About the pre-train strategy in the code

[Loneseven]

At the moment the internvl_chat/internvl/train/internvl_chat_finetune.py and internvl_chat/internvl/train/internvl_chat_pretrain.py are almost the same. I am wondering if the ITM loss and Contrastive loss would be used. Am I miss something? Is generative loss alone enough?