addmm_ TypeError when running vae example with cuda

[hassony2]

addmm_ TypeError when running vae example with cuda

[apaszke]

It's not a problem with weights, if you look at the order of the arguments it's (0, 1, input, weight.t()) and it's the 3rd argument that's not on CUDA. I see the bug, the example is missing

if args.cuda:
    data = data.cuda()

in the training loop. I'll send a PR in a sec. Thanks for reporting it!

[apaszke]

Fixed in #52.

[hassony2]

Thank you for your answer and the fix !
I had tried to pass the data to cuda but this alone had not solved it for me.
With the additional change in the PR it works great :)

[apaszke]

Yes, the sampled noise had to be allocated on the GPU too. You can check the diff in my PR to see where was that. Thanks!