Using Loss Functions in Machine Learning Models

i'm a big fan of machine learning models and approaches that simplify things. The more hyper perameter tweaki you have to do, the more likely it is for you to have trouble training a good model. For every task, every pre training ent, we use the same learning rate, schedule and everythingn i think that makes a huge difference in practice for practitioners. We consider a decoter only language model, our incoter decoter, et cetera.