Which backend is chosen for training, CPU or GPU ?

[Sreerag-ibtl]

I am running this experiment on Google Collaboratory. GPU is chosen as runtime. But, Training is taking longer to complete. Is CPU chosen explicitly in the code or Is this something I should expect? TIA

[koenhelwegen]

What exactly do you mean by longer? Are you training on CIFAR-10 or Imagenet? What type of GPU are you using? Is it possible to see GPU utilization?

Training BNNs tends to be a bit slower than training real-valued networks. The device is set here - it should run on the GPU if available.

[Sreerag-ibtl]

@koenhelwegen Thanks for mentioning about slowness in training. I used CIFAR10 for training. Hardware used was GPU from Google Collaboratory.

[koenhelwegen]

How long does an epoch take?

[Sreerag-ibtl]

80 seconds.