Since this is an autoencoder, the output is same as input. That is, the y is the same as the X. Therefore, we have model.fit(X, X, …).