Typically, most large-scale language models (LLMs) perform the task of 'predicting the next word,' and output one piece of data (token) at a time. In response to this, Meta proposed an approach called ...