PyTorch: specify weights_only for torch.load()

[vl-dud]

It makes sense to specify parameter weights_only for torch.load explicitly. The default value has changed from False to True in the latest version (2.6).

[lululxvi]

Just to confirm, using weights_only=True, it still loads optimizer state, right?

[vl-dud]

I checked, model.opt.state_dict() is the same when using weights_only=True and weights_only=False