vllm.models.deepseek_v41.nvidia ¶

Modules:

• dspark –

  DSpark draft model for DeepSeek-V4.1 (semi-autoregressive speculative decoding).

• engram –

  NVIDIA Engram DP sharding, shared host storage, and asynchronous prefetch.

• flashinfer_sparse –

  DeepSeek V4 FlashInfer sparse MLA backend.

• flashmla –
• model –
• model_state –
• vl_model –

  DeepSeek-V4.1 vision variant.