PJMixers-Dev/proxy-logs-ReRolls-Minos

该数据集包含prompt和response的交互数据，主要用于训练和评估语言模型。数据集的特征包括prompt的哈希值、prompt的来源和内容、多个response的详细信息（如数据集名称、端点、minos置信度、模型、响应内容等），以及响应的统计信息（如平均token数、slop比率等）。数据集分为train分割，包含274,610个示例，其中non-refusal responses有845,186个，refusal responses有38,285个。

This dataset contains interaction data between prompts and responses, primarily used for training and evaluating language models. The features of the dataset include prompt hashes, prompt sources and values, detailed information about multiple responses (such as dataset name, endpoint, minos confidence, model, response content, etc.), and response statistics (such as average token count, slop ratio, etc.). The dataset is divided into a train split with 274,610 examples, including 845,186 non-refusal responses and 38,285 refusal responses.