HiTZ/bbq

BBQ数据集是一个用于评估英语问答任务中语言模型社会偏见的基准数据集。它包含针对9个人口统计学维度的社会偏见测试问题，每个问题都有模糊和去模糊两种版本。这些维度包括年龄、残疾状况、性别认同、国籍、外貌、种族和民族、宗教、社会经济状况和性取向。数据集分为测试集和训练集，其中测试集是主要的评估数据，训练集包含用于少量样本学习的4个示例（如果可用）。

The BBQ dataset is a benchmark for evaluating social biases in language models on English question-answering tasks. It contains questions designed to test for social biases across nine demographic dimensions, with each question available in both ambiguous and disambiguated versions. These dimensions include Age, Disability_status, Gender_identity, Nationality, Physical_appearance, Race_ethnicity, Religion, SES, and Sexual_orientation. The dataset is split into a test set, which is the main evaluation data, and a train set, which includes 4-shot examples for few-shot learning (when available).