-
jointTrainig-separateLosses8f948bd7 · ·
Joint training is working with no more NaN (lr has to be really low). Losses are still separate, validation losses do not have a jont covergence (places loss increase)
Joint training is working with no more NaN (lr has to be really low). Losses are still separate, validation losses do not have a jont covergence (places loss increase)
IIT