vllm

[Kernel] Fuse FP8 output quantization into merge_attn_states

#36518

Merged