What does "jacobian" mean in pytorch?

Ok, I understand what jacobian means mathematically. But when we say it in Pytorch, as in “Jacobian-vector product”, do we mean by the real Jacobian or the transpose of the real Jacobian?