why torch.autograd could not compute the full Jacobian directly while it is mathematically achiveable

Question

I am reading the pytorch tutorial at this page: https://pytorch.org/tutorials/beginner/blitz/autograd_tutorial.html

It says that "Now in this case y is no longer