haowu89/math-ai-bench-sources-middle-wo-answer

该数据集是一个用于评估和生成数学与科学问题解答的集合，包含多个配置，每个配置对应一个特定的问题来源，如AIME（美国数学邀请赛）、APEX、arXiv数学论文、CMIMC（卡内基梅隆大学数学竞赛）、GPQA钻石（研究生水平科学问题）、HMMT（哈佛-麻省理工数学竞赛）、IMObench（国际数学奥林匹克问题）、OlympiadBench（奥林匹克竞赛问题）和TheoremQA（定理证明问题）。每个数据条目包括问题文本、原始解答、答案、来源、索引、模型名称、生成解答列表、计数以及上下文元数据。上下文元数据提供了解答的正确性、链式思考令牌数、正确和错误数量、来源候选计数、来源模型列表、来源轨迹索引和总令牌数等信息。数据集仅包含训练分割，用于支持数学问题解答生成和评估的研究。

This dataset is a collection for evaluating and generating solutions to mathematical and scientific problems, comprising multiple configurations, each corresponding to a specific problem source such as AIME (American Invitational Mathematics Examination), APEX, arXiv mathematics papers, CMIMC (Carnegie Mellon University Mathematics Competition), GPQA Diamond (graduate-level science questions), HMMT (Harvard-MIT Mathematics Tournament), IMObench (International Mathematical Olympiad problems), OlympiadBench (Olympiad competition problems), and TheoremQA (theorem proving questions). Each data entry includes the problem text, original solution, answer, source, index, model name, list of generated solutions, count, and context metadata. Context metadata provides information on solution correctness, chain-of-thought token counts, numbers of correct and wrong answers, source candidate count, list of source models, source trajectory indices, and total token count. The dataset contains only a training split and is intended for research on mathematical problem solution generation and evaluation.