适可而止：思维链推理的自适应早停机制
Stop When Enough: Adaptive Early-Stopping for Chain-of-Thought Reasoning
AI总结：
本文提出无需训练的框架REFRAIN，通过两阶段停止判别器和SW-UCB控制器自适应决定推理停止时机，有效缓解LLM过度思考，在保持准确率的同时减少20-55%的token使用。
摘要：
思维链（Chain-of-Thought, CoT）推理通过外化中间步骤，推动了大型语言模型（LLMs）在推理密集型任务上的近期进展。然而，过度或冗余的推理（即所谓的过度思考）会增加推理成本，并导致LLMs得出错误结论。本文提出了REFRAIN（用于自适应推理的反思-冗余框架），这是一个无需训练的框架，能够自适应地决定何时停止推理以缓解过度思考。REFRAIN集成了一个两阶段停止判别器，用于识别具有反思性但冗余的推理；以及一个滑动窗口上置信界（SW-UCB）多臂老虎机控制器，用于在无监督或微调的情况下根据问题难度动态调整停止阈值。在四个代表性基准和两个模型家族上，与标准CoT提示相比，REFRAIN在保持或提高准确率的同时，将token使用量减少了20-55%。广泛的消融和鲁棒性分析证明了其在不同模型、评分器和提示变化下的稳定性。总之，我们的发现强调了“何时停止”作为测试时缩放的一个全新且实用的轴——使模型不仅能推理得更多，而且恰到好处。
英文摘要：
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.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
