Error: [quantize] The requested number of bits 3 is not supported. The supported bits are 2, 4 and 8.

#34
by Felladrin - opened
MLX Community org

The quantization options Q3 and Q6 were added in https://huggingface.co/spaces/mlx-community/mlx-my-repo/discussions/31

But when we try converting, it throws the error:

Error: [quantize] The requested number of bits 3 is not supported. The supported bits are 2, 4 and 8.

image.png

MLX Community org

Looking into it!

MLX Community org

Should be fixed as soon as @awni puts out a CPU build πŸ€—

MLX Community org

This should work now: I just created a Q3 to test: https://huggingface.co/reach-vb/Qwen2.5-0.5B-Instruct-Q3-mlx

reach-vb changed discussion status to closed
MLX Community org

Thank you for fixing it so fast! πŸ™

Sign up or log in to comment