The purpose of this project is generate and test machine learning potentials based on the atomic cluster expansion (ACE) formalism, and its recent expansion - graph ACE (GRACE). Specifically, we aim to generate binary potentials for the W-He and Y-H systems, which are of relevance for nuclear reactor applications. We have generated datasets for training and evaluation by means of DFT modelling already. With this application, we apply for time to use GPUs for the training and testing the potentials.