fix(fused_moe): restore gfx942 FLAT kernel support - #4628
Open
bingxche wants to merge 1 commit into
Open
Conversation
Contributor
🏷️ CI GuideRuns automatically on every PR:
Extended tests (opt-in via labels):
|
Contributor
Author
|
Validation runs for AITER head
Both workflows were dispatched from SGLang |
bingxche
marked this pull request as ready for review
August 7, 2026 15:43
Contributor
Author
|
Hi @yadaish, could you please review? Thanks! |
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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
gfx942to the FLAT FMoE architecture allowlist while retaining the unsupported-architecture guardgfx942flat_pf3kernel againRoot cause
AITER #4040 added the
gfx942FLAT kernel, tuned GLM rows, and thegfx942/gfx950allowlist. #4394 later narrowed only the guard back togfx950, while leaving thegfx942kernel and config in place. SGLang GLM-5.1 therefore selects the validgfx942kernel but fails before launch.Observed in both SGLang jobs:
Test plan
git diff --checkgfx942FLAT FMoE case for M=4/8, model_dim=6144, inter_dim=256, E=257, topk=9nightly-8-gpu-glm51nightly-8-gpu-glm51-rocm720Local Python/ROCm tests were not run because the authoring environment is Windows without Python or AMD GPUs.