ACL Findings2025
Self-Training Elicits Concise Reasoning in Large Language Models
Tergel Munkhbat, Namgyu Ho, Seo Hyun Kim, Yongjin Yang, Yujin Kim, Se-Young Yun
Abstract
Chain-of-thought (CoT) reasoning has enabled large language models (LLMs) to utilize additional computation through intermediate tokens to solve complex tasks. However, we posit that typical reasoning traces contain many redundant tokens, incurring extraneous inference costs. Upon examination of the output distribution of current LLMs, we find evidence on their latent ability to reason more concisely, relative to their default behavior. To elicit this capability, we propose simple fine-tuning methods which leverage self-generated concise reasoning paths obtained by best-of-N sampling and few-shot conditioning, in task-specific settings. Our combined method achieves a 30% reduction in output tokens on average, across five model families on GSM8K and MATH, while maintaining average accuracy. By exploiting the fundamental stochasticity and in-context learning capabilities of LLMs, our self-training approach robustly elicits concise reasoning on a wide range of models, including those with extensive post-training.
Ask Lune
Know this paper in minutes, not an afternoon.
Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.
46 top-tier papers have cited it since. Know what they found before your reviewers do.
Lune can draw on
the full paper · the top-tier papers that cite it · related work
Free to start. No credit card required.
Lineage
- Let LRMs Break Free from Overthinking via Self-Braking Tuning cites Dynamic Early Exit in Reasoning Models
- Let LRMs Break Free from Overthinking via Self-Braking Tuning cites Let's Verify Step by Step
- S-GRPO: Early Exit via Reinforcement Learning in Reasoning Models cites Dynamic Early Exit in Reasoning Models
- Dynamic Early Exit in Reasoning Models cites S-GRPO: Early Exit via Reinforcement Learning in Reasoning Models
- S-GRPO: Early Exit via Reinforcement Learning in Reasoning Models cites Let's Verify Step by Step
- Efficiently Scaling LLM Reasoning Programs with Certaindex cites Dynamic Early Exit in Reasoning Models
- Efficiently Scaling LLM Reasoning Programs with Certaindex cites Let's Verify Step by Step
- Efficiently Scaling LLM Reasoning Programs with Certaindex cites Chain-of-Thought Prompting Elicits Reasoning in Large Language Models
- Dynamic Early Exit in Reasoning Models cites Learn to Reason Efficiently with Adaptive Length-based Reward Shaping
- Dynamic Early Exit in Reasoning Models cites Chain-of-Thought Prompting Elicits Reasoning in Large Language Models
- Large Language Models are Zero-Shot Reasoners cites Language Models are Few-Shot Learners
- Large Language Models are Zero-Shot Reasoners cites Chain-of-Thought Prompting Elicits Reasoning in Large Language Models
- Let's Verify Step by Step cites Large Language Models are Zero-Shot Reasoners
- Let's Verify Step by Step cites Chain-of-Thought Prompting Elicits Reasoning in Large Language Models
- Language Models are Few-Shot Learners cites The Curious Case of Neural Text Degeneration
Cited by top-tier papers46
- Dynamic Early Exit in Reasoning ModelsChenxu Yang, Qingyi Si, Yongjie Duan, Zheliang Zhu et al.ICLR 2026 · 250 citations
- S-GRPO: Early Exit via Reinforcement Learning in Reasoning ModelsMuzhi Dai, Chenxu Yang, Qingyi SiNeurIPS 2025 · 100 citations
- Learn to Reason Efficiently with Adaptive Length-based Reward ShapingWei Liu, Ruochen Zhou, Yiyun Deng, Yuzhen Huang et al.ICLR 2026 · 88 citations
- Efficiently Scaling LLM Reasoning Programs with CertaindexYichao Fu, Junda Chen, Siqi Zhu, Zheyu Fu et al.NeurIPS 2025 · 42 citations
- Let LRMs Break Free from Overthinking via Self-Braking TuningHaoran Zhao, Yuchen Yan, Yongliang Shen, Haolei Xu et al.NeurIPS 2025 · 31 citations
Builds on22
- Language Models are Few-Shot LearnersTom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah et al.NeurIPS 2020 · 64,255 citations
- Chain-of-Thought Prompting Elicits Reasoning in Large Language ModelsJason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma et al.NeurIPS 2022 · 22,562 citations
- Large Language Models are Zero-Shot ReasonersTakeshi Kojima, Shixiang Shane Gu, Machel Reid, Yutaka Matsuo et al.NeurIPS 2022 · 8,168 citations
- The Curious Case of Neural Text DegenerationAri Holtzman, Jan Buys, Li Du, Maxwell Forbes et al.ICLR 2020 · 4,112 citations
- Let's Verify Step by StepHunter Lightman, Vineet Kosaraju, Yuri Burda, Harrison Edwards et al.ICLR 2024 · 3,045 citations
Related papers
- Activation Control for Efficiently Eliciting Long Chain-of-thought Ability of Language ModelsZekai Zhao, Qi Liu, Kun Zhou, Zihan Liu et al.NeurIPS 2025 · 10 citations
- Correct, Concise and Complete: Multi-stage Training For Adaptive ReasoningNathanaël Carraz Rakotonirina, Ren Pang, Neha Anna John, Michael Bohlke-Schneider et al.ACL Findings 2026
- CODI: Compressing Chain-of-Thought into Continuous Space via Self-DistillationZhenyi Shen, Hanqi Yan, Linhai Zhang, Zhanghao Hu et al.EMNLP 2025
- System-1.5 Reasoning: Traversal in Language and Latent Spaces with Dynamic ShortcutsXiaoqiang Wang, Suyuchen Wang, Yun Zhu, Bang LiuNeurIPS 2025 · 26 citations
- Fewer is More: Boosting Math Reasoning with Reinforced Context PruningXijie Huang, Li Lyna Zhang, Kwang-Ting Cheng, Fan Yang et al.EMNLP 2024 · 7 citations
