Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Sparse Quantization Example Clarification #2334

Merged
merged 7 commits into from
Jun 18, 2024
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 5 additions & 3 deletions examples/llama7b_sparse_quantized/README.md
Original file line number Diff line number Diff line change
Expand Up @@ -4,7 +4,8 @@ This example uses SparseML and Compressed-Tensors to create a 2:4 sparse and qua
The model is calibrated and trained with the ultachat200k dataset.
At least 75GB of GPU memory is required to run this example.

Follow the steps below, or to run the example as `python examples/llama7b_sparse_quantized/llama7b_sparse_w4a16.py`
Follow the steps below one by one in a code notebook, or run the full example script
as `python examples/llama7b_sparse_quantized/llama7b_sparse_w4a16.py`

## Step 1: Select a model, dataset, and recipe
In this step, we select which model to use as a baseline for sparsification, a dataset to
Expand Down Expand Up @@ -36,7 +37,8 @@ recipe = "2:4_w4a16_recipe.yaml"

## Step 2: Run sparsification using `apply`
The `apply` function applies the given recipe to our model and dataset.
The hardcoded kwargs may be altered based on each model's needs.
The hardcoded kwargs may be altered based on each model's needs. This code snippet should
be run in the same Python instance as step 1.
After running, the sparsified model will be saved to `output_llama7b_2:4_w4a16_channel`.

```python
Expand Down Expand Up @@ -67,7 +69,7 @@ apply(
### Step 3: Compression

The resulting model will be uncompressed. To save a final compressed copy of the model
run the following:
run the following in the same Python instance as the previous steps.

```python
import torch
Expand Down
6 changes: 3 additions & 3 deletions examples/llama7b_w8a8_quantization.py
Original file line number Diff line number Diff line change
Expand Up @@ -16,12 +16,12 @@
num_bits: 8
type: "int"
symmetric: true
strategy: "channel"
strategy: "tensor"
input_activations:
num_bits: 8
type: "int"
symmetric: true
dynamic: True
dynamic: true
strategy: "token"
targets: ["Linear"]
"""
Expand All @@ -37,7 +37,7 @@
dataset = "ultrachat-200k"

# save location of quantized model out
output_dir = "./output_llama7b_w8a8_channel_dynamic_compressed"
output_dir = "./output_llama7b_w8a8_dynamic_compressed"

# set dataset config parameters
splits = {"calibration": "train_gen[:5%]"}
Expand Down
Loading