Fully yank dropout in vllm fork of FA2 - #153
Merged
Merged
Conversation
janeyx99
force-pushed
the
vllm-yank-dropout-fa2
branch
from
June 24, 2026 16:18
f288c45 to
1ed425a
Compare
janeyx99
marked this pull request as ready for review
June 24, 2026 16:33
4 tasks
Member
|
I see Dao-AILab#2680 and Dao-AILab#2683 are merged. Do you think Dao-AILab#2669 and Dao-AILab#2688 can also get in? |
Author
|
@Harry-Chen I'm not so sure; we were hoping to get Tri's opinion on the direction as those are more imperative changes. |
Signed-off-by: Jane Xu <janeyx@meta.com>
janeyx99
force-pushed
the
vllm-yank-dropout-fa2
branch
from
July 29, 2026 21:42
2d668b6 to
39e86aa
Compare
Author
|
The PR is merged! |
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
2 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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Follows the example set in upstream Dao-AILab#2669 to expand FLASHATTENTION_DISABLE_DROPOUT support.
vLLM already defaults to using the DROPOUT free FA2 (cuz inference doesn't need dropout), and syncing with upstream means we will not compile the dropout related headers (RNG/Philox) at all. This enables us to get the vllm cuda wheel fully migrated to ABI stable by migrating FA2 to be fully ABI stable, will be be the next step!
We sync with upstream to maintain similarity as much as possible. Followup work could be done to sync the branches further (e.g., remove the commented out code as it wouldn't be used anyway), but I opted to not mess with pre-existing vLLM commits in this PR.
Test plan: