Print which tensors are passed into .backward()

Is there a way to print which tensors are passed into .backward()? The problem is that, when I execute loss.backward(), the code prints out RuntimeError: grad can be implicitly created only for scalar outputs, meaning that the dimension of tensor passed into .backward() is not the same as the dimension of the tensor whose gradient is being calculated. Because of this, I would like to print which are the tensors used as input of the backward() function. Is this possible?

can you post your code,it helps us what exactly is wrong