[Triton] Fix triton launch grids under multi gpu scenarios - #1003
Open
vasqu wants to merge 1 commit into
Open
Conversation
2 tasks
|
! |
Contributor
|
nice catch on the launch grid — can you add a tiny multi-gpu smoke test or at least a comment with the failing shape so this doesn't regress quietly? |
Contributor
Author
|
cc @tridao whenever you have time Seems like nemotron h tried to fix it in a wrong way so I'd rather have this upstream (they pinned the complete cuda stream) |
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
3 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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
It's essentially the same underlying issue as Dao-AILab/flash-attention#523
Triton can have issues to select the correct underlying device