TIGER-Lab/VisualWebInstruct-Seed

VisualWebInstruct-Seed数据集是一个用于多模态指令数据扩展的数据集，通过谷歌搜索进行构建。它包含了图像和文本信息，旨在用于问题和回答的任务，特别是图像-文本到文本的任务。数据集分为三个配置：reference、refined和LongCoT，每个配置都有不同的特征和训练集大小。

The VisualWebInstruct-Seed dataset is a dataset for scaling up multimodal instruction data through web search. It contains both image and text information, designed for question-answering tasks, especially image-text-to-text tasks. The dataset comes in three configurations: reference, refined, and LongCoT, each with different features and training set sizes.