vllm.models.qwen4_exp.nvidia.ops ¶

NVIDIA-only Qwen4Exp kernels; import leaf modules directly.

Modules:

• hc –

  NVIDIA HyperConnection kernels for Qwen4Exp.

• ple –

  Fused Qwen4Exp PLE kernels.

• qsa –

  Triton kernels for Qwen4Exp QSA sparse attention and cache updates.

• qsa_indexer –

  Triton kernels for Qwen4Exp QSA index selection.

• qsa_pre_indexer –

  Fused QSA pre-indexer kernel for Qwen4Exp.