通过监测神经元激活动态，实现大模型推理的无训练早停。
NEAT: Neuron-Based Early Exit for Large Reasoning Models
- 基于神经元激活模式识别可退出节点，无需额外训练。
- 在四个基准上平均减少22%~28%的计算token，保持准确率。
- 适合追求推理效率但不愿改模型结构的研究者。
大型推理模型常出现冗余推理（过思）现象，即正确答案已得出后仍生成多余步骤。现有早停方法多依赖输出层面启发式或训练探针模型，需额外推理计算或外部标注数据。本文提出NEAT——一种基于神经元的早期推理退出框架，通过监控推理过程中的神经元激活动态，实现无需训练的早停，不增加测试时计算开销。NEAT识别与退出相关的神经元，跟踪其激活模式，动态触发早停或抑制反思，从而减少无效推理并保持解题质量。在六个不同规模和架构的模型上，于四个推理基准上的实验表明，每种模型在四基准上平均减少22%至28%的token消耗，同时维持原有准确率。
原文摘要 · Abstract (English)
Large Reasoning Models (LRMs) often suffer from \emph{overthinking}, a phenomenon in which redundant reasoning steps are generated after a correct solution has already been reached. Existing early reasoning exit methods primarily rely on output-level heuristics or trained probing models to skip redundant reasoning steps, thereby mitigating overthinking. However, these approaches typically require additional rollout computation or externally labeled datasets. In this paper, we propose \textbf{NEAT}, a \textbf{N}euron-based \textbf{E}arly re\textbf{A}soning exi\textbf{T} framework that monitors neuron-level activation dynamics to enable training-free early exits, without introducing additional test-time computation. NEAT identifies exit-associated neurons and tracks their activation patterns during reasoning to dynamically trigger early exit or suppress reflection, thereby reducing unnecessary reasoning while preserving solution quality. Experiments on four reasoning benchmarks across six models with different scales and architectures show that, for each model, NEAT achieves an average token reduction of 22\% to 28\% when averaged over the four benchmarks, while maintaining accuracy.
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。