| tags: | |
| - mlx | |
| - transformers | |
| - quantization | |
| - dq4 | |
| # MiniMax-M2.5_dq4 | |
| This model is a DQ4 quantized version of the original model [MiniMax-M2.5](https://huggingface.co/MiniMaxAI/MiniMax-M2.5). | |
| It was quantized locally using the `mlx_lm` library. | |
| ## Quantization Methodology (DQ4) | |
| This model was quantized using the dynamic **DQ4** (4-bit / 5-bit / 6-bit / 8-bit mixed) approach, inspired by the methodology described in the [mlx-community/Kimi-K2.5-mlx-DQ3_K_M-q8](https://huggingface.co/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 4-bit. | |
| - The first 5 layers are kept at higher quality (6-bit). | |
| - Every 5th layer is medium quality (5-bit). | |
| - All other layers (e.g. attention, normalization) remain at 8-bit to serve as the "8-bit brain". | |