AmazonScience/document-haystack
收藏官方服务：
资源简介：
Document Haystack是一个全面的长篇文档视觉语言模型基准数据集，用于评估模型在长文档上下文检索和多模态文档理解方面的性能。数据集包含5到200页的文档，并在其中嵌入了纯文本或文本+图像的“针”，以挑战模型检索隐藏信息的能力。
Document Haystack is a comprehensive benchmark dataset for Vision Language Models (VLMs) designed to evaluate the performance on long-context retrieval and multimodal document understanding. The dataset consists of documents ranging from 5 to 200 pages with embedded needles—short key-value statements in pure text or as multimodal text+image snippets—to challenge the models ability to retrieve hidden information.
提供机构：
AmazonScience
