Questions about built-in method torch.xxx(ops)

1

Hi there.
Is there any specification about torch.xxx(ops), such as torch.max_pool2d,torch.conv2d? I cannot found it in https://pytorch.org/docs.

How do they differ from their torch.nn.functional.xxx counterpart?

2

You can see the different in nn.Linear() and nn.functional.linear(). In general speaking, one is class and another is base function interface. I remember that if you use nn.functional.linear(), you should set bias manually.

3

torch.conv2d is the exact same as torch.nn.functional.conv2d. You can check where they are defined in torch/nn/functional.py.