vllm.models.minimax_m3.nvidia ¶

Modules:

• indexer_msa –

  MSA (SM100/Blackwell) indexer impl for MiniMax M3.

• model –

  Inference-only MiniMax M3 (text backbone) model.

• msa_cutlass_sparse_decode –

  MiniMax CUTLASS sparse decode using per-query-token page indices.

• mtp –
• ops –
• sparse_attention_msa –

  MSA (SM100/Blackwell) block-sparse attention for MiniMax M3.