Let the users fine-tune LLM for classification tasks, so it can achieve good accuracy without high labeling efforts and competing with state-of-the-art performances