[Bugfix][Attention] Forward per-head FP8 descales through FA4 - #51363
Merged
Merged
Conversation
Signed-off-by: yiliu30 <yi4.liu@intel.com>
Co-authored-by: OpenAI Codex <noreply@openai.com> Signed-off-by: yiliu30 <yi4.liu@intel.com>
yiliu30
marked this pull request as ready for review
August 7, 2026 05:47
yiliu30
requested review from
AndreasKaratzas,
LucasWilkinson,
MatthewBonanni,
WoosukKwon,
mgoin,
tlrmchlsmth,
yewentao256 and
zyongye
as code owners
August 7, 2026 05:47
![claude[bot]](https://avatars.githubusercontent.com/in/1236702?s=60&v=4){kind=small}
kylesayrs
approved these changes
Aug 7, 2026
brian-dellabetta
approved these changes
Aug 7, 2026
brian-dellabetta
left a comment
brian-dellabetta
left a comment
Contributor
There was a problem hiding this comment.
Thanks @yiliu30 !
Co-authored-by: Wentao Ye <44945378+yewentao256@users.noreply.github.com> Signed-off-by: Yi Liu <yi4.liu@intel.com>
yewentao256
approved these changes
Aug 8, 2026
yewentao256
left a comment
yewentao256
left a comment
Member
There was a problem hiding this comment.
LGTM, thanks for the work!
Member
|
/ci run |
|
✅ @yiliu30, CI is now available for this PR.
|
|
✅ Triggered Buildkite CI #83008 for commit |
Contributor
Author
|
/ci run |
|
✅ Triggered Buildkite CI #83244 for commit |
Contributor
Author
|
/ci run |
|
✅ Triggered Buildkite CI #83254 for commit |
Signed-off-by: yiliu30 <yi4.liu@intel.com>
Contributor
Author
|
/ci run |
|
✅ Triggered Buildkite CI #83318 for commit |
Contributor
Author
|
Hi @yewentao256 the CI was passed. Could you help get it merge? Thanks! |
zyp2014
pushed a commit
to zyp2014/vllm
that referenced
this pull request
Aug 21, 2026
…roject#51363) Signed-off-by: yiliu30 <yi4.liu@intel.com> Signed-off-by: Yi Liu <yi4.liu@intel.com> Co-authored-by: OpenAI Codex <noreply@openai.com> Co-authored-by: Wentao Ye <44945378+yewentao256@users.noreply.github.com>
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.
Summary
Forward per-head FP8 Q/K/V descales from vLLM's FlashAttention interface into the FA4 CuTe forward kernel.
Root cause
The FA3 path forwarded
q_descale,k_descale, andv_descale, but the FA4 path dropped them. Per-head FP8 Q/K/V tensors were therefore consumed without their calibration scales, producing garbage output for llm-compressor attention/KV-quantized checkpoints.Changes
_flash_attn_fwdcall.Validation
Duplicate-work check
Related issue #40444 and open PR #40448 were checked. PR #40448 clarifies the FlashInfer limitation and does not fix FA4 descale forwarding, so this is materially different kernel work.
AI assistance
AI assistance was used to prepare this PR. The human submitter must review every changed line, understand the fix, and confirm the evaluation results before requesting review.
cc @brian-dellabetta