Reconstructing Galactic Gravitational Potentials from Stellar Kinematics with Physics-Informed Neural Networks: Datasets and Trained Models

This archive contains trained weights, fitted data scalers, and validation datasets for physics-informed neural network (PINN) models of gravitational potentials. The models span both static and time-evolving systems and include deterministic and Bayesian variants -- covering a triaxial NFW halo, the Milky Way + LMC system, and the FIRE m12b simulation. A drop-in loaders.py is included alongside the data: each model can be loaded and queried in three lines of Python, returning potentials and accelerations in physical units. See README.md for installation, model architectures, dataset layouts, and example usage. The codebase can be accessed on GitHub: galactoPINNs.