🐛 Describe the bug

The PyTorch code base contains several uses of torch.load internally, but not all of them set weights_only, which in PyTorch 2.4 raises a FutureWarning. Since these are internal, the user would see the warning but have no way of acting on it. PyTorch should set weights_only in all of its internal uses.

Some examples (but there are more if you search the code base):

There was a PR recently that resolved one of these important cases:
#130242

These warnings will show up in PyTorch 2.4.
This surfaced in our test suite (PyTorch Lightning) since we have configured them to fail if FutureWarning is raised.

If desired, I can send a PR to fix this.

Versions

torch=2.4.0

cc @ezyang @gchanan @zou3519 @kadeng @msaroufim @XilunWu @H-Huang @awgu @kwen2501 @wanchaol @fegin @fduwjj @wz337 @wconstab @d4l3k @c-p-i-o @LucasLLC