这是indexloc提供的服务,不要输入任何密码
Skip to content

Conversation

@jiqing-feng
Copy link
Contributor

@jiqing-feng jiqing-feng commented Oct 9, 2025

There has no need to check cpu specifically as CPU support dequantize bnb weights.

Hi @BenjaminBossan . Could you please review this PR? Thanks!

cc @yao-matrix

Signed-off-by: jiqing-feng <jiqing.feng@intel.com>
@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.

@BenjaminBossan
Copy link
Member

@jiqing-feng Could you please run make style?

@jiqing-feng
Copy link
Contributor Author

@jiqing-feng Could you please run make style?

Done, please review it. Thx.

Signed-off-by: jiqing-feng <jiqing.feng@intel.com>
Copy link
Member

@BenjaminBossan BenjaminBossan left a comment

Choose a reason for hiding this comment

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

Thanks for simplifying the dequantization logic. Indeed, it is not required to move the weight to the accelerator first. I think it was at some point, which is why the code is there, but I tested the last few bnb versions and they all worked with the new code, so we should be good.

The failing tests on the CI are unrelated and can be ignored.

@BenjaminBossan BenjaminBossan merged commit 879587f into huggingface:main Oct 10, 2025
5 of 13 checks passed
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