happy8825/MMLongBench_var1_2turn_seed5

MMLongBench是一个包含1073个样本的多模态长文本基准数据集，主要用于评估模型在不同证据来源（纯文本、图表、表格等）和不同页面长度（无页面、单页、多页）下的性能。数据集包含丰富的特征，如相关问题页面、证据页面、分数、文档类型、问题和答案等。MMLongBench is a multimodal long-text benchmark dataset containing 1073 samples, primarily used to evaluate model performance across different evidence sources (pure-text, figures, tables, etc.) and varying page lengths (no pages, single page, multiple pages). The dataset includes rich features such as relevant pages, evidence pages, scores, document types, questions, and answers.