Design a fine-tuning pipeline — PEFT config, dataset format, training loop, and evaluation. Use when asked to "fine-tune a model", "set up a LoRA config", or "should we fine-tune or prompt".
