I would like to look at the computational graph at different stages of it. For specific reason I have to use autograd.grad which seems doesn't update data.json at all. It seems that torchexplorer uses handle_step() to update the data.json which in turn is called in these two hooks:
I think those two hooks are never called when you use autograd.grad. I'd like to be able to check out the computational graph at any time, so I was wondering if there's a way to manually update data.json?
I'm not immediately sure how something like this should be handled, and unfortunately I'm too swamped at the moment to look into it myself. If you figure something out definitely feel free to update.
Hello,
I would like to look at the computational graph at different stages of it. For specific reason I have to use
autograd.grad
which seems doesn't updatedata.json
at all. It seems thattorchexplorer
useshandle_step()
to update thedata.json
which in turn is called in these two hooks:I think those two hooks are never called when you use
autograd.grad
. I'd like to be able to check out the computational graph at any time, so I was wondering if there's a way to manually updatedata.json
?