ChatDoctor is an AI model designed to act as a medical companion, answering patient health queries and providing personalized advice. It targets patients seeking accessible medical information and aims to offer accurate, conversational medical guidance by leveraging a large language model fine-tuned on extensive medical dialogue datasets.

How It Works

ChatDoctor is built upon the LLaMA architecture and fine-tuned using a combination of real patient-doctor conversations (from HealthCareMagic.com and icliniq.com) and synthetically generated medical data. This approach allows the model to understand colloquial symptom descriptions and provide contextually relevant medical advice, simulating a natural doctor-patient interaction. It can also optionally integrate with external knowledge bases like Wikipedia or a disease database for more informed responses.

Quick Start & Requirements

• Install via pip install -r requirements.txt within a conda environment with PyTorch.
• Requires LLaMA model weights placed in the ./pretrained folder.
• An interactive demo is available on Hugging Face Spaces: https://huggingface.co/spaces/kenton-li/chatdoctor_csv

Highlighted Details

• Fine-tuned on 100k real patient-doctor conversations from HealthCareMagic.com and 10k from icliniq.com.
• Can be further fine-tuned using provided scripts for full fine-tuning or LoRA.
• Supports autonomous knowledge retrieval from external sources like Wikipedia or a disease database.
• Demonstrates improved accuracy in medical diagnosis compared to general LLMs like ChatGPT in provided examples.

Maintenance & Community

• The project is associated with authors from UT Southwestern Medical Center, University of Illinois at Urbana-Champaign, Ohio State University, and Hangzhou Dianzi University.
• No specific community channels (Discord/Slack) or roadmap are mentioned in the README.

Licensing & Compatibility

• The project inherits LLaMA's non-commercial license.
• Explicitly prohibited for commercial use and clinical use due to licensing and lack of sufficient security measures.

Limitations & Caveats

The model is for academic research only and is not guaranteed to be 100% accurate; it should not be used in real clinical scenarios. The non-commercial license of the base LLaMA model restricts its use.