brendel-group/MATH-Beyond
收藏Hugging Face2025-10-14 更新2025-11-01 收录
下载链接:
https://hf-mirror.com/datasets/brendel-group/MATH-Beyond
下载链接
链接失效反馈官方服务:
资源简介:
MATH-Beyond是一个数学问题基准数据集,旨在评估强化学习方法在解决超出基础模型能力的挑战性数学问题上的表现。包含181个数学问题,这些问题设计用来击败参数量达8B的常见开源模型,即使在大量采样预算下(如pass@1024)。数据集的问题来源于DAPO-Math-17K和DeepScaleR数据集的子集,并包含21个基础模型的未解决问题子集。
MATH-Beyond is a benchmark dataset for mathematics problems designed to evaluate reinforcement learning methods on challenging problems that exceed the capabilities of base models. It contains 181 mathematical problems sourced from subsets of the DAPO-Math-17K and DeepScaleR datasets, and includes unsolved subsets for 21 base models.
提供机构:
brendel-group



