Transformer models - characterized by self-attention, positional encoding and parallel processing - have become a cornerstone in language understanding, translation and generation.