Getting individual partials before they are added

When a variable is connected to the final scalar loss via two or more separate paths, the partial from each path is added. Is there a way to get the individual partials? An example would be dL/da in this one: https://towardsdatascience.com/getting-started-with-pytorch-part-1-understanding-how-automatic-differentiation-works-5008282073ec