in
-  stubs_squeeze_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let squeeze_copy self =
-  let out__ = CArray.make t 1 in
-  stubs_squeeze_copy (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let sqrt self = stubs_sqrt self |> with_tensor_gc
+let sqrt_ self = stubs_sqrt_ self |> with_tensor_gc
+let sqrt_out ~out self = stubs_sqrt_out out self |> with_tensor_gc
+let square self = stubs_square self |> with_tensor_gc
+let square_ self = stubs_square_ self |> with_tensor_gc
+let square_out ~out self = stubs_square_out out self |> with_tensor_gc
+let squeeze self = stubs_squeeze self |> with_tensor_gc
+let squeeze_ self = stubs_squeeze_ self |> with_tensor_gc
+let squeeze_copy self = stubs_squeeze_copy self |> with_tensor_gc
 
 let squeeze_copy_dim self ~dim =
-  let out__ = CArray.make t 1 in
-  stubs_squeeze_copy_dim (CArray.start out__) self (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_squeeze_copy_dim self (Int64.of_int dim) |> with_tensor_gc
 ;;
 
 let squeeze_copy_dim_out ~out self ~dim =
-  let out__ = CArray.make t 1 in
-  stubs_squeeze_copy_dim_out (CArray.start out__) out self (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_squeeze_copy_dim_out out self (Int64.of_int dim) |> with_tensor_gc
 ;;
 
 let squeeze_copy_dims self ~dim =
-  let out__ = CArray.make t 1 in
   stubs_squeeze_copy_dims
-    (CArray.start out__)
     self
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
-    (List.length dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length dim)
+  |> with_tensor_gc
 ;;
 
 let squeeze_copy_dims_out ~out self ~dim =
-  let out__ = CArray.make t 1 in
   stubs_squeeze_copy_dims_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
-    (List.length dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let squeeze_copy_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_squeeze_copy_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let squeeze_dim self ~dim =
-  let out__ = CArray.make t 1 in
-  stubs_squeeze_dim (CArray.start out__) self (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length dim)
+  |> with_tensor_gc
 ;;
 
-let squeeze_dim_ self ~dim =
-  let out__ = CArray.make t 1 in
-  stubs_squeeze_dim_ (CArray.start out__) self (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let squeeze_copy_out ~out self = stubs_squeeze_copy_out out self |> with_tensor_gc
+let squeeze_dim self ~dim = stubs_squeeze_dim self (Int64.of_int dim) |> with_tensor_gc
+let squeeze_dim_ self ~dim = stubs_squeeze_dim_ self (Int64.of_int dim) |> with_tensor_gc
 
 let squeeze_dims self ~dim =
-  let out__ = CArray.make t 1 in
   stubs_squeeze_dims
-    (CArray.start out__)
     self
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
-    (List.length dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length dim)
+  |> with_tensor_gc
 ;;
 
 let squeeze_dims_ self ~dim =
-  let out__ = CArray.make t 1 in
   stubs_squeeze_dims_
-    (CArray.start out__)
     self
     (List.map Int64.of_int dim |> CArray.of_list int64_t |> CArray.start)
-    (List.length dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length dim)
+  |> with_tensor_gc
 ;;
 
-let sspaddmm self ~mat1 ~mat2 =
-  let out__ = CArray.make t 1 in
-  stubs_sspaddmm (CArray.start out__) self mat1 mat2;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let sspaddmm self ~mat1 ~mat2 = stubs_sspaddmm self mat1 mat2 |> with_tensor_gc
 
 let sspaddmm_out ~out self ~mat1 ~mat2 =
-  let out__ = CArray.make t 1 in
-  stubs_sspaddmm_out (CArray.start out__) out self mat1 mat2;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_sspaddmm_out out self mat1 mat2 |> with_tensor_gc
 ;;
 
 let stack tensors ~dim =
-  let out__ = CArray.make t 1 in
   stubs_stack
-    (CArray.start out__)
-    (CArray.of_list t tensors |> CArray.start)
+    (CArray.of_list gc_tensor tensors |> CArray.start)
     (List.length tensors)
-    (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int dim)
+  |> with_tensor_gc
 ;;
 
 let stack_out ~out tensors ~dim =
-  let out__ = CArray.make t 1 in
   stubs_stack_out
-    (CArray.start out__)
     out
-    (CArray.of_list t tensors |> CArray.start)
+    (CArray.of_list gc_tensor tensors |> CArray.start)
     (List.length tensors)
-    (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int dim)
+  |> with_tensor_gc
 ;;
 
-let std self ~unbiased =
-  let out__ = CArray.make t 1 in
-  stubs_std (CArray.start out__) self (if unbiased then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let std self ~unbiased = stubs_std self (if unbiased then 1 else 0) |> with_tensor_gc
 
 let std_correction self ~dim ~correction ~keepdim =
-  let out__ = CArray.make t 1 in
   stubs_std_correction
-    (CArray.start out__)
     self
     (match dim with
      | None -> from_voidp int64_t null
@@ -30211,16 +19098,12 @@ let std_correction self ~dim ~correction ~keepdim =
      | None -> -1
      | Some v -> List.length v)
     correction
-    (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if keepdim then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let std_correction_out ~out self ~dim ~correction ~keepdim =
-  let out__ = CArray.make t 1 in
   stubs_std_correction_out
-    (CArray.start out__)
     out
     self
     (match dim with
@@ -30230,16 +19113,12 @@ let std_correction_out ~out self ~dim ~correction ~keepdim =
      | None -> -1
      | Some v -> List.length v)
     correction
-    (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if keepdim then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let std_dim self ~dim ~unbiased ~keepdim =
-  let out__ = CArray.make t 1 in
   stubs_std_dim
-    (CArray.start out__)
     self
     (match dim with
      | None -> from_voidp int64_t null
@@ -30248,24 +19127,20 @@ let std_dim self ~dim ~unbiased ~keepdim =
      | None -> -1
      | Some v -> List.length v)
     (if unbiased then 1 else 0)
-    (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if keepdim then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let std_mean self ~unbiased =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_std_mean (CArray.start out__) self (if unbiased then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let std_mean_correction self ~dim ~correction ~keepdim =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_std_mean_correction
     (CArray.start out__)
     self
@@ -30277,15 +19152,13 @@ let std_mean_correction self ~dim ~correction ~keepdim =
      | Some v -> List.length v)
     correction
     (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let std_mean_correction_out ~out0 ~out1 self ~dim ~correction ~keepdim =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_std_mean_correction_out
     (CArray.start out__)
     out0
@@ -30299,15 +19172,13 @@ let std_mean_correction_out ~out0 ~out1 self ~dim ~correction ~keepdim =
      | Some v -> List.length v)
     correction
     (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let std_mean_dim self ~dim ~unbiased ~keepdim =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_std_mean_dim
     (CArray.start out__)
     self
@@ -30319,17 +19190,13 @@ let std_mean_dim self ~dim ~unbiased ~keepdim =
      | Some v -> List.length v)
     (if unbiased then 1 else 0)
     (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let std_out ~out self ~dim ~unbiased ~keepdim =
-  let out__ = CArray.make t 1 in
   stubs_std_out
-    (CArray.start out__)
     out
     self
     (match dim with
@@ -30339,17 +19206,13 @@ let std_out ~out self ~dim ~unbiased ~keepdim =
      | None -> -1
      | Some v -> List.length v)
     (if unbiased then 1 else 0)
-    (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if keepdim then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let stft self ~n_fft ~hop_length ~win_length ~window ~normalized ~onesided ~return_complex
   =
-  let out__ = CArray.make t 1 in
   stubs_stft
-    (CArray.start out__)
     self
     (Int64.of_int n_fft)
     (match hop_length with
@@ -30366,13 +19229,11 @@ let stft self ~n_fft ~hop_length ~win_length ~window ~normalized ~onesided ~retu
      | None -> 1)
     (match window with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (if normalized then 1 else 0)
     (if onesided then 1 else 0)
-    (if return_complex then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if return_complex then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let stft_center
@@ -30387,9 +19248,7 @@ let stft_center
   ~onesided
   ~return_complex
   =
-  let out__ = CArray.make t 1 in
   stubs_stft_center
-    (CArray.start out__)
     self
     (Int64.of_int n_fft)
     (match hop_length with
@@ -30406,119 +19265,31 @@ let stft_center
      | None -> 1)
     (match window with
      | Some v -> v
-     | None -> null)
+     | None -> none_gc_tensor)
     (if center then 1 else 0)
     pad_mode
     (if normalized then 1 else 0)
     (if onesided then 1 else 0)
-    (if return_complex then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if return_complex then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let stride self ~dim = stubs_stride self (Int64.of_int dim)
-
-let sub self other =
-  let out__ = CArray.make t 1 in
-  stubs_sub (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let sub_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_sub_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let sub_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_sub_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let sub_scalar self other =
-  let out__ = CArray.make t 1 in
-  stubs_sub_scalar (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let sub_scalar_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_sub_scalar_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let sub_scalar_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_sub_scalar_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let subtract self other =
-  let out__ = CArray.make t 1 in
-  stubs_subtract (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let subtract_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_subtract_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let subtract_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_subtract_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let subtract_scalar self other =
-  let out__ = CArray.make t 1 in
-  stubs_subtract_scalar (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let subtract_scalar_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_subtract_scalar_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let sum self ~dtype =
-  let out__ = CArray.make t 1 in
-  stubs_sum (CArray.start out__) self (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let sub self other = stubs_sub self other |> with_tensor_gc
+let sub_ self other = stubs_sub_ self other |> with_tensor_gc
+let sub_out ~out self other = stubs_sub_out out self other |> with_tensor_gc
+let sub_scalar self other = stubs_sub_scalar self other |> with_tensor_gc
+let sub_scalar_ self other = stubs_sub_scalar_ self other |> with_tensor_gc
+let sub_scalar_out ~out self other = stubs_sub_scalar_out out self other |> with_tensor_gc
+let subtract self other = stubs_subtract self other |> with_tensor_gc
+let subtract_ self other = stubs_subtract_ self other |> with_tensor_gc
+let subtract_out ~out self other = stubs_subtract_out out self other |> with_tensor_gc
+let subtract_scalar self other = stubs_subtract_scalar self other |> with_tensor_gc
+let subtract_scalar_ self other = stubs_subtract_scalar_ self other |> with_tensor_gc
+let sum self ~dtype = stubs_sum self (Kind.packed_to_int dtype) |> with_tensor_gc
 
 let sum_dim_intlist self ~dim ~keepdim ~dtype =
-  let out__ = CArray.make t 1 in
   stubs_sum_dim_intlist
-    (CArray.start out__)
     self
     (match dim with
      | None -> from_voidp int64_t null
@@ -30527,16 +19298,12 @@ let sum_dim_intlist self ~dim ~keepdim ~dtype =
      | None -> -1
      | Some v -> List.length v)
     (if keepdim then 1 else 0)
-    (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Kind.packed_to_int dtype)
+  |> with_tensor_gc
 ;;
 
 let sum_intlist_out ~out self ~dim ~keepdim ~dtype =
-  let out__ = CArray.make t 1 in
   stubs_sum_intlist_out
-    (CArray.start out__)
     out
     self
     (match dim with
@@ -30546,50 +19313,37 @@ let sum_intlist_out ~out self ~dim ~keepdim ~dtype =
      | None -> -1
      | Some v -> List.length v)
     (if keepdim then 1 else 0)
-    (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Kind.packed_to_int dtype)
+  |> with_tensor_gc
 ;;
 
 let sum_out ~out self ~dtype =
-  let out__ = CArray.make t 1 in
-  stubs_sum_out (CArray.start out__) out self (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_sum_out out self (Kind.packed_to_int dtype) |> with_tensor_gc
 ;;
 
 let sum_to_size self ~size =
-  let out__ = CArray.make t 1 in
   stubs_sum_to_size
-    (CArray.start out__)
     self
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
-    (List.length size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length size)
+  |> with_tensor_gc
 ;;
 
 let svd self ~some ~compute_uv =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs_svd
     (CArray.start out__)
     self
     (if some then 1 else 0)
     (if compute_uv then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
 let svd_u ~u ~s ~v self ~some ~compute_uv =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs_svd_u
     (CArray.start out__)
     u
@@ -30598,91 +19352,36 @@ let svd_u ~u ~s ~v self ~some ~compute_uv =
     self
     (if some then 1 else 0)
     (if compute_uv then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
 let swapaxes self ~axis0 ~axis1 =
-  let out__ = CArray.make t 1 in
-  stubs_swapaxes (CArray.start out__) self (Int64.of_int axis0) (Int64.of_int axis1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_swapaxes self (Int64.of_int axis0) (Int64.of_int axis1) |> with_tensor_gc
 ;;
 
 let swapaxes_ self ~axis0 ~axis1 =
-  let out__ = CArray.make t 1 in
-  stubs_swapaxes_ (CArray.start out__) self (Int64.of_int axis0) (Int64.of_int axis1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_swapaxes_ self (Int64.of_int axis0) (Int64.of_int axis1) |> with_tensor_gc
 ;;
 
 let swapdims self ~dim0 ~dim1 =
-  let out__ = CArray.make t 1 in
-  stubs_swapdims (CArray.start out__) self (Int64.of_int dim0) (Int64.of_int dim1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_swapdims self (Int64.of_int dim0) (Int64.of_int dim1) |> with_tensor_gc
 ;;
 
 let swapdims_ self ~dim0 ~dim1 =
-  let out__ = CArray.make t 1 in
-  stubs_swapdims_ (CArray.start out__) self (Int64.of_int dim0) (Int64.of_int dim1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let tr self =
-  let out__ = CArray.make t 1 in
-  stubs_tr (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_swapdims_ self (Int64.of_int dim0) (Int64.of_int dim1) |> with_tensor_gc
 ;;
 
-let t_ self =
-  let out__ = CArray.make t 1 in
-  stubs_t_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let t_copy self =
-  let out__ = CArray.make t 1 in
-  stubs_t_copy (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let t_copy_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_t_copy_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let take self ~index =
-  let out__ = CArray.make t 1 in
-  stubs_take (CArray.start out__) self index;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let tr self = stubs_tr self |> with_tensor_gc
+let t_ self = stubs_t_ self |> with_tensor_gc
+let t_copy self = stubs_t_copy self |> with_tensor_gc
+let t_copy_out ~out self = stubs_t_copy_out out self |> with_tensor_gc
+let take self ~index = stubs_take self index |> with_tensor_gc
 
 let take_along_dim self ~indices ~dim =
-  let out__ = CArray.make t 1 in
   stubs_take_along_dim
-    (CArray.start out__)
     self
     indices
     (match dim with
@@ -30690,16 +19389,12 @@ let take_along_dim self ~indices ~dim =
      | Some v -> Int64.of_int v)
     (match dim with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let take_along_dim_out ~out self ~indices ~dim =
-  let out__ = CArray.make t 1 in
   stubs_take_along_dim_out
-    (CArray.start out__)
     out
     self
     indices
@@ -30708,83 +19403,26 @@ let take_along_dim_out ~out self ~indices ~dim =
      | Some v -> Int64.of_int v)
     (match dim with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let take_out ~out self ~index =
-  let out__ = CArray.make t 1 in
-  stubs_take_out (CArray.start out__) out self index;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let tan self =
-  let out__ = CArray.make t 1 in
-  stubs_tan (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let tan_ self =
-  let out__ = CArray.make t 1 in
-  stubs_tan_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let tan_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_tan_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let tanh self =
-  let out__ = CArray.make t 1 in
-  stubs_tanh (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
-let tanh_ self =
-  let out__ = CArray.make t 1 in
-  stubs_tanh_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let take_out ~out self ~index = stubs_take_out out self index |> with_tensor_gc
+let tan self = stubs_tan self |> with_tensor_gc
+let tan_ self = stubs_tan_ self |> with_tensor_gc
+let tan_out ~out self = stubs_tan_out out self |> with_tensor_gc
+let tanh self = stubs_tanh self |> with_tensor_gc
+let tanh_ self = stubs_tanh_ self |> with_tensor_gc
 
 let tanh_backward ~grad_output ~output =
-  let out__ = CArray.make t 1 in
-  stubs_tanh_backward (CArray.start out__) grad_output output;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_tanh_backward grad_output output |> with_tensor_gc
 ;;
 
 let tanh_backward_grad_input ~grad_input ~grad_output ~output =
-  let out__ = CArray.make t 1 in
-  stubs_tanh_backward_grad_input (CArray.start out__) grad_input grad_output output;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_tanh_backward_grad_input grad_input grad_output output |> with_tensor_gc
 ;;
 
-let tanh_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_tanh_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let tanh_out ~out self = stubs_tanh_out out self |> with_tensor_gc
 
 let tensor_split self ~sections ~dim =
   stubs_tensor_split self (Int64.of_int sections) (Int64.of_int dim) |> to_tensor_list
@@ -30808,203 +19446,114 @@ let tensor_split_tensor_indices_or_sections self ~tensor_indices_or_sections ~di
 ;;
 
 let tensordot self other ~dims_self ~dims_other =
-  let out__ = CArray.make t 1 in
   stubs_tensordot
-    (CArray.start out__)
     self
     other
     (List.map Int64.of_int dims_self |> CArray.of_list int64_t |> CArray.start)
     (List.length dims_self)
     (List.map Int64.of_int dims_other |> CArray.of_list int64_t |> CArray.start)
-    (List.length dims_other);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length dims_other)
+  |> with_tensor_gc
 ;;
 
 let tensordot_out ~out self other ~dims_self ~dims_other =
-  let out__ = CArray.make t 1 in
   stubs_tensordot_out
-    (CArray.start out__)
     out
     self
     other
     (List.map Int64.of_int dims_self |> CArray.of_list int64_t |> CArray.start)
     (List.length dims_self)
     (List.map Int64.of_int dims_other |> CArray.of_list int64_t |> CArray.start)
-    (List.length dims_other);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length dims_other)
+  |> with_tensor_gc
 ;;
 
 let threshold self ~threshold ~value =
-  let out__ = CArray.make t 1 in
-  stubs_threshold (CArray.start out__) self threshold value;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_threshold self threshold value |> with_tensor_gc
 ;;
 
 let threshold_ self ~threshold ~value =
-  let out__ = CArray.make t 1 in
-  stubs_threshold_ (CArray.start out__) self threshold value;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_threshold_ self threshold value |> with_tensor_gc
 ;;
 
 let threshold_backward ~grad_output self ~threshold =
-  let out__ = CArray.make t 1 in
-  stubs_threshold_backward (CArray.start out__) grad_output self threshold;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_threshold_backward grad_output self threshold |> with_tensor_gc
 ;;
 
 let threshold_backward_grad_input ~grad_input ~grad_output self ~threshold =
-  let out__ = CArray.make t 1 in
-  stubs_threshold_backward_grad_input
-    (CArray.start out__)
-    grad_input
-    grad_output
-    self
-    threshold;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_threshold_backward_grad_input grad_input grad_output self threshold
+  |> with_tensor_gc
 ;;
 
 let threshold_out ~out self ~threshold ~value =
-  let out__ = CArray.make t 1 in
-  stubs_threshold_out (CArray.start out__) out self threshold value;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_threshold_out out self threshold value |> with_tensor_gc
 ;;
 
 let tile self ~dims =
-  let out__ = CArray.make t 1 in
   stubs_tile
-    (CArray.start out__)
     self
     (List.map Int64.of_int dims |> CArray.of_list int64_t |> CArray.start)
-    (List.length dims);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length dims)
+  |> with_tensor_gc
 ;;
 
-let to_ self ~device =
-  let out__ = CArray.make t 1 in
-  stubs_to_ (CArray.start out__) self (Device.to_int device);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let to_ self ~device = stubs_to_ self (Device.to_int device) |> with_tensor_gc
 
 let to_dense self ~dtype ~masked_grad =
-  let out__ = CArray.make t 1 in
-  stubs_to_dense
-    (CArray.start out__)
-    self
-    (Kind.packed_to_int dtype)
-    (if masked_grad then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_to_dense self (Kind.packed_to_int dtype) (if masked_grad then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let to_dense_backward ~grad input ~masked_grad =
-  let out__ = CArray.make t 1 in
-  stubs_to_dense_backward (CArray.start out__) grad input (if masked_grad then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_to_dense_backward grad input (if masked_grad then 1 else 0) |> with_tensor_gc
 ;;
 
 let to_device self ~device ~dtype ~non_blocking ~copy =
-  let out__ = CArray.make t 1 in
   stubs_to_device
-    (CArray.start out__)
     self
     (Device.to_int device)
     (Kind.packed_to_int dtype)
     (if non_blocking then 1 else 0)
-    (if copy then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if copy then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let to_dtype self ~dtype ~non_blocking ~copy =
-  let out__ = CArray.make t 1 in
   stubs_to_dtype
-    (CArray.start out__)
     self
     (Kind.packed_to_int dtype)
     (if non_blocking then 1 else 0)
-    (if copy then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if copy then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let to_dtype_layout self ~options ~non_blocking ~copy =
-  let out__ = CArray.make t 1 in
   stubs_to_dtype_layout
-    (CArray.start out__)
     self
     (Kind.packed_to_int (fst options))
     (Device.to_int (snd options))
     (if non_blocking then 1 else 0)
-    (if copy then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if copy then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let to_mkldnn self ~dtype =
-  let out__ = CArray.make t 1 in
-  stubs_to_mkldnn (CArray.start out__) self (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_to_mkldnn self (Kind.packed_to_int dtype) |> with_tensor_gc
 ;;
 
-let to_mkldnn_backward ~grad input =
-  let out__ = CArray.make t 1 in
-  stubs_to_mkldnn_backward (CArray.start out__) grad input;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let to_mkldnn_backward ~grad input = stubs_to_mkldnn_backward grad input |> with_tensor_gc
 
 let to_mkldnn_out ~out self ~dtype =
-  let out__ = CArray.make t 1 in
-  stubs_to_mkldnn_out (CArray.start out__) out self (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_to_mkldnn_out out self (Kind.packed_to_int dtype) |> with_tensor_gc
 ;;
 
 let to_other self other ~non_blocking ~copy =
-  let out__ = CArray.make t 1 in
-  stubs_to_other
-    (CArray.start out__)
-    self
-    other
-    (if non_blocking then 1 else 0)
-    (if copy then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_to_other self other (if non_blocking then 1 else 0) (if copy then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let to_padded_tensor self ~padding ~output_size =
-  let out__ = CArray.make t 1 in
   stubs_to_padded_tensor
-    (CArray.start out__)
     self
     padding
     (match output_size with
@@ -31012,16 +19561,12 @@ let to_padded_tensor self ~padding ~output_size =
      | Some v -> List.map Int64.of_int v |> CArray.of_list int64_t |> CArray.start)
     (match output_size with
      | None -> -1
-     | Some v -> List.length v);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | Some v -> List.length v)
+  |> with_tensor_gc
 ;;
 
 let to_padded_tensor_out ~out self ~padding ~output_size =
-  let out__ = CArray.make t 1 in
   stubs_to_padded_tensor_out
-    (CArray.start out__)
     out
     self
     padding
@@ -31030,14 +19575,12 @@ let to_padded_tensor_out ~out self ~padding ~output_size =
      | Some v -> List.map Int64.of_int v |> CArray.of_list int64_t |> CArray.start)
     (match output_size with
      | None -> -1
-     | Some v -> List.length v);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | Some v -> List.length v)
+  |> with_tensor_gc
 ;;
 
 let topk self ~k ~dim ~largest ~sorted =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_topk
     (CArray.start out__)
     self
@@ -31045,15 +19588,13 @@ let topk self ~k ~dim ~largest ~sorted =
     (Int64.of_int dim)
     (if largest then 1 else 0)
     (if sorted then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let topk_values ~values ~indices self ~k ~dim ~largest ~sorted =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_topk_values
     (CArray.start out__)
     values
@@ -31063,120 +19604,51 @@ let topk_values ~values ~indices self ~k ~dim ~largest ~sorted =
     (Int64.of_int dim)
     (if largest then 1 else 0)
     (if sorted then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let totype self ~scalar_type =
-  let out__ = CArray.make t 1 in
-  stubs_totype (CArray.start out__) self (Kind.packed_to_int scalar_type);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_totype self (Kind.packed_to_int scalar_type) |> with_tensor_gc
 ;;
 
-let trace self =
-  let out__ = CArray.make t 1 in
-  stubs_trace (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let trace self = stubs_trace self |> with_tensor_gc
 
 let trace_backward ~grad ~sizes =
-  let out__ = CArray.make t 1 in
   stubs_trace_backward
-    (CArray.start out__)
     grad
     (List.map Int64.of_int sizes |> CArray.of_list int64_t |> CArray.start)
-    (List.length sizes);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length sizes)
+  |> with_tensor_gc
 ;;
 
-let trace_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_trace_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let trace_out ~out self = stubs_trace_out out self |> with_tensor_gc
 
 let transpose self ~dim0 ~dim1 =
-  let out__ = CArray.make t 1 in
-  stubs_transpose (CArray.start out__) self (Int64.of_int dim0) (Int64.of_int dim1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_transpose self (Int64.of_int dim0) (Int64.of_int dim1) |> with_tensor_gc
 ;;
 
 let transpose_ self ~dim0 ~dim1 =
-  let out__ = CArray.make t 1 in
-  stubs_transpose_ (CArray.start out__) self (Int64.of_int dim0) (Int64.of_int dim1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_transpose_ self (Int64.of_int dim0) (Int64.of_int dim1) |> with_tensor_gc
 ;;
 
 let transpose_copy self ~dim0 ~dim1 =
-  let out__ = CArray.make t 1 in
-  stubs_transpose_copy (CArray.start out__) self (Int64.of_int dim0) (Int64.of_int dim1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_transpose_copy self (Int64.of_int dim0) (Int64.of_int dim1) |> with_tensor_gc
 ;;
 
 let transpose_copy_int_out ~out self ~dim0 ~dim1 =
-  let out__ = CArray.make t 1 in
-  stubs_transpose_copy_int_out
-    (CArray.start out__)
-    out
-    self
-    (Int64.of_int dim0)
-    (Int64.of_int dim1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let trapezoid ~y ~dim =
-  let out__ = CArray.make t 1 in
-  stubs_trapezoid (CArray.start out__) y (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let trapezoid_x ~y ~x ~dim =
-  let out__ = CArray.make t 1 in
-  stubs_trapezoid_x (CArray.start out__) y x (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let trapz ~y ~x ~dim =
-  let out__ = CArray.make t 1 in
-  stubs_trapz (CArray.start out__) y x (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_transpose_copy_int_out out self (Int64.of_int dim0) (Int64.of_int dim1)
+  |> with_tensor_gc
 ;;
 
-let trapz_dx ~y ~dx ~dim =
-  let out__ = CArray.make t 1 in
-  stubs_trapz_dx (CArray.start out__) y dx (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let trapezoid ~y ~dim = stubs_trapezoid y (Int64.of_int dim) |> with_tensor_gc
+let trapezoid_x ~y ~x ~dim = stubs_trapezoid_x y x (Int64.of_int dim) |> with_tensor_gc
+let trapz ~y ~x ~dim = stubs_trapz y x (Int64.of_int dim) |> with_tensor_gc
+let trapz_dx ~y ~dx ~dim = stubs_trapz_dx y dx (Int64.of_int dim) |> with_tensor_gc
 
 let triangular_solve self ~a ~upper ~transpose ~unitriangular =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_triangular_solve
     (CArray.start out__)
     self
@@ -31184,15 +19656,13 @@ let triangular_solve self ~a ~upper ~transpose ~unitriangular =
     (if upper then 1 else 0)
     (if transpose then 1 else 0)
     (if unitriangular then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let triangular_solve_x ~x ~m self ~a ~upper ~transpose ~unitriangular =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_triangular_solve_x
     (CArray.start out__)
     x
@@ -31202,68 +19672,35 @@ let triangular_solve_x ~x ~m self ~a ~upper ~transpose ~unitriangular =
     (if upper then 1 else 0)
     (if transpose then 1 else 0)
     (if unitriangular then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
-let tril self ~diagonal =
-  let out__ = CArray.make t 1 in
-  stubs_tril (CArray.start out__) self (Int64.of_int diagonal);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let tril_ self ~diagonal =
-  let out__ = CArray.make t 1 in
-  stubs_tril_ (CArray.start out__) self (Int64.of_int diagonal);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let tril self ~diagonal = stubs_tril self (Int64.of_int diagonal) |> with_tensor_gc
+let tril_ self ~diagonal = stubs_tril_ self (Int64.of_int diagonal) |> with_tensor_gc
 
 let tril_indices ~row ~col ~offset ~options =
-  let out__ = CArray.make t 1 in
   stubs_tril_indices
-    (CArray.start out__)
     (Int64.of_int row)
     (Int64.of_int col)
     (Int64.of_int offset)
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let tril_indices_out ~out ~row ~col ~offset =
-  let out__ = CArray.make t 1 in
-  stubs_tril_indices_out
-    (CArray.start out__)
-    out
-    (Int64.of_int row)
-    (Int64.of_int col)
-    (Int64.of_int offset);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_tril_indices_out out (Int64.of_int row) (Int64.of_int col) (Int64.of_int offset)
+  |> with_tensor_gc
 ;;
 
 let tril_out ~out self ~diagonal =
-  let out__ = CArray.make t 1 in
-  stubs_tril_out (CArray.start out__) out self (Int64.of_int diagonal);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_tril_out out self (Int64.of_int diagonal) |> with_tensor_gc
 ;;
 
 let triplet_margin_loss ~anchor ~positive ~negative ~margin ~p ~eps ~swap ~reduction =
-  let out__ = CArray.make t 1 in
   stubs_triplet_margin_loss
-    (CArray.start out__)
     anchor
     positive
     negative
@@ -31271,264 +19708,129 @@ let triplet_margin_loss ~anchor ~positive ~negative ~margin ~p ~eps ~swap ~reduc
     p
     eps
     (if swap then 1 else 0)
-    (Reduction.to_int reduction |> Int64.of_int);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let triu self ~diagonal =
-  let out__ = CArray.make t 1 in
-  stubs_triu (CArray.start out__) self (Int64.of_int diagonal);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Reduction.to_int reduction |> Int64.of_int)
+  |> with_tensor_gc
 ;;
 
-let triu_ self ~diagonal =
-  let out__ = CArray.make t 1 in
-  stubs_triu_ (CArray.start out__) self (Int64.of_int diagonal);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let triu self ~diagonal = stubs_triu self (Int64.of_int diagonal) |> with_tensor_gc
+let triu_ self ~diagonal = stubs_triu_ self (Int64.of_int diagonal) |> with_tensor_gc
 
 let triu_indices ~row ~col ~offset ~options =
-  let out__ = CArray.make t 1 in
   stubs_triu_indices
-    (CArray.start out__)
     (Int64.of_int row)
     (Int64.of_int col)
     (Int64.of_int offset)
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
 let triu_indices_out ~out ~row ~col ~offset =
-  let out__ = CArray.make t 1 in
-  stubs_triu_indices_out
-    (CArray.start out__)
-    out
-    (Int64.of_int row)
-    (Int64.of_int col)
-    (Int64.of_int offset);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_triu_indices_out out (Int64.of_int row) (Int64.of_int col) (Int64.of_int offset)
+  |> with_tensor_gc
 ;;
 
 let triu_out ~out self ~diagonal =
-  let out__ = CArray.make t 1 in
-  stubs_triu_out (CArray.start out__) out self (Int64.of_int diagonal);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let true_divide self other =
-  let out__ = CArray.make t 1 in
-  stubs_true_divide (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_triu_out out self (Int64.of_int diagonal) |> with_tensor_gc
 ;;
 
-let true_divide_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_true_divide_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let true_divide self other = stubs_true_divide self other |> with_tensor_gc
+let true_divide_ self other = stubs_true_divide_ self other |> with_tensor_gc
 
 let true_divide_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_true_divide_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_true_divide_out out self other |> with_tensor_gc
 ;;
 
-let true_divide_scalar self other =
-  let out__ = CArray.make t 1 in
-  stubs_true_divide_scalar (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let true_divide_scalar self other = stubs_true_divide_scalar self other |> with_tensor_gc
 
 let true_divide_scalar_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_true_divide_scalar_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let trunc self =
-  let out__ = CArray.make t 1 in
-  stubs_trunc (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let trunc_ self =
-  let out__ = CArray.make t 1 in
-  stubs_trunc_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let trunc_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_trunc_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let type_as self other =
-  let out__ = CArray.make t 1 in
-  stubs_type_as (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_true_divide_scalar_ self other |> with_tensor_gc
 ;;
 
+let trunc self = stubs_trunc self |> with_tensor_gc
+let trunc_ self = stubs_trunc_ self |> with_tensor_gc
+let trunc_out ~out self = stubs_trunc_out out self |> with_tensor_gc
+let type_as self other = stubs_type_as self other |> with_tensor_gc
 let unbind self ~dim = stubs_unbind self (Int64.of_int dim) |> to_tensor_list
 let unbind_copy self ~dim = stubs_unbind_copy self (Int64.of_int dim) |> to_tensor_list
 
 let unbind_copy_int_out ~out self ~dim =
   stubs_unbind_copy_int_out
-    (CArray.of_list t out |> CArray.start)
+    (CArray.of_list gc_tensor out |> CArray.start)
     (List.length out)
     self
     (Int64.of_int dim)
 ;;
 
 let unflatten self ~dim ~sizes =
-  let out__ = CArray.make t 1 in
   stubs_unflatten
-    (CArray.start out__)
     self
     (Int64.of_int dim)
     (List.map Int64.of_int sizes |> CArray.of_list int64_t |> CArray.start)
-    (List.length sizes);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length sizes)
+  |> with_tensor_gc
 ;;
 
 let unflatten_dense_tensors ~flat tensors =
   stubs_unflatten_dense_tensors
     flat
-    (CArray.of_list t tensors |> CArray.start)
+    (CArray.of_list gc_tensor tensors |> CArray.start)
     (List.length tensors)
   |> to_tensor_list
 ;;
 
 let unfold self ~dimension ~size ~step =
-  let out__ = CArray.make t 1 in
-  stubs_unfold
-    (CArray.start out__)
-    self
-    (Int64.of_int dimension)
-    (Int64.of_int size)
-    (Int64.of_int step);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_unfold self (Int64.of_int dimension) (Int64.of_int size) (Int64.of_int step)
+  |> with_tensor_gc
 ;;
 
 let unfold_backward ~grad_in ~input_sizes ~dim ~size ~step =
-  let out__ = CArray.make t 1 in
   stubs_unfold_backward
-    (CArray.start out__)
     grad_in
     (List.map Int64.of_int input_sizes |> CArray.of_list int64_t |> CArray.start)
     (List.length input_sizes)
     (Int64.of_int dim)
     (Int64.of_int size)
-    (Int64.of_int step);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int step)
+  |> with_tensor_gc
 ;;
 
 let unfold_backward_out ~out ~grad_in ~input_sizes ~dim ~size ~step =
-  let out__ = CArray.make t 1 in
   stubs_unfold_backward_out
-    (CArray.start out__)
     out
     grad_in
     (List.map Int64.of_int input_sizes |> CArray.of_list int64_t |> CArray.start)
     (List.length input_sizes)
     (Int64.of_int dim)
     (Int64.of_int size)
-    (Int64.of_int step);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int step)
+  |> with_tensor_gc
 ;;
 
 let unfold_copy self ~dimension ~size ~step =
-  let out__ = CArray.make t 1 in
-  stubs_unfold_copy
-    (CArray.start out__)
-    self
-    (Int64.of_int dimension)
-    (Int64.of_int size)
-    (Int64.of_int step);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_unfold_copy self (Int64.of_int dimension) (Int64.of_int size) (Int64.of_int step)
+  |> with_tensor_gc
 ;;
 
 let unfold_copy_out ~out self ~dimension ~size ~step =
-  let out__ = CArray.make t 1 in
   stubs_unfold_copy_out
-    (CArray.start out__)
     out
     self
     (Int64.of_int dimension)
     (Int64.of_int size)
-    (Int64.of_int step);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let uniform self ~from ~to_ =
-  let out__ = CArray.make t 1 in
-  stubs_uniform (CArray.start out__) self from to_;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Int64.of_int step)
+  |> with_tensor_gc
 ;;
 
-let uniform_ self ~from ~to_ =
-  let out__ = CArray.make t 1 in
-  stubs_uniform_ (CArray.start out__) self from to_;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let uniform self ~from ~to_ = stubs_uniform self from to_ |> with_tensor_gc
+let uniform_ self ~from ~to_ = stubs_uniform_ self from to_ |> with_tensor_gc
 
 let uniform_out ~out self ~from ~to_ =
-  let out__ = CArray.make t 1 in
-  stubs_uniform_out (CArray.start out__) out self from to_;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_uniform_out out self from to_ |> with_tensor_gc
 ;;
 
 let unique_consecutive self ~return_inverse ~return_counts ~dim =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs_unique_consecutive
     (CArray.start out__)
     self
@@ -31540,17 +19842,14 @@ let unique_consecutive self ~return_inverse ~return_counts ~dim =
     (match dim with
      | Some _ -> 0
      | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
 let unique_consecutive_out ~out0 ~out1 ~out2 self ~return_inverse ~return_counts ~dim =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs_unique_consecutive_out
     (CArray.start out__)
     out0
@@ -31565,17 +19864,14 @@ let unique_consecutive_out ~out0 ~out1 ~out2 self ~return_inverse ~return_counts
     (match dim with
      | Some _ -> 0
      | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
 let unique_dim self ~dim ~sorted ~return_inverse ~return_counts =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs_unique_dim
     (CArray.start out__)
     self
@@ -31583,34 +19879,28 @@ let unique_dim self ~dim ~sorted ~return_inverse ~return_counts =
     (if sorted then 1 else 0)
     (if return_inverse then 1 else 0)
     (if return_counts then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
 let unique_dim_consecutive self ~dim ~return_inverse ~return_counts =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs_unique_dim_consecutive
     (CArray.start out__)
     self
     (Int64.of_int dim)
     (if return_inverse then 1 else 0)
     (if return_counts then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
 let unique_dim_consecutive_out ~out0 ~out1 ~out2 self ~dim ~return_inverse ~return_counts =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs_unique_dim_consecutive_out
     (CArray.start out__)
     out0
@@ -31620,17 +19910,14 @@ let unique_dim_consecutive_out ~out0 ~out1 ~out2 self ~dim ~return_inverse ~retu
     (Int64.of_int dim)
     (if return_inverse then 1 else 0)
     (if return_counts then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
 let unique_dim_out ~out0 ~out1 ~out2 self ~dim ~sorted ~return_inverse ~return_counts =
-  let out__ = CArray.make t 3 in
+  let out__ = CArray.make raw_tensor 3 in
   stubs_unique_dim_out
     (CArray.start out__)
     out0
@@ -31641,12 +19928,9 @@ let unique_dim_out ~out0 ~out1 ~out2 self ~dim ~sorted ~return_inverse ~return_c
     (if sorted then 1 else 0)
     (if return_inverse then 1 else 0)
     (if return_counts then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
-  let t2 = CArray.get out__ 2 in
-  Gc.finalise C.Tensor.free t2;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
+  let t2 = CArray.get out__ 2 |> with_tensor_gc in
   t0, t1, t2
 ;;
 
@@ -31660,7 +19944,7 @@ let unsafe_split self ~split_size ~dim =
 
 let unsafe_split_tensor_out ~out self ~split_size ~dim =
   stubs_unsafe_split_tensor_out
-    (CArray.of_list t out |> CArray.start)
+    (CArray.of_list gc_tensor out |> CArray.start)
     (List.length out)
     self
     (Int64.of_int split_size)
@@ -31678,7 +19962,7 @@ let unsafe_split_with_sizes self ~split_sizes ~dim =
 
 let unsafe_split_with_sizes_out ~out self ~split_sizes ~dim =
   stubs_unsafe_split_with_sizes_out
-    (CArray.of_list t out |> CArray.start)
+    (CArray.of_list gc_tensor out |> CArray.start)
     (List.length out)
     self
     (List.map Int64.of_int split_sizes |> CArray.of_list int64_t |> CArray.start)
@@ -31686,42 +19970,19 @@ let unsafe_split_with_sizes_out ~out self ~split_sizes ~dim =
     (Int64.of_int dim)
 ;;
 
-let unsqueeze self ~dim =
-  let out__ = CArray.make t 1 in
-  stubs_unsqueeze (CArray.start out__) self (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let unsqueeze_ self ~dim =
-  let out__ = CArray.make t 1 in
-  stubs_unsqueeze_ (CArray.start out__) self (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let unsqueeze self ~dim = stubs_unsqueeze self (Int64.of_int dim) |> with_tensor_gc
+let unsqueeze_ self ~dim = stubs_unsqueeze_ self (Int64.of_int dim) |> with_tensor_gc
 
 let unsqueeze_copy self ~dim =
-  let out__ = CArray.make t 1 in
-  stubs_unsqueeze_copy (CArray.start out__) self (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_unsqueeze_copy self (Int64.of_int dim) |> with_tensor_gc
 ;;
 
 let unsqueeze_copy_out ~out self ~dim =
-  let out__ = CArray.make t 1 in
-  stubs_unsqueeze_copy_out (CArray.start out__) out self (Int64.of_int dim);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_unsqueeze_copy_out out self (Int64.of_int dim) |> with_tensor_gc
 ;;
 
 let upsample_bicubic2d self ~output_size ~align_corners ~scales_h ~scales_w =
-  let out__ = CArray.make t 1 in
   stubs_upsample_bicubic2d
-    (CArray.start out__)
     self
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
     (List.length output_size)
@@ -31733,10 +19994,8 @@ let upsample_bicubic2d self ~output_size ~align_corners ~scales_h ~scales_w =
     (Option.value scales_w ~default:0.0)
     (match scales_w with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let upsample_bicubic2d_backward
@@ -31747,9 +20006,7 @@ let upsample_bicubic2d_backward
   ~scales_h
   ~scales_w
   =
-  let out__ = CArray.make t 1 in
   stubs_upsample_bicubic2d_backward
-    (CArray.start out__)
     grad_output
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
     (List.length output_size)
@@ -31763,10 +20020,8 @@ let upsample_bicubic2d_backward
     (Option.value scales_w ~default:0.0)
     (match scales_w with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let upsample_bicubic2d_backward_grad_input
@@ -31778,9 +20033,7 @@ let upsample_bicubic2d_backward_grad_input
   ~scales_h
   ~scales_w
   =
-  let out__ = CArray.make t 1 in
   stubs_upsample_bicubic2d_backward_grad_input
-    (CArray.start out__)
     grad_input
     grad_output
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
@@ -31795,16 +20048,12 @@ let upsample_bicubic2d_backward_grad_input
     (Option.value scales_w ~default:0.0)
     (match scales_w with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let upsample_bicubic2d_out ~out self ~output_size ~align_corners ~scales_h ~scales_w =
-  let out__ = CArray.make t 1 in
   stubs_upsample_bicubic2d_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
@@ -31817,16 +20066,12 @@ let upsample_bicubic2d_out ~out self ~output_size ~align_corners ~scales_h ~scal
     (Option.value scales_w ~default:0.0)
     (match scales_w with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let upsample_bicubic2d_vec input ~output_size ~align_corners ~scale_factors =
-  let out__ = CArray.make t 1 in
   stubs_upsample_bicubic2d_vec
-    (CArray.start out__)
     input
     (match output_size with
      | None -> from_voidp int64_t null
@@ -31836,16 +20081,12 @@ let upsample_bicubic2d_vec input ~output_size ~align_corners ~scale_factors =
      | Some v -> List.length v)
     (if align_corners then 1 else 0)
     (scale_factors |> CArray.of_list double |> CArray.start)
-    (List.length scale_factors);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length scale_factors)
+  |> with_tensor_gc
 ;;
 
 let upsample_bilinear2d self ~output_size ~align_corners ~scales_h ~scales_w =
-  let out__ = CArray.make t 1 in
   stubs_upsample_bilinear2d
-    (CArray.start out__)
     self
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
     (List.length output_size)
@@ -31857,10 +20098,8 @@ let upsample_bilinear2d self ~output_size ~align_corners ~scales_h ~scales_w =
     (Option.value scales_w ~default:0.0)
     (match scales_w with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let upsample_bilinear2d_backward
@@ -31871,9 +20110,7 @@ let upsample_bilinear2d_backward
   ~scales_h
   ~scales_w
   =
-  let out__ = CArray.make t 1 in
   stubs_upsample_bilinear2d_backward
-    (CArray.start out__)
     grad_output
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
     (List.length output_size)
@@ -31887,10 +20124,8 @@ let upsample_bilinear2d_backward
     (Option.value scales_w ~default:0.0)
     (match scales_w with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let upsample_bilinear2d_backward_grad_input
@@ -31902,9 +20137,7 @@ let upsample_bilinear2d_backward_grad_input
   ~scales_h
   ~scales_w
   =
-  let out__ = CArray.make t 1 in
   stubs_upsample_bilinear2d_backward_grad_input
-    (CArray.start out__)
     grad_input
     grad_output
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
@@ -31919,16 +20152,12 @@ let upsample_bilinear2d_backward_grad_input
     (Option.value scales_w ~default:0.0)
     (match scales_w with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let upsample_bilinear2d_out ~out self ~output_size ~align_corners ~scales_h ~scales_w =
-  let out__ = CArray.make t 1 in
   stubs_upsample_bilinear2d_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
@@ -31941,16 +20170,12 @@ let upsample_bilinear2d_out ~out self ~output_size ~align_corners ~scales_h ~sca
     (Option.value scales_w ~default:0.0)
     (match scales_w with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let upsample_bilinear2d_vec input ~output_size ~align_corners ~scale_factors =
-  let out__ = CArray.make t 1 in
   stubs_upsample_bilinear2d_vec
-    (CArray.start out__)
     input
     (match output_size with
      | None -> from_voidp int64_t null
@@ -31960,16 +20185,12 @@ let upsample_bilinear2d_vec input ~output_size ~align_corners ~scale_factors =
      | Some v -> List.length v)
     (if align_corners then 1 else 0)
     (scale_factors |> CArray.of_list double |> CArray.start)
-    (List.length scale_factors);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length scale_factors)
+  |> with_tensor_gc
 ;;
 
 let upsample_linear1d self ~output_size ~align_corners ~scales =
-  let out__ = CArray.make t 1 in
   stubs_upsample_linear1d
-    (CArray.start out__)
     self
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
     (List.length output_size)
@@ -31977,10 +20198,8 @@ let upsample_linear1d self ~output_size ~align_corners ~scales =
     (Option.value scales ~default:0.0)
     (match scales with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let upsample_linear1d_backward
@@ -31990,9 +20209,7 @@ let upsample_linear1d_backward
   ~align_corners
   ~scales
   =
-  let out__ = CArray.make t 1 in
   stubs_upsample_linear1d_backward
-    (CArray.start out__)
     grad_output
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
     (List.length output_size)
@@ -32002,10 +20219,8 @@ let upsample_linear1d_backward
     (Option.value scales ~default:0.0)
     (match scales with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let upsample_linear1d_backward_grad_input
@@ -32016,9 +20231,7 @@ let upsample_linear1d_backward_grad_input
   ~align_corners
   ~scales
   =
-  let out__ = CArray.make t 1 in
   stubs_upsample_linear1d_backward_grad_input
-    (CArray.start out__)
     grad_input
     grad_output
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
@@ -32029,16 +20242,12 @@ let upsample_linear1d_backward_grad_input
     (Option.value scales ~default:0.0)
     (match scales with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let upsample_linear1d_out ~out self ~output_size ~align_corners ~scales =
-  let out__ = CArray.make t 1 in
   stubs_upsample_linear1d_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
@@ -32047,16 +20256,12 @@ let upsample_linear1d_out ~out self ~output_size ~align_corners ~scales =
     (Option.value scales ~default:0.0)
     (match scales with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let upsample_linear1d_vec input ~output_size ~align_corners ~scale_factors =
-  let out__ = CArray.make t 1 in
   stubs_upsample_linear1d_vec
-    (CArray.start out__)
     input
     (match output_size with
      | None -> from_voidp int64_t null
@@ -32066,32 +20271,24 @@ let upsample_linear1d_vec input ~output_size ~align_corners ~scale_factors =
      | Some v -> List.length v)
     (if align_corners then 1 else 0)
     (scale_factors |> CArray.of_list double |> CArray.start)
-    (List.length scale_factors);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length scale_factors)
+  |> with_tensor_gc
 ;;
 
 let upsample_nearest1d self ~output_size ~scales =
-  let out__ = CArray.make t 1 in
   stubs_upsample_nearest1d
-    (CArray.start out__)
     self
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
     (List.length output_size)
     (Option.value scales ~default:0.0)
     (match scales with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let upsample_nearest1d_backward ~grad_output ~output_size ~input_size ~scales =
-  let out__ = CArray.make t 1 in
   stubs_upsample_nearest1d_backward
-    (CArray.start out__)
     grad_output
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
     (List.length output_size)
@@ -32100,10 +20297,8 @@ let upsample_nearest1d_backward ~grad_output ~output_size ~input_size ~scales =
     (Option.value scales ~default:0.0)
     (match scales with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let upsample_nearest1d_backward_grad_input
@@ -32113,9 +20308,7 @@ let upsample_nearest1d_backward_grad_input
   ~input_size
   ~scales
   =
-  let out__ = CArray.make t 1 in
   stubs_upsample_nearest1d_backward_grad_input
-    (CArray.start out__)
     grad_input
     grad_output
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
@@ -32125,16 +20318,12 @@ let upsample_nearest1d_backward_grad_input
     (Option.value scales ~default:0.0)
     (match scales with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let upsample_nearest1d_out ~out self ~output_size ~scales =
-  let out__ = CArray.make t 1 in
   stubs_upsample_nearest1d_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
@@ -32142,16 +20331,12 @@ let upsample_nearest1d_out ~out self ~output_size ~scales =
     (Option.value scales ~default:0.0)
     (match scales with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let upsample_nearest1d_vec input ~output_size ~scale_factors =
-  let out__ = CArray.make t 1 in
   stubs_upsample_nearest1d_vec
-    (CArray.start out__)
     input
     (match output_size with
      | None -> from_voidp int64_t null
@@ -32160,16 +20345,12 @@ let upsample_nearest1d_vec input ~output_size ~scale_factors =
      | None -> -1
      | Some v -> List.length v)
     (scale_factors |> CArray.of_list double |> CArray.start)
-    (List.length scale_factors);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length scale_factors)
+  |> with_tensor_gc
 ;;
 
 let upsample_nearest2d self ~output_size ~scales_h ~scales_w =
-  let out__ = CArray.make t 1 in
   stubs_upsample_nearest2d
-    (CArray.start out__)
     self
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
     (List.length output_size)
@@ -32180,16 +20361,12 @@ let upsample_nearest2d self ~output_size ~scales_h ~scales_w =
     (Option.value scales_w ~default:0.0)
     (match scales_w with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let upsample_nearest2d_backward ~grad_output ~output_size ~input_size ~scales_h ~scales_w =
-  let out__ = CArray.make t 1 in
   stubs_upsample_nearest2d_backward
-    (CArray.start out__)
     grad_output
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
     (List.length output_size)
@@ -32202,10 +20379,8 @@ let upsample_nearest2d_backward ~grad_output ~output_size ~input_size ~scales_h
     (Option.value scales_w ~default:0.0)
     (match scales_w with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let upsample_nearest2d_backward_grad_input
@@ -32216,9 +20391,7 @@ let upsample_nearest2d_backward_grad_input
   ~scales_h
   ~scales_w
   =
-  let out__ = CArray.make t 1 in
   stubs_upsample_nearest2d_backward_grad_input
-    (CArray.start out__)
     grad_input
     grad_output
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
@@ -32232,16 +20405,12 @@ let upsample_nearest2d_backward_grad_input
     (Option.value scales_w ~default:0.0)
     (match scales_w with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let upsample_nearest2d_out ~out self ~output_size ~scales_h ~scales_w =
-  let out__ = CArray.make t 1 in
   stubs_upsample_nearest2d_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
@@ -32253,16 +20422,12 @@ let upsample_nearest2d_out ~out self ~output_size ~scales_h ~scales_w =
     (Option.value scales_w ~default:0.0)
     (match scales_w with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let upsample_nearest2d_vec input ~output_size ~scale_factors =
-  let out__ = CArray.make t 1 in
   stubs_upsample_nearest2d_vec
-    (CArray.start out__)
     input
     (match output_size with
      | None -> from_voidp int64_t null
@@ -32271,16 +20436,12 @@ let upsample_nearest2d_vec input ~output_size ~scale_factors =
      | None -> -1
      | Some v -> List.length v)
     (scale_factors |> CArray.of_list double |> CArray.start)
-    (List.length scale_factors);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length scale_factors)
+  |> with_tensor_gc
 ;;
 
 let upsample_nearest3d self ~output_size ~scales_d ~scales_h ~scales_w =
-  let out__ = CArray.make t 1 in
   stubs_upsample_nearest3d
-    (CArray.start out__)
     self
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
     (List.length output_size)
@@ -32295,10 +20456,8 @@ let upsample_nearest3d self ~output_size ~scales_d ~scales_h ~scales_w =
     (Option.value scales_w ~default:0.0)
     (match scales_w with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let upsample_nearest3d_backward
@@ -32309,9 +20468,7 @@ let upsample_nearest3d_backward
   ~scales_h
   ~scales_w
   =
-  let out__ = CArray.make t 1 in
   stubs_upsample_nearest3d_backward
-    (CArray.start out__)
     grad_output
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
     (List.length output_size)
@@ -32328,10 +20485,8 @@ let upsample_nearest3d_backward
     (Option.value scales_w ~default:0.0)
     (match scales_w with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let upsample_nearest3d_backward_grad_input
@@ -32343,9 +20498,7 @@ let upsample_nearest3d_backward_grad_input
   ~scales_h
   ~scales_w
   =
-  let out__ = CArray.make t 1 in
   stubs_upsample_nearest3d_backward_grad_input
-    (CArray.start out__)
     grad_input
     grad_output
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
@@ -32363,16 +20516,12 @@ let upsample_nearest3d_backward_grad_input
     (Option.value scales_w ~default:0.0)
     (match scales_w with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let upsample_nearest3d_out ~out self ~output_size ~scales_d ~scales_h ~scales_w =
-  let out__ = CArray.make t 1 in
   stubs_upsample_nearest3d_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
@@ -32388,16 +20537,12 @@ let upsample_nearest3d_out ~out self ~output_size ~scales_d ~scales_h ~scales_w
     (Option.value scales_w ~default:0.0)
     (match scales_w with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let upsample_nearest3d_vec input ~output_size ~scale_factors =
-  let out__ = CArray.make t 1 in
   stubs_upsample_nearest3d_vec
-    (CArray.start out__)
     input
     (match output_size with
      | None -> from_voidp int64_t null
@@ -32406,16 +20551,12 @@ let upsample_nearest3d_vec input ~output_size ~scale_factors =
      | None -> -1
      | Some v -> List.length v)
     (scale_factors |> CArray.of_list double |> CArray.start)
-    (List.length scale_factors);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length scale_factors)
+  |> with_tensor_gc
 ;;
 
 let upsample_trilinear3d self ~output_size ~align_corners ~scales_d ~scales_h ~scales_w =
-  let out__ = CArray.make t 1 in
   stubs_upsample_trilinear3d
-    (CArray.start out__)
     self
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
     (List.length output_size)
@@ -32431,10 +20572,8 @@ let upsample_trilinear3d self ~output_size ~align_corners ~scales_d ~scales_h ~s
     (Option.value scales_w ~default:0.0)
     (match scales_w with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let upsample_trilinear3d_backward
@@ -32446,9 +20585,7 @@ let upsample_trilinear3d_backward
   ~scales_h
   ~scales_w
   =
-  let out__ = CArray.make t 1 in
   stubs_upsample_trilinear3d_backward
-    (CArray.start out__)
     grad_output
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
     (List.length output_size)
@@ -32466,10 +20603,8 @@ let upsample_trilinear3d_backward
     (Option.value scales_w ~default:0.0)
     (match scales_w with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let upsample_trilinear3d_backward_grad_input
@@ -32482,9 +20617,7 @@ let upsample_trilinear3d_backward_grad_input
   ~scales_h
   ~scales_w
   =
-  let out__ = CArray.make t 1 in
   stubs_upsample_trilinear3d_backward_grad_input
-    (CArray.start out__)
     grad_input
     grad_output
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
@@ -32503,10 +20636,8 @@ let upsample_trilinear3d_backward_grad_input
     (Option.value scales_w ~default:0.0)
     (match scales_w with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let upsample_trilinear3d_out
@@ -32518,9 +20649,7 @@ let upsample_trilinear3d_out
   ~scales_h
   ~scales_w
   =
-  let out__ = CArray.make t 1 in
   stubs_upsample_trilinear3d_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int output_size |> CArray.of_list int64_t |> CArray.start)
@@ -32537,16 +20666,12 @@ let upsample_trilinear3d_out
     (Option.value scales_w ~default:0.0)
     (match scales_w with
      | Some _ -> 0
-     | None -> 1);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+     | None -> 1)
+  |> with_tensor_gc
 ;;
 
 let upsample_trilinear3d_vec input ~output_size ~align_corners ~scale_factors =
-  let out__ = CArray.make t 1 in
   stubs_upsample_trilinear3d_vec
-    (CArray.start out__)
     input
     (match output_size with
      | None -> from_voidp int64_t null
@@ -32556,55 +20681,27 @@ let upsample_trilinear3d_vec input ~output_size ~align_corners ~scale_factors =
      | Some v -> List.length v)
     (if align_corners then 1 else 0)
     (scale_factors |> CArray.of_list double |> CArray.start)
-    (List.length scale_factors);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length scale_factors)
+  |> with_tensor_gc
 ;;
 
 let value_selecting_reduction_backward ~grad ~dim ~indices ~sizes ~keepdim =
-  let out__ = CArray.make t 1 in
   stubs_value_selecting_reduction_backward
-    (CArray.start out__)
     grad
     (Int64.of_int dim)
     indices
     (List.map Int64.of_int sizes |> CArray.of_list int64_t |> CArray.start)
     (List.length sizes)
-    (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let values self =
-  let out__ = CArray.make t 1 in
-  stubs_values (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let values_copy self =
-  let out__ = CArray.make t 1 in
-  stubs_values_copy (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if keepdim then 1 else 0)
+  |> with_tensor_gc
 ;;
 
-let values_copy_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_values_copy_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let values self = stubs_values self |> with_tensor_gc
+let values_copy self = stubs_values_copy self |> with_tensor_gc
+let values_copy_out ~out self = stubs_values_copy_out out self |> with_tensor_gc
 
 let vander ~x ~n ~increasing =
-  let out__ = CArray.make t 1 in
   stubs_vander
-    (CArray.start out__)
     x
     (match n with
      | None -> Int64.zero
@@ -32612,24 +20709,14 @@ let vander ~x ~n ~increasing =
     (match n with
      | Some _ -> 0
      | None -> 1)
-    (if increasing then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if increasing then 1 else 0)
+  |> with_tensor_gc
 ;;
 
-let var self ~unbiased =
-  let out__ = CArray.make t 1 in
-  stubs_var (CArray.start out__) self (if unbiased then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let var self ~unbiased = stubs_var self (if unbiased then 1 else 0) |> with_tensor_gc
 
 let var_correction self ~dim ~correction ~keepdim =
-  let out__ = CArray.make t 1 in
   stubs_var_correction
-    (CArray.start out__)
     self
     (match dim with
      | None -> from_voidp int64_t null
@@ -32638,16 +20725,12 @@ let var_correction self ~dim ~correction ~keepdim =
      | None -> -1
      | Some v -> List.length v)
     correction
-    (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if keepdim then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let var_correction_out ~out self ~dim ~correction ~keepdim =
-  let out__ = CArray.make t 1 in
   stubs_var_correction_out
-    (CArray.start out__)
     out
     self
     (match dim with
@@ -32657,16 +20740,12 @@ let var_correction_out ~out self ~dim ~correction ~keepdim =
      | None -> -1
      | Some v -> List.length v)
     correction
-    (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if keepdim then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let var_dim self ~dim ~unbiased ~keepdim =
-  let out__ = CArray.make t 1 in
   stubs_var_dim
-    (CArray.start out__)
     self
     (match dim with
      | None -> from_voidp int64_t null
@@ -32675,24 +20754,20 @@ let var_dim self ~dim ~unbiased ~keepdim =
      | None -> -1
      | Some v -> List.length v)
     (if unbiased then 1 else 0)
-    (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if keepdim then 1 else 0)
+  |> with_tensor_gc
 ;;
 
 let var_mean self ~unbiased =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_var_mean (CArray.start out__) self (if unbiased then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let var_mean_correction self ~dim ~correction ~keepdim =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_var_mean_correction
     (CArray.start out__)
     self
@@ -32704,15 +20779,13 @@ let var_mean_correction self ~dim ~correction ~keepdim =
      | Some v -> List.length v)
     correction
     (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let var_mean_correction_out ~out0 ~out1 self ~dim ~correction ~keepdim =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_var_mean_correction_out
     (CArray.start out__)
     out0
@@ -32726,15 +20799,13 @@ let var_mean_correction_out ~out0 ~out1 self ~dim ~correction ~keepdim =
      | Some v -> List.length v)
     correction
     (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let var_mean_dim self ~dim ~unbiased ~keepdim =
-  let out__ = CArray.make t 2 in
+  let out__ = CArray.make raw_tensor 2 in
   stubs_var_mean_dim
     (CArray.start out__)
     self
@@ -32746,17 +20817,13 @@ let var_mean_dim self ~dim ~unbiased ~keepdim =
      | Some v -> List.length v)
     (if unbiased then 1 else 0)
     (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  let t1 = CArray.get out__ 1 in
-  Gc.finalise C.Tensor.free t1;
+  let t0 = CArray.get out__ 0 |> with_tensor_gc in
+  let t1 = CArray.get out__ 1 |> with_tensor_gc in
   t0, t1
 ;;
 
 let var_out ~out self ~dim ~unbiased ~keepdim =
-  let out__ = CArray.make t 1 in
   stubs_var_out
-    (CArray.start out__)
     out
     self
     (match dim with
@@ -32766,143 +20833,63 @@ let var_out ~out self ~dim ~unbiased ~keepdim =
      | None -> -1
      | Some v -> List.length v)
     (if unbiased then 1 else 0)
-    (if keepdim then 1 else 0);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let vdot self other =
-  let out__ = CArray.make t 1 in
-  stubs_vdot (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (if keepdim then 1 else 0)
+  |> with_tensor_gc
 ;;
 
-let vdot_out ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_vdot_out (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let vdot self other = stubs_vdot self other |> with_tensor_gc
+let vdot_out ~out self other = stubs_vdot_out out self other |> with_tensor_gc
 
 let view self ~size =
-  let out__ = CArray.make t 1 in
   stubs_view
-    (CArray.start out__)
     self
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
-    (List.length size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let view_as self other =
-  let out__ = CArray.make t 1 in
-  stubs_view_as (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let view_as_complex self =
-  let out__ = CArray.make t 1 in
-  stubs_view_as_complex (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length size)
+  |> with_tensor_gc
 ;;
 
-let view_as_complex_copy self =
-  let out__ = CArray.make t 1 in
-  stubs_view_as_complex_copy (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let view_as self other = stubs_view_as self other |> with_tensor_gc
+let view_as_complex self = stubs_view_as_complex self |> with_tensor_gc
+let view_as_complex_copy self = stubs_view_as_complex_copy self |> with_tensor_gc
 
 let view_as_complex_copy_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_view_as_complex_copy_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_view_as_complex_copy_out out self |> with_tensor_gc
 ;;
 
-let view_as_real self =
-  let out__ = CArray.make t 1 in
-  stubs_view_as_real (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let view_as_real_copy self =
-  let out__ = CArray.make t 1 in
-  stubs_view_as_real_copy (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let view_as_real self = stubs_view_as_real self |> with_tensor_gc
+let view_as_real_copy self = stubs_view_as_real_copy self |> with_tensor_gc
 
 let view_as_real_copy_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_view_as_real_copy_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_view_as_real_copy_out out self |> with_tensor_gc
 ;;
 
 let view_copy self ~size =
-  let out__ = CArray.make t 1 in
   stubs_view_copy
-    (CArray.start out__)
     self
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
-    (List.length size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length size)
+  |> with_tensor_gc
 ;;
 
 let view_copy_dtype self ~dtype =
-  let out__ = CArray.make t 1 in
-  stubs_view_copy_dtype (CArray.start out__) self (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_view_copy_dtype self (Kind.packed_to_int dtype) |> with_tensor_gc
 ;;
 
 let view_copy_dtype_out ~out self ~dtype =
-  let out__ = CArray.make t 1 in
-  stubs_view_copy_dtype_out (CArray.start out__) out self (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_view_copy_dtype_out out self (Kind.packed_to_int dtype) |> with_tensor_gc
 ;;
 
 let view_copy_out ~out self ~size =
-  let out__ = CArray.make t 1 in
   stubs_view_copy_out
-    (CArray.start out__)
     out
     self
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
-    (List.length size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length size)
+  |> with_tensor_gc
 ;;
 
 let view_dtype self ~dtype =
-  let out__ = CArray.make t 1 in
-  stubs_view_dtype (CArray.start out__) self (Kind.packed_to_int dtype);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_view_dtype self (Kind.packed_to_int dtype) |> with_tensor_gc
 ;;
 
 let vsplit self ~sections = stubs_vsplit self (Int64.of_int sections) |> to_tensor_list
@@ -32916,195 +20903,82 @@ let vsplit_array self ~indices =
 ;;
 
 let vstack tensors =
-  let out__ = CArray.make t 1 in
-  stubs_vstack
-    (CArray.start out__)
-    (CArray.of_list t tensors |> CArray.start)
-    (List.length tensors);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_vstack (CArray.of_list gc_tensor tensors |> CArray.start) (List.length tensors)
+  |> with_tensor_gc
 ;;
 
 let vstack_out ~out tensors =
-  let out__ = CArray.make t 1 in
   stubs_vstack_out
-    (CArray.start out__)
     out
-    (CArray.of_list t tensors |> CArray.start)
-    (List.length tensors);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (CArray.of_list gc_tensor tensors |> CArray.start)
+    (List.length tensors)
+  |> with_tensor_gc
 ;;
 
 let where ~condition = stubs_where condition |> to_tensor_list
 
 let where_scalar ~condition self other =
-  let out__ = CArray.make t 1 in
-  stubs_where_scalar (CArray.start out__) condition self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_where_scalar condition self other |> with_tensor_gc
 ;;
 
 let where_scalarother ~condition self other =
-  let out__ = CArray.make t 1 in
-  stubs_where_scalarother (CArray.start out__) condition self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_where_scalarother condition self other |> with_tensor_gc
 ;;
 
 let where_scalarself ~condition self other =
-  let out__ = CArray.make t 1 in
-  stubs_where_scalarself (CArray.start out__) condition self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_where_scalarself condition self other |> with_tensor_gc
 ;;
 
 let where_self ~condition self other =
-  let out__ = CArray.make t 1 in
-  stubs_where_self (CArray.start out__) condition self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_where_self condition self other |> with_tensor_gc
 ;;
 
 let where_self_out ~out ~condition self other =
-  let out__ = CArray.make t 1 in
-  stubs_where_self_out (CArray.start out__) out condition self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let xlogy self other =
-  let out__ = CArray.make t 1 in
-  stubs_xlogy (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_where_self_out out condition self other |> with_tensor_gc
 ;;
 
-let xlogy_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_xlogy_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let xlogy self other = stubs_xlogy self other |> with_tensor_gc
+let xlogy_ self other = stubs_xlogy_ self other |> with_tensor_gc
 
 let xlogy_outscalar_other ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_xlogy_outscalar_other (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_xlogy_outscalar_other out self other |> with_tensor_gc
 ;;
 
 let xlogy_outscalar_self ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_xlogy_outscalar_self (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_xlogy_outscalar_self out self other |> with_tensor_gc
 ;;
 
 let xlogy_outtensor ~out self other =
-  let out__ = CArray.make t 1 in
-  stubs_xlogy_outtensor (CArray.start out__) out self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_xlogy_outtensor out self other |> with_tensor_gc
 ;;
 
-let xlogy_scalar_other self other =
-  let out__ = CArray.make t 1 in
-  stubs_xlogy_scalar_other (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let xlogy_scalar_other self other = stubs_xlogy_scalar_other self other |> with_tensor_gc
 
 let xlogy_scalar_other_ self other =
-  let out__ = CArray.make t 1 in
-  stubs_xlogy_scalar_other_ (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let xlogy_scalar_self self other =
-  let out__ = CArray.make t 1 in
-  stubs_xlogy_scalar_self (CArray.start out__) self other;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let zero self =
-  let out__ = CArray.make t 1 in
-  stubs_zero (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let zero_ self =
-  let out__ = CArray.make t 1 in
-  stubs_zero_ (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+  stubs_xlogy_scalar_other_ self other |> with_tensor_gc
 ;;
 
-let zero_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_zero_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let xlogy_scalar_self self other = stubs_xlogy_scalar_self self other |> with_tensor_gc
+let zero self = stubs_zero self |> with_tensor_gc
+let zero_ self = stubs_zero_ self |> with_tensor_gc
+let zero_out ~out self = stubs_zero_out out self |> with_tensor_gc
 
 let zeros ~size ~options =
-  let out__ = CArray.make t 1 in
   stubs_zeros
-    (CArray.start out__)
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
     (List.length size)
     (Kind.packed_to_int (fst options))
-    (Device.to_int (snd options));
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
-
-let zeros_like self =
-  let out__ = CArray.make t 1 in
-  stubs_zeros_like (CArray.start out__) self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (Device.to_int (snd options))
+  |> with_tensor_gc
 ;;
 
-let zeros_like_out ~out self =
-  let out__ = CArray.make t 1 in
-  stubs_zeros_like_out (CArray.start out__) out self;
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
-;;
+let zeros_like self = stubs_zeros_like self |> with_tensor_gc
+let zeros_like_out ~out self = stubs_zeros_like_out out self |> with_tensor_gc
 
 let zeros_out ~out ~size =
-  let out__ = CArray.make t 1 in
   stubs_zeros_out
-    (CArray.start out__)
     out
     (List.map Int64.of_int size |> CArray.of_list int64_t |> CArray.start)
-    (List.length size);
-  let t0 = CArray.get out__ 0 in
-  Gc.finalise C.Tensor.free t0;
-  t0
+    (List.length size)
+  |> with_tensor_gc
 ;;
diff --git a/third_party/pytorch/dune b/third_party/pytorch/dune
new file mode 100644
index 0000000..e69de29
diff --git a/torch.opam b/torch.opam
index 7ae43ef..d62c920 100644
--- a/torch.opam
+++ b/torch.opam
@@ -16,6 +16,7 @@ depends: [
   "ppx_bench"
   "ppx_inline_test"
   "ppx_jane"
+  "ppx_string"
   "stdio"
   "ctypes"              {>= "0.18.0"}
   "ctypes-foreign"