This dataset supports our publication Dynamic Phase Transitions Dictate the Size Effect and Activity of Supported Gold Catalysts. It includes the Au-Ce-C-O model, a machine learning force field (MLFF) developed for simulating supported gold catalysts on Ceria surface in CO and O2 atmosphere, along with the data and files used in the training and simulation processes. Freezed and compressed Au-Ce-C-O model (graph.pb and graph-compress.pb): The fully trained MLFF model used for production simulations in our study. Au-Ce-C-O dataset (dataset.tgz): Contains the training datasets for Au/CeO2 system with CO and O2 , in DeePMD-kit format (dataset.tgz). Each image is labeled with coordinations (coord.npy) in Å, total energies (energy.npy) in eV, force (force.npy) in eV/Å, and cell parameters (box.npy) in Å. DeePMD-kit files (input-auceco.json): Input file example utilized for training using DeePMD-kit. LAMMPS files (in.rigid): Input file example utilized for NVT Molecular dynamics simulation. VASP files (VASP-au10ceo2-3co-aimd.tgz): Input file example utilized for AIMD simulation. DPMD trajectaries (traj.tgz): The full trajectories of the Au/CeO2 system, with and without CO exposure, with temperature increasing in increments of 25 K.