Medical-Reasoning-SFT-Baichuan-M3-235B

# Medical-Reasoning-SFT-Baichuan-M3-235B

A large-scale medical reasoning dataset generated using [baichuan-inc/Baichuan-M3-235B](https://huggingface.co/baichuan-inc/Baichuan-M3-235B), containing over 124,000 samples with detailed chain-of-thought reasoning for medical and healthcare questions.

**Baichuan-M3-235B is ranked #1 on HealthBench Total leaderboard and achieves state-of-the-art performance on medical reasoning benchmarks.**

## Dataset Overview

| Metric | Value |
|--------|-------|
| **Model** | baichuan-inc/Baichuan-M3-235B |
| **Total Samples** | 124,520 |
| **Samples with Reasoning** | 124,520 (100%) |
| **Estimated Tokens** | ~255 Million |
| **Content Tokens** | ~160 Million |
| **Reasoning Tokens** | ~95 Million |
| **Language** | English |

## Why Baichuan-M3-235B?

Baichuan-M3-235B is a purpose-built medical AI model with exceptional health evaluation results:

### HealthBench Performance
- **#1 on HealthBench Total Leaderboard** - Top-ranked model globally
- **HealthBench-Hard: 44.4%** - A 28-point improvement over M2, surpassing GPT-5.2
- **Industry-lowest hallucination rate: 3.5%** - Achieved through innovative Fact-Aware RL training

### Clinical Benchmarks
- **SCAN-Bench: First Place** - Across all three dimensions:
- Clinical Inquiry
- Lab Testing
- Final Diagnosis
- **SPAR Algorithm** - Segmented Pipeline Reinforcement Learning specifically designed for clinical decision-making

### Model Architecture
- **Parameters**: 235B
- **Base**: Qwen3-235B-A22B
- **License**: Apache 2.0

## Schema

Each sample follows the conversational messages format with reasoning content:

```json
{
"messages": [
{
"role": "system",
"content": "You are a medical expert...",
"reasoning_content": null
},
{
"role": "user",
"content": "What are the symptoms of diabetes?",
"reasoning_content": null
},
{
"role": "assistant",
"content": "The main symptoms of diabetes include...",
"reasoning_content": "Let me think through this systematically. Diabetes affects blood sugar regulation, so I should consider symptoms related to hyperglycemia..."
}
]
}
```

### Fields

| Field | Type | Description |
|-------|------|-------------|
| `messages` | list | Array of message objects in the conversation |
| `messages[].role` | string | Either "system", "user", or "assistant" |
| `messages[].content` | string | The main message content |
| `messages[].reasoning_content` | string or null | Chain-of-thought reasoning (assistant messages only) |

## Usage

### Loading with Datasets Library

```python
from datasets import load_dataset

dataset = load_dataset("OpenMed/Medical-Reasoning-SFT-Baichuan-M3-235B")
```

### Accessing Samples

```python
# Get a sample
sample = dataset['train'][0]

# Access messages
for msg in sample['messages']:
print(f"Role: {msg['role']}")
print(f"Content: {msg['content'][:100]}...")
if msg['reasoning_content']:
print(f"Reasoning: {msg['reasoning_content'][:100]}...")
```

### Filtering by Reasoning

```python
# Get samples with reasoning content
samples_with_reasoning = dataset['train'].filter(
lambda x: x['messages'][-1]['reasoning_content'] is not None
)
```

## Intended Use

This dataset is designed for:

- **Fine-tuning medical reasoning models**: Train LLMs to provide detailed, step-by-step medical reasoning
- **Chain-of-thought training**: Develop models that show their thinking process
- **Medical QA systems**: Build question-answering systems for healthcare applications
- **Research**: Study reasoning patterns in medical domain AI

## Limitations and Considerations

- This dataset is generated by an AI model and should not be used as a substitute for professional medical advice
- Responses may contain inaccuracies and should be validated by medical professionals
- Not intended for clinical decision-making without expert review
- The reasoning traces reflect the model's approach, not necessarily optimal clinical reasoning

## Citation

If you use this dataset, please cite:

```bibtex
@dataset{medical_reasoning_sft_baichuan_m3_235b,
title={Medical-Reasoning-SFT-Baichuan-M3-235B},
author={OpenMed},
year={2025},
publisher={Hugging Face},
url={https://huggingface.co/datasets/OpenMed/Medical-Reasoning-SFT-Baichuan-M3-235B}
}
```

## License

Apache 2.0

# 医疗推理SFT-Baichuan-M3-235B

本数据集为基于[baichuan-inc/Baichuan-M3-235B](https://huggingface.co/baichuan-inc/Baichuan-M3-235B)生成的大规模医疗推理数据集，包含超过12.4万条针对医疗健康问题的、带有详细思维链（Chain-of-Thought）推理过程的样本。

**Baichuan-M3-235B在HealthBench总排行榜中位列第一，并在医疗推理基准测试中取得了当前最优（state-of-the-art）性能。**

## 数据集概览

| 指标 | 数值 |
|--------|-------|
| **模型** | baichuan-inc/Baichuan-M3-235B |
| **总样本数** | 124,520 |
| **带推理过程的样本数** | 124,520 (100%) |
| **估算Token数** | ~2.55亿 |
| **内容Token数** | ~1.60亿 |
| **推理Token数** | ~9500万 |
| **语言** | 英语 |

## 为何选择Baichuan-M3-235B？

Baichuan-M3-235B是一款专为医疗场景打造的AI模型，具备卓越的健康评估性能：

### HealthBench性能表现
- **HealthBench总排行榜全球第一** - 全球排名最高的医疗AI模型
- **HealthBench-Hard任务准确率：44.4%** - 较M2提升28个百分点，超越GPT-5.2
- **行业最低幻觉率（hallucination rate）：3.5%** - 通过创新性的事实感知强化学习（Fact-Aware RL）训练实现

### 临床基准测试
- **SCAN-Bench排名第一** - 覆盖三大维度：
- 临床问诊
- 实验室检测
- 最终诊断
- **SPAR算法**：专为临床决策设计的分段流水线强化学习（Segmented Pipeline Reinforcement Learning）

### 模型架构
- **参数量**：2350亿
- **基础模型**：Qwen3-235B-A22B
- **许可证**：Apache 2.0

## 数据格式规范

每条样本采用带推理内容的对话消息格式：

json
{
"messages": [
{
"role": "system",
"content": "您是一名医疗专家……",
"reasoning_content": null
},
{
"role": "user",
"content": "糖尿病的症状有哪些？",
"reasoning_content": null
},
{
"role": "assistant",
"content": "糖尿病的主要症状包括……",
"reasoning_content": "让我系统地梳理一下思路。糖尿病会影响血糖调节，因此我需要考虑与高血糖相关的症状……"
}
]
}

### 字段说明

| 字段 | 类型 | 描述 |
|-------|------|-------------|
| `messages` | 列表 | 对话消息对象数组 |
| `messages[].role` | 字符串 | 角色，可选值为"system"、"user"或"assistant" |
| `messages[].content` | 字符串 | 消息主体内容 |
| `messages[].reasoning_content` | 字符串或空 | 思维链推理过程（仅助手消息包含此字段） |

## 使用方法

### 使用Datasets库加载数据集

python
from datasets import load_dataset

dataset = load_dataset("OpenMed/Medical-Reasoning-SFT-Baichuan-M3-235B")

### 获取样本

python
# 获取单条样本
sample = dataset['train'][0]

# 遍历对话消息
for msg in sample['messages']:
print(f"角色：{msg['role']}")
print(f"内容：{msg['content'][:100]}……")
if msg['reasoning_content']:
print(f"推理过程：{msg['reasoning_content'][:100]}……")

### 按推理过程过滤样本

python
# 获取带有推理过程的样本
samples_with_reasoning = dataset['train'].filter(
lambda x: x['messages'][-1]['reasoning_content'] is not None
)

## 预期用途

本数据集适用于：
- **医疗推理模型微调**：训练大语言模型以生成详细的分步医疗推理过程
- **思维链训练**：开发可展示自身思考过程的AI模型
- **医疗问答系统**：构建面向医疗健康场景的问答系统
- **学术研究**：探究医疗领域AI的推理模式与规律

## 局限性与注意事项
- 本数据集由AI模型生成，不得替代专业医疗建议
- 生成内容可能存在不准确之处，需由医疗专业人员验证
- 未经专家审核，不得用于临床决策
- 推理轨迹仅反映模型自身的推理路径，未必代表最优临床推理逻辑

## 引用方式
若您使用本数据集，请引用如下文献：

bibtex
@dataset{medical_reasoning_sft_baichuan_m3_235b,
title={Medical-Reasoning-SFT-Baichuan-M3-235B},
author={OpenMed},
year={2025},
publisher={Hugging Face},
url={https://huggingface.co/datasets/OpenMed/Medical-Reasoning-SFT-Baichuan-M3-235B}
}

## 许可证
Apache 2.0