[doc] nn.Module.forward documentation unclear

[henon]

📚 Documentation

This note is totally unclear to me. What does it mean? I don't know what should be called afterward. The link doesn't point to a function, just to torch.nn.Module.

[ssnl]

module_instance(inputs)

[henon]

I searched the PyTorch docs and didn't find any hits for module_instance. Also class Module doesn't seem to contain such a method when looking at the source. What am I missing?

[ssnl]

I was just referring to a generic instance of the module class..

[yf225]

@henon An example for this would be:

module_instance = torch.nn.Linear(10, 20)
input = torch.randn(10)
output = module_instance(input)

We should add a simple example in the nn.Module.forward doc about this.

[henon]

@yf225 thanks for the explanation. So is it true to say that any Module is in fact a Callable? Would it be a good idea to add that fact to the description of torch.nn.Module as well?

[apaszke]

@henon yes, every module is a callable, and calling it is (modulo some bookkeeping that we do automatically) equivalent to doing module.forward(input) (which you should never do!!).