AlignmentResearch/NestedCiphers

该数据集包含了多种配置的训练和验证数据，每个配置都有八个字段：completion, instructions, answer_prompt, content, clf_label, proxy_clf_label, gen_target, 和 proxy_gen_target。clf_label和proxy_clf_label字段用于分类，分为Benign（良性）和Harmful（有害）两类。每个配置的数据量和示例数量各不相同。The dataset consists of multiple configurations, each including training and validation sets. Each configuration has eight fields: completion, instructions, answer_prompt, content, clf_label, proxy_clf_label, gen_target, and proxy_gen_target. The clf_label and proxy_clf_label fields are used for classification, divided into Benign and Harmful categories. Each configuration has a different amount of data and number of examples.