Running a trained large language model to generate predictions or responses without further training.