lmarena-ai/arena-human-preference-140k

该数据集包含用户在文本类别中对两个模型（model_a和model_b）的投票，每个条目代表对一次用户对话的评估，包括完整的对话历史和元数据。数据集字段包括唯一的反馈ID，评估会话ID，评估顺序，获胜模型信息，两个模型的对话内容，完整的对话历史，对话元数据，分类标签，是否包含代码信息，以及对话的风格控制标记和统计信息。

This dataset contains user votes on two models (model_a and model_b) in the text-only category. Each entry represents an evaluation of a user conversation, including the full conversation history and metadata. The dataset fields include a unique feedback ID, evaluation session ID, evaluation order, winner information, conversation content for both models, full conversation history, conversation metadata, category tags, code involvement information, and style control tokens and statistics.