Top AI Repos — open-source AI, indexed and scored

Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.

FudanDISC/DISC-MedLLM

Repository of DISC-MedLLM, it is a comprehensive solution that leverages Large Language Models (LLMs) to provide accurate and truthful medical response in end-to-end conversational healthcare services.

43Apache-2.0Python

View on GitHub

Stars: 565
Forks: 53
Watchers: 3
Open issues: 16
Contributors: 3

Star history, last 90 days— over 3 snapshots

2026-07-31 · 5652026-08-06 · 565

Star history, last 90 days
Date	Stars
2026-07-31	565
2026-08-05	565
2026-08-06	565

Today

— stars today

This week

— stars this week

This month

— stars this month

Momentum

0.0

growth rate 0.00%/day

From the README

# DISC-MedLLM

<div align="center">
  
[![Generic badge](https://img.shields.io/badge/🤗-Huggingface%20Repo-green.svg)](https://huggingface.co/Flmc/DISC-MedLLM)
[![license](https://img.shields.io/github/license/modelscope/modelscope.svg)](https://github.com/FudanDISC/DICS-MedLLM/blob/main/LICENSE)
<br>
</div>
<div align="center">

[Demo](http://med.fudan-disc.com) | [技术报告](https://arxiv.org/abs/2308.14346)
<br>
中文 | [EN](https://github.com/FudanDISC/DISC-MedLLM/blob/main/README_EN.md)
</div>
  
DISC-MedLLM 是一个专门针对医疗健康对话式场景而设计的医疗领域大模型，由[复旦大学数据智能与社会计算实验室 (Fudan-DISC)](http://fudan-disc.com) 开发并开源。

该项目包含下列开源资源:
* [DISC-Med-SFT 数据集](https://huggingface.co/datasets/Flmc/DISC-Med-SFT) (不包括行为偏好训练数据)
* DISC-MedLLM 的[模型权重](https://huggingface.co/Flmc/DISC-MedLLM)

您可以通过访问这个[链接](http://med.fudan-disc.com)来试用我们的模型。

## 概述

DISC-MedLLM 是一个专为医疗健康对话场景而打造的领域大模型，它可以满足您的各种医疗保健需求，包括疾病问诊和治疗方案咨询等，为您提供高质量的健康支持服务。

DISC-MedLLM 有效地对齐了医疗场景下的人类偏好，弥合了通用语言模型输出与真实世界医疗对话之间的差距，这一点在实验结果中有所体现。

得益于我们以目标为导向的策略，以及基于真实医患对话数据和知识图谱，引入LLM in the loop 和 Human in the loop的多元数据构造机制，DISC-MedLLM 有以下几个特点：

* **可靠丰富的专业知识**，我们以医学知识图谱作为信息源，通过采样三元组，并使用通用大模型的语言能力进行对话样本的构造。
* **多轮对话的问询能力**，我们以真实咨询对话纪录作为信息源，使用大模型进行对话重建，构建过程中要求模型完全对齐对话中的医学信息。
* **对齐人类偏好的回复**，病人希望在咨询的过程中获得更丰富的支撑信息和背景知识，但人类医生的回答往往简练；我们通过人工筛选，构建符合人类偏好的高质量的小规模行为微调样本，对齐病人的需求。

<img src="https://github.com/FudanDISC/DISC-MedLLM/blob/main/images/data_construction.png" alt="data-construction" width="85%"/>

## 模型效果演示
### 疾病问诊
<img src="https://github.com/FudanDISC/DISC-MedLLM/blob/main/images/consultation.gif" alt="sample1" width="60%"/>

### 治疗方案咨询
<img src="https://github.com/FudanDISC/DISC-MedLLM/blob/main/images/advice.gif" alt="sample2" width="60%"/>

## 数据集

为了训练 DISC-MedLLM ，我们构建了一个高质量的数据集，命名为 DISC-Med-SFT，其中包含了超过47万个衍生于现有的医疗数据集重新构建得到的样本。我们采用了目标导向的策略，通过对于精心选择的几个数据源进行重构来得到SFT数据集。这些数据的作用在于帮助模型学习医疗领域知识，将行为模式与人类偏好对齐，并对齐真实世界在线医疗对话的分布情况。

<!-- <style type="text/css">
.tg  {border-collapse:collapse;border-spacing:0;}
.tg td{border-color:black;border-style:solid;border-width:1px;font-family:Arial, sans-serif;font-size:14px;
  overflow:hidden;padding:10px 5px;word-break:normal;}
.tg th{border-color:black;border-style:solid;border-width:1px;font-family:Arial, sans-serif;font-size:14px;
  font-weight:normal;overflow:hidden;padding:10px 5px;word-break:normal;}
.tg .tg-9wq8{border-color:inherit;text-align:center;vertical-align:middle}
.tg .tg-c3ow{border-color:inherit;text-align:center;vertical-align:top}
</style> -->
<table class="tg" style="undefined;table-layout: fixed; width: 442px">
<colgroup>
<col style="width: 204.428571px">
<col style="width: 135.428571px">
<col style="width: 102.428571px">
</colgroup>
<thead>
  <tr>
    <th class="tg-9wq8" rowspan="2"><br>数据集</th>
    <th class="tg-9wq8" rowspan="2"><br>数据来源</th>
    <th class="tg-9wq8" rowspan="2"><br>样本量</th>
  </tr>
  <tr>
  </tr>
</thead>
<tbody>
  <tr>
    <td class="tg-9wq8" rowspan="2">重构AI医患对话</td>
    <td class="tg-9wq8">MedDialog</td>
    <td class="tg-9wq8">400k</td>
  </tr>
  <tr>
    <td class="tg-9wq8">cMedQA2</td>
    <td class="tg-c3ow">20k</td>
  </tr>
  <tr>
    <td class="tg-c3ow">知识图谱问答对</td>
    <td class="tg-9wq8">CMeKG</td>
    <td class="tg-9wq8">50k</td>
  </tr>
  <tr>
    <td class="tg-c3ow">行为偏好数据集</td>
    <td class="tg-9wq8">人为筛选</td>
    <td class="tg-9wq8">2k</td>
  </tr>
  <tr>
    <td class="tg-9wq8" rowspan="3">其他</td>
    <td class="tg-c3ow">MedMCQA</td>
    <td class="tg-c3ow">8k</td>
  </tr>
  <tr>
    <td class="tg-c3ow">MOSS-SFT</td>
    <td class="tg-c3ow">33k</td>
  </tr>
  <tr>
    <td class="tg-c3ow">Alpaca-GPT4-zh</td>
    <td class="tg-c3ow">1k</td>
  </tr>
</tbody>
</table>


<br>


### 下载

我们总共发布了近47万条训练数据，其中包括重构AI医患对话和知识图谱问答对。您可以访问这个[链接](https://huggingface.co/datasets/Flmc/DISC-Med-SFT)下载数据集。

<br>


## 部署

当前版本的 DISC-MedLLM 是基于[Baichuan-13B-Base](https://github.com/baichuan-inc/Baichuan-13B)训练得到的。

Excerpt of 13,045 characters

Read on GitHub

Top contributors3 shown

F1mc
42
lemuria-wchen
Lemuria Chen
2
eltociear
Ikko Eltociear Ashimine · Japan
1

Quality score

Would you bet a product on this? Bounded 0–100 and slow moving.

43/10043

Maintenance: 5%; How recently the default branch was pushed to
Releases: 25%; Recency and cadence of tagged releases
Community: 13%; Contributor breadth and bus factor
Issue backlog: 85%; Open issues measured against the size of the audience
Documentation: 86%; README depth, plus a homepage and topics
Licensing: 100%; How freely the code can be adopted

What to know before adopting

No commits in over a year
No tagged releases
Bus factor of 1 — one contributor dominates
Usable in commercial products with attribution.

Details

Created: 3 years ago
Last push: 3 years ago
Last release: none
Release tag: —
Releases in the last year: 0
Default branch: main
Size: 11k KB
Owner type: Organization
Owner location: China
Discovered via: phrase:language model
First indexed: 7 days ago
Last synced: 17 hours ago

FudanDISC/DISC-MedLLM

43Apache-2.0Python

View on GitHub

Stars: 565
Forks: 53
Watchers: 3
Open issues: 16
Contributors: 3

Star history, last 90 days— over 3 snapshots

2026-07-31 · 5652026-08-06 · 565

Star history, last 90 days
Date	Stars
2026-07-31	565
2026-08-05	565
2026-08-06	565

Today

— stars today

This week

— stars this week

This month

— stars this month

Momentum

0.0

growth rate 0.00%/day

From the README

# DISC-MedLLM

<div align="center">
  
[![Generic badge](https://img.shields.io/badge/🤗-Huggingface%20Repo-green.svg)](https://huggingface.co/Flmc/DISC-MedLLM)
[![license](https://img.shields.io/github/license/modelscope/modelscope.svg)](https://github.com/FudanDISC/DICS-MedLLM/blob/main/LICENSE)
<br>
</div>
<div align="center">

[Demo](http://med.fudan-disc.com) | [技术报告](https://arxiv.org/abs/2308.14346)
<br>
中文 | [EN](https://github.com/FudanDISC/DISC-MedLLM/blob/main/README_EN.md)
</div>
  
DISC-MedLLM 是一个专门针对医疗健康对话式场景而设计的医疗领域大模型，由[复旦大学数据智能与社会计算实验室 (Fudan-DISC)](http://fudan-disc.com) 开发并开源。

该项目包含下列开源资源:
* [DISC-Med-SFT 数据集](https://huggingface.co/datasets/Flmc/DISC-Med-SFT) (不包括行为偏好训练数据)
* DISC-MedLLM 的[模型权重](https://huggingface.co/Flmc/DISC-MedLLM)

您可以通过访问这个[链接](http://med.fudan-disc.com)来试用我们的模型。

## 概述

DISC-MedLLM 是一个专为医疗健康对话场景而打造的领域大模型，它可以满足您的各种医疗保健需求，包括疾病问诊和治疗方案咨询等，为您提供高质量的健康支持服务。

DISC-MedLLM 有效地对齐了医疗场景下的人类偏好，弥合了通用语言模型输出与真实世界医疗对话之间的差距，这一点在实验结果中有所体现。

得益于我们以目标为导向的策略，以及基于真实医患对话数据和知识图谱，引入LLM in the loop 和 Human in the loop的多元数据构造机制，DISC-MedLLM 有以下几个特点：

* **可靠丰富的专业知识**，我们以医学知识图谱作为信息源，通过采样三元组，并使用通用大模型的语言能力进行对话样本的构造。
* **多轮对话的问询能力**，我们以真实咨询对话纪录作为信息源，使用大模型进行对话重建，构建过程中要求模型完全对齐对话中的医学信息。
* **对齐人类偏好的回复**，病人希望在咨询的过程中获得更丰富的支撑信息和背景知识，但人类医生的回答往往简练；我们通过人工筛选，构建符合人类偏好的高质量的小规模行为微调样本，对齐病人的需求。

<img src="https://github.com/FudanDISC/DISC-MedLLM/blob/main/images/data_construction.png" alt="data-construction" width="85%"/>

## 模型效果演示
### 疾病问诊
<img src="https://github.com/FudanDISC/DISC-MedLLM/blob/main/images/consultation.gif" alt="sample1" width="60%"/>

### 治疗方案咨询
<img src="https://github.com/FudanDISC/DISC-MedLLM/blob/main/images/advice.gif" alt="sample2" width="60%"/>

## 数据集

为了训练 DISC-MedLLM ，我们构建了一个高质量的数据集，命名为 DISC-Med-SFT，其中包含了超过47万个衍生于现有的医疗数据集重新构建得到的样本。我们采用了目标导向的策略，通过对于精心选择的几个数据源进行重构来得到SFT数据集。这些数据的作用在于帮助模型学习医疗领域知识，将行为模式与人类偏好对齐，并对齐真实世界在线医疗对话的分布情况。

<!-- <style type="text/css">
.tg  {border-collapse:collapse;border-spacing:0;}
.tg td{border-color:black;border-style:solid;border-width:1px;font-family:Arial, sans-serif;font-size:14px;
  overflow:hidden;padding:10px 5px;word-break:normal;}
.tg th{border-color:black;border-style:solid;border-width:1px;font-family:Arial, sans-serif;font-size:14px;
  font-weight:normal;overflow:hidden;padding:10px 5px;word-break:normal;}
.tg .tg-9wq8{border-color:inherit;text-align:center;vertical-align:middle}
.tg .tg-c3ow{border-color:inherit;text-align:center;vertical-align:top}
</style> -->
<table class="tg" style="undefined;table-layout: fixed; width: 442px">
<colgroup>
<col style="width: 204.428571px">
<col style="width: 135.428571px">
<col style="width: 102.428571px">
</colgroup>
<thead>
  <tr>
    <th class="tg-9wq8" rowspan="2"><br>数据集</th>
    <th class="tg-9wq8" rowspan="2"><br>数据来源</th>
    <th class="tg-9wq8" rowspan="2"><br>样本量</th>
  </tr>
  <tr>
  </tr>
</thead>
<tbody>
  <tr>
    <td class="tg-9wq8" rowspan="2">重构AI医患对话</td>
    <td class="tg-9wq8">MedDialog</td>
    <td class="tg-9wq8">400k</td>
  </tr>
  <tr>
    <td class="tg-9wq8">cMedQA2</td>
    <td class="tg-c3ow">20k</td>
  </tr>
  <tr>
    <td class="tg-c3ow">知识图谱问答对</td>
    <td class="tg-9wq8">CMeKG</td>
    <td class="tg-9wq8">50k</td>
  </tr>
  <tr>
    <td class="tg-c3ow">行为偏好数据集</td>
    <td class="tg-9wq8">人为筛选</td>
    <td class="tg-9wq8">2k</td>
  </tr>
  <tr>
    <td class="tg-9wq8" rowspan="3">其他</td>
    <td class="tg-c3ow">MedMCQA</td>
    <td class="tg-c3ow">8k</td>
  </tr>
  <tr>
    <td class="tg-c3ow">MOSS-SFT</td>
    <td class="tg-c3ow">33k</td>
  </tr>
  <tr>
    <td class="tg-c3ow">Alpaca-GPT4-zh</td>
    <td class="tg-c3ow">1k</td>
  </tr>
</tbody>
</table>


<br>


### 下载

我们总共发布了近47万条训练数据，其中包括重构AI医患对话和知识图谱问答对。您可以访问这个[链接](https://huggingface.co/datasets/Flmc/DISC-Med-SFT)下载数据集。

<br>


## 部署

当前版本的 DISC-MedLLM 是基于[Baichuan-13B-Base](https://github.com/baichuan-inc/Baichuan-13B)训练得到的。

Excerpt of 13,045 characters

Read on GitHub

Top contributors3 shown

F1mc
42
lemuria-wchen
Lemuria Chen
2
eltociear
Ikko Eltociear Ashimine · Japan
1

Quality score

Would you bet a product on this? Bounded 0–100 and slow moving.

43/10043

Maintenance: 5%; How recently the default branch was pushed to
Releases: 25%; Recency and cadence of tagged releases
Community: 13%; Contributor breadth and bus factor
Issue backlog: 85%; Open issues measured against the size of the audience
Documentation: 86%; README depth, plus a homepage and topics
Licensing: 100%; How freely the code can be adopted

What to know before adopting

No commits in over a year
No tagged releases
Bus factor of 1 — one contributor dominates
Usable in commercial products with attribution.

Details

Created: 3 years ago
Last push: 3 years ago
Last release: none
Release tag: —
Releases in the last year: 0
Default branch: main
Size: 11k KB
Owner type: Organization
Owner location: China
Discovered via: phrase:language model
First indexed: 7 days ago
Last synced: 17 hours ago

Top AI Repos — open-source AI, indexed and scored

FudanDISC/DISC-MedLLM

From the README

Top contributors3 shown

Quality score

What to know before adopting

Categories

Details

FudanDISC/DISC-MedLLM

From the README

Top contributors3 shown

Quality score

What to know before adopting

Categories

Details

Top AI Repos — open-source AI, indexed and scored

Trend

From the README

Top contributors3 shown

Quality score

What to know before adopting

Categories

Details

Trend

From the README

Top contributors3 shown

Quality score

What to know before adopting

Categories

Details