Full-parameter fine-tuning yields better results but at a higher cost, while parameter-efficient methods such as LoRA can achieve good results under limited resources.