27. Tune large models with LoRA and adapters
Parameter-efficient tuning updates small parts of a large model instead of retraining everything. This chapter covers LoRA, QLoRA, adapters, rank choices, memory savings, merge decisions, and practical limits.