Each large language model only has a specific quantity of memory, so it might only settle for a particular amount of tokens as input. A model could be pre-properly trained possibly to predict how the section proceeds, or precisely what is lacking during the segment, supplied a segment from it