Skip to content
This repository was archived by the owner on May 11, 2025. It is now read-only.

Conversation

@younesbelkada
Copy link
Collaborator

What does this PR do?

Fixes current issues with transformers main (next release) and autoawq + fused modules. The caching mechanism has been refactored thus requiring some changes on AutoAWQ side to make things compatible.

Related PR: huggingface/transformers#26681

cc @casper-hansen

@casper-hansen
Copy link
Owner

Thanks for fixing the breakage once again @younesbelkada. Do we need to add more testing into transformers to avoid these larger updates that break AutoAWQ?

@younesbelkada
Copy link
Collaborator Author

Thanks @casper-hansen , we do have tests in transformers to test this feature - https://github.com/huggingface/transformers/blob/main/tests/quantization/autoawq/test_awq.py#L231 this is how I was able to flag there was an issue with transformers main

@casper-hansen casper-hansen merged commit ae24f42 into main Dec 11, 2023
@younesbelkada younesbelkada deleted the younesbelkada-cache-patch branch December 11, 2023 12:40
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants