attention is all you need (transformer models)

A novel model architecture based solely on attention mechanisms and feed-forward networks, without any recurrence or convolutions.