[论文] OmniVerifier-C1: Decomposed Graded Verifier for Rule-Governed Generati...

论文概要

研究领域: NLP
作者: Bowei Liu, Xinchen Zhang, Xuhuan Li, Tianlin Zhang, Kaian Jiang, Jingwei Liu, Congyang Zhao, Yifei Qian, Junhong Liu, Zhiheng Li, Yujiu Yang, Yang Cai, Ling Yang
发布时间: 2026-10-15
arXiv: 2610.22355

中文摘要

生成式视觉验证器在推进多模态智能方面发挥着关键作用。为增强视觉验证器在涉及组合需求的复杂多模态场景中的鲁棒性，我们超越传统的仅产生 True/False 离散判断的二元验证器，引入一种能自动将复杂提示或环境规则分解为细粒度组件并分配逐点分数的验证器，提供连续且更具信息量的评判信号。我们引入解耦 RL 训练策略，联合优化子问题级二元判断和全局对齐评分。实验表明，所提出的逐点评分验证器在传统文本到图像场景中显著优于二元验证器。从理论上表明当有真值图像可用时，成对验证范式比逐点表述产生更准确的评判。我们的发现揭示了一个关键洞察：从生成模型到结构化世界建模智能体的转变，关键取决于准确的成对评分视觉验证器的可用性。

原文摘要

Generative visual verifiers play a crucial role in advancing multimodal intelligence. To enhance the robustness of visual verifiers in complex multimodal scenarios involving compositional requirements, we move beyond conventional binary verifiers that produce only True/False discrete judgments and introduce a verifier that automatically decomposes complex prompts or environmental rules into fine-grained components and assigns pointwise scores to provide a continuous and more informative critique signal. Our experiments further demonstrate that these structured and interpretable outputs improve the verifier's generalization to out-of-domain multimodal critic tasks. To train this verifier, we introduce a Decoupled RL training strategy that jointly optimizes sub-question-level binary judgment...

自动采集于 2026-10-17

#论文 #arXiv #NLP #小凯