---
title: Medical-Reasoning-SFT-GPT-OSS-120B-Small
canonical_url: "https://www.modelscope.cn/datasets/OpenMed/Medical-Reasoning-SFT-GPT-OSS-120B-Small"
md_url: "https://www.modelscope.cn/datasets/OpenMed/Medical-Reasoning-SFT-GPT-OSS-120B-Small.md"
repository: OpenMed/Medical-Reasoning-SFT-GPT-OSS-120B-Small
last_updated: 2026-01-05
license: apache-2.0
storage_size: "265 MB"
downloads: 116
stars: 0
---

# Medical-Reasoning-SFT-GPT-OSS-120B-Small

> Medical-Reasoning-SFT-GPT-OSS-120B-Small - OpenMed 在 ModelScope 开源的数据集。Medical-Reasoning-SFT-GPT-OSS-120B-Small

OpenMed/Medical-Reasoning-SFT-GPT-OSS-120B-Small 是 ModelScope 魔搭社区上的数据集，存储大小 265 MB，采用 apache-2.0 许可。

- **Repository**: OpenMed/Medical-Reasoning-SFT-GPT-OSS-120B-Small
- **License**: apache-2.0
- **Storage size**: 265 MB
- **Downloads**: 116
- **Stars**: 0
- **Last updated**: 2026-01-05

Source: https://www.modelscope.cn/datasets/OpenMed/Medical-Reasoning-SFT-GPT-OSS-120B-Small

---

# Medical-Reasoning-SFT-GPT-OSS-120B-Small

A filtered and processed version of [OpenMed/Medical-Reasoning-SFT-GPT-OSS-120B](https://huggingface.co/datasets/OpenMed/Medical-Reasoning-SFT-GPT-OSS-120B) optimized for training efficiency.

## Dataset Description

This dataset contains high-quality medical reasoning conversations with the following modifications:

1. **Length Filtering**: Only includes samples where assistant responses are between 1000 and 10000 characters
2. **Reasoning Extraction**: Reasoning content from `<think>` tags has been extracted into a separate `reasoning_content` field

## Statistics

- **Original dataset size**: 200,193 samples
- **Filtered dataset size**: 131,728 samples (65.8%)
- **Samples with reasoning**: 131,728
- **Samples without reasoning**: 0

## Filtering Criteria

- **Minimum assistant response length**: 1000 characters
- **Maximum assistant response length**: 10000 characters
- **Removed**: 13,664 samples (too short) + 54,801 samples (too long)

## Dataset Structure

Each sample contains:
- `messages`: List of conversation messages with roles (user/assistant)
  - Assistant messages now have two fields:
    - `content`: The actual response (without `<think>` tags)
    - `reasoning_content`: The extracted reasoning from `<think>` tags

## Usage

```python
from datasets import load_dataset

dataset = load_dataset("OpenMed/Medical-Reasoning-SFT-GPT-OSS-120B-Small")

# Access a sample
sample = dataset['train'][0]
messages = sample['messages']

for msg in messages:
    if msg['role'] == 'assistant':
        reasoning = msg.get('reasoning_content', '')
        response = msg.get('content', '')
        print(f"Reasoning: {reasoning}")
        print(f"Response: {response}")
```

## Source Dataset

This dataset is derived from [OpenMed/Medical-Reasoning-SFT-GPT-OSS-120B](https://huggingface.co/datasets/OpenMed/Medical-Reasoning-SFT-GPT-OSS-120B), which was generated using OpenAI's gpt-oss-120B model.

## License

Apache 2.0 (inherited from source dataset)

## Citation

If you use this dataset, please cite the original dataset:

```
@dataset{medical_reasoning_sft_gpt_oss_120b,
  title={Medical-Reasoning-SFT-GPT-OSS-120B},
  author={OpenMed},
  year={2024},
  publisher={Hugging Face},
  url={https://huggingface.co/datasets/OpenMed/Medical-Reasoning-SFT-GPT-OSS-120B}
}
```