vllm

[main][BugFix] Fixed an accuracy bug of Qwen3-next-MTP when batched inferring

#30632

Merged