Illegal instruction jetson nano - yolov8

[user35592]

I get an illegal instruction error (without core dumped).
I tried to run python -c ‘import numpy’,
python -c ‘import torch’,
python -c ‘import torchvision’,
python -c ‘import cv2’
I already export OPENBLAS_CORETYPE=ARMV8
everything runs normally without error, but when going to run the model(image_path) , there is an illegal instruction error

my jetson nano:

• Module: Jetson Nano 4GB (production)
• AI Performance: 472 GFLOPS
• GPU: 128-core NVIDIA Maxwell™ architecture GPU
• GPU Max Frequency: 921MHz
• CPU: Quad-core ARM® Cortex®-A57 MPCore processor
• CPU Max Frequency: 1.43GHz
• Memory: 4GB 64-bit LPDDR4 25.6GB/s
• ubuntu 18.04

python default 3.6.9 but upgrade to python3.10.5

[AastaLLL]

Hi,

Do you use Ultralytics frameworks to run YOLOv8?
If yes, please check if your PyTorch has CUDA enabled.

Thanks.

[user35592]

yes i use ultralytics framework, from pip install ultralytics.
my torch=2.1.2 and torchvision=0.16.2
after downgrade torch=1.12.0 and torchvision==0.13.0 and succes to run yolo,
but torch cuda still disabled.
Does this affect Yolo’s performance (detection time) when detecting images?
what version of torch and torchvision is compatible with cuda10.2?

[AastaLLL]

Hi,

You can find the package with CUDA enabled in the below document:

Installing PyTorch for Jetson Platform - NVIDIA Docs

Please note that we only provide prebuilt for the default Python version.
For JetPack 4, it is Python 3.6.

Below is some info about the “Illegal instruction” error for your reference as well:
https://elinux.org/Jetson/L4T/TRT_Customized_Example#.22Illegal_instruction_.28core_dumped.29.22

Thanks.