How to use Pytorch 1.5 without installing CUDA

C++
1

I was wondering how we could use PyTorch with just cpus. There were previous builds which separated gpu/cpu, but using 1.5 allows for more complex indexing, so I wanted to find a way to use the most updated stable version of PyTorch without installing cuda (won’t be able to use GPUs based on hardware constraints).

Thanks!

2

For the CPU only version, you would have to select the CUDA None option on the website.

This command would install 1.5 without CUDA:

conda install pytorch torchvision cpuonly -c pytorch
3

To clarify, I’m trying to use libtorch. Can I edit the libtorch cmake file (i.e. TorchConfig.cmake) in order to not care about CUDA installation?

4

The CPU libtorch package shouldn’t care about CUDA, right?
What kind of error are you getting and how are you building your project?

5

Oh my… I should go to sleep. I didn’t even see the None option for CUDA when downloading libtorch. I’m so sorry!!!

6

Haha, no worries! Let me know, if it still yields errors. :slight_smile: