Elamur Research⌕
☰
home / papers / 2025 / DRP: Distilled Reasoning Pruning with Skill-aware Step Decomposition for Efficient Large Reasoning Models

DRP: Distilled Reasoning Pruning with Skill-aware Step Decomposition for Efficient Large Reasoning Models

Paper · 2025 · Yuxuan Jiang, Dawei Li and Francis Ferraro

Summary

Large Reasoning Models tend to "overthink": they produce very long chains of thought full of redundant checks and backtracking, which wastes compute. Distilling from a concise teacher can backfire because the student's long-form reasoning style doesn't match the teacher's short outputs, creating a learnability gap. The authors propose Distilled Reasoning Pruning (DRP), which merges pruning and distillation. The student (e.g., R1-Distill-Qwen-7B) first writes its own traces on GSM8K (train split) and PRM12K problems. A teacher LLM (e.g., GPT-4o) then splits each trace into steps tagged with the reasoning skill they use, and edits each step. Finally, the student is fine-tuned on these shortened versions of its own reasoning. DRP cuts token usage substantially while maintaining or improving accuracy, including on harder out-of-domain benchmarks. The broader lesson is that training traces should be compact but still match the student's own reasoning structure. This makes DRP a practical recipe for cheaper inference with small reasoning models.

Key points

  • Pipeline: DRP runs in three stages: skill-based step decomposition, teacher-guided pruning, and SFT of the student. For each step, the teacher chooses to keep, delete, or rewrite it (the overview figure also shows merging).
  • Main results (7B student, GPT-4o teacher):
    • GSM8K: tokens drop from 917 to 328, and accuracy rises from 91.7% to 94.1%.
    • MATH500: tokens drop from 2486 to 1781, and accuracy rises from 92.4% to 93.0%.
    • AIME24: tokens fall about 43% (8674 to 4966), with accuracy unchanged at 15/30.
    • AMC: accuracy improves from 31/40 to 33/40, and tokens fall from 4845 to 3258.
  • Comparison with direct distillation: Training directly on GPT-4o's own short CoTs (about 186 tokens, versus about 330 for DRP traces) hurts accuracy. GSM8K drops to 90.7%, and the decline is larger out of domain (88.6% on MATH500; 13/30 on AIME24).
  • Teacher choice: GPT-4o, Gemini 2.0 Flash, ChatGPT and DeepSeek-V3 all improve over the base model. GPT-4o compresses tokens most, and the teacher's model family appears not to matter much.
  • Availability and limitations: Code is released at github.com/YuxuanJiang1/DRP. The authors note that few strong small LRMs are publicly available, so generalization to other architectures remains untested.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 worksComputer Vision and Pattern Recognition · 3,486 works

🧭 Topics 2Research subjects and languages the work is about.Reasoning · 1,685 worksEfficient models & compression · 297 works

Cite

Cite as Yuxuan Jiang, Dawei Li and Francis Ferraro. 2025. DRP: Distilled Reasoning Pruning with Skill-aware Step Decomposition for Efficient Large Reasoning Models. arXiv:2505.13975.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2505.13975 · arXiv 2505.13975

@misc{jiang2025drp,
  title = {{DRP: Distilled Reasoning Pruning with Skill-aware Step Decomposition for Efficient Large Reasoning Models}},
  author = {Yuxuan Jiang, Dawei Li and Francis Ferraro},
  year = {2025},
  doi = {10.48550/arxiv.2505.13975},
  eprint = {2505.13975},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2505.13975}
}

References

38 references, 20 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Tingxu Han, Zhenting Wang, Chunrong Fang, Shiyu Zhao, Shiqing Ma, and Zhenyu Chen. 2024. Token-budget-aware llm reasoning. arXiv preprint arXiv:2412.18547.
  2. AI-MO Team. 2024a. AIMO Validation Set - AIME Subset. https://huggingface.co/datasets/ AI-MO/aimo-validation-aime . ↗
  3. AI-MO Team. 2024b. AIMO Validation Set - AMC Subset. https://huggingface.co/datasets/ AI-MO/aimo-validation-amc . ↗
  4. Dan Hendrycks, Collin Burns, Saurav Kadavath, Akul Arora, Steven Basart, Eric Tang, Dawn Song, and Jacob Steinhardt. 2021. Measuring mathematical problem solving with the math dataset. arXiv preprint arXiv:2103.03874.
  5. Qiguang Chen, Libo Qin, Jinhao Liu, Dengyun Peng, Jiannan Guan, Peng Wang, Mengkang Hu, Yuhang Zhou, Te Gao, and Wanxiang Che. 2025. Towards reasoning era: A survey of long chain-of-thought for reasoning large language models. arXiv preprint arXiv:2503.09567.
  6. Bairu Hou, Yang Zhang, Jiabao Ji, Yujian Liu, Kaizhi Qian, Jacob Andreas, and Shiyu Chang. 2025. Thinkprune: Pruning long chain-of-thought of llms via reinforcement learning. arXiv preprint arXiv:2504.01296.
  7. Xingyu Chen, Jiahao Xu, Tian Liang, Zhiwei He, Jianhui Pang, Dian Yu, Linfeng Song, Qiuzhi Liu, Mengfei Zhou, Zhuosheng Zhang, et al. 2024. Do not think that much for 2+ 3=? on the overthinking of o1-like llms. arXiv preprint arXiv:2412.21187.
  8. Yuxuan Jiang and Francis Ferraro. 2024. Memorization over reasoning? exposing and mitigating verbatim memorization in large language models’ character understanding evaluation. arXiv preprint arXiv:2412.14368. — [arXiv:2412.14368](https://arxiv.org/abs/2412.14368) find
  9. Karl Cobbe, Vineet Kosaraju, Mohammad Bavarian, Mark Chen, Heewoo Jun, Lukasz Kaiser, Matthias Plappert, Jerry Tworek, Jacob Hilton, Reiichiro Nakano, et al. 2021. Training verifiers to solve math word problems. arXiv preprint arXiv:2110.14168.
  10. Dawei Li, Bohan Jiang, Liangjie Huang, Alimohammad Beigi, Chengshuai Zhao, Zhen Tan, Amrita Bhattacharjee, Yuxuan Jiang, Canyu Chen, Tianhao Wu, et al. 2024a. From generation to judgment: Opportunities and challenges of llm-as-a-judge. arXiv preprint arXiv:2411.16594. find
  11. Yingqian Cui, Pengfei He, Jingying Zeng, Hui Liu, Xianfeng Tang, Zhenwei Dai, Yan Han, Chen Luo, Jing Huang, Zhen Li, et al. 2025. Stepwise perplexity-guided refinement for efficient chain-of-thought reasoning in large language models. arXiv preprint arXiv:2502.13260. find
  12. Dawei Li, Renliang Sun, Yue Huang, Ming Zhong, Bohan Jiang, Jiawei Han, Xiangliang Zhang, Wei Wang, and Huan Liu. 2025a. Preference leakage: A contamination problem in llm-as-a-judge. arXiv preprint arXiv:2502.01534. — [arXiv:2502.01534](https://arxiv.org/abs/2502.01534) find
  13. Muzhi Dai, Chenxu Yang, and Qingyi Si. 2025. S-grpo: Early exit via reinforcement learning in reasoning models. arXiv preprint arXiv:2505.07686. — [arXiv:2505.07686](https://arxiv.org/abs/2505.07686) find
  14. Introducing https://blog.google/technology/google-deepmind/ gemini-model-updates-february-2025/ . ↗
  15. Ming Li, Lichang Chen, Jiuhai Chen, Shwai He, Jiuxiang Gu, and Tianyi Zhou. 2024b. Selective reflection-tuning: Student-selected data recycling for https://chatgpt.com/llm instruction-tuning. In Findings of the Association for Computational Linguistics ACL 2024, pages 16189–16211. ↗
  16. Zhen Tan, Dawei Li, Song Wang, Alimohammad Beigi, Bohan Jiang, Amrita Bhattacharjee, Mansooreh Karami, Jundong Li, Lu Cheng, and Huan Liu. 2024. Large language models for data annotation and synthesis: A survey. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processin…
  17. Yuetai Li, Xiang Yue, Zhangchen Xu, Fengqing Jiang, Luyao Niu, Bill Yuchen Lin, Bhaskar Ramasubramanian, and Radha Poovendran. 2025b. Small models struggle to learn from strong reasoners. arXiv preprint arXiv:2502.12143.
  18. Kimi Team, Angang Du, Bofei Gao, Bowei Xing, Changjiu Jiang, Cheng Chen, Cheng Li, Chenjun Xiao, Chenzhuang Du, Chonghua Liao, et al. 2025. Kimi k1. 5: Scaling reinforcement learning with llms. arXiv preprint arXiv:2501.12599. — Kimi k1.5: Scaling Reinforcement Learning with LLMs
  19. Zhong-Zhi Li, Duzhen Zhang, Ming-Liang Zhang, Jiaxin Zhang, Zengyan Liu, Yuxuan Yao, Haotian Xu, Junhao Zheng, Pei-Jie Wang, Xiuyi Chen, et al. 2025c. From system 1 to system 2: A survey of reasoning large language models. arXiv preprint arXiv:2502.17419.
  20. Yongqi Tong, Dawei Li, Sizhe Wang, Yujia Wang, Fei Teng, and Jingbo Shang. 2024. Can llms learn from previous mistakes? investigating llms’ errors to boost for reasoning. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 3065–30…
  21. Hunter Lightman, Vineet Kosaraju, Yuri Burda, Harrison Edwards, Bowen Baker, Teddy Lee, Jan Leike, John Schulman, Ilya Sutskever, and Karl Cobbe. 2023. Let’s verify step by step. In The Twelfth International Conference on Learning Representations.
  22. Heming Xia, Yongqi Li, Chak Tou Leong, Wenjie Wang, and Wenjie Li. 2025. Tokenskip: Controllable chain-of-thought compression in llms. arXiv preprint arXiv:2502.12067.
  23. Aixin Liu, Bei Feng, Bing Xue, Bingxuan Wang, Bochao Wu, Chengda Lu, Chenggang Zhao, Chengqi Deng, Chenyu Zhang, Chong Ruan, et al. 2024. Deepseek-v3 technical report. arXiv preprint arXiv:2412.19437. find
  24. Fengli Xu, Qianyue Hao, Zefang Zong, Jingwei Wang, Yunke Zhang, Jingyi Wang, Xiaochong Lan, Jiahui Gong, Tianjian Ouyang, Fanjin Meng, et al. 2025a. Towards large reasoning models: A survey of reinforced reasoning with large language models. arXiv preprint arXiv:2501.09686.
  25. Xinyin Ma, Guangnian Wan, Runpeng Yu, Gongfan Fang, and Xinchao Wang. 2025. Cot-valve: Length-compressible chain-of-thought tuning. arXiv preprint arXiv:2502.09601.
  26. Jingxian Xu, Mengyu Zhou, Weichang Liu, Hanbing Liu, Shi Han, and Dongmei Zhang. 2025b. Twt: Thinking without tokens by habitual reasoning distillation with multi-teachers’ guidance. arXiv preprint arXiv:2503.24198. find
  27. Aman Madaan, Niket Tandon, Prakhar Gupta, Skyler Hallinan, Luyu Gao, Sarah Wiegreffe, Uri Alon, Nouha Dziri, Shrimai Prabhumoye, Yiming Yang, et al. 2023. Self-refine: Iterative refinement with self-feedback. Advances in Neural Information Processing Systems, 36:46534–46594. find
  28. Zhangchen Xu, Fengqing Jiang, Luyao Niu, Bill Yuchen Lin, and Radha Poovendran. 2024. Stronger models are not stronger teachers for instruction tuning. arXiv preprint arXiv:2411.07133. — [arXiv:2411.07133](https://arxiv.org/abs/2411.07133) find
  29. Niklas Muennighoff, Zitong Yang, Weijia Shi, Xiang Lisa Li, Li Fei-Fei, Hannaneh Hajishirzi, Luke Zettlemoyer, Percy Liang, Emmanuel Candès, and Tatsunori Hashimoto. 2025. s1: Simple test-time scaling. arXiv preprint arXiv:2501.19393.
  30. Chenxu Yang, Qingyi Si, Yongjie Duan, Zheliang Zhu, Chenyu Zhu, Zheng Lin, Li Cao, and Weiping Wang. 2025. Dynamic early exit in reasoning models. arXiv preprint arXiv:2504.15895.
  31. Tergel Munkhbat, Namgyu Ho, Seo Hyun Kim, Yongjin Yang, Yujin Kim, and Se-Young Yun. 2025. Self-training elicits concise reasoning in large language models. arXiv preprint arXiv:2502.20122. find
  32. Yiyao Yu, Yuxiang Zhang, Dongdong Zhang, Xiao Liang, Hengyuan Zhang, Xingxing Zhang, Ziyi Yang, Mahmoud Khademi, Hany Awadalla, Junjie Wang, et al. 2025. Chain-of-reasoning: Towards unified mathematical reasoning in large language models via a multi-paradigm perspective. arXiv preprint arXiv:2501.11…
  33. OpenAI. 2022. Chatgpt: Openai’s conversational ai. https://chat.openai.com/ . ↗
  34. Gpt-4o: Openai’s multimodal https://openai.com/index/hello-gpt-4o/ . ↗
  35. Zhiyuan Zeng, Qinyuan Cheng, Zhangyue Yin, Yunhua Zhou, and Xipeng Qiu. 2025. Revisiting the test-time scaling of o1-like models: Do they truly possess test-time scaling capabilities? arXiv preprint arXiv:2502.12215.
  36. OpenAI. 2024b. https://openai.com/index/learning-to-reason-with-llms/. ↗
  37. Yang Sui, Yu-Neng Chuang, Guanchu Wang, Jiamu Zhang, Tianyi Zhang, Jiayi Yuan, Hongyi Liu, Andrew Wen, Shaochen Zhong, Hanjie Chen, et al. 2025. Stop overthinking: A survey on efficient reasoning for large language models. arXiv preprint arXiv:2503.16419.
  38. Xunyu Zhu, Jian Li, Can Ma, and Weiping Wang. 2024. Improving mathematical reasoning capabilities of small language models via feedback-driven distillation. arXiv preprint arXiv:2411.14698. — [arXiv:2411.14698](https://arxiv.org/abs/2411.14698) find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/jiang-2025-drp-distilled-reasoning-pruning-skill/