When compared to typically applied Decoder-only Transformer models, seq2seq architecture is a lot more well suited for training generative LLMs offered more robust bidirectional interest towards the context. AlphaCode [132] A set of large language models, ranging from 300M to 41B paramet