Using YOLOv5 on AGX uses the CPU and not the GPU

Hello, I’ve installed Yolov5 on an AGX using the latest Jetpack, and followed the instructions to install Pytorch using this answer: LINK.

My problem is that when I run a detection it says its using CPU. I confirmed that CUDA is installed however when I check in python, by importing Pytorch, if CUDA is present it says False.

How do I get Pytorch to use the GPU? When using Yolov5 on my desktop this didn’t happen so I’ve no idea how to fix it. Thanks.

Also I don’t know if this is an issue but I get a warning with OpenBLAS, it went away by doing what was said here : LINK2

however when I rebooted it came back. Is there a permanent way to fix this? (Sorry if I’m not supposed to ask two things in one post.)


Could you follow below topic to install PyTorch?
The package is built with CUDA and Jetson GPU architecture.