AlignmentResearch/ClearHarm

这是一个包含文本数据的分类数据集，用于区分内容是否为良性或有害。数据集包含四个配置：default、neg、pos和rep40。每个配置都有clf_label（分类标签）、instructions（指令）、content（内容）、answer_prompt（答案提示）、proxy_clf_label（代理分类标签）、gen_target（生成目标）和proxy_gen_target（代理生成目标）等特征。clf_label有两个可能的值：良性（Benign）和有害（Harmful）。数据集分为训练集和验证集，但某些配置的验证集大小为零。数据集的总大小和下载大小在README中给出。

This is a text classification dataset for distinguishing between Benign and Harmful content. The dataset consists of four configurations: default, neg, pos, and rep40. Each configuration includes features such as clf_label (classification label), instructions, content, answer_prompt, proxy_clf_label, gen_target, and proxy_gen_target. The clf_label feature has two possible values: Benign and Harmful. The dataset is split into training and validation sets, although the validation set size for some configurations is zero. The total size and download size of the dataset are provided in the README.