GPU Compatible Version?

[bhorowitz]

I'm a bit of a pytorch newbie, but was curious how difficult it would be to make a GPU compatible version of this code? My attempts of just adding ".to(device)" at various points of the code when defining new variables have been mostly going in circles so perhaps there is something deeper going on or an easier way within the structure of the code... Perhaps if someone could give guidance I could make the needed changes.

[aaronkl]

Hmm, I am afraid not. In order to run it on the GPU, one needs to move all input tensors and the network to the GPU by calling .cuda() (see here).
May I ask why you want to run it on GPU? Based on my experience, the small network architectures that we use inside pybnn are usually much faster on CPU than GPU.