Custom Autograd.Functions to GPU

autograd
1

I am just starting out, and I need to create a custom activation function just like how they did it here:
https://pytorch.org/tutorials/beginner/examples_autograd/two_layer_net_custom_function.html

So my question is does relu(input) run on the cpu or gpu? I am guessing that it doesn’t run on the gpu, since anytime I want something to run on the gpu, I always had to call .to(device).

I tried that with the custom activation, but I was unable to do that. So if I am correct in that the activation does not run on the gpu, then how do I do that.

If I am wrong, can you please explain how it knows to send to gpu?

2

Hi,

These Functions don’t have any parameters, so they will work with whatever inputs are given. Just give it inputs on the gpu and it will run on the gpu.