Recommendations for job resources

Hello! I am interested in learning how to optimize execution of geNomad. I'm working on an HPC with some gpu capacity. It looks like some of the XGBoost/tf steps can benefit from execution on gpus, but I don't know what parameters would need to be set to take advantage of that. Is there an optimum number of threads? Do any env variables (like `TF_NUM_INTRAOP_THREADS` or something) need to be set? Any advice would be appreciated, thanks!