无需标注数据，用逻辑规则提升大模型推理能力
Logic-Regularized Verifier Elicits Reasoning from LLMs
- 通过逻辑约束设计无监督验证器，利用模型内部激活
- 在10个数据集上达到有监督验证器95%的平均性能
- 适合希望低成本提升模型推理能力的研究者
验证器是提升现代大语言模型推理能力的关键组件。传统验证器依赖昂贵且多样性的标注数据构建，存在资源消耗大、数据多样性不足的问题。本文提出无监督验证器LOVER，基于逻辑规则进行正则化。LOVER将验证器视为二值隐变量，利用模型内部激活，并对多条推理路径施加三种逻辑约束：否定一致性、组内一致性和组间一致性（按最终答案分组）。通过引入逻辑规则作为先验知识，LOVER可有效利用未标注样本，且与任意现成大模型兼容。在10个数据集上的实验表明，LOVER显著优于无监督基线，在平均性能上达到有监督验证器的95%水平。代码已公开于https://github.com/wangxinyufighting/llm-lover。
原文摘要 · Abstract (English)
Verifiers are crucial components for enhancing modern LLMs' reasoning capability. Typicalverifiers require resource-intensive superviseddataset construction, which is costly and faceslimitations in data diversity. In this paper, wepropose LOVER, an unsupervised verifier regularized by logical rules. LOVER treats theverifier as a binary latent variable, utilizinginternal activations and enforcing three logical constraints on multiple reasoning paths:negation consistency, intra-group consistency,and inter-group consistency (grouped by thefinal answer). By incorporating logical rulesas priors, LOVER can leverage unlabeled examples and is directly compatible with any offthe-shelf LLMs. Experiments on 10 datasetsdemonstrate that LOVER significantly outperforms unsupervised baselines, achieving performance comparable to the supervised verifier(reaching its 95% level on average). The sourcecode is publicly available at https://github.com/wangxinyufighting/llm-lover.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。