Skip to content

vllm.models.deepseek_v4.cpu

Modules:

  • cpu_compressor –

    CPU DeepSeek-V4 compressor subclass and dispatch: partial-state-cache

  • cpu_mla –

    CPU DeepSeek-V4 sparse-MLA backend descriptor.

  • cpu_sparse –

    CPU DeepSeek-V4 attention subclass.

  • cpu_utils –

    Shared helpers for DeepSeek-V4's CPU-ported kernels.

  • dspark –

    CPU DeepSeek-V4 DSpark (speculative decoding) — deferred, not implemented.

  • model –
  • mtp –

    CPU DeepSeek-V4 MTP (speculative decoding) — deferred, not implemented.