Larger version?
#2
by
Carthin
- opened
I noticed that this is significantly smaller than the mlx version of r1, despite the base model being more parameters. Is that a choice, or just how it works out? I would like something in the range of 400-425 gb rather than 350 if possible, in MLX.
Edit: I understand that MLX supports Q4_1 quantization, which I would prefer over this that I assume is Q4_0
I would just download the full and create the mlx yourself. It looks like you can create a mixture 4 and 6 bit. I plan on trying this out once my studio arrives.