Introduction to the Sunsimiao Project
The Sunsimiao project, named after the esteemed Tang Dynasty physician and Taoist Sun Simiao, aims to revolutionize the accessibility and reliability of Chinese medical knowledge through advanced language modeling. The project aspires to mirror its namesake's emphasis on incorporating folk medicine experiences and creating a widely beneficial medical resource.
Project Progress
Sunsimiao is an ever-evolving project, with significant advancements achieved recently:
- July 6, 2023: The initial release of the Sunsimiao Chinese medical large model signified the project's promising beginnings.
- July 23, 2024: The open-sourcing of the 7B-parameter model, Sunsimiao-7B, marked a milestone, showcasing capabilities in medical queries and examination scenarios refined through the high-caliber Qwen2-7B model.
- September 9, 2024: An online interface for the Sunsimiao-7B model was launched, allowing broader user interaction.
- September 12, 2024: Sunsimiao-7B became available for download from platforms like WiseModel for more extensive integration.
Model Overview
The model draws inspiration from Sun Simiao's historical contributions and strives to offer a dependable and comprehensive Chinese medical model. The Sunsimiao-7B model, representing the project's flagship, has achieved state-of-the-art results in the CMB-Exam, demonstrating proficiency in national medical, pharmaceutical, and nursing licensure examinations in China.
Model List
The Sunsimiao project encompasses various model versions, including:
- Sunsimiao-7B: Available on ModelScope, WiseModel, and HuggingFace platforms.
- Sunsimiao-01M series: Including Chat and 6B variations, accessible for different use cases and research purposes.
Dataset Details
The foundation of Sunsimiao lies in a meticulously curated medical data pool. This data corpus integrates diverse medical literature, textbooks, diagnostic records from multiple departments, extensive medical consultations, knowledge Q&A, and clinical case analyses. Data is sourced from open resources and generated using GPT-4, enriched through manual annotation and automated analysis.
Model Evaluation
The Sunsimiao-7B model has shown remarkable performance in the CMB-Exam and various medical qualification exams, achieving high accuracy rates, such as 79.23% in the National Practitioner Examination. This performance underscores the model's training depth and practical utility.
Usage Guide
Utilizing the Sunsimiao models involves integrating them within text generation pipelines for responses to medical inquiries. Here's a basic Python example for usage:
from modelscope.pipelines import pipeline
from modelscope.utils.constant import Tasks
pipe = pipeline(task=Tasks.text_generation, model='AI-ModelScope/Sunsimiao', model_revision='v1.0.0')
query = 'What should I do if a child has a fever?'
prompt="Below is an instruction that describes a task. Write a response that appropriately completes the request."
prompt+="### Instruction:\n{}\n\n### Response:\n".format(query)
result = pipe(prompt)
print(result)
Acknowledgements
The Sunsimiao project was initiated by Associate Professor Dong Xue from East China University of Science and Technology and carries the support of various partners, including LLaMA, OpenI, and HuggingFace, who provide essential tools and computational facilities.
Disclaimer
The Sunsimiao models are offered without warranties and should not be used as a sole decision-making resource in critical scenarios. Users are cautioned about potential misuses and assume responsibility for interpretation and application.
Citation
Researchers and developers utilizing Sunsimiao in projects are encouraged to cite the following repository:
@misc{Sunsimiao,
author={Xin Yan, Dong Xue*},
title = {Sunsimiao: Chinese Medicine LLM},
year = {2023},
publisher = {GitHub},
journal = {GitHub repository},
howpublished = {\url{https://github.com/thomas-yanxin/Sunsimiao}},
}
The Sunsimiao project is committed to expanding the frontiers of Chinese medical knowledge and delivering a model that echoes the legacy of Sun Simiao in promoting health practices rooted in historical wisdom.