Large foundation models, vision and language multimodal, and pretraining and self-supervised training