Best practices for training Deep Ensembles with PL

[CompRhys]

I am looking at training an ensemble of models in PL and was wondering whether there are any established best practices for how to manage and distribute the training of an ensemble when using PL to organize the code and train the model.