PulseAugur
实时 11:17:42
English(EN) When Does Learning to Stop Help? A Cost-Aware Study of Early Exits in Reasoning Models

新研究探讨推理语言模型的学习停止规则

一篇新研究论文探讨了推理语言模型的学习停止规则的有效性,并介绍了一种名为LearnStop的方法。该技术分析各种在线特征,如答案置信度、熵和前缀稳定性,以在固定的计算预算下预测正确性。研究发现,学习停止主要在自由形式的数学任务中提供优势,提高了性能,优于简单的标量退出。然而,对于多项选择题或非常困难的任务,传统的标量置信度或收敛规则仍然具有竞争力或更优,这表明学习停止的价值是任务依赖的。 AI

影响 这项研究可能通过使推理模型在有信心预测正确答案时停止处理,从而更有效地利用计算资源。

排序理由 该集群包含一篇详细介绍推理模型新方法的学术论文。[lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.CL 阅读 →

AI 生成摘要 · Google Gemini · 来自 2 个来源。 我们如何撰写摘要 →

新研究探讨推理语言模型的学习停止规则

报道来源 [2]

  1. arXiv cs.AI TIER_1 English(EN) · Zhe Dong (University of Maine at Presque Isle), Fang Qin (Stanford University), Manish Shah (Independent Researcher) ·

    When Does Learning to Stop Help? A Cost-Aware Study of Early Exits in Reasoning Models

    arXiv:2606.30852v1 Announce Type: new Abstract: Reasoning models spend different amounts of useful computation across instances, but it remains unclear when a learned stopping rule improves over simple confidence or convergence thresholds. We study this question with LearnStop, a…

  2. arXiv cs.CL TIER_1 English(EN) · Manish Shah ·

    When Does Learning to Stop Help? A Cost-Aware Study of Early Exits in Reasoning Models

    Reasoning models spend different amounts of useful computation across instances, but it remains unclear when a learned stopping rule improves over simple confidence or convergence thresholds. We study this question with LearnStop, a hidden-state-free checkpoint stopper for reason…