Where is the input value for backpropagation saved?

Hello dear Pytorch-User,
i have the following question:

suppose you have a very simple network with:

input * w1 = output
MSE(output,target) = loss

Then you need the value from input for the backpropagation. But i can t find the place where it is saved. I allready read a little bit about the backpropagation system from pytorch, but it didnt give me the answer to this question. I am just curious, its not very important.

Hi,

It is saved within the computational graph in c++ (it saves a reference to input).
There are no easy way for you to access that reference from python though.

1 Like

hi, thank you very much for your answer