Skip to content

vllm.models.deepseek_v4.nvidia

Modules:

  • dspark –

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

  • flashinfer_sparse –

    DeepSeek V4 FlashInfer sparse MLA backend.

  • flashmla –
  • model –
  • mtp –

    MTP draft model for DeepSeek V4 (internal codename: DeepseekV4).

  • ops –

    NVIDIA-only (cutedsl/cutlass) kernels for DeepSeek V4.

  • vl_model –

    Compatibility imports for the platform-neutral DeepSeek-V4 vision model.