lgy0404/

MemGUI-3K

Paper | Project Page | GitHub MemGUI-3K is a memory-intensive mobile GUI agent trajectory dataset for training and analyzing agents that proactively manage long-horizon context. It contains teacher rollouts from MemGUI-Agent using the ConAct Context-as-Action paradigm, where the agent emits both GUI actions and context actions for history folding and UI memory management. Code, data processing scripts, model training scripts, and evaluation tools are available in the… See the full description on the dataset page: https://huggingface.co/datasets/lgy0404/MemGUI-3K.