@inproceedings{wen-etal-2025-light,
    title = "Light-R1: Curriculum {SFT}, {DPO} and {RL} for Long {COT} from Scratch and Beyond",
    author = "Wen, Liang  and
      Cai, Yunke  and
      Xiao, Fenrui  and
      He, Xin  and
      An, Qi  and
      Duan, Zhenyu  and
      Du, Yimin  and
      Liu, Junchen  and
      Tanglifu, Tanglifu  and
      Lv, Xiaowei  and
      Zou, Haosheng  and
      Deng, Yongchao  and
      Jia, Shousheng  and
      Zhang, Xiangzheng",
    editor = "Rehm, Georg  and
      Li, Yunyao",
    booktitle = "Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 6: Industry Track)",
    month = jul,
    year = "2025",
    address = "Vienna, Austria",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/ingestion-acl-25/2025.acl-industry.24/",
    pages = "318--327",
    ISBN = "979-8-89176-288-6",
    abstract = "This paper introduces Light-R1, an opensource suite for training long reasoning modelsusing reproducible and cost-effective methodology. Given the proprietary nature of data usedin the DeepSeek-R1 series, we develop an alternative approach leveraging exclusively publicdata and models. Our curriculum training progressively increases data difficulty, combinedwith multi-staged post-training. Our LightR1-32B model, trained from Qwen2.5-32BInstruct, outperforms DeepSeek-R1-DistillQwen-32B in math reasoning. Experimental results show that this curriculum approachbecomes more effective when distinct, diverse datasets are available for different training stages: fine-tuning DeepSeek-R1-Distilledmodels (pre-tuned by DeepSeek team on proprietary data) with 3,000 challenging examplesfrom our curriculum dataset yielded state-ofthe-art 7B and 14B models, while the 32Bmodel, Light-R1-32B-DS performed comparably to QwQ-32B and DeepSeek-R1. Furthermore, we extend our work by applying GRPOon long reasoning models. Our final Light-R1-14B-DS achieves SOTA performance among14B models in math, with AIME24 {\&} 25 scoresof 74.0 and 60.2 respectively, surpassing many32B models and DeepSeek-R1-Distill-Llama70B. Despite math-focused training, Light-R1-14B-DS demonstrates strong cross-domain generalization. Light-R1 represents a significantadvancement in making sophisticated reasoning models more accessible and implementablein real-world applications. Our models, training data and code have been made available."
}