vllm.v1.attention.backends.mla ¶

Modules:

• flashinfer_mla_sparse –

  FlashInfer sparse MLA attention backend.

• flashinfer_mla_sparse_sm120 –

  SM120 implementation variant for FLASHINFER_MLA_SPARSE_SM120.

• flashmla_sparse –
• indexer –
• prefill –
• rocm_aiter_mla –
• rocm_aiter_mla_sparse –
• sparse_swa –
• sparse_utils –

  Utility functions for sparse MLA backends.

• tokenspeed_mla –

  TokenSpeed CuTe DSL MLA decode backend (Blackwell, FP8 KV cache only).