Breaking change in upcoming PyTorch version for weight norm and loading pretrained models

Probably to be fixed around here: https://github.com/huggingface/transformers/blob/bbf3090848cf0ceff98f9465691e9ecce63684a1/src/transformers/modeling_utils.py#L3016

See this issue on PyTorch:
https://github.com/pytorch/pytorch/issues/102999#issuecomment-1623975562

