Commit History
-
inference: Qwen3.5 support (VL unwrap, Gated-DeltaNet LoRA targets, linear_attn walk)
Verified signaturecc06a04
-
Phase A serving speedup: zero-copy state-cache expand (no deepcopy per decision)
Verified signature3617254 -
Serving-path speedup: state-KV LRU cache + suffix diet (see inference/bench_serve.py)
Verified signature61830b8 -
Serving-path speedup: state-KV LRU cache + suffix diet (see inference/bench_serve.py)
Verified signatured7738b8
-
Add inference/: minimal self-contained Typical inference package
Verified signatureac4bc1e