I’m trying to create callback that will use current state of model when training ends. I’m using “on_fit_end” method and calling model(data) to create prediction. Should I also switch model mode to “eval()” or model is in that state on this stage of training?