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

Add bits and sym parameters to the OV quantization config #560

Merged
merged 13 commits into from
Feb 15, 2024

Conversation

echarlaix
Copy link
Collaborator

No description provided.

@HuggingFaceDocBuilderDev

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

@AlexKoff88
Copy link
Collaborator

@echarlaix, all the failures in CI are related to the fact that the GitHub version of NNCF enables weight quantization of Conv layers by default while the officially released version does not have it. Realistically, we will have a new NNCF version released by the end of Feb but we will create a release branch this week, so we can just install NNCF from the release branch and no changes in the test references are required.

@echarlaix echarlaix marked this pull request as ready for review February 15, 2024 14:42
@echarlaix echarlaix requested a review from AlexKoff88 February 15, 2024 15:22
Comment on lines 604 to 607
default_config = _check_default_4bit_configs(config)

if default_config:
logger.info(f"For the given model, we recommend the following `quantization_config` : {default_config}")
Copy link
Collaborator Author

@echarlaix echarlaix Feb 15, 2024

Choose a reason for hiding this comment

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

I don't think we should overwrite the quantization_config as it's given by the user, so here we are just adding a warning for the user, wdyt @AlexKoff88 ?

Copy link
Collaborator

Choose a reason for hiding this comment

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

Actually, this was an idea behind the combination of load_in_4bit + quantizaiton_config. If the later one is None we use the pre-defined default config, otherwise we override it an let the user to use a custom config.

@echarlaix echarlaix changed the title Quant config nncf Add bits and sym parameters to the OV quantization config Feb 15, 2024
@echarlaix echarlaix merged commit 6c8fa79 into main Feb 15, 2024
10 of 12 checks passed
@echarlaix echarlaix deleted the quant-config-nncf branch February 15, 2024 16:37
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants