A Medical Diagnostic Assistant Based on LLM

With the advent of ChatGPT, large language models (LLMs) have received extensive attention because of their excellent instruction comprehension and generation capabilities. However, LLMs are not specifically designed for the healthcare domain and still lack accuracy in answering specialized healthcare-related questions. In this paper, we mainly used some approaches to improve the performance of large language models in the medical domain. First, we analyzed and processed data to ensure high quality and consistency. Second, we used the model’s excellent ability to generate inference process to the training data. Finally, the data with the explanation and inference process, which are helpful in guiding the thinking and improving the inference ability of the model, are used for training. In terms of model training, we used ChatGLM2-6B as the base model, and the large language model was fine-tuned using the QLoRA framework. To guide the model to generate compliant outputs better, we also explored and carefully constructed appropriate prompts. Overall, our approachs enable the model to achieve the F1 value of 0.433 in this task.