Doesn't make much semantic sense to determine (and certainly not return) the loss in that function. Instead might make more sense to have user specify it in config, and let Trainer parse it and determine whether the requested loss function makes sense for the given model.
Doesn't make much semantic sense to determine (and certainly not return) the loss in that function. Instead might make more sense to have user specify it in config, and let Trainer parse it and determine whether the requested loss function makes sense for the given model.