Comment by moinnadeem

What do you think about the train test discrepancy? ie. will practitioners have to fine-tune Nubia's models on their training dataset in order to evaluate on their test dataset?