The LLM is sampled to crank out just one-token continuation in the context. Specified a sequence of tokens, only one token is drawn fr