Fix flex_attention in training mode #35605
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
MekkCyber
approved these changes
Jan 10, 2025
|
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. |
ArthurZucker
pushed a commit
that referenced
this pull request
Jan 10, 2025
* fix flex * add test * style
|
thank you so much @ArthurZucker @Cyrilvallez ! I will test it out later today! |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
What does this PR do?
As per the title.
In training mode, the in-place operation would produce
RuntimeError: FakeTensor is wrapped to wrong device, found cpu, expected cuda:1. It is however working when gradients are not required (i.e. withgenerate), which is the reason why it wasn't detected before.I am not entirely sure what's happening behind the scene and the root of the issue, but switching to explicit operation fixes the issue.