[Bugfix]Reject invalid FlashInfer MNNVL workspaces - #49043
Merged
Merged
Conversation
Signed-off-by: lengrongfu <lenronfu@gmail.com>
lengrongfu
marked this pull request as ready for review
July 18, 2026 15:39
![claude[bot]](https://avatars.githubusercontent.com/in/1236702?s=60&v=4){kind=small}
Contributor
Author
Member
|
According to #48075, TRTLLM is ~10× slower than no-fuse |
mgoin
approved these changes
Jul 25, 2026
mgoin
added
performance
2 tasks
Open
1 task
9 tasks
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
3 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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Purpose
Fixs: #49041
On topologies without symmetric-memory multicast support, FlashInfer can construct an MNNVL allreduce workspace whose mc_ptr is null. vLLM treated the workspace as initialized and passed it into fused allreduce kernels, which dereferenced the null multicast mapping and failed with a CUDA illegal memory access.
Validate mc_ptr immediately after workspace creation. Destroy invalid MNNVL workspaces and return None so the existing backend selection can fall back to TRTLLM on supported single-node configurations, or disable the fusion safely when no fallback is available.
https://docs.nvidia.com/cuda/cuda-driver-api/group__CUDA__MULTICAST.html
Test Plan
Rerun using the previous command.
Test Result
Essential Elements of an Effective PR Description Checklist
supported_models.mdandexamplesfor a new model.