MMTU

该数据集名为MMTU，是一个大规模的基准测试，包含了超过30,000个问题，覆盖了25个现实世界中的表格任务。它旨在评估模型在专家级别对真实表格的理解、推理和操作能力。该数据集包含了来自计算机科学领域数十年对表格数据研究的多样化复杂任务。任务的范畴涉及表格的理解、推理以及在多个任务中的操作。

This dataset, named MMTU, is a large-scale benchmark containing over 30,000 questions covering 25 real-world table-related tasks. It aims to evaluate models' expert-level capabilities of understanding, reasoning over, and manipulating real tables. This dataset includes diverse and complex tasks derived from decades of table data research in the field of computer science. The scope of these tasks covers table understanding, reasoning, and manipulation across various task scenarios.