position IDs
Contrary to RNNs that have the position of each token embedded within them, transformers are unaware of the position of
each token.