WikiMixQA

该数据集名为WikiMixQA，是一个包含1000个多项选择题的基准测试，旨在评估模型在跨越7000个维基百科页面中提取的表格和图表上进行跨模态推理的能力。这些页面涵盖了七个不同的话题。该数据集强调复杂的推理能力，要求模型从多种模态中综合信息，并包含一个用于文档分类的自定义分类法。其规模包括7258个文档和3528个问题-答案对，任务类型为问答。

This dataset, named WikiMixQA, is a benchmark comprising 1,000 multiple-choice questions, designed to evaluate models' cross-modal reasoning capabilities on tables and charts extracted from over 7,000 Wikipedia pages. These pages cover seven distinct topics. This dataset emphasizes complex reasoning abilities, requiring models to synthesize information across multiple modalities, and includes a custom taxonomy for document classification. In terms of scale, it contains 7,258 documents and 3,528 question-answer pairs, with the task type being question answering.