This repository contains the implementation of an equivariant neural network architecture for molecular and atomic multipole prediction. Large parts are taken from the mlff repository, in particular the network layer implementation of the So3krates network. My personal contributions are most of the observable modules in /src/nn/observable/observable.py and other small helper functions. The class CeqNet is an adaptation of the StackNet class from mlff to the multipole prediction task. Example trainings implementations will follow soon.