zhanwenchen/personalgui-accel-data

GUI-agent evaluation: code, per-episode records, and captured media This repository holds everything behind a study of whether a language model can operate real software when its only input is a video feed of the screen and its only output is a keyboard and mouse. The rendered report lives in a separate Space; this is the material it is computed from. What the setup is Two machines, joined by two cables and no software link. One runs the agent. The other runs the… See the full description on the dataset page: https://huggingface.co/datasets/zhanwenchen/personalgui-accel-data.