MING: A Medical Inquiry Model for Chinese Language
MING (pronounced as "Ming Yi" in Chinese) stands for a comprehensive and open-source project designed to enhance medical consultations using a finely-tuned Chinese language model. The purpose of the project is to address and analyze medical inquiries, providing automated, intelligent responses in the Chinese language.
Key Features of MING
The core functions of the MING model are diverse, focusing primarily on:
- Medical Question-Answering: MING provides detailed answers to medical questions, including analyzing symptoms and offering case studies.
- Intelligent Consultation: It supports multi-turn dialogue, where users can interact with MING through a series of questions and responses. This feature helps deliver relevant diagnostic results and suggestions after thoroughly understanding the patient's input.
Research Backing
MING is built upon cutting-edge research in the field of medical artificial intelligence. Several key papers have been published to document the foundations and advancements related to its development:
- MING-MOE Technical Report: This paper explores enhancements in medical multi-task learning through the use of a sparse mixture of low-rank adapter experts.
- Automatic Interactive Evaluation Framework: Discusses an evaluation framework for assessing large language models using state-aware patient simulators.
- Two-Stage Decoupled Learning for Clinical Model Alignment: Advances in aligning clinical models through a decoupled learning approach.
- Reflection-Aware Tool-Augmented Clinical Agents: Focuses on adaptive learning and multi-dimensional evaluation benchmarks in medical AI.
Open Source Models
MING offers several pre-trained models that serve different purposes and capacities, available on Hugging Face hub. Major variants include:
- MING-7B: Based on the bloomz-7b1-mt platform for detailed and elaborate responses.
- MING-1.8B and MING-MOE Series: These models leverage Qwen architecture, ranging from 1.8B to 14B parameters, each fine-tuned for enhanced performance in producing informative and contextually aware responses.
How to Get Started
To utilize MING effectively, users are encouraged to follow these initial setup steps:
- Configure Environment: Use python 3.9 or higher, along with PyTorch 2.0.1.
- Install Dependencies: Clone the repository from GitHub and install the necessary packages.
- Run Models: Choose a model variant and follow specific instructions to start the dialogue interface, ensuring the correct configuration for your hardware capacity.
Practical Examples and Applications
- Multi-Session Dialogue: Supports conversation over multiple turns allowing for extensive and coherent exchanges.
- Reset Dialogue: Users can start a new conversation cycle using specific keywords to reset the interaction context.
Contributions and Acknowledgments
This project has been a joint effort from the Shanghai Jiao Tong University's Future Media Network Collaboration Center along with the Wisdom Medical Center of Shanghai AI laboratory. The team, led by Assistant Professor Yu Wang, is responsible for developing the system, model data, and underlying codebase.
Disclaimer
While MING is an advanced tool for generating informative medical advice, it should not be used as a substitute for professional medical advice or decision-making due to inherent biases and limitations in its training datasets. Users are advised to verify any information provided through MING independently.
Citation
For those referencing this project, appropriate citation formats are provided to acknowledge the work and contributions made by the team in creating and developing MING.
By presenting a sophisticated model in an easy-to-understand manner, MING aims to revolutionize how medical inquiries are handled in the Chinese language, fostering accessibility and accuracy in medical communications.