CompassJudger-2：通过可验证奖励迈向通用评判模型
CompassJudger-2: Towards Generalist Judge Model via Verifiable Rewards
AI总结：
提出通用评判模型 CompassJudger-2，利用可验证奖励和边际策略梯度损失增强鲁棒性，其 7B 模型媲美超大模型，并发布 JudgerBenchV2 基准以标准化评估。
摘要：
近期，LLM-as-judge 在评估大语言模型中的作用日益凸显。然而，当前的评判模型存在专业化过窄和鲁棒性有限的问题，削弱了其进行全面评估的能力。在本研究中，我们提出了 CompassJudger-2，一种新型通用评判模型，通过任务驱动的多领域数据筛选策略克服了这些局限。我们方法的核心是使用可验证奖励监督评判任务，通过拒绝采样引导内在的批判性推理，从而培养鲁棒且可泛化的评判能力。我们引入了带有边际策略梯度损失改进学习目标以增强性能。在实验上，CompassJudger-2 在多个评判和奖励基准上取得了优异结果，我们的 7B 模型展示出与 DeepSeek-V3 和 Qwen3-235B-A22B 等显著更大模型相当的评判准确度。此外，我们提出了 JudgerBenchV2，这是一个评估跨领域评判准确度和排名一致性的综合基准，旨在标准化评判模型评估。这些贡献推进了鲁棒、可扩展的 LLM 评判，并确立了新的性能和评估标准。
英文摘要：
Recently, the role of LLM-as-judge in evaluating large language models has gained prominence. However, current judge models suffer from narrow specialization and limited robustness, undermining their capacity for comprehensive evaluations. In this work, we present CompassJudger-2, a novel generalist judge model that overcomes these limitations via a task-driven, multi-domain data curation strategy. Central to our approach is supervising judgment tasks with verifiable rewards, guiding intrinsic critical reasoning through rejection sampling to foster robust, generalizable judgment capabilities. We introduce a refined learning objective with margin policy gradient loss to enhance performance. Empirically, CompassJudger-2 achieves superior results across multiple judge and reward benchmarks, and our 7B model demonstrates competitive judgment accuracy with significantly larger models like DeepSeek-V3 and Qwen3-235B-A22B. Additionally, we propose JudgerBenchV2, a comprehensive benchmark evaluating cross-domain judgment accuracy and rank consistency to standardize judge model evaluation. These contributions advance robust, scalable LLM judgment and establish new performance and evaluation standards.
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。
