Researchers have developed a family of large language models, ranging from 4 to 14 billion parameters, specifically designed for traditional Chinese medicine (TCM) clinical reasoning. These models utilize a novel staged clinical alignment framework that incorporates continued pretraining on TCM data, complementary reasoning supervision, and prescription-centered reinforcement learning. The 14 billion parameter model, after reinforcement learning, achieved a prescription generation F1 score of 37.0 on a public benchmark and a herb-level F1 of 52.8 on a private clinical dataset, outperforming the larger DeepSeek-R1 model on certain metrics. AI
IMPACT These models could advance AI's role in specialized medical fields, offering clinical decision support for traditional medicine practices.
RANK_REASON The cluster describes a research paper detailing the development of new LLMs for a specific domain. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
