ageppert/world-model-transitions
收藏资源简介:
--- language: - en license: mit tags: - world-model - computer-use - transition-prediction --- # World Model Transition Dataset Derived from the [AgentNet](https://huggingface.co/datasets/xlangai/AgentNet) Ubuntu 5K dataset for training a world model (transition predictor) for computer-use agents. ## Format Each example is a chat conversation: - **System**: World model role description - **User**: Current screen state observation + planned action - **Assistant**: Predicted outcome (transition description) ## Source Tier 1 extraction from AgentNet: steps with non-empty reflections, correct actions, non-redundant, non-terminal. See `preprocessing_stats.txt` for detailed filtering statistics. ## Citation If you use this dataset, please cite the original OpenCUA paper: ```bibtex @misc{wang2025opencua, title={OpenCUA: Open Foundations for Computer-Use Agents}, author={Xinyuan Wang and others}, year={2025}, eprint={2508.09123}, archivePrefix={arXiv}, } ```
语言: - 英语 许可协议:MIT 标签: - 世界模型(world model) - 计算机交互(computer-use) - 转移预测(transition-prediction) # 世界模型转移数据集 本数据集源自[AgentNet](https://huggingface.co/datasets/xlangai/AgentNet)的Ubuntu 5K数据集,用于训练面向计算机交互智能体的世界模型(transition predictor,转移预测器)。 ## 数据格式 每条样本均为一段对话: - **系统(System)**:世界模型角色说明 - **用户(User)**:当前屏幕状态观测结果与计划执行的动作 - **助手(Assistant)**:预测的结果(转移过程描述) ## 数据源 本数据集从AgentNet中进行一级提取:选取带有非空反思、有效动作、无冗余且非终止的步骤。 详细的过滤统计信息请参阅`preprocessing_stats.txt`文件。 ## 引用规范 若使用本数据集,请引用原始OpenCUA论文: bibtex @misc{wang2025opencua, title={OpenCUA:面向计算机交互智能体的开源基础}, author={王新元等}, year={2025}, eprint={2508.09123}, archivePrefix={arXiv}, primaryClass={cs.AI} }



