---
title: OpenScience-Chinese-Reasoning-SFT
canonical_url: "https://www.modelscope.cn/datasets/zake7749/OpenScience-Chinese-Reasoning-SFT"
md_url: "https://www.modelscope.cn/datasets/zake7749/OpenScience-Chinese-Reasoning-SFT.md"
repository: zake7749/OpenScience-Chinese-Reasoning-SFT
chinese_name: "中文科學知識 MCQA 問答數據"
last_updated: 2026-03-14
license: cc-by-4.0
storage_size: "276 MB"
downloads: 26
stars: 0
---

# OpenScience-Chinese-Reasoning-SFT

> OpenScience-Chinese-Reasoning-SFT - zake7749 在 ModelScope 开源的数据集。A Chinese multiple-choice science QA dataset with chain-of-thought reasoning, derived from nvidia/OpenScience through translation and rejection sampling.

zake7749/OpenScience-Chinese-Reasoning-SFT 是 ModelScope 魔搭社区上的数据集，存储大小 276 MB，采用 cc-by-4.0 许可。

- **Repository**: zake7749/OpenScience-Chinese-Reasoning-SFT
- **License**: cc-by-4.0
- **Storage size**: 276 MB
- **Downloads**: 26
- **Stars**: 0
- **Last updated**: 2026-03-14

Source: https://www.modelscope.cn/datasets/zake7749/OpenScience-Chinese-Reasoning-SFT

---

# OpenScience-Chinese

A Chinese multiple-choice science QA dataset with chain-of-thought reasoning, derived from [nvidia/OpenScience](https://huggingface.co/datasets/nvidia/OpenScience) through translation and rejection sampling.

* Key Features:
  * Scale: 50,000 high-quality instances.
  * Reasoning: Built-in Chain-of-Thought (`<think>` tags) for interpretable AI.
  * Reliability: Rejection sampling ensures 100% alignment with ground truth.

## Data Source

The questions originate from [nvidia/OpenScience](https://huggingface.co/datasets/nvidia/OpenScience), a large-scale science QA dataset covering diverse domains including physics, chemistry, biology, medicine, computer science, and engineering. The original English questions were translated into Chinese and polished for fluency.

## Construction Pipeline

```
nvidia/OpenScience (English MCQA)
    │
    ▼ Sampling & Translation & Polishing
    │ Translated to Traditional Chinese with terminology review
    │
    ▼ Rejection Sampling (3 rejections per prompt)
    │ temperature=0.6, top_p=0.95, max_tokens=16384
    │
    ▼ Answer Extraction & Verification
    │ Extract model's chosen answer → compare with ground truth
    │ Keep only correct responses
    │
    ▼ Sampling & Quality Filtering
    │ 
    ▼ Final Dataset (50,000 examples)
```

### Models Used

| Model | Samples | Accuracy (num_rejections=3) |
|-------|---------|-------|
| Kyara-3-Flash | 35,000 | 88.32% |
| Luoyang-48B-A3B | 15,000 | 89.48% |

### Rejection Sampling Details

- Each question was sampled 3 times (num_rejections=3) with `temperature=0.6` and `top_p=0.95`
- Model-extracted answer choices were compared against the ground truth from OpenScience.
- Only responses where the model's final answer matched the ground truth were retained.

## Format

Each example follows a standard `messages` format with reasoning in `<think>` tags:

```json
{
  "messages": [
    {
      "role": "user",
      "content": "以下哪一個古代帝國以其在數學與天文學發展上的重要貢獻而聞名，但卻經常被更為人熟知的文明所掩蓋？\n\nA: 孔雀王朝 (Maurya Empire)\nB: 印加帝國 (Inca Empire)\n..."
    },
    {
      "role": "assistant",
      "content": "<think>\n這個問題是問哪個古代帝國在數學和天文學上有重要貢獻，但經常被更知名的文明掩蓋。選項有A到J，我需要仔細想想每個選項的情況。\n\n首先，題目中的選項包括孔雀王朝、印加帝國、阿克蘇姆、薩珊王朝、塞伐那王朝、貴霜帝國、阿拔斯王朝、羅馬帝國、笈多王朝、拜占庭帝國。我需要回憶每個帝國在數學和天文學方面的貢獻，然後看哪個可能被其他更出名的文明所掩蓋。...</think>\n在眾多古代帝國中，**笈多王朝（Gupta Empire）**（選項 I）以其在數學與天文學上的重要貢獻而聞名，卻常被更為人熟知的文明（如古希臘、羅馬或阿拉伯帝國）所掩蓋。..."
    }
  ],
  "metadata": {
    "model": "Luoyang-48B-A3B",
    "temperature": 0.6,
    "top_p": 0.95,
    "num_rejections": 3,
    "max_tokens": 16384
  }
}
```

## Usage

```python
from datasets import load_dataset

ds = load_dataset("zake7749/OpenScience-Chinese")
print(ds["train"][0]["messages"])
```

## Limitations

- The dataset is focused on Traditional Chinese.
- All responses have been verified for answer correctness, but the reasoning paths are not individually verified for logical soundness.
- Domain coverage follows the distribution of the source OpenScience dataset, which is skewed toward STEM subjects.

## Citation

If you use this dataset, you may cite the dataset as:

```bibtex
@misc{yang_openscience_chinese_2026,
  title={OpenScience-Chinese},
  author={Yang, Kai-Chou},
  year={2026},
  publisher={Hugging Face},
  url={https://huggingface.co/datasets/zake7749/OpenScience-Chinese}
}
```

And please cite the original OpenScience dataset:

```bibtex
@misc{nvidia_openscience,
  title={OpenScience},
  author={NVIDIA},
  year={2025},
  publisher={Hugging Face},
  url={https://huggingface.co/datasets/nvidia/OpenScience}
}
```