Skip to content

vllm.models.minimax_m3.amd.ops.norm_rope_nvfp4_insert

Modules:

  • host –

    Launch wrapper of the fused norm / RoPE / NVFP4 insert.

  • kernel –

    Gemma norm + RoPE of q / k / index-q / index-k and the NVFP4 K/V + index-K