vllm

[Core] Support `min_tokens` with speculative decoding

#32642

Merged