[v0.13.0][BugFix][Cherry Pick] Fix input parameter bug of dispatch_gmm_combine_decode#5931
Merged
wangxiyuan merged 1 commit intovllm-project:releases/v0.13.0from Jan 22, 2026
Conversation
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Code Review
This pull request correctly fixes a critical bug in the dispatch_gmm_combine_decode operator. The global_bs parameter was previously calculated incorrectly as it did not account for tensor parallelism, which could lead to runtime errors or incorrect results. The change ensures global_bs is calculated consistently with other MoE operators by factoring in the tensor parallel size. The removal of the now-unused fused_global_bs attribute is also a good code cleanup. The changes are correct and improve the robustness of the fused MoE implementation.
Signed-off-by: wangqiankun <wangqiankun13@huawei.com>
wangqiankun13
force-pushed
the
v0.13.0-fix-global-bs-bug
branch
from
02f41ca to
fa6bc4e
Compare
wangqiankun13
changed the title
wangqiankun13
changed the title
weijinqian0
approved these changes
Jan 21, 2026
wangxiyuan
added
ready
845473182
pushed a commit
to 845473182/vllm-ascend
that referenced
this pull request
Jan 22, 2026
…lm-ascend into FIA_v0.13.0 * 'releases/v0.13.0' of https://github.com/vllm-project/vllm-ascend: [Feature][Cherry Pick]Enable DispatchGmmCombineDecode when eagle is moe with w8a8, or not moe (vllm-project#6081) [v0.13.0][BugFix][Cherry Pick] Fix input parameter bug of dispatch_gmm_combine_decode (vllm-project#5931) [0.13.0][Bugfix] Fix Triton operator usage for multimodal models based on `the mrope_interleaved` parameter (vllm-project#6074) [v0.13.0][CI] Upgrade to CANN 8.5.0 (vllm-project#6101)
tangtiangu
pushed a commit
to tangtiangu/jiusi-vllm-ascend
that referenced
this pull request
Feb 24, 2026
…m_combine_decode (vllm-project#5931) ### What this PR does / why we need it? This PR is cherry-picked from [PR5932](vllm-project#5932). In vllm-project#5040, the dispatch_gmm_combine_decode operator was configured with an incorrect global_bs parameter. This PR is to fix the bug. The global_bs provided as input should have the same meaning as in the moe_distributed_dispatch operator, specifically: (the maximum batch size across all cards) * (expert parallel world size). However, the implementation incorrectly used the variable max_num_tokens, which does not account for tensor parallelism. This error likely resulted in an unnecessarily large (overestimated) value. More info about this operator, please refer to RFC: issue vllm-project#5476 Signed-off-by: wangqiankun <wangqiankun13@huawei.com>
tangtiangu
pushed a commit
to tangtiangu/jiusi-vllm-ascend
that referenced
this pull request
Feb 24, 2026
…m_combine_decode (vllm-project#5931) ### What this PR does / why we need it? This PR is cherry-picked from [PR5932](vllm-project#5932). In vllm-project#5040, the dispatch_gmm_combine_decode operator was configured with an incorrect global_bs parameter. This PR is to fix the bug. The global_bs provided as input should have the same meaning as in the moe_distributed_dispatch operator, specifically: (the maximum batch size across all cards) * (expert parallel world size). However, the implementation incorrectly used the variable max_num_tokens, which does not account for tensor parallelism. This error likely resulted in an unnecessarily large (overestimated) value. More info about this operator, please refer to RFC: issue vllm-project#5476 Signed-off-by: wangqiankun <wangqiankun13@huawei.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
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.
What this PR does / why we need it?
This PR is cherry-picked from PR5932.
In #5040, the dispatch_gmm_combine_decode operator was configured with an incorrect global_bs parameter. This PR is to fix the bug.
The global_bs provided as input should have the same meaning as in the moe_distributed_dispatch operator, specifically: (the maximum batch size across all cards) * (expert parallel world size).
However, the implementation incorrectly used the variable max_num_tokens, which does not account for tensor parallelism. This error likely resulted in an unnecessarily large (overestimated) value.
More info about this operator, please refer to RFC: issue #5476
Does this PR introduce any user-facing change?
How was this patch tested?