ReasonFlux-F1-SFT

# Dataset Card for ReasonFlux-F1-SFT

> ReasonFlux-F1-SFT consists of 1k high-quality problems from [s1k](https://huggingface.co/datasets/simplescaling/s1K). We use ReasonFlux-Zero to generate template augmented reasoning trajectories for each problem and transform them into Long-CoT format to finetune reasoning LLMs.

* Github Repository: [Gen-Verse/ReasonFlux](https://github.com/Gen-Verse/ReasonFlux)
* Paper:[ReasonFlux: Hierarchical LLM Reasoning via Scaling Thought Templates](https://arxiv.org/abs/2502.06772)
* Model: [Gen-Verse/ReasonFlux-F1](https://huggingface.co/Gen-Verse/ReasonFlux-F1)

## Citation

```bash
@article{yang2025reasonflux,
title={ReasonFlux: Hierarchical LLM Reasoning via Scaling Thought Templates},
author={Yang, Ling and Yu, Zhaochen and Cui, Bin and Wang, Mengdi},
journal={arXiv preprint arXiv:2502.06772},
year={2025}
}
```

# ReasonFlux-F1-SFT 数据集卡片

> ReasonFlux-F1-SFT 包含来自[s1k](https://huggingface.co/datasets/simplescaling/s1K)的1000道高质量问题。我们使用ReasonFlux-Zero为每道问题生成经过模板增强的推理轨迹，并将其转换为长思维链（Long Chain of Thought，Long-CoT）格式，以用于推理型大语言模型（Large Language Model，LLM）的微调。

* GitHub 仓库：[Gen-Verse/ReasonFlux](https://github.com/Gen-Verse/ReasonFlux)
* 论文：[ReasonFlux：基于扩展思维模板的分层大语言模型推理](https://arxiv.org/abs/2502.06772)
* 模型：[Gen-Verse/ReasonFlux-F1](https://huggingface.co/Gen-Verse/ReasonFlux-F1)

## 引用

bash
@article{yang2025reasonflux,
title={ReasonFlux: Hierarchical LLM Reasoning via Scaling Thought Templates},
author={Yang, Ling and Yu, Zhaochen and Cui, Bin and Wang, Mengdi},
journal={arXiv preprint arXiv:2502.06772},
year={2025}
}

搜集汇总

背景与挑战

背景概述

ReasonFlux-F1-SFT数据集包含1000个高质量推理问题，这些数据源自s1k。该数据集通过ReasonFlux-Zero方法生成模板增强的推理轨迹，并转换为Long-CoT格式，专门用于微调推理大语言模型。