transformers

add VisionTextDualEncoder and CLIP fine-tuning script

#15701

Merged