Pytorch with cuda for YOLOv9 object detection

Robotics & Edge Computing Jetson & Embedded Systems Jetson Orin Nano
1

I booted my Jetson Orin Nano with JetPack 6.0 and want to train a YOLOv9 object detection model. After I booted my device, I ran the following commands: sudo apt update, sudo apt upgrade, sudo apt install nvidia-jetpack.
However, when I ran nvidia-smi, it shows my driver version as N/A. Does this matter? And if it does how do I fix it?
Additionally, when I ran nvcc --version, I got an error message saying nvcc command not found. How do I fix this?
Furthermore, What version of torch, torchaudio, and torchvision should I install for CUDA 12.2? And what are the commands to install them?

3

Hi,

Could you share your nvidia-smi output with us?
If you see similar to the one below, it is expected due to the limited support for Jetson.

For JetPack 6, you can find our prebuilt PyTorch/TorchVision/TorchAudio in the link below:

Thanks.

4

Hello,

My nvidia-smi output is exactly the same as the one you shared above.
Do you know why nvcc doesn’t work?
Additionally, I have tried using those commands previously and they have never worked (I have probably done it incorrectly but I don’t where I am going wrong). I always got some different error, and I don’t just want to try the same thing again because it takes a really long time for me to reflash my sd card to get a fresh start.
If it is not too much trouble, could you please describe exactly how I can use those commands? Thank You so much if you can and I understand if you cannot.
Edit: I fixed the nvcc issue by adding CUDA to path. Turned out to be a pretty trivial fix

5

@AastaLLL
In the second link that you provided, am I supposed to change this command to match my torch version?

pip3 install numpy torch-1.8.0-cp36-cp36m-linux_aarch64.whl
6

@AastaLLL Everytime I try to install torch using the link above, it install torch-1.8.0 instead of torch-2.3.0 which I actually want. I ran it all correctly and made the appropriate changes before running the commands. Is there another way?

7

Hi,

Please install the packages built on the same JetPack version for compatibility.

For JetPack 6.0, the available one is v2.3.0.
If you need other versions, please build it from the source since the v1.8 package is built for the JetPack 4 environment.

The building from the source commands is also included in the abovementioned topic.

Thanks.

9

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.