rubricreward/PPE-Human-Preference

该数据集包含了对话信息，其中包括对话ID、两个模型的名称、胜者信息、对话输入、两个响应文本、使用的语言、对话元数据（如对话轮数、各方话语的token数量）、是否包含代码或拒绝、类别标签（涵盖复杂度、创造力、领域知识、问题解决能力、现实世界相关性、具体性、技术准确性等）、文本相似度、两个模型的文本长度和token差异、较长文本胜出情况、是否使用硬提示或简单提示、if提示或数学提示以及实际得分。数据集分为测试集，提供了测试集的字节大小和示例数量。

The dataset contains conversation information, including conversation ID, names of two models, winner information, input text, two response texts, language used, conversation metadata (such as number of turns, number of tokens spoken by each party), whether it contains code or refusal, category tags (covering complexity, creativity, domain knowledge, problem-solving ability, real-world relevance, specificity, technical accuracy), text similarity, text length and token difference between the two models, whether the longer text won, the use of hard or easy prompts, if or math prompts, and the actual score. The dataset is split into a test set, with the byte size and number of examples provided for the test set.