DISC-MedLLM
| Property | Value |
|---|---|
| Base Model | Baichuan-13B-Base |
| Language | Chinese |
| License | Apache 2.0 |
| Research Paper | arXiv:2308.14346 |
What is DISC-MedLLM?
DISC-MedLLM is a specialized medical domain large language model developed by Fudan-DISC lab, designed specifically for conversational healthcare scenarios. Built upon the Baichuan-13B architecture, it has been trained on a comprehensive dataset of over 470,000 medical examples to bridge the gap between general language models and real-world medical consultations.
Implementation Details
The model leverages a goal-oriented training strategy using the DISC-Med-SFT dataset, which combines real-world conversations, knowledge graph-derived QA pairs, and human preference-aligned data. The implementation utilizes PyTorch and Transformers frameworks, supporting both full-parameter and quantized inference options.
- Training dataset includes 420k real-world conversations, 50k knowledge graph QA pairs, and 2k human preference examples
- Implements a specialized dialogue format with specific user and assistant tokens (195 and 196)
- Supports both full-parameter and quantized (int8, int4) deployment options
Core Capabilities
- Knowledge-intensive medical consultations and treatment inquiries
- Multi-turn medical dialogue support
- Human preference alignment in responses
- Integration with medical knowledge graphs
- Reliable and context-aware medical information delivery
Frequently Asked Questions
Q: What makes this model unique?
DISC-MedLLM stands out through its specialized focus on Chinese medical consultations, comprehensive training data incorporating real doctor-patient dialogues, and integration with medical knowledge graphs. Its goal-oriented training strategy ensures high-quality health support services while maintaining alignment with human preferences.
Q: What are the recommended use cases?
The model is specifically designed for research and testing in medical consultation scenarios, including patient inquiries, treatment discussions, and health support services. However, it should not be used as a replacement for professional medical advice, and all outputs should be critically assessed.
