Skip to content

vllm.models.glm5next.common

Modules:

  • attention –
  • kda –

    GLM-5.3-Flash KDA layer with separate convolutions and a bounded safe gate.

  • model –
  • mtp –
  • multimodal –

    GLM-5.3-Flash vision tower and multimodal processor.

  • sparse_indexer –

    Shared helpers for the glm5next sparse attention indexer (kpool) layers.