Skip to content

vllm.models.deepseek_v41.nvidia

Modules:

  • decoder_replay_cudagraph –

    CUDA graphs of the decoder replay layers on a step's replay batch.

  • dspark –

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

  • flash_mla_mega_attn –

    DeepSeek V4.1 attention on FlashMLA's mega-attention kernel (SM100).

  • flashinfer_sparse –

    DeepSeek V4 FlashInfer sparse MLA backend.

  • flashmla –
  • model –
  • model_state –
  • ops –
  • pipeline –

    Pipeline dependency planning, cache replicas and intermediate tensor transport.

  • quantization –

    DeepSeek-V4.1 quantization methods on NVIDIA GPUs before SM100.

  • specialized_decode –

    DeepSeek-V4.1 specialized decode on NVIDIA GPUs: a decode executor over the

  • vl_model –

    DeepSeek-V4.1 vision variant.