openai/healthbench-professional

包含用于HealthBench Professional评估的数据。每个示例包含：对话列表（用户/助手消息，以用户消息结尾）、评分项列表（每个包含criterion_text和points）、用例类型（consult、writing或research）、类型（good_faith或red_teaming）、难度等级（由医生评定的difficult或typical）、专业领域（医学专业或子专业）以及医生写的回复。数据集要求不公开示例内容以防止数据污染或作弊。

Contains the data for the HealthBench Professional eval. Each example contains: conversation (list of user / assistant messages, ending in a user message), rubric_items (list of rubric items, each containing criterion_text and points), use_case (one of consult, writing, or research), type (one of good_faith or red_teaming), difficulty (physician-assigned difficulty rating), specialty (medical specialty or sub-specialty), and physician_response (response written by a physician). The dataset requests not to reveal examples to prevent contamination or cheating.