update docs to recommend __call__ for forward passes

[jeremyjordan]

📚 Documentation

We should update the docs to recommend usage of self(x) for calculating the forward pass rather than self.forward(x). Calling forward() directly can cause issues when you're using PyTorch model hooks (eg. see the additional logic in nn.Module.__call__).

Although most people don't play around with hooks and using self.forward() will work fine for them, we should probably follow the best practice (as I understand) of using __call__ for calculating forward passes.

Related issues: #632

[awaelchli]

Very good idea! Then the examples should also be updated I guess?

[jeremyjordan]

yes! we should update:

• docs pages
• lightning examples
• colab notebooks

i'm happy to tackle the first two, i think @williamFalcon is planning to do the third. will link to a WIP PR soon.

[williamFalcon]

@jeremyjordan i'll do colabs!