Large Language Model in Medical Domain
prompt Learning, RLHF related, downstream tasks
We released PULSE, a Chinese medical large language model and its related applications.
Model Description
We collected a dataset consists of textbooks, guidelines, EHR, medical & generic domain instruction tuning task, Q&A tasks, multi-round dialog, plugins to fine-tune a large language model (LLM) in medical domain.
A self-evaluation prompt is added in the reward model training and standard PPO framework are further optimized for better performance. Plugins for the downstream applications are under development.