[Bugfix] Fix FA4 OOM with diff headdim#123
Merged
LucasWilkinson merged 4 commits intovllm-project:mainfrom Mar 6, 2026
Merged
Conversation
5 tasks
Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
Collaborator
LucasWilkinson
left a comment
LucasWilkinson
left a comment
There was a problem hiding this comment.
Does it make more sense to adjust tile sizes instead?
Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
Author
|
Turns out the best approach is a heuristic that either disables splitting or adjusts tile size. Benchmarked with vLLM attention benchmarks with 16 q heads (deepseek V2): |
MatthewBonanni
changed the title
LucasWilkinson
approved these changes
Mar 6, 2026
MatthewBonanni
changed the title
This was referenced Mar 11, 2026
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Using split KV with diff headdim (e.g. 192/128 like DeepSeek MLA prefill) leads to exceeding SMEM. This PR uses a heuristic to either disable splitting or adjust tile size in this case