Can someone refresh my memory difference between VV and V in core of fastai. I know V is variable. But what is VV?
2 Likes
VV creates Variable with volatile set to True. Basically you use VV when you don’t care to calculate Gradients w.r.t that inputs to save some GPU memory. Typically this is done for Y (Targets) because you don’t usually calculate gradients w.r.t targets (its done w.r.t inputs). But when you are running evaluation on the model, VV(X) and VV(Y) are also used.
in PyTorch 0.4, they are deprecating this volatile flag, so this might change in the future.
10 Likes
Thank you @ramesh !