---
title: "Deconstructing Long Chain-of-Thought: A Structured Reasoning"
description: "This offers practical guidance for curating reasoning-distillation data. 27 references, cited by 14 papers here."
authors: ["Yijia Luo", "Yulin Song", "Xingyao Zhang", "Jiaheng Liu", "Weixun Wang", "GengRu Chen", "Wenbo Su", "Bo Zheng"]
year: 2025
venue: "arXiv:2503.16385"
source: "https://arxiv.org/abs/2503.16385"
doi: "10.48550/arxiv.2503.16385"
arxiv: "2503.16385"
licence: "CC0 1.0 (arXiv)"
licence_url: "http://creativecommons.org/publicdomain/zero/1.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/luo-2025-deconstructing-long-chain-thought-structured/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/luo-2025-deconstructing-long-chain-thought-structured/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2025](https://md.research.elamur.ai/free/en/papers/2025/page/8/) / Deconstructing Long Chain-of-Thought: A Structured Reasoning Optimization Framework for Long CoT Distillation

# Deconstructing Long Chain-of-Thought: A Structured Reasoning Optimization Framework for Long CoT Distillation

**Authors:** Yijia Luo, Yulin Song, [Xingyao Zhang](https://md.research.elamur.ai/free/en/authors/xingyao-zhang/), [Jiaheng Liu](https://md.research.elamur.ai/free/en/authors/jiaheng-liu/), [Weixun Wang](https://md.research.elamur.ai/free/en/authors/weixun-wang/), [GengRu Chen](https://md.research.elamur.ai/free/en/authors/gengru-chen/), [Wenbo Su](https://md.research.elamur.ai/free/en/authors/wenbo-su/) and [Bo Zheng](https://md.research.elamur.ai/free/en/authors/bo-zheng/) · **Year:** 2025 · **Venue:** arXiv:2503.16385 · **Licence:** [CC0 1.0 (arXiv)](http://creativecommons.org/publicdomain/zero/1.0/)

**Links:** [Publisher's page](https://arxiv.org/abs/2503.16385) · [PDF](https://arxiv.org/pdf/2503.16385) · [DOI 10.48550/arxiv.2503.16385](https://doi.org/10.48550/arxiv.2503.16385) · [arXiv 2503.16385](https://arxiv.org/abs/2503.16385) · [OpenAlex](https://openalex.org/W4415289310)

**Citations:** 0 citations (OpenAlex, 2026-10-14) · [14](https://md.research.elamur.ai/free/en/papers/luo-2025-deconstructing-long-chain-thought-structured/cited-by/) in our library, of the 149,439 papers whose bibliographies we read

## Summary
Distilling long chain-of-thought (CoT) reasoning from large teachers into smaller models is an inexpensive way to build strong reasoners, but it is unclear why it works and how well it transfers. The authors fine-tuned Qwen2.5 and Llama3.1 students on about 6K correct NuminaMath solutions generated by DeepSeek-R1 and QwQ-32B-Preview. QwQ-generated data helped Qwen-family students far more than Llama, so distillation data is less universal than often assumed. R1 data transferred better across model families.

The authors observed that long CoTs form linear, tree or network structures around a core "trunk" and often contain overthinking. To address this, they built DLCoT, an automated pipeline that segments CoTs, prunes redundant or unsolvable branches, and corrects intermediate errors. On AIME2024, MATH500 and GSM8K, DLCoT generally raised accuracy and reduced output length. The key finding is that a diverse, non-redundant set of solution approaches matters more than flawless individual steps. This offers practical guidance for curating reasoning-distillation data.

## Key points
- Students included Qwen2.5-Math-7B, Qwen2.5-14B/32B, Llama3.1-8B and Llama3.1-70B. The R1-distilled Qwen2.5-14B reached accuracy comparable to the QwQ-distilled Qwen2.5-32B.
- DLCoT has three modules: Intelligent Segmentation, Redundancy Elimination and Error Correction. Variants include DLCoT-multi1, -multi2 and -multiall. The authors claim token efficiency improves by at least 5%.
- With R1 data, Qwen2.5-14B improved on AIME2024 from 46.7 to 53.3 using DLCoT-multiall. Its average output length on that benchmark fell from about 28.5K to 18.8K tokens. With QwQ data, the same model rose from 33.3 to 40.0.
- Llama3.1-8B (R1 data) produced far fewer tokens on GSM8K under DLCoT-multiall, dropping from about 16.6K to 3.8K.
- Deleting incorrect approaches or calculation errors hurt performance. R1-distilled Qwen2.5-14B fell from 53.3 to 40.0 on AIME2024, because this deletion damages self-reflection and reduces approach diversity. Code is announced at the SODE GitHub repository.

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

📚 Shelves (4): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 19,503 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 18,805 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 17,959 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 17,789 works

🧭 Topics (2): [🧮 Reasoning](https://md.research.elamur.ai/free/en/topics/reasoning/) · 1,948 works · [🪶 Efficient models & compression](https://md.research.elamur.ai/free/en/topics/efficient-models/) · 328 works

## Cite

Cite as: Yijia Luo, Yulin Song, Xingyao Zhang, Jiaheng Liu, Weixun Wang, GengRu Chen, Wenbo Su and Bo Zheng. 2025. Deconstructing Long Chain-of-Thought: A Structured Reasoning Optimization Framework for Long CoT Distillation. arXiv:2503.16385.

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@misc{luo2025deconstructing,
  title = {{Deconstructing Long Chain-of-Thought: A Structured Reasoning Optimization Framework for Long CoT Distillation}},
  author = {Yijia Luo, Yulin Song, Xingyao Zhang, Jiaheng Liu, Weixun Wang, GengRu Chen, Wenbo Su and Bo Zheng},
  year = {2025},
  doi = {10.48550/arxiv.2503.16385},
  eprint = {2503.16385},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2503.16385}
}
```

- [← previous Comparative Study of Multilingual Idioms and Similes in Large Language Models](https://md.research.elamur.ai/free/en/papers/khoshtab-2025-comparative-study-multilingual-idioms-similes/)
- [next → Does Context Matter? ContextualJudgeBench for Evaluating LLM-based Judges in Contextual Settings](https://md.research.elamur.ai/free/en/papers/xu-2025-does-context-matter-contextualjudgebench-evaluating/)
- [↑ back to papers of 2025](https://md.research.elamur.ai/free/en/papers/2025/page/8/)

## References

27 references, 19 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Bespoke, 2025. [https://huggingface.co/datasets/bespokelabs/Bespoke-Stratos-17k](https://huggingface.co/datasets/bespokelabs/Bespoke-Stratos-17k). Bespoke-Stratos-17k. [↗](https://huggingface.co/datasets/bespokelabs/Bespoke-Stratos-17k)
2. Bradley Brown, Jordan Juravsky, Ryan Ehrlich, Ronald Clark, Quoc V. Le, Christopher Ré, and Azalia Mirhoseini. 2024. [Large language monkeys: Scaling inference compute with repeated sampling](https://md.research.elamur.ai/free/en/papers/brown-2024-large-language-monkeys-scaling-inference/). Preprint, arXiv:2407.21787.
3. Z. Z. Ren, Zehui Ren, Zhangli Sha, Zhe Fu, Zhean Xu, Zhenda Xie, Zhengyan Zhang, Zhewen Hao, Zhicheng Ma, Zhigang Yan, Zhiyu Wu, Zihui Gu, Zijia Zhu, Zijun Liu, Zilin Li, Ziwei Xie, Ziyang Song, Zizheng Pan, Zhen Huang, Zhipeng Xu, Zhongyu Zhang, and Zhen Zhang. 2025. [Deepseek-r1: Incentivizing reas…](https://md.research.elamur.ai/free/en/papers/deepseek-ai-2025-deepseek-r1-incentivizing-reasoning-capability/)
4. Xingyu Chen, Jiahao Xu, Tian Liang, Zhiwei He, Jianhui Pang, Dian Yu, Linfeng Song, Qiuzhi Liu, Mengfei Zhou, Zhuosheng Zhang, Rui Wang, Zhaopeng Tu, Haitao Mi, and Dong Yu. 2025. [Do not think that much for 2+3=? on the overthinking of o1-like llms](https://md.research.elamur.ai/free/en/papers/chen-2024-think-that-much-2-3/). Preprint, arXiv:2412.21187.
5. Tingxu Han, Zhenting Wang, Chunrong Fang, Shiyu Zhao, Shiqing Ma, and Zhenyu Chen. 2024. [Token-budget-aware llm reasoning](https://md.research.elamur.ai/free/en/papers/han-2024-token-budget-aware-llm-reasoning/). Preprint, arXiv:2412.18547.
6. Karl Cobbe, Vineet Kosaraju, Mohammad Bavarian, Mark Chen, Heewoo Jun, Lukasz Kaiser, Matthias Plappert, Jerry Tworek, Jacob Hilton, Reiichiro Nakano, Christopher Hesse, and John Schulman. 2021. [Training verifiers to solve math word problems](https://md.research.elamur.ai/free/en/papers/cobbe-2021-training-verifiers-solve-math-word/). Preprint, arXiv:2110.14168.
7. 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](https://md.research.elamur.ai/free/en/papers/hendrycks-2021-measuring-mathematical-problem-solving-math/). Preprint, arXiv:2103.03874.
8. DeepSeek-AI, Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Ruoyu Zhang, Runxin Xu, Qihao Zhu, Shirong Ma, Peiyi Wang, Xiao Bi, Xiaokang Zhang, Xingkai Yu, Yu Wu, Z. F. Wu, Zhibin Gou, Zhihong Shao, Zhuoshu Li, Ziyi Gao, Aixin Liu, Bing Xue, Bingxuan Wang, Bochao Wu, Bei Feng, Chengda Lu, Chengg… — [Distilling the Knowledge in a Neural Network](https://md.research.elamur.ai/free/en/papers/hinton-2015-distilling-knowledge-neural-network/)
9. Zhen Huang, Haoyang Zou, Xuefeng Li, Yixiu Liu, Yuxiang Zheng, Ethan Chern, Shijie Xia, Yiwei Qin, Weizhe Yuan, and Pengfei Liu. 2024. [O1 replication journey – part 2: Surpassing o1-preview through simple distillation, big progress or bitter lesson?](https://md.research.elamur.ai/free/en/papers/huang-2024-o1-replication-journey-part-2/) Preprint, arXiv:2411.16489.
10. Jinhao Jiang, Zhipeng Chen, Yingqian Min, Jie Chen, Xiaoxue Cheng, Jiapeng Wang, Yiru Tang, Haoxiang Sun, Jia Deng, Wayne Xin Zhao, Zheng Liu, Dong Yan, Jian Xie, Zhongyuan Wang, and Ji-Rong Wen. 2024. Enhancing llm reasoning with reward-guided tree search. Preprint, arXiv:2411.11694. — [arXiv:2411.11694](https://arxiv.org/abs/2411.11694) [find](https://scholar.google.com/scholar?q=Jinhao+Jiang%2C+Zhipeng+Chen%2C+Yingqian+Min%2C+Jie+Chen%2C+Xiaoxue+Cheng%2C+Jiapeng+Wang%2C+Yiru+Tang%2C+Haoxiang+Sun%2C+Jia+Deng%2C+Wayne+Xin+Zhao%2C+Zheng+Liu%2C+Dong+Yan%2C+Jian+Xie%2C+Zhongyuan+Wang%2C+and+Ji%2DRong+Wen%2E+2024)
11. Ryo Kamoi, Yusen Zhang, Nan Zhang, Jiawei Han, and Rui Zhang. 2024. [When can LLMs actually correct their own mistakes? a critical survey of self-correction of LLMs](https://md.research.elamur.ai/free/en/papers/kamoi-2024-when-can-llms-actually-correct/). Transactions of the Association for Computational Linguistics, 12:1417–1440.
12. Bespoke Labs. 2025. Bespoke-stratos: The unreasonable effectiveness of reasoning distillation. [https://www.bespokelabs.ai/blog/bespoke-stratos-the-unreasonable-effectiveness-of-reasoning-distillation](https://www.bespokelabs.ai/blog/bespoke-stratos-the-unreasonable-effectiveness-of-reasoning-distillation). Accessed: 2025-01-22. [↗](https://www.bespokelabs.ai/blog/bespoke-stratos-the-unreasonable-effectiveness-of-reasoning-distillation)
13. Dacheng Li, Shiyi Cao, Tyler Griggs, Shu Liu, Xiangxi Mo, Shishir G. Patil, Matei Zaharia, Joseph E. Gonzalez, and Ion Stoica. 2025. [Llms can easily learn to reason from demonstrations structure, not content, is what matters!](https://md.research.elamur.ai/free/en/papers/li-2025-llms-easily-learn-reason-demonstrations/) Preprint, arXiv:2502.07374.
14. Jia Li, Edward Beeching, Lewis Tunstall, Ben Lipkin, Roman Soletskyi, Shengyi Huang, Kashif Rasul, Longhui Yu, Albert Q Jiang, Ziju Shen, et al. 2024. Numinamath: The largest public dataset in ai4maths with 860k pairs of competition math problems and solutions. Hugging Face repository, 13:9. [find](https://scholar.google.com/scholar?q=Jia+Li%2C+Edward+Beeching%2C+Lewis+Tunstall%2C+Ben+Lipkin%2C+Roman+Soletskyi%2C+Shengyi+Huang%2C+Kashif+Rasul%2C+Longhui+Yu%2C+Albert+Q+Jiang%2C+Ziju+Shen%2C+et+al%2E+2024%2E+Numinamath%3A+The+largest+public+dataset+in+ai4math)
15. Tian Liang, Zhiwei He, Wenxiang Jiao, Xing Wang, Yan Wang, Rui Wang, Yujiu Yang, Shuming Shi, and Zhaopeng Tu. 2024. [Encouraging divergent thinking in large language models through multi-agent debate](https://md.research.elamur.ai/free/en/papers/liang-2023-encouraging-divergent-thinking-large-language/). In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 17…
16. Mayer, Andrew Mayne, Bob McGrew, Scott Mayer McKinney, Christine McLeavey, Paul McMillan, Jake McNeil, David Medina, Aalok Mehta, Jacob Menick, Luke Metz, Andrey Mishchenko, Pamela Mishkin, Vinnie Monaco, Evan Morikawa, Daniel Mossing, Tong Mu, Mira Murati, Oleg Murk, David Mély, Ashvin Nair, Reiich… — [GPT-4 Technical Report](https://md.research.elamur.ai/free/en/papers/openai-2023-gpt-4-technical-report/)
17. Arindam Mitra, Hamed Khanpour, Corby Rosset, and Ahmed Awadallah. 2024. [Orca-math: Unlocking the potential of slms in grade school math](https://md.research.elamur.ai/free/en/papers/mitra-2024-orca-math-unlocking-potential-slms/). Preprint, arXiv:2402.14830.
18. Maxwell Nye, Anders Johan Andreassen, Guy Gur-Ari, Henryk Michalewski, Jacob Austin, David Bieber, David Dohan, Aitor Lewkowycz, Maarten Bosma, David Luan, Charles Sutton, and Augustus Odena. 2021. [Show your work: Scratchpads for intermediate computation with language models](https://md.research.elamur.ai/free/en/papers/al-2021-show-your-work-scratchpads-intermediate/). Preprint, arXiv:2112.00…
19. OpenAI, Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, Red Avila, Igor Babuschkin, Suchir Balaji, Valerie Balcom, Paul Baltescu, Haiming Bao, Mohammad Bavarian, Jeff Belgum, Irwan Bello, Ja… [find](https://scholar.google.com/scholar?q=OpenAI%2C+Josh+Achiam%2C+Steven+Adler%2C+Sandhini+Agarwal%2C+Lama+Ahmad%2C+Ilge+Akkaya%2C+Florencia+Leoni+Aleman%2C+Diogo+Almeida%2C+Janko+Altenschmidt%2C+Sam+Altman%2C+Shyamal+Anadkat%2C+Red+Avila%2C+Igor+Babuschkin%2C+Suchir)
20. Yiwei Qin, Xuefeng Li, Haoyang Zou, Yixiu Liu, Shijie Xia, Zhen Huang, Yixin Ye, Weizhe Yuan, Hector Liu, Yuanzhi Li, and Pengfei Liu. 2024. [O1 replication journey: A strategic progress report – part 1](https://md.research.elamur.ai/free/en/papers/qin-2024-o1-replication-journey-strategic-progress/). Preprint, arXiv:2410.18982.
21. Qwen, :, An Yang, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chengyuan Li, Dayiheng Liu, Fei Huang, Haoran Wei, Huan Lin, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Yang, Jiaxi Yang, Jingren Zhou, Junyang Lin, Kai Dang, Keming Lu, Keqin Bao, Kexin Yang, Le Yu, Mei Li, Mingfeng… [find](https://scholar.google.com/scholar?q=Qwen%2C+%3A%2C+An+Yang%2C+Baosong+Yang%2C+Beichen+Zhang%2C+Binyuan+Hui%2C+Bo+Zheng%2C+Bowen+Yu%2C+Chengyuan+Li%2C+Dayiheng+Liu%2C+Fei+Huang%2C+Haoran+Wei%2C+Huan+Lin%2C+Jian+Yang%2C+Jianhong+Tu%2C+Jianwei+Zhang%2C+Jianxin+Yang%2C+Jiaxi+)
22. Qwen Team, 2024d. [https://qwenlm.git](https://qwenlm.git) hub.io/blog/qwq-32b-preview/. QwQ: Reflect Deeply on the Boundaries of the Unknown. [↗](https://qwenlm.git)
23. Kimi Team, Angang Du, Bofei Gao, Bowei Xing, Changjiu Jiang, Cheng Chen, Cheng Li, Chenjun Xiao, Chenzhuang Du, Chonghua Liao, Chuning Tang, Congcong Wang, Dehao Zhang, Enming Yuan, Enzhe Lu, Fengxiang Tang, Flood Sung, Guangda Wei, Guokun Lai, Haiqing Guo, Han Zhu, Hao Ding, Hao Hu, Hao Yang, Hao Z… — [Kimi k1.5: Scaling Reinforcement Learning with LLMs](https://md.research.elamur.ai/free/en/papers/kimi-team-2025-kimi-k1-5-scaling-reinforcement/)
24. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Brian Ichter, Fei Xia, Ed Chi, Quoc Le, and Denny Zhou. 2023. [Chain-of-thought prompting elicits reasoning in large language models](https://md.research.elamur.ai/free/en/papers/wei-2022-chain-thought-prompting-elicits-reasoning/). Preprint, arXiv:2201.11903.
25. Zhiheng Xi, Dingwen Yang, Jixuan Huang, Jiafu Tang, Guanyu Li, Yiwen Ding, Wei He, Boyang Hong, Shihan Do, Wenyu Zhan, Xiao Wang, Rui Zheng, Tao Ji, Xiaowei Shi, Yitao Zhai, Rongxiang Weng, Jingang Wang, Xunliang Cai, Tao Gui, Zuxuan Wu, Qi Zhang, Xipeng Qiu, Xuanjing Huang, and Yu-Gang Jiang. 2024.… — [arXiv:2411.16579](https://arxiv.org/abs/2411.16579) [find](https://scholar.google.com/scholar?q=Zhiheng+Xi%2C+Dingwen+Yang%2C+Jixuan+Huang%2C+Jiafu+Tang%2C+Guanyu+Li%2C+Yiwen+Ding%2C+Wei+He%2C+Boyang+Hong%2C+Shihan+Do%2C+Wenyu+Zhan%2C+Xiao+Wang%2C+Rui+Zheng%2C+Tao+Ji%2C+Xiaowei+Shi%2C+Yitao+Zhai%2C+Rongxiang+Weng%2C+Jingang+Wa)
26. Yixin Ye, Zhen Huang, Yang Xiao, Ethan Chern, Shijie Xia, and Pengfei Liu. 2025. [Limo: Less is more for reasoning](https://md.research.elamur.ai/free/en/papers/ye-2025-limo-less-more-reasoning/). Preprint, arXiv:2502.03387.
27. Zhiyuan Zeng, Qinyuan Cheng, Zhangyue Yin, Bo Wang, Shimin Li, Yunhua Zhou, Qipeng Guo, Xuanjing Huang, and Xipeng Qiu. 2024. [Scaling of search and learning: A roadmap to reproduce o1 from reinforcement learning perspective](https://md.research.elamur.ai/free/en/papers/zeng-2024-scaling-search-learning-roadmap-reproduce/). Preprint, arXiv:2412.14135.

*Updated 3 November 2026*