Elamur Research⌕
☰
home / papers / 2026 / Efficient Self-Evaluation for Diffusion Language Models via Sequence Regeneration

Efficient Self-Evaluation for Diffusion Language Models via Sequence Regeneration

Paper · 2026 · Linhao Zhong, Linyu Wu, Wen Wang et al.

Summary

Diffusion language models (dLLMs) such as LLaDA generate text by iteratively unmasking tokens in no fixed order. Because of this, they cannot score their own outputs the way left-to-right models do. The usual workaround is Monte Carlo masking, which is slow and often inaccurate. The lack of a built-in confidence signal also forces dLLMs to produce outputs of a predetermined length.

The authors propose DiSE, a training-free confidence score. It passes the complete prompt-plus-response sequence back through the model and averages the log-probabilities of reproducing a chosen subset of the existing tokens. They justify the approach through the model's generalization: dLLMs can recover the correct token even when the input at that position is corrupted. They also show that DiSE tracks both semantic coherence and answer correctness. The score is applied to:

  • likelihood estimation,
  • uncertainty quantification,
  • a flexible-length decoding scheme that searches for, assesses, and stops at a suitable output length.

Across these tasks, DiSE is cheaper and more accurate than Monte Carlo estimation, and it gives dLLMs a self-assessment capability previously limited to autoregressive LLMs.

Key points

  • DiSE is the mean log-probability of regenerating selected token positions given the full sequence. It needs a single forward pass, compared with many masked samples for Monte Carlo estimation.
  • In a robustness test of 20,000 random-token substitutions, the ground-truth token ranked in the top 5 in 88.85% of cases and in the top 10 in 96.14%.
  • Flexible-length generation was tested on Countdown, GSM8K, MATH500, and SVAMP. For the first reported model (LLaDA-Instruct-8B), average accuracy rose from 52.24 (fixed-length baseline) and 52.38 (max-length baseline) to 53.79.
  • For the second model (LLaDA-1.5-8B), DiSE-flexible averaged 54.92, compared with 53.37 and 53.52 for the two fixed-length baselines.
  • Stated limitations: the method has not been tested on semi-autoregressive hybrid models, and the best token subset for computing DiSE may vary by task.

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

🎓National University of SingaporeSingapore · Singapore180 works · 34,304 citations🎓Zhejiang UniversityHangzhou · China172 works · 12,988 citations🎓Zhejiang University of TechnologyHangzhou · China6 works · 98 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 2Research subjects and languages the work is about.Language models & pretraining · 3,320 worksEvaluation & benchmarks · 1,220 works

Cite

Cite as Linhao Zhong, Linyu Wu, Wen Wang, Yuling Xi, Chenchen Jing, Jiaheng Zhang, Hao Chen and Chunhua Shen. 2026. Efficient Self-Evaluation for Diffusion Language Models via Sequence Regeneration. Proceedings of the 64th Annual Meeting 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.acl-long.298

@inproceedings{zhong2026efficient,
  title = {{Efficient Self-Evaluation for Diffusion Language Models via Sequence Regeneration}},
  author = {Linhao Zhong, Linyu Wu, Wen Wang, Yuling Xi, Chenchen Jing, Jiaheng Zhang, Hao Chen and Chunhua Shen},
  booktitle = {Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
  year = {2026},
  doi = {10.18653/v1/2026.acl-long.298},
  url = {https://doi.org/10.18653/v1/2026.acl-long.298}
}

References

40 references, 17 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Jacob Austin, Daniel D Johnson, Jonathan Ho, Daniel Tarlow, and Rianne Van Den Berg. 2021. Structured denoising diffusion models in discrete state-spaces. Advances in neural information processing systems, 34:17981–17993.
  2. Yavuz Faruk Bakman, Duygu Nur Yaldiz, Baturalp Buyukates, Chenyang Tao, Dimitrios Dimitriadis, and Salman Avestimehr. 2024. Mars: Meaning-aware response scoring for uncertainty estimation in generative llms. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (… find
  3. Yupeng Chang, Xu Wang, Jindong Wang, Yuan Wu, Linyi Yang, Kaijie Zhu, Hao Chen, Xiaoyuan Yi, Cunxiang Wang, Yidong Wang, and 1 others. 2024. A survey on evaluation of large language models. ACM transactions on intelligent systems and technology, 15(3):1–45.
  4. Peter Clark, Isaac Cowhey, Oren Etzioni, Tushar Khot, Ashish Sabharwal, Carissa Schoenick, and Oyvind Tafjord. 2018. Think you have solved question answering? try arc, the ai2 reasoning challenge. arXiv:1803.05457v1.
  5. Karl Cobbe, Vineet Kosaraju, Mohammad Bavarian, Mark Chen, Heewoo Jun, Lukasz Kaiser, Matthias Plappert, Jerry Tworek, Jacob Hilton, Reiichiro Nakano, and 1 others. 2021. Training verifiers to solve math word problems. arXiv preprint arXiv:2110.14168.
  6. Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Amy Yang, Angela Fan, and 1 others. 2024. The llama 3 herd of models. arXiv e-prints, pages arXiv–2407.
  7. Ekaterina Fadeeva, Aleksandr Rubashevskii, Artem Shelmanov, Sergey Petrakov, Haonan Li, Hamdy Mubarak, Evgenii Tsymbalov, Gleb Kuzmin, Alexander Panchenko, Timothy Baldwin, and 1 others. 2024. Fact-checking the output of large language models via token-level uncertainty quantification. arXiv preprin…
  8. Jiahui Geng, Fengyu Cai, Yuxia Wang, Heinz Koeppl, Preslav Nakov, and Iryna Gurevych. 2023. A survey of confidence estimation and calibration in large language models. arXiv preprint arXiv:2311.08298. find
  9. Shansan Gong, Shivam Agarwal, Yizhe Zhang, Jiacheng Ye, Lin Zheng, Mukai Li, Chenxin An, Peilin Zhao, Wei Bi, Jiawei Han, and 1 others. 2024. Scaling diffusion language models via adaptation from autoregressive models. arXiv preprint arXiv:2410.17891. — [arXiv:2410.17891](https://arxiv.org/abs/2410.17891) find
  10. Robert L Harrison. 2010. Introduction to monte carlo simulation. In AIP conference proceedings, volume 1204, page 17. find
  11. Wenchong He, Zhe Jiang, Tingsong Xiao, Zelin Xu, and Yukun Li. 2023. A survey on uncertainty quantification methods for deep learning. arXiv preprint arXiv:2302.13425. — [arXiv:2302.13425](https://arxiv.org/abs/2302.13425) find
  12. Jonathan Ho, Ajay Jain, and Pieter Abbeel. 2020. Denoising diffusion probabilistic models. Advances in Neural Information Processing Systems, 33:6840– 6851. find
  13. Yukun Huang, Yixin Liu, Raghuveer Thirukovalluru, Arman Cohan, and Bhuwan Dhingra. 2024. Calibrating long-form generations from large language models. arXiv preprint arXiv:2402.06544. find
  14. Lorenz Kuhn, Yarin Gal, and Sebastian Farquhar. 2023. Semantic uncertainty: Linguistic invariances for uncertainty estimation in natural language generation. arXiv preprint arXiv:2302.09664. find
  15. Jinsong Li, Xiaoyi Dong, Yuhang Zang, Yuhang Cao, Jiaqi Wang, and Dahua Lin. 2025. Beyond fixed: Training-free variable-length denoising for diffusion large language models. arXiv preprint arXiv:2508.00819. — [arXiv:2508.00819](https://arxiv.org/abs/2508.00819) find
  16. 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.
  17. Stephanie Lin, Jacob Hilton, and Owain Evans. 2022. Teaching models to express their uncertainty in words. arXiv preprint arXiv:2205.14334. find
  18. Alexander Quinn Nichol and Prafulla Dhariwal. 2021. Improved denoising diffusion probabilistic models. In International Conference on Machine Learning, pages 8162–8171. PMLR. find
  19. Shen Nie, Fengqi Zhu, Zebin You, Xiaolu Zhang, Jingyang Ou, Jun Hu, Jun Zhou, Yankai Lin, Ji-Rong Wen, and Chongxuan Li. 2025. Large language diffusion models. arXiv preprint arXiv:2502.09992.
  20. Jiayi Pan, Junjie Zhang, Xingyao Wang, Lifan Yuan, Hao Peng, and Alane Suhr. 2025. Tinyzero. https://github.com/Jiayi-Pan/TinyZero. Accessed: 2025-01-24. ↗
  21. Arkil Patel, Satwik Bhattamishra, and Navin Goyal. 2021. Are nlp models really able to solve simple math word problems? arXiv preprint arXiv:2103.07191.
  22. Dustin Podell, Zion English, Kyle Lacey, Andreas Blattmann, Tim Dockhorn, Jonas Müller, Joe Penna, and Robin Rombach. 2023. Sdxl: Improving latent diffusion models for high-resolution image synthesis. arXiv preprint arXiv:2307.01952.
  23. David Rein, Betty Li Hou, Asa Cooper Stickland, Jackson Petty, Richard Yuanzhe Pang, Julien Dirani, Julian Michael, and Samuel R Bowman. 2024. Gpqa: A graduate-level google-proof q&a benchmark. In First Conference on Language Modeling.
  24. Jie Ren, Yao Zhao, Tu Vu, Peter J Liu, and Balaji Lakshminarayanan. 2023. Self-evaluation improves selective generation in large language models. In Proceedings on, pages 49–64. PMLR. find
  25. Subham Sahoo, Marianne Arriola, Yair Schiff, Aaron Gokaslan, Edgar Marroquin, Justin Chiu, Alexander Rush, and Volodymyr Kuleshov. 2024. Simple and effective masked diffusion language models. Advances in Neural Information Processing Systems, 37:130136–130184.
  26. Jiaxin Shi, Kehang Han, Zhe Wang, Arnaud Doucet, and Michalis Titsias. 2024. Simplified and generalized masked diffusion for discrete data. Advances in neural information processing systems, 37:103131– 103167. find
  27. Ola Shorinwa, Zhiting Mei, Justin Lidard, Allen Z Ren, and Anirudha Majumdar. 2025. A survey on uncertainty quantification of large language models: Taxonomy, open research challenges, and future directions. ACM Computing Surveys. find
  28. Jiaming Song, Chenlin Meng, and Stefano Ermon. 2020. Denoising diffusion implicit models. arXiv preprint arXiv:2010.02502.
  29. Elias Stengel-Eskin, Peter Hase, and Mohit Bansal. 2024. Lacie: Listener-aware finetuning for calibration in large language models. Advances in Neural Information Processing Systems, 37:43080–43106. find
  30. Jaesung Tae, Hamish Ivison, Sachin Kumar, and Arman Cohan. 2025. Tess 2: A large-scale generalist diffusion language model. arXiv preprint arXiv:2502.13917.
  31. Roman Vashurin, Ekaterina Fadeeva, Artem Vazhentsev, Lyudmila Rvanova, Akim Tsvigun, Daniil Vasilev, Rui Xing, Abdelrahman Boda Sadallah, Kirill Grishchenkov, Sergey Petrakov, and 1 others. 2024. Benchmarking uncertainty quantification methods for large language models with lm-polygraph. arXiv prepr…
  32. Johnathan Xie, Annie S Chen, Yoonho Lee, Eric Mitchell, and Chelsea Finn. 2024. Calibrating language models with adaptive temperature scaling. arXiv preprint arXiv:2409.19817. find
  33. Tianyang Xu, Shujin Wu, Shizhe Diao, Xiaoze Liu, Xingyao Wang, Yangyi Chen, and Jing Gao. 2024. Sayself: Teaching llms to express confidence with self-reflective rationales. arXiv preprint arXiv:2405.20974. find
  34. Jiacheng Ye, Zhihui Xie, Lin Zheng, Jiahui Gao, Zirui Wu, Xin Jiang, Zhenguo Li, and Lingpeng Kong. 2025. Dream 7b: Diffusion large language models. arXiv preprint arXiv:2508.15487. find
  35. Jiasheng Ye, Zaixiang Zheng, Yu Bao, Lihua Qian, and Quanquan Gu. 2023. Diffusion language models can perform many tasks with scaling and instruction-finetuning. arXiv preprint arXiv:2308.12219. — [arXiv:2308.12219](https://arxiv.org/abs/2308.12219) find
  36. Runpeng Yu, Qi Li, and Xinchao Wang. 2025. Discrete diffusion in large language and multimodal models: A survey. arXiv preprint arXiv:2506.13759. — [arXiv:2506.13759](https://arxiv.org/abs/2506.13759) find
  37. Lin Zheng, Jianbo Yuan, Lei Yu, and Lingpeng Kong. 2023. A reparameterized discrete diffusion model for text generation. arXiv preprint arXiv:2302.05737. — [arXiv:2302.05737](https://arxiv.org/abs/2302.05737) find
  38. Linhao Zhong, Fan Li, Yi Huang, Jianzhuang Liu, Renjing Pei, and Fenglong Song. 2025. Outdreamer: Video outpainting with a diffusion transformer. arXiv preprint arXiv:2506.22298. — [arXiv:2506.22298](https://arxiv.org/abs/2506.22298) find
  39. Linhao Zhong, Linyu Wu, Bozhen Fang, Tianjian Feng, Chenchen Jing, Wen Wang, Jiaheng Zhang, Hao Chen, and Chunhua Shen. 2026. Beyond hard masks: Progressive token evolution for diffusion language models. arXiv preprint arXiv:2601.07351.
  40. Fengqi Zhu, Rongzhen Wang, Shen Nie, Xiaolu Zhang, Chunwei Wu, Jun Hu, Jun Zhou, Jianfei Chen, Yankai Lin, Ji-Rong Wen, and 1 others. 2025. Llada 1.5: Variance-reduced preference optimization for large language diffusion models. arXiv preprint arXiv:2505.19223.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/zhong-2026-efficient-self-evaluation-diffusion-language/