MathScape
收藏资源简介:
MathScape是由南开大学、北京大学、百川公司和中国科学院大学联合开发的多模态数学问题数据集,旨在评估多模态大语言模型在数学问题解决中的应用。该数据集包含1325张图像,涵盖从小学到高中的数学问题,难度从易到难,涉及多种题型和知识领域。数据集的创建过程包括数学文档的图像转换、真实场景的图像捕捉和严格的质量检查与知识分类。MathScape主要用于评估模型的数学推理能力和复杂多步量化推理能力,特别是在视觉上下文中的应用。
MathScape is a multimodal mathematical problem dataset jointly developed by Nankai University, Peking University, Baichuan Inc., and the University of Chinese Academy of Sciences, aiming to evaluate the application of multimodal large language models in mathematical problem-solving. This dataset contains 1,325 images, covering mathematical problems from primary to high school levels, with difficulty ranging from easy to hard, involving multiple question types and knowledge domains. The dataset's construction process includes image conversion of mathematical documents, image capture of real-world scenarios, as well as strict quality inspection and knowledge classification. MathScape is primarily used to assess the mathematical reasoning capabilities and complex multi-step quantitative reasoning capabilities of models, especially their applications in visual contexts.




