mlx-community/KAT-Coder-V2.5-Dev-OptiQ-4bit-REAP-18B
71.8k
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].
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`.
card: fix summary table split by omitted rows
card: drop empty benchmark table
OptiQ REAP: 128 of 256 experts retained
initial commit
