Greygre1n/WMDP

WMDP（大规模杀伤性武器代理）基准是一个多项选择题数据集，用作生物安全、网络安全和化学安全领域危险知识的代理测量。该数据集有两个主要作用：一是评估大语言模型中的危险知识，二是作为遗忘方法移除此类危险知识的基准。具体包括三个子集：WMDP-Bio包含1,273个多项选择题，WMDP-Cyber包含1,987个多项选择题，WMDP-Chem包含408个多项选择题。数据集特征包括答案（整数类型）、问题（字符串类型）和选项（字符串序列）。数据集分割为bio、chem和cyber三个部分，每个部分都有相应的字节数和示例数量。The Weapons of Mass Destruction Proxy (WMDP) benchmark is a dataset of multiple-choice questions that serve as a proxy measurement of hazardous knowledge in biosecurity, cybersecurity, and chemical security. WMDP serves two roles: first, as an evaluation for hazardous knowledge in LLMs, and second, as a benchmark for unlearning methods to remove such hazardous knowledge. WMDP includes: a collection of 1,273 multiple-choice questions in WMDP-Bio, a collection of 1,987 multiple-choice questions in WMDP-Cyber, and a collection of 408 multiple-choice questions in WMDP-Chem. The dataset features include answer (int64), question (string), and choices (sequence of strings). It is split into bio, chem, and cyber sections, each with specific byte sizes and example counts.