topic Re: How force Pytorch to use CPU instead of GPU? in ArcGIS Image Analyst Questions

How force Pytorch to use CPU instead of GPU?

MaryamBarzegar — Wed, 14 Apr 2021 05:40:02 GMT

Hello, I have a 2GB GPU and it's not enough for training the model and I get CUDA out of memory error every time (when running model.Ir_find ()). Is there any way to force Pytorch to use only CPU? For some reasons I can't clone the default Python environment either and update the ArcGIS API to see I'll get an error in other versions or not. I'm using ArcGIS API 1.8.3.

Re: How force Pytorch to use CPU instead of GPU?

MaryamBarzegar — Wed, 14 Apr 2021 07:01:22 GMT

Obviously I've done that before and none of the solutions worked and that's why I posted my question here. For instance, I tried

device = torch.device("cpu")

and,

os.environ["CUDA_VISIBLE_DEVICES"] =""

both didn't work.

Re: How force Pytorch to use CPU instead of GPU?

Tim_McGinnes — Wed, 14 Apr 2021 12:31:30 GMT

Try this:

import torch torch.cuda.is_available = lambda : False device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

It's definitely using CPU on my system as shown in screenshot.

BTW, I am also getting an error trying to update the Python API, details here - is it the same for you?

Unable to install\upgrade via conda: InvalidSpecError: Invalid spec: >=

Re: How force Pytorch to use CPU instead of GPU?

MaryamBarzegar — Wed, 14 Apr 2021 23:28:19 GMT

Thank you so much! It worked.

I tried upgrading packages on another laptop using a cloned environment and it worked but on this laptop, I couldn't even clone the default environment.