class VAE(nn.Module):
    def __init__(self):
        super(VAE, self).__init__()

        self.fc1 = nn.Linear(784, 400)
        self.fc21 = nn.Linear(400, 20)

    def forward(self, x):
        x = x.view(-1, 784)
        h1 = self.fc1(x)
        h2 = self.fc21(h1)
        return torch.sigmoid(h2)


# helper function to get sum of List[Tensor]
def _sum_of_list(tensorlist):
    s = 0
    for t in tensorlist:
        if isinstance(t, torch.Tensor):
            s += t.sum()
    return s

def clone_inputs(arg):
    input = arg.detach().clone().requires_grad_()
    return input, input


input_tensor = torch.rand((128, 1, 28, 28), requires_grad=True)
traced = torch.jit.trace(VAE(), input_tensor)


recording_inputs, recording_tensors = clone_inputs(input_tensor)
outputs = traced(recording_inputs)
l1 = _sum_of_list(outputs)
grads = torch.autograd.grad(l1, recording_tensors, create_graph=True, allow_unused=True)

l2 = (_sum_of_list(grads) * l1)
grads2 = torch.autograd.grad(l2, recording_tensors, create_graph=True, allow_unused=True)