Skip to content

vllm.models.minimax_m3.common

Modules:

  • encoder_cudagraph –

    Encoder CUDA graph support for the MiniMax M3 vision tower.

  • indexer –

    MiniMax M3 lightning indexer: side cache, metadata, and impl.

  • mm_preprocess –
  • ops –

    Cross-platform (Triton) kernels for MiniMax M3 sparse attention.

  • sparse_attention –

    Main block-sparse GQA attention for MiniMax M3 sparse layers.

  • vision_tower –