vllm.models.kimi_k3.nvidia ¶

Modules:

• dspark_mla –

  K3 dense MLA draft model for DSpark speculative decoding.

• kda –
• kda_metadata –

  Kimi-K3 specialization of GDN attention metadata.

• latent_moe_runner –
• low_latency_gemm –

  Kimi-K3 decode GEMM selection for unquantized BF16 on SM103.

• mla –

  Clean Multi-head Latent Attention for Kimi-K3 (NVIDIA).

• model –

  Kimi-K3 multimodal model implementation for vLLM.

• mtp –

  Inference-only Kimi-K3 Multi-Token-Prediction (MTP) draft model.

• ops –