Fine-tuning and Utilization Methods of Domain-specific LLMs