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

[MarktHart]

Probably to be fixed around here:

transformers/src/transformers/modeling_utils.py

Line 3016 in bbf3090

expected_keys = list(model_state_dict.keys())

See this issue on PyTorch:
pytorch/pytorch#102999 (comment)

[ydshieh]

Hi, what's your transformers version?

If you use a dev version with the commit of this PR ##24030 included, it should be fine. Let me know if not, thanks.