The size of PyTorch checkpoint of bart-large https://huggingface.co/facebook/bart-large/tree/main is 971M while that of the tf_model is 1.51G. And the size of Pytorch checkpoint of bart-large-cnn https://huggingface.co/facebook/bart-large-cnn is also 1.51G. Why is the Pytorch bart-large checkpoint much smaller than others? @patrickvonplaten