vllm.models.hy_v4.nvidia ¶

Modules:

• attention –

  MLA attention and lightning indexer for HY V4 (NVIDIA).

• flashmla_sparse –

  Sink-capable FlashMLA sparse backend for HY V4 (NVIDIA).

• hc –

  iHC (independent Hyper-Connections) layers for HY V4 (NVIDIA).

• model –

  Inference-only HY V4 model compatible with HuggingFace weights (NVIDIA).

• moe –

  Dense FFN and MoE blocks for HY V4 (NVIDIA).

• mtp –

  Multi-token prediction (MTP) head for HY V4 (NVIDIA).