ibm-research/AttaQ-JA

AttaQ-JA数据集是一个评估大型语言模型（LLM）在生成有害或不期望回应方面的能力的日文数据集。它由1402个精心设计的对抗性问题组成，是AttaQ数据集的日文版本，通过人工翻译和精心制作而成。由于数据集内容可能具有攻击性，阅读时需要根据个人风险承受能力进行。

The AttaQ-JA dataset is a Japanese version of the AttaQ dataset, designed to evaluate the tendency of Large Language Models (LLMs) to generate harmful or undesirable responses. It consists of 1402 carefully crafted adversarial questions. The dataset contains content that may be offensive and should be read with consideration to personal risk tolerance.