hello everyone
I’m new here and this is my first topic.
I have problem with my code.
I’ve searched and didn’t find a solution.
The problem is I have a function that has an output shape different than input
for example if input shape be (batch_no,1), my output shape will be (batch_no,2,10).
the torch.autograd.Function makes the grad shape as the output but excpects as input.
I thought this should not be an issue because we should have n_output*n_input gradians.
i share a dummy code here which makes the same error.
I would appreciate any help
class test2(torch.autograd.Function):
@staticmethod
def forward(ctx,x):
ctx.save_for_backward(x)
result = [np.ones((5)) for i in x]
return torch.tensor(result)
@staticmethod
def backward(ctx, grad):
x = ctx.saved_tensors
return grad
inp = torch.tensor([0.5,0.5,0.5,0.5], requires_grad=True)
x = test2.apply(inp)
x.shape
x.sum().backward()
and i get this error:
RuntimeError: Function test2Backward returned an invalid gradient at index 0 - got [4, 5] but expected shape compatible with [4]
