基于方面成对锦标赛的测试时选择方法
原标题：Aspect-wise Pairwise Tournaments for Test-Time Selection
AI 摘要
该论文提出 Aspect-wise Pairwise Tournaments（APT），一种无需训练的测试时选择方法：从生成器 LLM 采样 n 个候选解，通过单败淘汰赛两两比较，每场比赛由多个现成的方面验证器按各自维度投票决定胜者。在五个领域、三个生成器 LLM 上，APT 在 15 组生成器-领域组合中的 14 组优于自一致性、奖励模型验证和二元多维度批准方法，在开放式任务上提升最大。由于方面以自然语言指定，用户可在推理时添加并加权方面来引导选择，无需重新生成候选或训练验证器。
正文节选
Aspect-wise Pairwise Tournaments for Test-Time Selection Abstract Scaling test-time compute has proven highly effective at improving LLM reasoning. A popular method is best-of-\(n\) sampling, where a model generates many candidate solutions and one is selected. This motivates the question: how can we select the best candidate? When humans make choices, we find it more natural to compare two options than judge each in isolation. Additionally, when the options become complex, we like to judge the