WSL: Training on CPU even when having compatible NVIDIA GPU

Issue description

Ivadomed doing training on cpu even though i have a compatible nvidia gpu and installed via requirements_gpu.txt

Current behavior

Training on cpu

Expected behavior

Training on gpu

Steps to reproduce

All steps in https://ivadomed.org/en/latest/installation.html with approach 2

Environment

Windows 10 Pro : (WSL1 Ubuntu 20.04 with python 3.7) and (WSL2 Ubuntu 18.04 with python 3.8) Nvidia Geforce gtx 1050

Solutions tried

Tried https://docs.nvidia.com/cuda/wsl-user-guide/index.html, but the current insider preview build (which is a version of Windows 11) is not stable (crashing at every startup), had to do a clean installation of Windows 10 afterwards.

[dyt811]

Yup. Will update this ASAP. Windows Torch CPU based IvadoMed has been running and well tested through GitHub Action CI processes for a long time post PR #757 but the Windows GPU based approaches has not undergone the same systematic tests. Going to test it out shortly before the documentation update.

[jcohenadad]

Hey @dyt811, I didn’t know ivadomed supports windows natively because I read in https://ivadomed.org/en/latest/installation.html “Currently, we only support MacOS and Linux operating systems. Windows users have the possibility to install and use ivadomed via Windows Subsystem for Linux (WSL)”.

@dyt811 if windows is supported natively, this is something we need to fix in the doc