Skip to content

fix a typo in flax T5 attention - attention_mask variable is misnamed#26663

Merged
ArthurZucker merged 2 commits intohuggingface:mainfrom
giganttheo:fix_typo_flax_t5_attn
Oct 10, 2023
Merged

fix a typo in flax T5 attention - attention_mask variable is misnamed#26663
ArthurZucker merged 2 commits intohuggingface:mainfrom
giganttheo:fix_typo_flax_t5_attn

Conversation

@giganttheo
Copy link
Contributor

@giganttheo giganttheo commented Oct 7, 2023 •
edited by sanchit-gandhi
Loading

What does this PR do?

Fixes a typo in the Flax code for T5 model.

There is a typo in the Attention module of the Flax version of T5, where the attention_mask updated by the _concatenate_to_cache method should override the previous attention_mask but does not because of a misnamed variable.

Fixes #26564

Before submitting

  • This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).

Who can review?

@sanchit-gandhi

sanchit-gandhi
sanchit-gandhi approved these changes Oct 9, 2023
View reviewed changes
Copy link
Contributor

@sanchit-gandhi sanchit-gandhi left a comment

Choose a reason for hiding this comment

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

Very nice @giganttheo! Thanks for identifying the bug and proposing the fix 🤗 Confirming that the slow tests pass following the fix? As per #26564 (comment) If so, then this all LGTM!

@HuggingFaceDocBuilderDev
Copy link

HuggingFaceDocBuilderDev commented Oct 9, 2023

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint.

@giganttheo
Copy link
Contributor Author

giganttheo commented Oct 10, 2023

Very nice @giganttheo! Thanks for identifying the bug and proposing the fix 🤗 Confirming that the slow tests pass following the fix? As per #26564 (comment) If so, then this all LGTM!

The slow tests are passing for t5 and longt5:

RUN_SLOW=1 pytest -sv tests/models/t5/test_modeling_flax_t5.py::FlaxT5ModelIntegrationTests

outputs: ================== 6 passed, 4 warnings in 331.38s (0:05:31) ===================

and for the longT5 version:

RUN_SLOW=1 pytest -sv tests/models/longt5/test_modeling_flax_longt5.py::FlaxLongT5ModelIntegrationTests

outputs: =================== 1 passed, 1 warning in 401.61s (0:06:41) ===================

@sanchit-gandhi
Copy link
Contributor

sanchit-gandhi commented Oct 10, 2023

Awesome - thanks for confirming! Requesting a final review from @ArthurZucker

ArthurZucker
ArthurZucker approved these changes Oct 10, 2023
View reviewed changes
Copy link
Collaborator

@ArthurZucker ArthurZucker left a comment

Choose a reason for hiding this comment

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

LGTM! Thanks for catching 🤗

@ArthurZucker ArthurZucker merged commit 975003e into huggingface:main Oct 10, 2023
@giganttheo giganttheo deleted the fix_typo_flax_t5_attn branch May 17, 2024 11:25
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@ArthurZucker ArthurZucker ArthurZucker approved these changes

+1 more reviewer

@sanchit-gandhi sanchit-gandhi sanchit-gandhi approved these changes

Reviewers whose approvals may not affect merge requirements

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

Flax T5 model - code typo during AutoRegressive decoding?

4 participants

@giganttheo @HuggingFaceDocBuilderDev @sanchit-gandhi @ArthurZucker