Hi all,
I wanted to ask if it is possible to not just measure the time of the backward pass as a whole, but also to measure the time of specific parts of the backwards pass. I.e., I have different models within my algorithm and I would want to find out if there is any particular model of these which is requiring particularly much time for backpropagation.
Or would you just assume that the time of for backpropagation is proportional to the forward pass in each separate model?