Fix synced multi-GPU generation with LLMs and VLMs#35893
Merged
zucchini-nlp merged 3 commits intohuggingface:mainfrom Feb 5, 2025
Merged
Fix synced multi-GPU generation with LLMs and VLMs#35893zucchini-nlp merged 3 commits intohuggingface:mainfrom
zucchini-nlp merged 3 commits intohuggingface:mainfrom
Conversation
This was referenced Feb 3, 2025
zucchini-nlp
approved these changes
Feb 3, 2025
Member
zucchini-nlp
left a comment
zucchini-nlp
left a comment
There was a problem hiding this comment.
Sorry for later review and thanks for fixing! Indeed we need to use same exceptions when model overrides prepare_inputs
|
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. |
4 tasks
Contributor
|
@ManukyanD Thank you for the fix 💛 |
MekkCyber
pushed a commit
that referenced
this pull request
Feb 7, 2025
* Fix synced multi-GPU generation * fix copies --------- Co-authored-by: Davit Manukyan <ManukyanD> Co-authored-by: Raushan Turganbay <raushan@huggingface.co>
elvircrn
pushed a commit
to elvircrn/transformers
that referenced
this pull request
Feb 13, 2025
* Fix synced multi-GPU generation * fix copies --------- Co-authored-by: Davit Manukyan <ManukyanD> Co-authored-by: Raushan Turganbay <raushan@huggingface.co>
sbucaille
pushed a commit
to sbucaille/transformers
that referenced
this pull request
Feb 16, 2025
* Fix synced multi-GPU generation * fix copies --------- Co-authored-by: Davit Manukyan <ManukyanD> Co-authored-by: Raushan Turganbay <raushan@huggingface.co>
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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?
Generation with some LLMs and VLMs in synced multi-GPU settings crashes because the cache_position goes out of bounds. The issue was solved for Gemma2 in this PR. I have noticed that this issue is also present in several other LLMs, as well as several VLMs. This PR fixes the issue for Bamba, Bloom, Chameleon, Cohere2, Jamba, MLlama, Qwen2 VL, Qwen2.5 VL and Zamba.
Before submitting
Pull Request section?
to it if that's the case.
documentation guidelines, and
here are tips on formatting docstrings.
Who can review?
@zucchini-nlp
@gante
@ArthurZucker