PyTorch Geometric (PyG)New

PyTorch Geometric (PyG) is a PyTorch-based library for building graph neural networks. It provides a large catalog of GNN layers and message-passing operators (GCN, GAT, GINE, Schnet, GemNet and more), along with graph samplers, standard datasets and heterogeneous-graph support.

In physics AI it is the de facto standard for graph representation learning on molecules, crystals and physical systems — powering GNN interatomic potentials, molecular property prediction and topology-aware surrogate models.