[vLLM][MoE] Update kernel grid to use autotune selected BLOCK_M and BLOCK_N values - #7776
Open
sarah12121212 wants to merge 20 commits into
Open
[vLLM][MoE] Update kernel grid to use autotune selected BLOCK_M and BLOCK_N values#7776sarah12121212 wants to merge 20 commits into
sarah12121212 wants to merge 20 commits into
Conversation
sarah12121212
changed the title
sarah12121212
changed the title
sarah12121212
changed the title
…ne-key # Conflicts: # scripts/skiplist/default/vllm_moe.txt # scripts/skiplist/xe2/vllm_moe.txt
sarah12121212
added
run-vllm-tests
run-vllm-benchmarks
and removed
run-vllm-tests
labels
sarah12121212
marked this pull request as ready for review
August 21, 2026 14:49
quinnlp
requested changes
Aug 21, 2026
sarah12121212
marked this pull request as draft
August 21, 2026 15:20
sarah12121212
added
run-vllm-tests
and removed
run-vllm-tests
run-vllm-benchmarks
labels
sarah12121212
marked this pull request as ready for review
August 24, 2026 12:55
quinnlp
approved these changes
Aug 24, 2026
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
2 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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Batched MoE Kernel launch grid still used the old vLLM
BLOCK_SIZE_MandBLOCK_SIZE_Nvalues returned by vLLM's configuration lookup, instead of theBLOCK_MandBLOCK_Nvalues chosen by the Triton autotuner. This lead to accuracy failures, for example if too few programs were launched compared to the number of programs required for the selected autotuned dimensions.This PR creates the grid from the selected autotune metadata so that the launch MxN dimension matches the kernel’s actual tile sizes.
All bmg and pvc vllm_tdesc tests pass: https://github.com/intel/intel-xpu-backend-for-triton/actions/runs/32497175599/job/96828939774
Closes #7753