# 明医 (MING)：中文医疗问诊大模型
<p align="center">
<img src=".\img\bgimage.png" width=800px/>
</p>
<div align="center"><img src="https://img.shields.io/badge/Version-1.3--alpha-brightgreen"> <img src="https://img.shields.io/badge/Code%20License-Apache_2.0-green.svg"> <img src="https://img.shields.io/badge/python-3.9+-blue.svg"></div>
## 🌐项目简介
本项目开源了基于医疗指令微调的中文医疗问诊模型：**明医 (MING)**。目前模型的主要功能如下：
<!DOCTYPE html>
<html>
<body>
<table style="width: 100%;">
<tr style="border-collapse: collapse; border: transparent;">
<td style="width: 50%; border-collapse: collapse;border: transparent;"><img src=".\img\demo1.gif" alt="demo1"/></td>
<td style="width: 50%; border-collapse: collapse;border: transparent;"><img src=".\img\demo2.gif" alt="demo2"/></td>
</tr>
<tr style="border-collapse: collapse; border: transparent;">
<td style="width: 50%; border-collapse: collapse;border: transparent;" ><div align="center"><strong>医疗问答</strong>：对医疗问题进行解答，对案例进行分析。</div></td>
<td style="width: 50%; border-collapse: collapse;border: transparent;"><div align="center"><strong>智能问诊</strong>：多轮问诊后给出诊断结果和建议。</div></td>
</tr>
</table>
</body>
</html>
## 📄相关论文
* MING-MOE技术报告
[MING-MOE: Enhancing Medical Multi-Task Learning in Large Language Models with Sparse Mixture of Low-Rank Adapter Experts](https://arxiv.org/pdf/2404.09027.pdf)
* 基于多智能体交互的大语言模型多轮问诊自动评估框架
[Automatic Interactive Evaluation for Large Language Models with State Aware Patient Simulator](https://arxiv.org/pdf/2403.08495.pdf)
## 💫更新
* 🔥 [2024/04/14] 开源了基于Qwen1.5指令微调的专家混合模型MING-MOE
* [2024/03/14] 开源了基于Qwen1.5-1.8b指令微调的MING-1.8B
* [2023/07/25] 开源了基于bloomz-7b指令微调的MING-7B
* [2023/07/25] MedicalGPT-zh更名为**MING**
## 🔬开源模型
<!DOCTYPE html>
<html>
<head>
</head>
<body>
<table style="width: 80%;">
<tr>
<td style="width: 20%;"><div align="center"><strong>模型</strong></div></td>
<td style="width: 20%;"><div align="center"><strong>基座</strong></div></td>
<td style="width: 30%;"><div align="center"><strong>HuggingFace</strong></div></td>
</tr>
<tr>
<td><center>MING-7B</center></td>
<td><center><a href="https://huggingface.co/bigscience/bloomz-7b1-mt">bloomz-7b1-mt</a></center></td>
<td><center>🤗<a href="https://huggingface.co/BlueZeros/MING-7B">MING-7B</a></center></td>
</tr>
<tr>
<td><center>MING-1.8B</center></td>
<td><center><a href="https://huggingface.co/Qwen/Qwen1.5-1.8B-Chat">Qwen1.5-1.8B</a></center></td>
<td><center>🤗<a href="https://huggingface.co/BlueZeros/MING-1.8B">MING-1.8B</a></center></td>
</tr>
<tr>
<td><center>MING-MOE-1.8B</center></td>
<td><center><a href="https://huggingface.co/Qwen/Qwen1.5-1.8B-Chat">Qwen1.5-1.8B</a></center></td>
<td><center>🤗<a href="https://huggingface.co/BlueZeros/MING-MOE-1.8B">MING-MOE-1.8B</a></center></td>
</tr>
<tr>
<td><center>MING-MOE-4B</center></td>
<td><center><a href="https://huggingface.co/Qwen/Qwen1.5-4B-Chat">Qwen1.5-4B</a></center></td>
<td><center>🤗<a href="https://huggingface.co/BlueZeros/MING-MOE-4B">MING-MOE-4B</a></center></td>
</tr>
<tr>
<td><center>MING-MOE-7B</center></td>
<td><center><a href="https://huggingface.co/Qwen/Qwen1.5-7B-Chat">Qwen1.5-7B</a></center></td>
<td><center>🤗<a href="https://huggingface.co/BlueZeros/MING-MOE-7B">MING-MOE-7B</a></center></td>
</tr>
<tr>
<td><center>MING-MOE-14B</center></td>
<td><center><a href="https://huggingface.co/Qwen/Qwen1.5-14B-Chat">Qwen1.5-14B</a></center></td>
<td><center>🤗<a href="https://huggingface.co/BlueZeros/MING-MOE-14B">MING-MOE-14B</a></center></td>
</tr>
</table>
</body>
</html>
## ⚡快速开始
1. 配置环境（测试环境如下，具体版本可以根据实际需求配置）
* python==3.9.16
* pytorch==2.0.1+cu117
* peft==0.9.0
2. 安装项目依赖
```bash
git clone https://github.com/MediaBrain-SJTU/MING
cd MING
pip install -e .
```
2. 下载模型参数并运行（要求单卡显存 >= 15G）
* MING-MOE
```bash
CUDA_VISIBLE_DEVICES=0 python -m ming/serve/cli.py \
--model_path {path_to_checkpoint} \ # 模型路径
--model_base {path_to_base_model} \ # 基座模型路径
--max_new_token 3072 # 输出最大长度
```
* MING-1.8B
```bash
CUDA_VISIBLE_DEVICES=0 python -m ming/serve/cli.py \
--model_path {path_to_checkpoint} \ # 模型路径
--max_new_token 2048 # 输出最大长度
```
* MING-7B
```bash
CUDA_VISIBLE_DEVICES=0 python -m ming/serve/cli.py \
--model_path {path_to_checkpoint} \ # 模型路径
--conv_template bloom \ # prompt
--max_new_token 512 \ # 输出最大长度
--beam_size 3 \ # beam search宽度
--temperature 1.2 # 采样温度
```
* 注：由于transformers库的问题，当beam-size > 1时，需要满足temperature>=1.0，否则会报错。
4. 命令行运行实例
* 对话支持多轮
* 对话中输入关键词 `new chat` 能够开启新一轮对话。
## 🧭测试样例
<p align="center">
<img src=".\img\case1.png" width=800px/>
</p>
<p align="center">
<img src=".\img\case2.png" width=800px/>
</p>
## 🪶贡献
本项目由上海交通大学未来媒体网络协同创新中心和上海人工智能实验室智慧医疗中心合作研发。模型数据系统主要由廖育生，江书洋，刘泓呈，孟昱同完成，指导教师为[王钰](https://mediabrain.sjtu.edu.cn/yuwang/)副教授。
## 免责声明
预训练模型是基于大量语料库和算法模型进行训练的，并且在训练过程中可能存在偏差、错误和不完整的信息。因此，本项目提供的预训练模型仅供参考和研究使用，并不能保证其准确性和可靠性。使用预训练模型产生的结果可能存在误差和偏差，不能用于实际应用或决策。本项目不对使用预训练模型所产生的结果承担任何责任，也不对因使用预训练模型所产生的任何损失承担责任。使用者在使用预训练模型时应自行承担风险并进行自我验证。
## 引用
如果你使用了本项目的数据或者代码，请声明引用
```latex
@article{liao2024ming,
title={MING-MOE: Enhancing Medical Multi-Task Learning in Large Language Models with Sparse Mixture of Low-Rank Adapter Experts},
author={Liao, Yusheng and Jiang, Shuyang and Wang, Yu and Wang, Yanfeng},
journal={arXiv preprint arXiv:2404.09027},
year={2024}
}
```
```latex
@misc{MING,
author={Yusheng Liao, Yutong Meng, Hongcheng Liu, Yu Wang, Yanfeng Wang},
title = {明医 (MING)：中文医疗问诊大模型},
year = {2023},
publisher = {GitHub},
journal = {GitHub repository},
howpublished = {\url{https://github.com/MediaBrain-SJTU/MING}},
}
```
《AI大模型应用》-明医 (MING)：中文医疗问诊大模型.zip （110个子文件） 
cmexam_test_01_clinical_depart.csv 554KB cmexam_test_01_difficulty_level.csv 546KB
demo2.gif 11.7MB demo1.gif 6.31MB .gitignore 58B zero3_offload.json 1KB zero3.json 801B zero2.json 556B mmedbench-05-english-tiny.jsonl 1.52MB mmedbench-02-chinese-tiny.jsonl 711KB LICENSE 11KB README.md 7KB README.md 6KB PKG-INFO 8KB
case1.png 6.31MB case2.png 4.51MB bgimage.png 4.17MB guideline_qa.png 600KB utils.py 27KB train.py 25KB post_generate_process.py 20KB eval_em.py 18KB evaluate.py 17KB modeling_molora_qwen.py 14KB trainer.py 12KB conversations.py 12KB model_diverse_gen.py 12KB text2dt_eval_func.py 11KB get_attention.py 10KB self_distill.py 10KB huatuo_med_qa.py 9KB evaluators.py 9KB model_mmedbench_qa.py 8KB eval_gpt4.py 8KB builder.py 8KB model_med_qa.py 7KB rouge.py 6KB cli.py 5KB gpt4_evaluator.py 5KB inference.py 5KB ming_flash_attn_monkey_patch.py 5KB average_attention.py 2KB gen_tiny_mmedbench.py 2KB eval_humaneval.py 2KB gt_evaluator.py 2KB gen_tiny_cmexam.py 2KB merge_lora_molora_weights.py 2KB utils.py 756B inferece.py 666B train_mem.py 498B app.py 127B __init__.py 70B constants.py 20B __init__.py 0B __init__.py 0B __init__.py 0B __init__.py 0B modeling_phi.cpython-38.pyc 39KB train.cpython-310.pyc 18KB train.cpython-38.pyc 18KB train.cpython-39.pyc 18KB utils.cpython-310.pyc 15KB utils.cpython-39.pyc 15KB eval_em.cpython-310.pyc 11KB utils.cpython-38.pyc 11KB modeling_molora_qwen.cpython-310.pyc 9KB modeling_molora_qwen.cpython-39.pyc 9KB trainer.cpython-38.pyc 9KB trainer.cpython-310.pyc 9KB model_diverse_gen.cpython-310.pyc 8KB conversations.cpython-310.pyc 8KB conversations.cpython-39.pyc 8KB model_diverse_gen.cpython-39.pyc 8KB evaluate.cpython-310.pyc 8KB conversations.cpython-38.pyc 8KB eval_em.cpython-39.pyc 7KB post_generate_process.cpython-310.pyc 6KB eval_gpt4.cpython-310.pyc 6KB get_attention.cpython-310.pyc 6KB text2dt_eval_func.cpython-310.pyc 6KB evaluators.cpython-310.pyc 5KB builder.cpython-39.pyc 4KB builder.cpython-310.pyc 4KB inference.cpython-310.pyc 4KB inference.cpython-39.pyc 4KB builder.cpython-38.pyc 4KB average_attention.cpython-310.pyc 2KB utils.cpython-310.pyc 1KB utils.cpython-38.pyc 1KB utils.cpython-39.pyc 1KB __init__.cpython-39.pyc 252B __init__.cpython-310.pyc 245B __init__.cpython-38.pyc 177B __init__.cpython-39.pyc 162B __init__.cpython-310.pyc 160B __init__.cpython-39.pyc 156B __init__.cpython-310.pyc 155B __init__.cpython-310.pyc 154B __init__.cpython-310.pyc 149B get_attention_parallel.sh 1KB共 110 条
- 1
- 2
资源评论
