ServiceNow-AI/DNRBench
收藏Hugging Face2025-04-18 更新2025-05-31 收录
下载链接：
https://hf-mirror.com/datasets/ServiceNow-AI/DNRBench
下载链接
链接失效反馈官方服务：
资源简介：
DNR Bench数据集是一个新颖的基准测试，旨在揭示当前大型语言模型（RLM）的一个漏洞：它们在尝试解决无法解决的问题时倾向于过度推理，导致响应过长。该数据集包含150个对抗性构建的提示，分为五个不同的类别：虚构参照、无关紧要、数学问题、冗余信息和无法回答的问题。每个类别都是为了针对推理优化的大语言模型中观察到的特定失败模式而设计的。
The DNR Bench dataset is a novel benchmark designed to expose a vulnerability in current RLMs: their tendency to over-reason by attempting to solve unsolvable problems, leading to excessively long responses. The dataset contains 150 adversarially crafted prompts divided into five distinct categories: Imaginary Reference, Indifferent, Math, Redundant, and Unanswerable. Each category is designed to target a specific failure mode observed in reasoning-optimized LLMs.
提供机构：
ServiceNow-AI
