Stride has 2 1s in it

1

I have a tensor called grads with stride:

(1866240, 155520, 64, 1)

and shape:

shape: torch.Size([1, 12, 2430, 64])

But, when I do a backwards pass (using a custom autograd function), like this:

result.backwards(grads)

In other words:

        @staticmethod
        def backward(ctx, grad_output):

The stride of grad_output becomes (1, 155520, 64, 1), even though the stride of grads is (1866240, 155520, 64, 1).

The stride of this tensor becomes: (1, 155520, 64, 1), even though the shape hasn’t changed. Why does this happen?