MMTU
收藏资源简介:
MMTU是一个大规模的基准数据集,包含超过3万个问题,跨越25个现实世界的表格任务。该数据集旨在全面评估模型在理解、推理和操作真实表格方面的能力,旨在解决现实世界中专业用户面临的复杂任务。数据集的问题均来源于计算机科学领域几十年的研究,重点关注专业用户面临的复杂任务。MMTU数据集的创建过程包括文献调研、任务选择、数据标准化和整理等步骤,最终形成了一个全面评估模型在表格理解和推理方面的能力的数据集。
MMTU is a large-scale benchmark dataset containing over 30,000 questions spanning 25 real-world table-related tasks. This dataset is designed to comprehensively evaluate models' capabilities in understanding, reasoning over, and manipulating real-world tables, and to address complex tasks faced by professional users in real-world scenarios. The questions within the dataset are sourced from decades of research in the field of computer science, with a focus on complex tasks encountered by professional users. The development process of the MMTU dataset includes steps such as literature review, task selection, data standardization and curation, ultimately resulting in a comprehensive benchmark for evaluating models' table understanding and reasoning abilities.




