arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2603.01580cs.CL
收藏

马尔可夫ODE引导的评分可以评估语言模型离线推理轨迹的质量

Markovian ODE-guided scoring can assess the quality of offline reasoning traces in language models

Department of Electrical Engineering, Indian Institute of Technology Delhi(电子工程系,印度理工学院德里);Indian Institute of Technology Delhi(印度理工学院德里);Yardi School of Artificial Intelligence, Indian Institute of Technology Delhi(人工智能学院,印度理工学院德里)

Arghodeep Nandi, Ojasva Saxena, Tanmoy Chakraborty

AI总结:

MarODE通过马尔可夫模型和常微分方程评估语言模型推理轨迹质量,有效提升评估性能。

摘要:

生成语言模型产生的推理轨迹越来越多地用于从数学问题解决到自动事实核查在内的各种任务。然而,现有的评估方法仍然大多机械,无法以一种在各种变化和逐渐退化的推理中都能通用的方式捕捉人类中心的推理质量观念。我们介绍了MarODE,一个离线评估框架,它为推理轨迹分配质量评分。其有效性通过人类中心的扰动和人类判断来评估,共同评估评估指标的基本维度——良好性和正确性。该方法基于推理进展的马尔可夫式 formulation 和基于常微分方程的轨迹动态特征化,使推理质量的高效评估成为可能。在大规模评估中,MarODE在Somers' D相关性下比现有基线高出超过250%。我们的结果强调了理论驱动的评估框架的价值,随着推理轨迹成为语言模型系统的核心部分,其价值愈加凸显。

英文摘要:

Reasoning traces produced by generative language models are increasingly used for tasks ranging from mathematical problem solving to automated fact checking. However, existing evaluation methods remain largely mechanical and fail to capture human-centric notions of reasoning quality in a way that generalizes across varied and progressively degraded reasoning. We introduce MarODE, an offline evaluation framework that assigns quality scores to reasoning traces. Its effectiveness is assessed using human-centric perturbations and human judgments, which jointly evaluate the fundamental dimensions of an evaluation metric - goodness and soundness. The approach is grounded in a Markovian formulation of reasoning progression and an ordinary differential equation based characterization of trace dynamics, enabling efficient evaluation of reasoning quality. In a large-scale evaluation, MarODE outperforms existing baselines by over 250% under Somers' D correlation. Our results emphasize the value of theory-driven evaluation frameworks as reasoning traces become central to language model-based systems.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑