Skip to content

Update AutoRound layer_config usage #2331

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.

Sign up for GitHub

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

Open
wants to merge 3 commits into
base: master
Choose a base branch
from
Open

Update AutoRound layer_config usage #2331

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
ba1ebc7
update AutoRound layer_config usage
Kaihui-intel Nov 7, 2025
2d76c7f
align format
Kaihui-intel Nov 7, 2025
f9d3ac9
update example
Kaihui-intel Nov 7, 2025
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
19 changes: 19 additions & 0 deletions docs/source/3x/PT_WeightOnlyQuant.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -178,6 +178,8 @@ model = convert(model, config) # after this step, the model is ready for W4A8 i
| not_use_best_mse (bool) | Whether to use mean squared error | False |
| dynamic_max_gap (int) | The dynamic maximum gap | -1 |
| scale_dtype (str) | The data type of quantization scale to be used, different kernels have different choices | "float16" |
| scheme (str) | A preset scheme that defines the quantization configurations. | "W4A16" |
| layer_config (dict) | Layer-wise quantization config | None |

``` python
# Quantization code
Expand Down Expand Up @@ -283,6 +285,23 @@ quant_config = RTNConfig()
lm_head_config = RTNConfig(dtype="fp32")
quant_config.set_local("lm_head", lm_head_config)
```
3. Example of using `layer_config` for AutoRound
Copy link
Contributor

@thuang6 thuang6 Nov 7, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

If 'set_local' does not work in current implementation, we should call out this, let user know, AutoRound specific 'layer_config' instead of 'set_local' API should be used. since in AutoRoundConfig, not any one of 3 items work.

Copy link
Contributor Author

@Kaihui-intel Kaihui-intel Nov 7, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Code will be raised in another PR.

Copy link
Contributor

@thuang6 thuang6 Nov 7, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

do u mean you will implement 'set_local' support by converting it to layer_config so any of 3 options is valid after your another PR merged?

Copy link
Contributor Author

@Kaihui-intel Kaihui-intel Nov 7, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

sure. Currently, only the third option is supported. option `1, 2 'will be implemented in phase 2.

```python
# layer_config = {
# "layer1": {
# "data_type": "int",
# "bits": 3,
# "group_size": 128,
# "sym": True,
# },
# "layer2": {
# "W8A16"
# }
# }
# Use the AutoRound specific 'layer_config' instead of the 'set_local' API.
layer_config = {"lm_head": {"data_type": "int"}}
quant_config = AutoRoundConfig(layer_config=layer_config)
```

### Saving and Loading

Expand Down