MolmoWeb-HumanSkills

MolmoWeb-HumanSkills 是一个包含人类收集的网络导航技能的数据集，其中每个技能代表一个低级别任务（如查找并打开、填写表单等）的轨迹。每个样本包含一个指令与一系列网页截图及对应的代理操作（点击、输入、滚动等）。数据集的主要字段包括：sample_id（唯一标识符）、instruction（JSON编码的任务指令）、trajectory（JSON编码的轨迹，包含代理操作和截图文件名）、images（截图原始数据列表）和image_paths（截图路径列表）。轨迹中的每个步骤包含截图文件名、代理操作（可解析的操作字符串、自然语言描述和结构化输出）、浏览器状态（当前URL、页面索引等）以及操作时间戳。数据集分为训练集（115,637个样本）和预览集（10个样本），适用于研究网络导航、人机交互等任务。数据集采用ODC-BY 1.0许可，仅供研究和教育用途。

MolmoWeb-HumanSkills is a dataset containing human-collected web navigation skills, where each skill represents a trace of a low-level task (e.g., find and open, fill out forms, etc.). Each sample includes an instruction, a series of web screenshots, and corresponding agent actions such as clicks, inputs, scrolling and more. The main fields of the dataset are as follows: sample_id (unique identifier), instruction (JSON-encoded task instruction), trajectory (JSON-encoded trajectory containing agent actions and screenshot filenames), images (list of raw screenshot data), and image_paths (list of screenshot paths). Each step in the trajectory contains the screenshot filename, agent action (including parsable action strings, natural language descriptions and structured outputs), browser state (current URL, page index and other relevant information), and operation timestamp. The dataset is divided into a training set with 115,637 samples and a preview set with 10 samples, which is applicable to research on web navigation, human-computer interaction and other related tasks. The dataset is licensed under ODC-BY 1.0 and is for research and educational purposes only.