BianQue

Introduction to the BianQue Project

BianQue is an innovative project centered on proactive health, focusing on key characteristics such as proactivity, prevention, precision, personalization, co-creation, and self-discipline. Developed by the School of Future Technology at South China University of Technology, in collaboration with the Guangdong Digital Twin Lab, BianQue serves as a foundational Chinese-language model for living space proactive health, known as ProactiveHealthGPT.

Components of ProactiveHealthGPT

ProactiveHealthGPT consists of two primary models:

• BianQue: A health dialogue model fine-tuned with large-scale Chinese health dialogue data.
• SoulChat: A model for mental health, fine-tuned with extensive long-text instructions and empathetic dialogue data.

Recent Updates

BianQue has undergone several updates to improve its functionalities and accuracy:

• July 7, 2023: SoulChat model's beta version was released online.
• July 1, 2023: BianQue was included in the China Model List for its multi-turn inquiry and advice capabilities, marking it as a pioneering open-source health model in China.
• June 6, 2023: Release of BianQue-2.0, enhancing its empathy and listening skills through detailed fine-tuning.
• April 22, 2023: Launch of BianQue-1.0, focusing on improving the questioning ability of medical chat models.

BianQue Health Big Data (BianQueCorpus)

BianQueCorpus addresses a critical issue in healthcare dialogue—users often fail to clearly describe their problems in a single interaction. Many existing models focus on addressing single-turn user descriptions, neglecting the aspect where users' descriptions might be insufficient. Unlike these models, BianQue employs a "Chain of Questioning (CoQ)" approach, where continuous inquiries are made by the doctor based on the user's descriptions, ensuring comprehensive information is gathered before providing advice.

To develop BianQueCorpus, BianQue integrates various open-source Chinese medical dialogue datasets along with proprietary data collected over time. This comprehensive dataset allows models to simulate the in-depth questioning process typical in real medical scenarios.

How to Use BianQue

Users can easily access and utilize BianQue by following these steps:

1. Clone the Project:

   cd ~
   git clone https://github.com/scutcyr/BianQue.git
   

2. Install Dependencies: Ensure that the appropriate version of torch is installed based on your server's CUDA version.

3. Set Up on Windows:

   • Follow specific instructions for setting up the environment and CUDA-11.6 on Windows.

4. Invoke BianQue Model in Python:

   • Load the model and tokenizer.
   • Execute single or multi-turn dialogues using the chat function.

5. Start the Service:

   • Use Streamlit to run BianQue's service and access it through a web interface.

Advancements and Features

BianQue has advanced through iterations, notably BianQue-2.0, initialized with the ChatGLM-6B model, and fine-tuned to enhance advice and knowledge retrieval capabilities. It incorporates various data, including medicinal instruction commands and medical encyclopedia knowledge.

Joint Usage

BianQue-2.0 and BianQue-1.0 can be used together to leverage strengths in multi-turn questioning and health advisory, offering a comprehensive proactive health service.

Community and Collaboration

The BianQue project encourages contributions and collaborations from educational institutions, hospitals, research labs, and companies. By collaborating, the project aims to advance the development of future iterations that integrate more sophisticated capabilities in line with the project's vision.

For those interested in contributing or collaborating on BianQue, communication and problem reporting are encouraged via the GitHub page or through direct email correspondence for more private discussions.