Thanks! I'd like to ask how to fine-tune v1p2 model? The pipeline code in finetune.py is incompatible with v1p1.