Generate code training data and fine‑tune a JAX model on CPU
Fine-tune Qwen3-0.6B with LoRA using custom data