MidGUI/Mid-Training_data_of_separate_domains

GUIMid数据集是用于训练GUI代理的任务泛化官方数据集，包含9个多样化的任务，其中有7个视觉和语言任务以及4个仅语言任务。这些任务包括图表/文档问答、代码I/O、GUI感知、数学指导、中间GUI、多模态数学、多轮视觉对话、非GUI代理、非GUI感知、奥数数学和网页知识库等。数据集支持任务类别有文本生成和问答，语言为英语，大小类别在10万到100万之间。

The GUIMid dataset is the official dataset for training GUI agents through task generalization, containing 9 diverse tasks, including 7 vision and language tasks and 4 language-only tasks. These tasks include Chart/Document QA, CodeIO, GUI Perception, MathInstruct, MidGUI, Multi-modal Math, Multi-round Visual Conversation, Non-GUI Agents, Non-GUI Perception, Olympiad Math, and Web Knowledge Base. The dataset supports task categories of text generation and question-answering, with language in English and size category between 100K and 1M.