让大模型提前停止冗余推理，提速3.7倍还保持准确
ESTAR: Early-Stopping Token-Aware Reasoning For Efficient Inference
- 用轨迹分类器判断何时可安全停止推理
- 推理长度从4799降到1290，准确率仅微降0.7%
- 适合追求高效推理的AI研究者和部署工程师
大型推理模型（LRMs）通过生成长链条思维实现顶尖性能，但常在得出正确答案后仍进行冗余计算。本文提出早期停止的词元感知推理方法（ESTAR），通过检测并减少此类冗余以提升效率而不损失精度。该方法结合三部分：(i) 基于轨迹的分类器识别可安全停止的时机；(ii) 监督微调使模型学会自动生成<stop>信号；(iii) <stop>-感知强化学习，在自生成停止点截断推理过程，并采用计算感知奖励。在四个推理数据集上的实验表明，ESTAR将推理长度平均缩短3.7倍（从4,799降至1,290），同时保持准确率（74.9%对比74.2%），且具备强跨领域泛化能力。结果表明，早期停止是提升大型推理模型效率的简单而有效机制。
原文摘要 · Abstract (English)
Large reasoning models (LRMs) achieve state-of-the-art performance by generating long chains-of-thought, but often waste computation on redundant reasoning after the correct answer has already been reached. We introduce Early-Stopping for Token-Aware Reasoning (ESTAR), which detects and reduces such reasoning redundancy to improve efficiency without sacrificing accuracy. Our method combines (i) a trajectory-based classifier that identifies when reasoning can be safely stopped, (ii) supervised fine-tuning to teach LRMs to propose self-generated <stop> signals, and (iii) <stop>-aware reinforcement learning that truncates rollouts at self-generated stop points with compute-aware rewards. Experiments on four reasoning datasets show that ESTAR reduces reasoning length by about 3.7x (from 4,799 to 1,290) while preserving accuracy (74.9% vs. 74.2%), with strong cross-domain generalization. These results highlight early stopping as a simple yet powerful mechanism for improving reasoning efficiency in LRMs.
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。