Is model.forward(x) the same as model.__call__(x)?

When I worked with Tensorflow, I used to define a model’s forward pass and other customizations under its def __call__(self, x) function. If I want to implement the same thing in PyTorch, should I do it in def forward(self,x) instead?

1 Like

It is just a naming convention. Why do you want to define call(x) again inside the forward function? It isn’t needed, just use the forward function to get the outputs for the input data. Check this tutorial

__call__ is already defined in nn.Module, will register all hooks and call your forward. That’s also the reason to call the module directly (output = model(data)) instead of model.forward(data).