Mixed-precision quantized vision language model with an effective 5.064 bits per weight. Combines the size and speed benefits of a 4-bit quant with higher precision where it matters most.

mlx_vlm.convert --quantize --q-group-size 32 --quant-predicate mixed_4_6

🛠️ Customizations

This quant includes a bugfix for tools calling. It is aware of the current date, and also enables thinking (if available). You may disable this behavior by deleting the following line from the chat template, or changing true to false:

{%- set enable_thinking = true %}

You may need to adjust your environment’s Reasoning Section Parsing to recognize <|channel>thought as the Start String, and <channel|> as the End String.

🖥️ Use with `mlx`

pip install -U mlx-vlm

mlx_vlm.generate --model zecanard/gemma-4-31B-it-Claude-Opus-Distilled-v2-MLX-4bit-mixed_4_6 --max-tokens 100 --temperature 0 --prompt "Describe this image." --image <path_to_image>

zecanard/gemma-4-31B-it-Claude-Opus-Distilled-v2-MLX-4bit-mixed_4_6

🦆 zecanard/gemma-4-31B-it-Claude-Opus-Distilled-v2-MLX-4bit-mixed_4_6

🌟 Quality

🛠️ Customizations

🖥️ Use with mlx

zecanard/
gemma-4-31B-it-Claude-Opus-Distilled-v2-MLX-4bit-mixed_4_6

🖥️ Use with `mlx`