Can Qwen3-0.6B-MLX-4bit be used in Supervised Fine-tuning?

#3
by color-rabbit - opened

Can Qwen3-0.6B-MLX-4bit be used in Supervised Fine-tuning? If so, what's the suggested framework to use? Is there any sample code?

yes, it can. I tried and succeeded, using the MLX community in GitHub.
E.g.

mlx_lm.lora \         
    --model Qwen/Qwen3-0.6B-MLX-4bit \
    --train \
    --data ./data \
    --fine-tune-type lora \
    --batch-size 4 \
    --iters 1000 \
    --adapter-path ./qwen3-0.6b-lora-out

Sign up or log in to comment