metadata
tags:
- mlx
- transformers
- quantization
- dq3
MiniMax-M2.5_dq3
This model is a DQ3 quantized version of the original model MiniMax-M2.5.
It was quantized locally using the mlx_lm library.
Quantization Methodology (DQ3)
This model was quantized using the dynamic DQ3 (3-bit / 4-bit / 8-bit mixed) approach, inspired by the methodology described in the mlx-community/Kimi-K2.5-mlx-DQ3_K_M-q8 repository.
The weights are mixed based on MLX layers:
- Expert layers (switch_mlp / mlp) are quantized to 3-bit.
- The first 5 layers are kept at higher quality (5-bit).
- Every 5th layer is medium quality (4-bit).
- All other layers (e.g. attention, normalization) remain at 8-bit to serve as the "8-bit brain".