Conversation
Contributor
|
Warning You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again! |
yichiche
approved these changes
Mar 3, 2026
Collaborator
yichiche
left a comment
yichiche
left a comment
There was a problem hiding this comment.
This PR fixes a crash in the CUDA graph replay path for non-MLA backends by properly initializing the forward metadata. Previously, custom_mask and mask_indptr were not consistently set outside the MLA path, which could lead to invalid memory access during replay. LGTM.
michaelzhang-ai
requested review from
Fridge003,
HaiShaw,
Qiaolin-Yu,
hebiao064,
ispobock and
merrymercy
as code owners
Contributor
|
Warning You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again! |
Collaborator
|
/tag-and-rerun-ci |
…d call #18882 added k_scale and v_scale as required positional parameters to extend_attention_fwd and updated triton_backend.py, but missed updating the call site in aiter_backend.py for non-MLA target_verify/draft_extend paths. This caused Qwen3-Coder-Next MTP to crash with: TypeError: extend_attention_fwd() missing 1 required positional argument: 'v_scale' Fixes: https://github.com/sgl-project/sglang/actions/runs/22636393480/job/65600256830
michaelzhang-ai
force-pushed
the
amd/fix-aiter-mla-replay-cuda-graph-v2
branch
from
6aedf5d to
9cf021b
Compare
michaelzhang-ai
changed the title
HaiShaw
approved these changes
Mar 4, 2026
Kangyan-Zhou
pushed a commit
to Kangyan-Zhou/sglang
that referenced
this pull request
Mar 4, 2026
…d_attention_fwd in aiter_backend (sgl-project#19736)
qeternity
pushed a commit
to qeternity/sglang
that referenced
this pull request
Mar 6, 2026
…d_attention_fwd in aiter_backend (sgl-project#19736)
This was referenced Mar 8, 2026
magicYang1573
pushed a commit
to magicYang1573/sglang
that referenced
this pull request
Mar 9, 2026
…d_attention_fwd in aiter_backend (sgl-project#19736)
Wangzheee
pushed a commit
to Wangzheee/sglang
that referenced
this pull request
Mar 21, 2026
…d_attention_fwd in aiter_backend (sgl-project#19736)
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
TypeError: extend_attention_fwd() missing 1 required positional argument: 'v_scale'crash in aiter_backend when running non-MLA speculative decoding (target_verify / draft_extend) paths: https://github.com/sgl-project/sglang/actions/runs/22648816293/job/65643197339#step:5:6363k_scale=1.0, v_scale=1.0positional args to theextend_attention_fwdcallMotivation
#18882 added
k_scaleandv_scaleas required positional parameters toextend_attention_fwdintriton_ops/extend_attention.pyand updatedtriton_backend.py, but missed the call site inaiter_backend.pyfor non-MLA target_verify/draft_extend paths (used by hybrid models like Qwen3-Coder-Next with MTP).Fixes
Nightly (AMD ROCm) failure: https://github.com/sgl-project/sglang/actions/runs/22648816293/job/65643197339#step:5:6363
Test plan
nightly-accuracy-8-gpu-mi35x/nightly-accuracy-8-gpu-mi35x-rocm720to confirm Qwen3-Coder-Next MTP passes:https://github.com/sgl-project/sglang/actions/runs/22649193163/job/65644407193