Commit History

  1. inference: Qwen3.5 support (VL unwrap, Gated-DeltaNet LoRA targets, linear_attn walk)

    Verified signature
    f91a401

  1. Phase A serving speedup: zero-copy state-cache expand (no deepcopy per decision)

    Verified signature
    5c02833
  2. Serving-path speedup: state-KV LRU cache + suffix diet (see inference/bench_serve.py)

    Verified signature
    5ca406a
  3. Serving-path speedup: state-KV LRU cache + suffix diet (see inference/bench_serve.py)

    Verified signature
    68ae554

  1. Add inference/: minimal self-contained Typical inference package

    Verified signature
    002c12c