Published: May 21, 2025
8
10
114
Qwen 3 0.6B fine-tune as wine classifier using MLX. 91% accuracy! Pure power in a tiny model! Settings below 👇
batch_size: 8 iters: 3200 optimizer_config: adamw: betas: [0.9, 0.98] eps: 1e-6 weight_decay: 0.05 bias_correction: true lora_parameters: keys: [self_attn.k_proj", "self_attn.v_proj", "self_attn.q_proj", "self_attn.o_proj", "mlp.gate_proj",