Compared to normally applied Decoder-only Transformer models, seq2seq architecture is more well suited for schooling generative LLMs presented