R1-Distill-SFT

# 🔉 𝗦𝗟𝗔𝗠 𝗹𝗮𝗯 - 𝗥𝟭-𝗗𝗶𝘀𝘁𝗶𝗹𝗹-𝗦𝗙𝗧 Dataset

Lewis Tunstall, Ed Beeching, Loubna Ben Allal, Clem Delangue 🤗 and others at Hugging Face announced today that they are - 𝗼𝗽𝗲𝗻𝗹𝘆 𝗿𝗲𝗽𝗿𝗼𝗱𝘂𝗰𝗶𝗻𝗴 𝗥𝟭 🔥

We at 𝗦𝗟𝗔𝗠 𝗹𝗮𝗯 (ServiceNow Language Models) have been cooking up something as well.

Inspired by Open-r1, we have decided to open source the data **stage-by-stage** to support the open source community.

𝗕𝗼𝗼𝗸𝗺𝗮𝗿𝗸 this page!

**KEY DETAILS**:
- ⚗️ Distilled with DeepSeek-R1-32b
- 📕 Generated using Numina-math and Tulu
- 🌡️ Sampled one response per prompt

# 𝗦𝗖𝗛𝗘𝗗𝗨𝗟𝗘:

- 🆕 [27 Jan] Release seed set of 170,000 samples
- 🛑 [28 Jan] Release the unfiltered / unverified dataset ~ 2 million samples
- 🟢 [TBD] Filtered and verified version to follow shortly after
- 🏁 [TBD] SFT Models released

**If you use our dataset, please cite us!**

```
@misc{slam-distillation-from-r1,
author = {Sathwik Tejaswi Madhusudhan and Shruthan Radhakrishna and Jash Mehta and Toby Liang},
title = {Millions scale dataset distilled from R1-32b},
howpublished = {https://huggingface.co/datasets/ServiceNow-AI/R1-Distill-SFT},
publisher = {SLAM - ServiceNow Language Models Lab}
year = {2025}
}
```

# 🔉 SLAM（ServiceNow语言模型）实验室——R1-Distill-SFT数据集

Hugging Face的Lewis Tunstall、Ed Beeching、Loubna Ben Allal、Clem Delangue及其他团队成员今日宣布，将开源复现R1模型🔥。

我们所在的SLAM（ServiceNow语言模型）实验室也正在推进相关研发工作。受Open-R1项目启发，我们决定分阶段开源本数据集，以助力开源社区发展。请收藏本页面！

**关键细节**：
- ⚗️ 基于DeepSeek-R1-32B进行模型蒸馏
- 📕 基于Numina-math与Tulu模型生成数据
- 🌡️ 每条提示仅采样一条回复

**发布计划**：
- 🆕 【1月27日】发布包含17万条样本的种子数据集
- 🛑 【1月28日】发布约200万条样本的未过滤/未验证数据集
- 🟢 【待定】后续将推出经过过滤与验证的数据集版本
- 🏁 【待定】将发布监督微调（Supervised Fine-Tuning，SFT）模型

**若使用本数据集，请引用以下文献！**

@misc{slam-distillation-from-r1,
author = {Sathwik Tejaswi Madhusudhan、Shruthan Radhakrishna、Jash Mehta、Toby Liang},
title = {基于R1-32B蒸馏的百万级规模数据集},
howpublished = {https://huggingface.co/datasets/ServiceNow-AI/R1-Distill-SFT},
publisher = {SLAM——ServiceNow语言模型实验室},
year = {2025}
}