arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2512.23213cs.CLcs.AI
收藏

评分、推理与选择最佳!通过同行评审过程进行大型语言模型的集成

Scoring, Reasoning, and Selecting the Best! Ensembling Large Language Models via a Peer-Review Process

Beihang University, Beijing, China(北京航空航天大学);Zhongguancun Laboratory, Beijing, China(中关村实验室);Beijing University of Posts and Telecommunications(北京邮电大学);Hong Kong University of Science and Technology(香港科学与技术大学);Xi'an Jiaotong University, Xi'an, China(西安交通大学);Tsinghua University, Beijing, China(清华大学);Singapore University of Technology and Design(新加坡科技设计大学)
Zhijun Chen, Zeyu Ji, Qianren Mao, Hao Wu, Jinhuan Song, Junhang Cheng, Bangjie Qin, Zhuoran Li, Jingzheng Li, Kai Sun, Zizhe Wang, Yikun Ban, Zhu Sun, Xiangyan… 展开作者

Zhijun Chen, Zeyu Ji, Qianren Mao, Hao Wu, Jinhuan Song, Junhang Cheng, Bangjie Qin, Zhuoran Li, Jingzheng Li, Kai Sun, Zizhe Wang, Yikun Ban, Zhu Sun, Xiangyang Ji, Hailong Sun, Xiao Huang

AI总结:

本文提出LLM-PeerReview方法,通过同行评审机制集成多个大型语言模型,提升响应质量。实验显示其在多个数据集上优于Smoothie-Global,适用于事实性问答、数学推理等任务。

摘要:

我们提出LLM-PeerReview,一种无监督的大型语言模型集成方法,通过多个模型生成的候选响应选择最优响应。该方法基于新颖的同行评审框架,提供透明和可解释的机制,同时保持无监督特性以实现灵活适应和泛化。具体分为三个阶段:评分阶段使用LLM-as-a-Judge技术评估每个响应;推理阶段通过简单平均策略或基于图模型的真理推断算法聚合分数;最终选择得分最高的响应作为集成输出。LLM-PeerReview概念简单且实证效果显著。在四个数据集上的结果表明,所提出方法的两个变体在多个任务类型上优于Smoothie-Global,提升了6.9%和7.3个百分点。

英文摘要:

We propose LLM-PeerReview, an unsupervised LLM Ensemble method that selects the most ideal response from multiple LLM-generated candidates for each query, harnessing the collective wisdom of multiple models with diverse strengths. LLM-PeerReview is built on a novel, peer-review-inspired framework that offers a transparent and interpretable mechanism, while remaining fully unsupervised for flexible adaptability and generalization. Specifically, it operates in three stages: For scoring, we use the emerging LLM-as-a-Judge technique to evaluate each response by reusing multiple LLMs at hand; For reasoning, we can apply a straightforward averaging strategy or a principled graphical model-based truth inference algorithm to aggregate multiple scores to produce a final score for each response; Finally, the highest-scoring response is selected as the best ensemble output. LLM-PeerReview is conceptually simple and empirically powerful. Our results across four datasets show that the two variants of the proposed approach outperform the advanced model Smoothie-Global by 6.9% and 7.3% points, cross diverse task types including factual recall QA, math reasoning, and instruction following. Notably, we also establish a carefully curated benchmark suite for LLM Ensemble, integrating 12 methods across four classic datasets and three task families, all evaluated under a rigorous and consistent protocol. We hope this repository will help researchers reproduce the LLM Ensemble baselines.

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。

↑