Torch.autograd.grad got None gradients for cascaded model

[InnovArul (Arul)]

tl: dr; use append() instead of extend().

Long story:

The reason for this behavior is due to how extend() works.
extend() iterates over the given batch variable and adds the rows one by one to x_list. By doing this, extend() creates its non-leaf nodes, which are unrelated to the computation graphs involving model1, model2.
Using append() leaves batch variable untouched.

You can find an explanation about leaf variables from @albanD here:

ValueError: can't optimize a non-leaf Tensor?

Hi, A leaf Variable is a variable that is at the beginning of the graph. That means that no operation tracked by the autograd engine created it. This is what you want when you optimize neural networks as it is usually your weights or input. So to be able to give weights to the optimizer, they should follow the definition of leaf variable above. a = torch.rand(10, requires_grad=True) # a is a leaf variable a = torch.rand(10, requires_grad=True).double() # a is NOT a leaf variable as it was cr…