Elamur Research⌕
☰
home / papers / 2026 / Enhancing Auto-regressive Chain-of-Thought through Loop-Aligned Reasoning

Enhancing Auto-regressive Chain-of-Thought through Loop-Aligned Reasoning

Paper · 2026 · Qifan Yu, Zhenyu He, Sijie Li et al.

Summary

Auto-regressive language models struggle to produce long, correct chain-of-thought (CoT) traces. Accuracy falls once reasoning runs longer than anything seen in training, and good step-by-step training data is costly to obtain. Looped Transformers reuse one parameter-shared block for a number of iterations that scales with problem difficulty, and they extrapolate to longer inputs much better. However, they are less flexible for general language tasks. The authors propose RELAY (REasoning through Loop Alignment iterativelY), which ties each loop iteration to one CoT step and supervises the intermediate iterations during training. A single looped model trained this way keeps its length generalization across several tasks and can also output explicit reasoning steps for problems longer than its training data. These synthetic traces are then used to fine-tune a standard auto-regressive CoT model. On Arithmetic, Edit Distance and Longest Increasing Subsequence, the fine-tuned model clearly beats the baseline on out-of-range lengths and sometimes slightly surpasses the looped teacher. This shows that extrapolation ability can be transferred without changing the model architecture.

Key points

  • Benchmarks and length splits:
    • Arithmetic: trained up to length 15, tested on 15–25.
    • Edit Distance (ED): trained up to 30, tested on 30–40.
    • LIS: trained up to 100, tested on 100–120.
  • Without RELAY, the AR-CoT model's accuracy drops rapidly beyond the training range, while a vanilla looped model stays considerably more accurate.
  • For LIS at length 105 (11 steps, each with 10 numbers plus a <sep> token), RELAY-generated traces reach close to 100% token accuracy at every step. The AR model's own traces degrade in later steps and mostly get only the format right.
  • A baseline that fine-tunes the AR model on its own traces gives only small gains. In this baseline, traces were filtered by final-answer correctness using either the looped model or ground-truth answers as the verifier.
  • Stated limitations:
    • The method is tested only on structured reasoning tasks.
    • Choosing the iteration count is non-trivial.
    • Variable CoT lengths complicate step-wise alignment.
  • Code is planned for release at github.com/qifanyu/RELAY.

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

 Institutions

🎓Peking UniversityBeijing · China244 works · 39,359 citations

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksComputer Vision and Pattern Recognition · 2,869 works

🧭 Topics 1Research subjects and languages the work is about.Reasoning · 1,426 works

Cite

Cite as Qifan Yu, Zhenyu He, Sijie Li, Zhou Xun, Jun Zhang, Jingjing Xu and Di He. 2026. Enhancing Auto-regressive Chain-of-Thought through Loop-Aligned Reasoning. Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics (Volume 1: Long Papers).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.eacl-long.97

@inproceedings{yu2026enhancing,
  title = {{Enhancing Auto-regressive Chain-of-Thought through Loop-Aligned Reasoning}},
  author = {Qifan Yu, Zhenyu He, Sijie Li, Zhou Xun, Jun Zhang, Jingjing Xu and Di He},
  booktitle = {Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics (Volume 1: Long Papers)},
  year = {2026},
  doi = {10.18653/v1/2026.eacl-long.97},
  url = {https://doi.org/10.18653/v1/2026.eacl-long.97}
}

References

40 references, 16 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Bo Chen, Xiaoyu Li, Yingyu Liang, Zhenmei Shi, and Zhao Song. 2024. Bypassing the exponential dependency: Looped transformers efficiently learn in-context by multi-step gradient descent. arXiv preprint arXiv:2410.11268. — [arXiv:2410.11268](https://arxiv.org/abs/2410.11268) find
  2. Shouyuan Chen, Sherman Wong, Liangjian Chen, and Yuandong Tian. 2023. Extending context window of large language models via positional interpolation. arXiv preprint arXiv:2306.15595. find
  3. Ta-Chung Chi, Ting-Han Fan, Peter J Ramadge, and Alexander Rudnicky. 2022. Kerple: Kernelized relative positional embedding for length extrapolation. Advances in Neural Information Processing Systems, 35:8386–8399. find
  4. Ta-Chung Chi, Ting-Han Fan, Alexander Rudnicky, and Peter Ramadge. 2023. Dissecting transformer length extrapolation via the lens of receptive field analysis. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 13522–13537. find
  5. Artur Back de Luca and Kimon Fountoulakis. 2024. Simulation of graph algorithms with looped transformers. In Forty-first International Conference on Machine Learning. find
  6. 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, and 181 others. 2025. Deepseek-r1: Incentivizing reasoning capability in ll…
  7. Mostafa Dehghani, Stephan Gouws, Oriol Vinyals, Jakob Uszkoreit, and Lukasz Kaiser. 2019. Universal transformers. In International Conference on Learning Representations. find
  8. Mingyu Jin, Qinkai Yu, Dong Shu, Haiyan Zhao, Wenyue Hua, Yanda Meng, Yongfeng Zhang, and Mengnan Du. 2024. The impact of reasoning step length on large language models. In Findings of the Association for Computational Linguistics: ACL 2024, pages 1830–1842. Association for Computational Linguistics…
  9. Tushar Khot, Harsh Trivedi, Matthew Finlayson, Yao Fu, Kyle Richardson, Peter Clark, and Ashish Sabharwal. 2022. Decomposed prompting: A modular approach for solving complex tasks. arXiv preprint arXiv:2210.02406. find
  10. Zhenzhong Lan, Mingda Chen, Sebastian Goodman, Kevin Gimpel, Piyush Sharma, and Radu Soricut. 2020. Albert: A lite bert for self-supervised learning of language representations. In International Conference on Learning Representations.
  11. Ying Fan, Yilun Du, Kannan Ramchandran, and Kangwook Lee. 2024. Looped transformers for length generalization. arXiv preprint arXiv:2409.15647. find
  12. Shanda Li, Chong You, Guru Guruganesh, Joshua Ainslie, Santiago Ontanon, Manzil Zaheer, Sumit Sanghai, Yiming Yang, Sanjiv Kumar, and Srinadh Bhojanapalli. 2024. Functional interpolation for relative positions improves long context transformers. In The Twelfth International Conference on Learning Re… find
  13. Guhao Feng, Bohang Zhang, Yuntian Gu, Haotian Ye, Di He, and Liwei Wang. 2024. Towards revealing the mystery behind chain of thought: a theoretical perspective. Advances in Neural Information Processing Systems, 36. find
  14. Hunter Lightman, Vineet Kosaraju, Yuri Burda, Harrison Edwards, Bowen Baker, Teddy Lee, Jan Leike, John Schulman, Ilya Sutskever, and Karl Cobbe. 2024. Let’s verify step by step. In The Twelfth International Conference on Learning Representations.
  15. Yihang Gao, Chuanyang Zheng, Enze Xie, Han Shi, Tianyang Hu, Yu Li, Michael K Ng, Zhenguo Li, and Zhaoqiang Liu. 2024. On the expressive power of a variant of the looped transformer. arXiv preprint arXiv:2402.13572. — [arXiv:2402.13572](https://arxiv.org/abs/2402.13572) find
  16. Yansheng Mao, Jiaqi Li, Fanxu Meng, Jing Xiong, Zilong Zheng, and Muhan Zhang. 2024. Lift: Improving long context understanding through long input fine-tuning. arXiv preprint arXiv:2412.13626. — [arXiv:2412.13626](https://arxiv.org/abs/2412.13626) find
  17. Khashayar Gatmiry, Nikunj Saunshi, Sashank J Reddi, Stefanie Jegelka, and Sanjiv Kumar. 2024. Can looped transformers learn to implement multi-step gradient descent for in-context learning? arXiv preprint arXiv:2410.08292. — [arXiv:2410.08292](https://arxiv.org/abs/2410.08292) find
  18. Angeliki Giannou, Shashank Rajput, Jy-Yong Sohn, Kangwook Lee, Jason D. Lee, and Dimitris Papailiopoulos. 2023. Looped transformers as programmable computers. In Proceedings of the 40th International Conference on Machine Learning, volume 202 of Proceedings of Machine Learning Research, pages 11398–… find
  19. 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. In Thirty-fifth Conference on Neural Information Processing Systems Datasets and Benchmarks Track (Round 2).
  20. Arian Hosseini, Xingdi Yuan, Nikolay Malkin, Aaron Courville, Alessandro Sordoni, and Rishabh Agarwal. 2024. V-STar: Training verifiers for self-taught reasoners. In First Conference on Language Modeling.
  21. William Merrill and Ashish Sabharwal. 2024. The expressive power of transformers with chain of thought. In The Twelfth International Conference on Learning Representations. find
  22. Debjit Paul, Robert West, Antoine Bosselut, and Boi Faltings. 2024. Making reasoning matter: Measuring and improving faithfulness of chain-of-thought reasoning. In Findings of the Association for Computational Linguistics: EMNLP 2024, pages 15012– 15032, Miami, Florida, USA. Association for Computat…
  23. Bowen Peng, Jeffrey Quesnelle, Honglu Fan, and Enrico Shippole. 2024. YaRN: Efficient context window extension of large language models. In The Twelfth International Conference on Learning Representations. find
  24. Ofir Press, Noah Smith, and Mike Lewis. 2022. Train short, test long: Attention with linear biases enables input length extrapolation. find
  25. Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J Liu. 2020. Exploring the limits of transfer learning with a unified text-to-text transformer. The Journal of Machine Learning Research, 21(1):5485–5551.
  26. Anian Ruoss, Grégoire Delétang, Tim Genewein, Jordi Grau-Moya, Róbert Csordás, Mehdi Bennani, Shane Legg, and Joel Veness. 2023. Randomized positional encodings boost length generalization of transformers. pages 1889–1903, Toronto, Canada. Association for Computational Linguistics. find
  27. Liu Yang, Kangwook Lee, Robert D Nowak, and Dimitris Papailiopoulos. 2024. Looped transformers are better at learning learning algorithms. In The Twelfth International Conference on Learning Representations. find
  28. Zheng Yuan, Hongyi Yuan, Chengpeng Li, Guanting Dong, Keming Lu, Chuanqi Tan, Chang Zhou, and Jingren Zhou. 2024. Scaling relationship on learning mathematical reasoning with large language models.
  29. Nikunj Saunshi, Nishanth Dikkala, Zhiyuan Li, Sanjiv Kumar, and Sashank J. Reddi. 2025. Reasoning with latent thoughts: On the power of looped transformers. Preprint, arXiv:2502.17416. find
  30. Eric Zelikman, Yuhuai Wu, Jesse Mu, and Noah Goodman. 2022. STar: Bootstrapping reasoning with reasoning. In Advances in Neural Information Processing Systems.
  31. Avi Singh, John D Co-Reyes, Rishabh Agarwal, Ankesh Anand, Piyush Patil, Xavier Garcia, Peter J Liu, James Harrison, Jaehoon Lee, Kelvin Xu, Aaron T Parisi, Abhishek Kumar, Alexander A Alemi, Alex Rizkowsky, Azade Nova, Ben Adlam, Bernd Bohnet, Gamaleldin Fathy Elsayed, Hanie Sedghi, and 21 others.… — Beyond Human Data: Scaling Self-Training for Problem-Solving with Language Models
  32. Chunting Zhou, Pengfei Liu, Puxin Xu, Srini Iyer, Jiao Sun, Yuning Mao, Xuezhe Ma, Avia Efrat, Ping Yu, LILI YU, Susan Zhang, Gargi Ghosh, Mike Lewis, Luke Zettlemoyer, and Omer Levy. 2023. LIMA: Less is more for alignment. In Thirty-seventh Conference on Neural Information Processing Systems. find
  33. Jianlin Su, Murtadha Ahmed, Yu Lu, Shengfeng Pan, Wen Bo, and Yunfeng Liu. 2024. Roformer: Enhanced transformer with rotary position embedding. Neurocomput., 568(C).
  34. Dawei Zhu, Nan Yang, Liang Wang, Yifan Song, Wenhao Wu, Furu Wei, and Sujian Li. 2024. PoSE: Efficient context window extension of LLMs via positional skip-wise training. In The Twelfth International Conference on Learning Representations.
  35. Yutao Sun, Li Dong, Barun Patra, Shuming Ma, Shaohan Huang, Alon Benhaim, Vishrav Chaudhary, Xia Song, and Furu Wei. 2023. A length-extrapolatable transformer. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). Association for Computat…
  36. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, and 1 others. 2022. Chain-of-thought prompting elicits reasoning in large language models. Advances in neural information processing systems, 35:24824– 24837.
  37. Mengzhou Xia, Sadhika Malladi, Suchin Gururangan, Sanjeev Arora, and Danqi Chen. 2024. LESS: Selecting influential data for targeted instruction tuning. In International Conference on Machine Learning (ICML).
  38. Changnan Xiao and Bing Liu. 2023. Conditions for length generalization in learning reasoning skills. arXiv preprint arXiv:2311.16173. — [arXiv:2311.16173](https://arxiv.org/abs/2311.16173) find
  39. Kevin Xu and Issei Sato. 2024. On expressive power of looped transformers: Theoretical analysis and enhancement via timestep encoding. arXiv preprint arXiv:2410.01405. — [arXiv:2410.01405](https://arxiv.org/abs/2410.01405) find
  40. Liu Yang, Kangwook Lee, Robert Nowak, and Dimitris Papailiopoulos. 2023. Looped transformers are better at learning learning algorithms. arXiv preprint arXiv:2311.12424. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/yu-2026-enhancing-auto-regressive-chain-thought/