What is model fine-tuning, and how does it differ from model distillation?