Hi! When using Keras and calling .fit() Keras there’s variable called validation_split which splits your training data to have a validation set. The default is 0.0, meaning the data won’t be split and you won’t have a validation set (unless you specify validation_data) - not a good idea.
Within PyTorch, there’s no fit style function so you need to write your own. I sense this is what is giving you the impression that there is no val in Keras as the train/test/val is a bit more clear in PyTorch’s case.
Does that help? It’s in Keras, just a bit more hidden.