The initial training of a model. Pretraining involves passing very large amounts of data to build a robust statistical model of language and knowledge.
