减少大模型推理中无效自检，提升效率不降精度
Self-Verification Dilemma: Experience-Driven Suppression of Overused Checking in LLM Reasoning
- 基于历史验证经验，动态判断是否需要重复检查
- 在多个模型和数据集上降低20.3%的令牌消耗，部分场景准确率还提升
- 适合追求高效推理、关注生成成本的研究者与应用开发者
大型推理模型通过生成带反思的长推理链实现优异表现。大规模实证分析发现，大量反思步骤为自我验证（重查），反复确认中间结果。这些重查在多个模型和基准上频繁出现，但绝大多数为确认性而非纠错性，极少发现错误或改变推理结果。这暴露出自检激活频率与其实际效用之间的错配。为此，我们提出一种新颖的经验驱动测试时框架，抑制过度验证。该方法检测重查行为，查询离线的历史验证结果池，通过高效检索估算重查是否必要；若历史经验表明无需重查，则发出抑制信号，引导模型继续推进。在多个模型和基准上，该方法将令牌使用量最多减少20.3%，同时保持准确率，部分数据集甚至提升准确率。
原文摘要 · Abstract (English)
Large Reasoning Models (LRMs) achieve strong performance by generating long reasoning traces with reflection. Through a large-scale empirical analysis, we find that a substantial fraction of reflective steps consist of self-verification (recheck) that repeatedly confirm intermediate results. These rechecks occur frequently across models and benchmarks, yet the vast majority are confirmatory rather than corrective, rarely identifying errors and altering reasoning outcomes. This reveals a mismatch between how often self-verification is activated and how often it is actually useful. Motivated by this, we propose a novel, experience-driven test-time framework that reduces the overused verification. Our method detects the activation of recheck behavior, consults an offline experience pool of past verification outcomes, and estimates whether a recheck is likely unnecessary via efficient retrieval. When historical experience suggests unnecessary, a suppression signal redirects the model to proceed. Across multiple model and benchmarks, our approach reduces token usage up to 20.3% while maintaining the accuracy, and in some datasets even yields accuracy improvements.
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。