Mixed-precision MLX quantization of huihui-ai/Huihui-gemma-4-12B-it-qat-q4_0-unquantized-abliterated, quantized with MLX Smart Quantize (MSQ) — my own sensitivity-based mixed-precision quantization method for Apple Silicon. It measures per-layer NMSE and assigns optimal bit widths automatically, combining architecture knowledge with measured data.