One Hot Encoding

No - nothing is being one-hot encoded. Instead, each categorical column is given it’s own embedding matrix in which it learns robust representations of each item. Check out lesson five or look at the lesson notes here: