vllm.models.minimax_m3.nvidia.ops

Modules:

• index_decode_score –

  CuteDSL MiniMax M3 index decode score kernel.

• sparse_prefill –

  Hopper query-tiled sparse prefill attention for MiniMax M3.