CoolFace
Modelpublic

mlx-community/KAT-Coder-V2.5-Dev-OptiQ-4bit-REAP-18B

sourceHugging Faceapache-2.0updated 10d agoView on Hugging Face
7likes1.8kdownloads
6 commits on main
b33dfde10d ago

Update kv_config.json: re-measured with `optiq kv-cache` float32 sensitivity (the previous config was ranked on bfloat16 rounding noise). 8-bit layers [23, 35] -> [11, 15].

codelion
87fd0f411d ago

Add kv_config.json: per-layer KV cache precision from `optiq kv-cache` (target 4.5 bits). Use with `optiq serve --kv-config kv_config.json`.

codelion
01d1e7528d ago

card: fix summary table split by omitted rows

codelion
b0375d328d ago

card: drop empty benchmark table

codelion
ed90bef28d ago

OptiQ REAP: 128 of 256 experts retained

codelion
95bcd7b28d ago

initial commit

codelion