mlx-community/Qwen3.6-35B-A3B-OptiQ-4bit

Correct BFCL + Capability: the AST checker could not match array-typed arguments

index: register the bf16 vision tower so stock VLM loaders (mlx-vlm, oMLX, LM Studio) can find it

Move OptiQ sidecars under optiq/ so *.safetensors loaders (mlx-vlm, LM Studio) skip them

docs: OptiQ brand spelling

card: remove em-dashes, ensure funnel

card: add mlx-optiq funnel banner + CTA

Restore vision_config + optiq_vision metadata for the vision sidecar

Add bf16 vision sidecar (Qwen3-VL tower, 333 tensors) for image+text inference

Add kv_config.json (per-layer mixed-precision KV cache, 5.0 BPW target from OptiQ kv-cache sensitivity analysis)

v0.1.0: add Qwen-recommended sampling defaults to generation_config.json so Lab can surface them

v0.1.0: 6-metric Capability Score + OptIQ vs U4 deltas

v0.1.0: re-quant with no-strip pipeline + attention-aware floor + 6-domain calibration

Re-eval GSM8K with chat-template + enable_thinking=False: OptIQ 89.5% / uniform 91.5% (-2.0pp)

Add GSM8K benchmark: OptiQ 72.0% vs uniform 4-bit 45.5% (+26.5pp)

Drop Apple Silicon notes section per convention (no notes blocks)

Initial upload: Qwen3.6-35B-A3B-OptiQ-4bit (MoE 256 experts, 4.88 BPW, 20 GB)

initial commit