iNeil77/PPE-Human-Preference-V1-EN

该数据集包含多个字段，用于记录两个模型（model_a和model_b）对同一提示词（prompt）的响应，并标注了胜者（winner）。还包括语言、对话元数据（如token数、轮次）、是否涉及代码、是否拒绝回答、类别标签（如复杂度、创造性、领域知识、问题解决、现实世界、具体性、技术准确性，以及指令遵循、数学相关标签）、相似度得分、响应长度、token差异、是否更长响应获胜、以及硬提示、简单提示、指令遵循提示、数学提示等布尔标记。数据集主要用于比较模型回答质量，可能用于训练奖励模型或偏好模型。This dataset contains fields for recording responses from two models (model_a and model_b) to a given prompt, with a winner annotation. It includes language, conversation metadata (e.g., token counts, turns), code/refusal flags, category tags (e.g., complexity, creativity, domain knowledge, problem solving, real world, specificity, technical accuracy, instruction following, math), similarity score, response lengths, token difference, whether longer response won, and boolean flags for hard, easy, instruction following, and math prompts. The dataset is primarily used for comparing model response quality, likely for training reward models or preference models.