finding1/DeepSeek-V3.1-Terminus-MLX-mixed_4_6-v0.28.2

This model finding1/DeepSeek-V3.1-Terminus-MLX-mixed_4_6-v0.28.2 was converted to MLX format from deepseek-ai/DeepSeek-V3.1-Terminus using mlx-lm version 0.28.2 mlx_lm.convert --quantize --q-bits 4 --quant-predicate mixed_4_6 --hf-path deepseek-ai/DeepSeek-V3.1-Terminus --mlx-path DeepSeek-V3.1-Terminus-MLX-mixed_4_6-v2. The console reported 4.811 bits per weight. The difference between this and finding1/DeepSeek-V3.1-Terminus-MLX-mixed_4_6 is using 6 bits instead of 4 bits for some self-attention tensors.