how to use pretrained clip model in train_decoder.py

question about how to integrate own-trained CLIP model in train_decoder.py, is there any documents about this