Setting PyTorch to `train` or `eval` modes globally instead of at nn.Module level

At PyTorch model that inherit from torch.nn.Module can be set to train or eval modes using model.train() or model.eval(). Is there a way to do this globally for the present PyTorch session, so that it applies to any op that is invoked, without having to update different models?

This could be useful in running PyTorch scripts in training or inference context as desired.