stochastic-parrots/rm-bench-lighteval

该数据集包含多个字段，如提示文本、选中选项、拒绝选项、领域、处理后的领域、难度、选中样式、拒绝样式等。数据集分为测试、聊天、代码、数学、安全拒绝、安全响应、正常、困难、简单等不同类型，每种类型又进一步细分为正常、困难、简单三种难度级别。数据集适用于多种场景，包括文本生成、安全响应生成等。

The dataset includes multiple fields such as prompt text, chosen option, rejected option, domain, sanitized domain, difficulty, chosen style, rejected style, etc. The dataset is divided into different types such as test, chat, code, math, safety refuse, safety response, normal, hard, easy, and each type is further subdivided into three difficulty levels: normal, hard, and easy. The dataset is suitable for various scenarios, including text generation, safety response generation, etc.