Huihui-gemma-4-12B-it-qat-q4_0-unquantized-abliterated — MLX 5.0 BPW
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.
Details
- Type: Vision (VLM)
- Average: 5.01 bits per weight
- Method: MLX Smart Quantize (MSQ)