MING (明医) is an open-source Chinese medical consultation large language model, designed for medical question answering, case analysis, and intelligent multi-turn diagnosis. It targets researchers and developers in the medical AI domain, offering specialized capabilities for Chinese medical dialogue.

How It Works

MING utilizes instruction fine-tuning on various base models, including Qwen and BLOOMZ, to achieve its medical dialogue capabilities. Recent updates introduce MING-MOE, a Mixture of Experts model leveraging sparse Mixture of Low-Rank Adapter Experts for enhanced multi-task learning in medical contexts. This approach aims to improve efficiency and performance on specialized medical tasks.

Quick Start & Requirements

• Install via pip install -e . after cloning the repository.
• Requires Python 3.9.16, PyTorch 2.0.1+cu117, and PEFT 0.9.0.
• Running models requires a GPU with at least 15GB VRAM.
• Example CLI commands are provided for MING-MOE, MING-1.8B, and MING-7B.
• Official documentation and technical reports are linked.

Highlighted Details

• Offers multiple model sizes and base architectures (e.g., MING-7B, MING-1.8B, MING-MOE variants).
• Supports multi-turn conversations with a "new chat" command for resetting context.
• Several related research papers are available, detailing evaluation frameworks and alignment methods.

Maintenance & Community

Developed collaboratively by Shanghai Jiao Tong University and Shanghai AI Laboratory. Key contributors and supervisors are listed.

Licensing & Compatibility

The repository does not explicitly state a license. The disclaimer warns against using the model for actual applications or decision-making, stating it's for reference and research use only, with users assuming all risks.

Limitations & Caveats

The project disclaimer strongly advises against using the models for practical applications or decision-making due to potential biases, errors, and incompleteness in the training data and algorithms. Users must perform their own validation.