Light R1 7B DS

🚀 Light-R1-7B-DS: SOTA 7B Math Model with only 3K Data

Light-R1-7B-DS is, to the best of our knowledge, the State-Of-The-Art open-source 7B math model, achieving scores of 59.1 and 44.3 on AIME24 and AIME25 respectively. It also performs well on GPQA without any specific training. Originated from DeepSeek-R1-Distill-Qwen-7B, it is further trained with only 3K SFT data that we've open-sourced, demonstrating the strong applicability of the released data.

🚀 Quick Start

Model Information

Model Comparison

Related Links

• Technical Report
• GitHub Page

💻 Usage Examples

Basic Usage

Same as DeepSeek-R1-Distill-Qwen-7B.

🔧 Technical Details

Data Decontamination

We carefully evaluated data contamination of several open-sourced datasets. While certain contamination may be inevitable during pre-training, it is unacceptable for post-training to compare on benchmarks. MATH-500 is somewhat compromised with tens of questions that are identical or only numbers changed. AIME 24 and 25 stay intact but we have to pay special attention when we incorporate AIME data up to 2023. Light-R1 did thorough decontamination with exact matching (excluding digits) and N-gram (N = 32) matching.

📄 License

The model is released under the apache-2.0 license.

📚 Documentation

Citation

@misc{lightr1proj,
      title={Light-R1: Curriculum SFT, DPO and RL for Long COT from Scratch and Beyond}, 
      author={Liang Wen, Yunke Cai, Fenrui Xiao, Xin He, Qi An, Zhenyu Duan, Yimin Du, Junchen Liu, Lifu Tang, Xiaowei Lv, Haosheng Zou, Yongchao Deng, Shousheng Jia, Xiangzheng Zhang},
      year={2025},
      eprint={},
      archivePrefix={},
      url={https://github.com/Qihoo360/Light-R1}, 
}