Merge branch 'master' of github.com:guptaaryan16/ignite

.github/workflows/binaries-nightly-release.yml

@@ -38,7 +38,7 @@ jobs:
           chmod +x ./conda.recipe/build_and_upload.sh
           ./conda.recipe/build_and_upload.sh
 
-      - name: Build and Publish PyPi binaries
+      - name: Build and Publish PyPI binaries
         shell: bash -l {0}
         run: |
           # workaround to fix https://github.com/pytorch/ignite/issues/2373

.github/workflows/docker-build.yml

@@ -5,12 +5,8 @@ on:
     paths:
       - docker/**
       - ".github/workflows/docker-build.yml"
-  push:
-    branches:
-      - master
-    paths:
-      - docker/**
-      - ".github/workflows/docker-build.yml"
+  release:
+    types: [published]
   workflow_dispatch:
 
 
@@ -32,11 +28,13 @@ jobs:
     steps:
       - uses: actions/checkout@v4
       - name: Changed Files Exporter
+        if: github.event_name == 'pull_request'
         id: files
         uses: umani/[email protected]
         with:
           repo-token: ${{ secrets.GITHUB_TOKEN }}
       - name: Get a list of modified files
+        if: github.event_name == 'pull_request'
         run: echo "modified=${{ steps.files.outputs.files_created }} ${{ steps.files.outputs.files_updated }}" >> $GITHUB_ENV
       - name: Set outputs
         id: set-modified
@@ -52,7 +50,7 @@ jobs:
   build-hvd:
     name: Build all Horovod flavoured PyTorch-Ignite images
     needs: setup
-    if: contains(needs.setup.outputs.modified, 'hvd/') || contains(needs.setup.outputs.modified, 'docker.cfg')
+    if: github.event_name != 'pull_request' || (contains(needs.setup.outputs.modified, 'hvd/') || contains(needs.setup.outputs.modified, 'docker.cfg'))
 
     # Cherry-picked from https://github.com/pytorch/test-infra/blob/main/.github/workflows/linux_job.yml
     env:
@@ -98,7 +96,7 @@ jobs:
           bash build.sh hvd hvd-nlp
 
       - name: Publish docker images
-        if: ${{ (github.ref == 'refs/heads/master' && github.event_name == 'push') || github.event_name == 'release' }}
+        if: github.event_name == 'workflow_dispatch' || github.event_name == 'release'
         env:
           DOCKER_TOKEN: ${{ secrets.DOCKER_TOKEN }}
           DOCKER_USER: ${{ secrets.DOCKER_USER }}
@@ -112,7 +110,7 @@ jobs:
   build-hvd-apex:
     name: Build all Horovod + Apex flavoured PyTorch-Ignite images
     needs: setup
-    if: contains(needs.setup.outputs.modified, 'hvd/') || contains(needs.setup.outputs.modified, 'docker.cfg')
+    if: github.event_name != 'pull_request' || (contains(needs.setup.outputs.modified, 'hvd/') || contains(needs.setup.outputs.modified, 'docker.cfg'))
 
     # Cherry-picked from https://github.com/pytorch/test-infra/blob/main/.github/workflows/linux_job.yml
     env:
@@ -158,7 +156,7 @@ jobs:
           bash build.sh hvd hvd-apex-nlp
 
       - name: Publish docker images
-        if: ${{ (github.ref == 'refs/heads/master' && github.event_name == 'push') || github.event_name == 'release' }}
+        if: github.event_name == 'workflow_dispatch' || github.event_name == 'release'
         env:
           DOCKER_TOKEN: ${{ secrets.DOCKER_TOKEN }}
           DOCKER_USER: ${{ secrets.DOCKER_USER }}
@@ -172,7 +170,7 @@ jobs:
   build-main:
     name: Build all PyTorch-Ignite images
     needs: setup
-    if: contains(needs.setup.outputs.modified, 'main/') || contains(needs.setup.outputs.modified, 'docker.cfg')
+    if: github.event_name != 'pull_request' || (contains(needs.setup.outputs.modified, 'main/') || contains(needs.setup.outputs.modified, 'docker.cfg'))
 
     # Cherry-picked from https://github.com/pytorch/test-infra/blob/main/.github/workflows/linux_job.yml
     env:
@@ -218,7 +216,7 @@ jobs:
           bash build.sh main nlp
 
       - name: Publish docker images
-        if: ${{ (github.ref == 'refs/heads/master' && github.event_name == 'push') || github.event_name == 'release' }}
+        if: github.event_name == 'workflow_dispatch' || github.event_name == 'release'
         env:
           DOCKER_TOKEN: ${{ secrets.DOCKER_TOKEN }}
           DOCKER_USER: ${{ secrets.DOCKER_USER }}
@@ -232,7 +230,7 @@ jobs:
   build-main-apex:
     name: Build all PyTorch-Ignite images with Apex
     needs: setup
-    if: contains(needs.setup.outputs.modified, 'main/') || contains(needs.setup.outputs.modified, 'docker.cfg')
+    if: github.event_name != 'pull_request' || (contains(needs.setup.outputs.modified, 'main/') || contains(needs.setup.outputs.modified, 'docker.cfg'))
 
     # Cherry-picked from https://github.com/pytorch/test-infra/blob/main/.github/workflows/linux_job.yml
     env:
@@ -278,7 +276,7 @@ jobs:
           bash build.sh main apex-nlp
 
       - name: Publish docker images
-        if: ${{ (github.ref == 'refs/heads/master' && github.event_name == 'push') || github.event_name == 'release' }}
+        if: github.event_name == 'workflow_dispatch' || github.event_name == 'release'
         env:
           DOCKER_TOKEN: ${{ secrets.DOCKER_TOKEN }}
           DOCKER_USER: ${{ secrets.DOCKER_USER }}

.github/workflows/mps-tests.yml

@@ -0,0 +1,135 @@
+name: Run unit tests on M1
+on:
+  push:
+    branches:
+      - master
+      - "*.*.*"
+    paths:
+      - "ignite/**"
+      - "tests/ignite/**"
+      - "tests/run_code_style.sh"
+      - "examples/**.py"
+      - "requirements-dev.txt"
+      - ".github/workflows/mps-tests.yml"
+  pull_request:
+    paths:
+      - "ignite/**"
+      - "tests/ignite/**"
+      - "tests/run_code_style.sh"
+      - "examples/**.py"
+      - "requirements-dev.txt"
+      - ".github/workflows/mps-tests.yml"
+  workflow_dispatch:
+
+concurrency:
+  # <workflow_name>-<branch_name>-<true || commit_sha (if branch is protected)>
+  group: mps-tests-${{ github.ref_name }}-${{ !(github.ref_protected) || github.sha }}
+  cancel-in-progress: true
+
+# Cherry-picked from 
+# - https://github.com/pytorch/vision/blob/main/.github/workflows/tests.yml
+# - https://github.com/pytorch/test-infra/blob/main/.github/workflows/macos_job.yml
+
+jobs:
+  mps-tests:
+    strategy:
+      matrix:
+        python-version: [3.8]
+        pytorch-channel: ["pytorch"]
+        skip-distrib-tests: [1]
+      fail-fast: false
+    runs-on: ["macos-m1-12"]
+    timeout-minutes: 60
+
+    steps:
+      - name: Clean workspace
+        run: |
+          echo "::group::Cleanup debug output"
+          sudo rm -rfv "${GITHUB_WORKSPACE}"
+          mkdir -p "${GITHUB_WORKSPACE}"
+          echo "::endgroup::"
+
+      - name: Checkout repository (pytorch/test-infra)
+        uses: actions/checkout@v3
+        with:
+          # Support the use case where we need to checkout someone's fork
+          repository: pytorch/test-infra
+          path: test-infra
+
+      - name: Checkout repository (${{ github.repository }})
+        uses: actions/checkout@v3
+        with:
+          # Support the use case where we need to checkout someone's fork
+          repository: ${{ github.repository }}
+          ref: ${{ github.ref }}
+          path: ${{ github.repository }}
+          fetch-depth: 1
+
+      - name: Setup miniconda
+        uses: ./test-infra/.github/actions/setup-miniconda
+        with:
+          python-version: ${{ matrix.python-version }}
+
+      - name: Install PyTorch
+        if: ${{ matrix.pytorch-channel == 'pytorch' }}
+        shell: bash -l {0}
+        run: |
+          conda shell.bash hook
+          conda activate $CONDA_ENV
+          pip install torch torchvision
+
+      - name: Install PyTorch (nightly)
+        if: ${{ matrix.pytorch-channel == 'pytorch-nightly' }}
+        shell: bash -l {0}
+        run: |
+          conda shell.bash hook
+          conda activate $CONDA_ENV
+          pip install --pre torch torchvision --index-url https://download.pytorch.org/whl/nightly/cpu
+
+      - name: Install dependencies
+        shell: bash -l {0}
+        working-directory: ${{ github.repository }}
+        run: |
+          conda activate $CONDA_ENV
+          # TODO: We add set -xe to explicitly fail the CI if one of the commands is failing.
+          # Somehow the step is passing even if a subcommand failed
+          set -xe
+          pip install -r requirements-dev.txt
+          echo "1 returned code: $?"
+          pip install -e .
+          echo "2 returned code: $?"
+          pip list
+          echo "3 returned code: $?"
+
+      # Download MNIST: https://github.com/pytorch/ignite/issues/1737
+      # to "/tmp" for unit tests
+      - name: Download MNIST
+        uses: pytorch-ignite/download-mnist-github-action@master
+        with:
+          target_dir: /tmp
+
+      # Copy MNIST to "." for the examples
+      - name: Copy MNIST
+        run: |
+          cp -R /tmp/MNIST .
+
+      - name: Run Tests
+        shell: bash -l {0}
+        working-directory: ${{ github.repository }}
+        run: |
+          conda activate $CONDA_ENV
+          SKIP_DISTRIB_TESTS=${{ matrix.skip-distrib-tests }} bash tests/run_cpu_tests.sh
+
+      - name: Upload coverage to Codecov
+        uses: codecov/codecov-action@v3
+        with:
+          file: ${{ github.repository }}/coverage.xml
+          flags: mps
+          fail_ci_if_error: false
+
+      - name: Run MNIST Examples
+        shell: bash -l {0}
+        working-directory: ${{ github.repository }}
+        run: |
+          conda activate $CONDA_ENV
+          python examples/mnist/mnist.py --epochs=1

.github/workflows/stable-release-pypi.yml

@@ -21,23 +21,15 @@ jobs:
         run: |
           conda install -y pytorch torchvision cpuonly -c pytorch
           pip install -r requirements-dev.txt
-          pip install --upgrade --no-cache-dir twine
-          python setup.py install
 
       - name: Build and Publish PyPI binaries
         shell: bash -l {0}
         run: |
+          # workaround to fix https://github.com/pytorch/ignite/issues/2373
+          pip uninstall -y twine pkginfo
+          pip install --upgrade --no-cache-dir twine 'pkginfo>=1.8.2'
           python setup.py sdist bdist_wheel
+          twine --version
           twine check dist/*
           TWINE_USERNAME="${{ secrets.PYPI_USER }}" TWINE_PASSWORD="${{ secrets.PYPI_TOKEN }}" twine upload --verbose dist/*
 
-  # docker-build-publish:
-  #   name: Trigger Build and Push Docker images to Docker Hub
-  #   needs: build-publish
-  #   runs-on: ubuntu-latest
-
-  #   steps:
-  #     - uses: actions/checkout@v4
-  #     - uses: actions/setup-python@v4
-  #       with:
-  #         python-version: "3.10"

.github/workflows/unit-tests.yml

@@ -93,6 +93,7 @@ jobs:
         run: |
           pip install -r requirements-dev.txt
           python setup.py install
+          pip list
 
       - name: Check code formatting
         run: |

docker/push_all.sh

@@ -27,7 +27,7 @@ echo $DOCKER_TOKEN | docker login --username=$DOCKER_USER --password-stdin
 set -xeu
 
 
-if [ ${folder_name} == "all" ]; then
+if [ ${push_selected_image} == "all" ]; then
 
     image_name="base"
     image_tag=`docker run --rm -i pytorchignite/${image_name}:latest python -c "import torch; import ignite; print(torch.__version__ + \"-\" + ignite.__version__, end=\"\")"`

docker/test_image.py

@@ -19,7 +19,9 @@ def check_package(package_name, expected_version=None):
     if expected_version is not None:
         assert hasattr(mod, "__version__"), f"Imported package {package_name} does not have __version__ attribute"
         version = mod.__version__
-        assert version == expected_version, f"Version mismatch for package {package_name}: got {version} but expected {expected_version}"
+        assert (
+            version == expected_version
+        ), f"Version mismatch for package {package_name}: got {version} but expected {expected_version}"
 
 
 if __name__ == "__main__":

examples/super_resolution/README.md

@@ -2,7 +2,7 @@
 
 ported from [pytorch-examples](https://github.com/pytorch/examples/tree/main/super_resolution)
 
-This example illustrates how to use the efficient sub-pixel convolution layer described in ["Real-Time Single Image and Video Super-Resolution Using an Efficient Sub-Pixel Convolutional Neural Network" - Shi et al.](https://arxiv.org/abs/1609.05158) for increasing spatial resolution within your network for tasks such as superresolution.
+This example illustrates how to use the efficient sub-pixel convolution layer described in ["Real-Time Single Image and Video Super-Resolution Using an Efficient Sub-Pixel Convolutional Neural Network" - Shi et al. 2016](https://arxiv.org/abs/1609.05158) for increasing spatial resolution within your network for tasks such as superresolution.
 
 ```
 usage: main.py [-h] --upscale_factor UPSCALE_FACTOR [--crop_size CROPSIZE] [--batch_size BATCHSIZE]

ignite/contrib/engines/common.py

@@ -167,9 +167,7 @@ def _setup_common_training_handlers(
 
     if lr_scheduler is not None:
         if isinstance(lr_scheduler, PyTorchLRScheduler):
-            trainer.add_event_handler(
-                Events.ITERATION_COMPLETED, lambda engine: cast(PyTorchLRScheduler, lr_scheduler).step()
-            )
+            trainer.add_event_handler(Events.ITERATION_COMPLETED, lambda engine: lr_scheduler.step())
         else:
             trainer.add_event_handler(Events.ITERATION_STARTED, lr_scheduler)
 

ignite/distributed/comp_models/base.py

@@ -3,6 +3,9 @@
 from typing import Any, Callable, cast, List, Optional, Union
 
 import torch
+from packaging.version import Version
+
+_torch_version_le_112 = Version(torch.__version__) > Version("1.12.0")
 
 
 class ComputationModel(metaclass=ABCMeta):
@@ -326,7 +329,7 @@ def get_node_rank(self) -> int:
     def device(self) -> torch.device:
         if torch.cuda.is_available():
             return torch.device("cuda")
-        if torch.backends.mps.is_available():
+        if _torch_version_le_112 and torch.backends.mps.is_available():
             return torch.device("mps")
         return torch.device("cpu")
 

ignite/engine/__init__.py

@@ -96,6 +96,8 @@ def supervised_training_step(
         Added `model_transform` to transform model's output
     .. versionchanged:: 0.4.13
         Added `model_fn` to customize model's application on the sample
+    .. versionchanged:: 0.4.14
+        Added support for ``mps`` device
     """
 
     if gradient_accumulation_steps <= 0:
@@ -391,9 +393,12 @@ def update(engine: Engine, batch: Sequence[torch.Tensor]) -> Union[Any, Tuple[to
 
 
 def _check_arg(
-    on_tpu: bool, amp_mode: Optional[str], scaler: Optional[Union[bool, "torch.cuda.amp.GradScaler"]]
+    on_tpu: bool, on_mps: bool, amp_mode: Optional[str], scaler: Optional[Union[bool, "torch.cuda.amp.GradScaler"]]
 ) -> Tuple[Optional[str], Optional["torch.cuda.amp.GradScaler"]]:
-    """Checking tpu, amp and GradScaler instance combinations."""
+    """Checking tpu, mps, amp and GradScaler instance combinations."""
+    if on_mps and amp_mode:
+        raise ValueError("amp_mode cannot be used with mps device. Consider using amp_mode=None or device='cuda'.")
+
     if on_tpu and not idist.has_xla_support:
         raise RuntimeError("In order to run on TPU, please install PyTorch XLA")
 
@@ -546,11 +551,14 @@ def output_transform_fn(x, y, y_pred, loss):
         Added ``model_transform`` to transform model's output
     .. versionchanged:: 0.4.13
         Added `model_fn` to customize model's application on the sample
+    .. versionchanged:: 0.4.14
+        Added support for ``mps`` device
     """
 
     device_type = device.type if isinstance(device, torch.device) else device
     on_tpu = "xla" in device_type if device_type is not None else False
-    mode, _scaler = _check_arg(on_tpu, amp_mode, scaler)
+    on_mps = "mps" in device_type if device_type is not None else False
+    mode, _scaler = _check_arg(on_tpu, on_mps, amp_mode, scaler)
 
     if mode == "amp":
         _update = supervised_training_step_amp(
@@ -791,10 +799,13 @@ def create_supervised_evaluator(
         Added ``model_transform`` to transform model's output
     .. versionchanged:: 0.4.13
         Added `model_fn` to customize model's application on the sample
+    .. versionchanged:: 0.4.14
+        Added support for ``mps`` device
     """
     device_type = device.type if isinstance(device, torch.device) else device
     on_tpu = "xla" in device_type if device_type is not None else False
-    mode, _ = _check_arg(on_tpu, amp_mode, None)
+    on_mps = "mps" in device_type if device_type is not None else False
+    mode, _ = _check_arg(on_tpu, on_mps, amp_mode, None)
 
     metrics = metrics or {}
     if mode == "amp":