30. Adapt models with fine-tuning and lightweight training
Customize models with supervised fine-tuning, adapters, LoRA, instruction data, preference data, and evaluation sets. You will know when fine-tuning is worth the cost and when prompting or retrieval is better.