让大模型学会适时停止推理，避免浪费算力还出错。
Stop When Enough: Adaptive Early-Stopping for Chain-of-Thought Reasoning
- 用两阶段判别器识别冗余思考，结合动态阈值控制停止时机。
- 在四个基准上减少20%-55%的token消耗，准确率不降反升。
- 无需训练或微调，适配不同模型和提示，实用性强。
链式思维（CoT）通过显式化中间推理步骤，推动大语言模型在推理密集型任务上的表现提升。然而，过度或冗余的推理——即“过度思考”——会增加推理开销，并可能导致错误结论。本文提出REFRAIN（反射性冗余自适应推理），一种无需训练的框架，可自适应决定何时停止推理以缓解过度思考。REFRAIN融合了两阶段停止判别器以识别具有反思性但冗余的推理过程，以及滑动窗口上置信度边界（SW-UCB）多臂赌博机控制器，根据问题难度动态调整停止阈值，且无需监督或微调。在四个代表性基准和两种模型家族上，REFRAIN将token使用量降低20%-55%，同时保持或提升准确率。大量消融实验与鲁棒性分析表明其在不同模型、评分器及提示变体下均稳定有效。综上，我们的研究揭示‘何时停止’是测试时扩展的一个新且实用维度——使模型不仅能推理更多，更能恰到好处。
原文摘要 · Abstract (English)
Chain-of-Thought (CoT) reasoning has driven recent gains of large language models (LLMs) on reasoning-intensive tasks by externalizing intermediate steps. However, excessive or redundant reasoning -- so-called overthinking -- can increase inference costs and lead LLMs toward incorrect conclusions. In this paper, we present REFRAIN ($\underline{REF}$lective-$\underline{R}$edundancy for $\underline{A}$daptive $\underline{IN}$ference), a training-free framework that adaptively determines when to stop reasoning to mitigate overthinking. REFRAIN integrates a two-stage stop discriminator to identify reflective yet redundant reasoning and a sliding-window Upper Confidence Bound (SW-UCB) multi-armed bandit controller to dynamically adjust stopping thresholds according to problem difficulty without supervision or fine-tuning. Across four representative benchmarks and two model families, REFRAIN reduces token usage by 20-55% while maintaining or improving accuracy compared to standard CoT prompting. Extensive ablation and robustness analyses demonstrate its stability across models, scorers, and prompt variations. In summary, our findings highlight when-to-stop as a new and practical axis of test-time scaling -- enabling models to reason not just more, but just enough.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。