WE-MATH
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/We-Math/We-Math
下载链接
链接失效反馈官方服务：
资源简介：
该数据集名为WE-MATH，包含了6,500个视觉数学问题，这些问题被细分为67个分层的知识概念，旨在评估大型多模态模型（LMMs）在视觉数学推理中的问题解决原则。此外，数据集还包含了一个测试子集，包含1,740个样本用于评估，这些样本分为一步、两步和三步问题。整个数据集规模为6,500个问题，任务重点在于视觉数学推理。
WE-MATH is a dataset consisting of 6,500 visual mathematical problems, which are categorized into 67 hierarchical knowledge concepts. It is designed to evaluate the problem-solving principles of Large Multimodal Models (LMMs) in visual mathematical reasoning. Furthermore, the dataset includes a test subset with 1,740 samples for evaluation, which are divided into one-step, two-step, and three-step mathematical problems. The entire dataset has a total of 6,500 problems, and the task focuses on visual mathematical reasoning.
提供机构：
WE-MATH Team
