让推理模型自动判断何时停止思考，避免无效计算。
Stop When Further Reasoning Won't Help: Attention-State Adaptive Generation in Reasoning Models
- 通过分析注意力分布判断推理状态，动态调整生成策略。
- 在Qwen3-8B上提升准确率3.2%，生成tokens减少近40%。
- 无需训练、即插即用，适配多种大模型和任务。
通过引入测试时计算扩展，大型推理模型（LRMs）可借助显式的思维链（CoT）过程解决复杂问题。然而，它们常出现过度思考现象，导致冗余文本输出并降低准确性。现有缓解方法受限：基于训练的方法需大量算力，而无训练方法依赖精心设计的提示或不可靠的置信度信号。本文从注意力分布角度研究早期停止，提出一种简单方法ASAG，通过推断模型推理状态并自适应调整生成策略。该框架无需训练且可即插即用，可无缝集成至现有LRMs中。在九个基准上的实验表明，其在不同参数规模的主流模型（包括DeepSeek-R1-Distill和Qwen3系列）上均取得一致提升。具体而言，ASAG在所有推理任务上使Qwen3-8B的平均准确率提升3.2%，生成的token数量减少近40%。
原文摘要 · Abstract (English)
By incorporating test-time compute scaling, large reasoning models (LRMs) can solve complex problems through explicit chain-of-thought (CoT) reasoning processes. However, they often suffer from overthinking, resulting in redundant token outputs and degraded accuracy. Current methods to mitigate this issue remain limited: training-based approaches require substantial computational resources, while training-free methods rely on well-crafted prompts or unreliable confidence signals. In this work, we investigate early stopping from the perspective of attention distributions and propose a simple method, ASAG, which infers the model's reasoning state and adaptively adjusts the generation strategy. The proposed framework is training-free and plug-and-play, enabling seamless integration into existing LRMs. Extensive experiments on nine benchmarks demonstrate consistent improvements across mainstream LRMs with varying parameter scales, including the DeepSeek-R1-Distill and Qwen3 series. Specifically, ASAG improves average accuracy by 3.2% while reducing the number of generated tokens by nearly 40% across all reasoning tasks on Qwen3-8B.
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。