iMeanAI/UATD

统一代理轨迹数据集（UATD）是一个用于代理和更多应用的统一轨迹数据集。它包含了自然语言指令、任务的状态列表、状态描述、对话（可选）、状态观察（包括视觉截图、HTML源代码、文本描述和截图字幕）、与状态相关的动作以及动作目标等信息。数据集由多个配置组成，每个配置包含默认split或其他split（如训练、测试等）的多个数据文件。数据集包含了来自Mind2Web、AutoWebGLM、WebArena、WebLINX和GUIAct等来源的任务和状态。

The Unified Agent Trajectory Dataset (UATD) is a unified trajectory dataset for Agents and more. It contains natural language instructions, a list of states in the task, state descriptions, optional dialogues, state observations (including visual screenshots, HTML source code, textual descriptions, and screenshot captions), actions associated with the state, and action targets. The dataset consists of multiple configurations, each with multiple data files for the default split or other splits (such as train, test, etc.). The dataset includes tasks and states from sources such as Mind2Web, AutoWebGLM, WebArena, WebLINX, and GUIAct.