Google’s technology. A large scale pretrained model that is first trained on very large amounts of unannotated data. The model is then transferred to an NLP task where it is fed another smaller task-specific dataset which is used to fine-tune the final model.