vllm
Add `moe_align_block_size_no_permute` for small batch size with large num_expert
#30280

Open

Add `moe_align_block_size_no_permute` for small batch size with large num_expert #30280

RunkaiTao wants to merge 4 commits into vllm-project:main from RunkaiTao:feat/unpermute-kernel

unpermute kernel cuda

e4fc343c

pre-commit

bb5ee065

disable-expert-map

84b1b202

gemini-code-assist commented on 2025-12-08

pre-commit

976bf1d1

mergify added needs-rebase

Reviewers

gemini-code-assist

Assignees

No one assigned

Labels

needs-rebase

Milestone

No milestone