Porting ML models to KIM TorchML driver# Contents: TorchML Model Driver for KIM API Dependencies and Environment Model locality and the need to port Design Introduction Portability Computation Supported models Required Signatures Basic porting steps Basic model file structure Generic Models Species Coordinates Neighbour list and number of neighbors Contributing atoms Descriptor Models Limitations Graph Neural Networks Cyclic graph convolutions Staged graph convolutions Periodic boundary conditions Generating staged graphs GNN model signature Dealing with unmodifiable PBC and lattice vectors Nequip to KIM API KIM-NequIP-port tool Introduction Caveats Installation Usage Inference GPU Support Appendix A1: Using layers from a TorchScript model A2: Pytorch implementation of Stillinger-Weber energy functions A3: Example descriptor.dat file Indices and tables# Index Module Index Search Page