mlx-community

/

Qwen1.5-0.5B-Chat-4bit

like 4

Follow

MLX Community 2.83k

I use this mlx command to convert the model to 4bits:
python -m mlx_lm.convert --hf-path Qwen/Qwen1.5-0.5B-Chat -q --q-bits 4
the result shows that the model.safetensors is 500+M size, but mlx-community/Qwen1.5-0.5B-Chat-4bit only got 261M

Is there any difference ?