Install cuda pytorch gpu operating environment on orin

https://blog.csdn.net/David_jiahuan/article/details/103903593?utm_medium=distribute.pc_relevant.none-task-blog-BlogCommendFromMachineLearnPai2-1.channel_param&depth_1-utm_source=distribute.pc_relevant.none-task-blog-BlogCommendFromMachineLearnPai2-1.cha1.nnel_param

1. Reinstall jetpack first

[Jetson Agx Orin] When executing the sudo apt install nvidia-jetpack command, an error message appears: E: Unable to locate package nvidia-jetpack

2. Check if there is /usr/local/cuda-11.4

jetson nano Check CUDA version: nvcc -V Error: bash: nvcc: command not found

At this time, switch to the ~ directory: cd ~;
then open the .bashrc file: vim .bashrc;
then press the i key to enter the editing state;
then add the following three lines of code at the end of the file:

export PATH=/usr/local/cuda-11.4/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/cuda-11.4/lib64:$LD_LIBRARY_PATH
export CUDA_ROOT=/usr/local/cuda-11.4

Then, press the Esc key, enter a colon, and then press wq! (which means force writing and exit)!

The last step is also a step that is easy to forget. Be sure to source this file:

source .bashrc

After everything above is OK, enter nvcc again and you can see the CUDA version information in the system:

3. Install torch 1.13.0 GPU version and torchvision

Install pytorch
jetson orin pytorch version download address
https://forums.developer.nvidia.com/t/pytorch-for-jetson/72048

Insert image description here

sudo apt-get install python3-pip libopenblas-base libopenmpi-dev libomp-dev
pip3 install Cython
pip3 install numpy torch-1.13.0-cp38-cp38-linux_aarch64.whl

b. Install torchvison

torchvision download URL:
https://github.com/pytorch/vision

Select main–>Tags 1.14.0 to download and unzip.

cd torchvision
export BUILD_VERSION=0.14.0  
python3 setup.py install --user
cd ../  
pip install 'pillow<7'

View version

python
import torch
import torchvision
torch.__version__
torchvision.__version__

Insert image description here

Possible errors

RuntimeError: Couldn’t load custom C++ ops. This can happen if your PyTorch and torchvision versions are incompatible

The specific error information is as follows

jk@jk-desktop:~/Desktop/work/python_project/yolov8_tracking$ python examples/track.py --source /home/jk/Desktop/work/python_project/yolov8_tracking/test_1.mp4 --tracking-method deepocsort --yolo-model /home/jk/Desktop/work/python_project/yolov8_tracking/yolov8/ultralytics/weights/yolov8s.pt
/home/jk/.local/lib/python3.8/site-packages/torchvision/io/image.py:13: UserWarning: Failed to load image Python extension:
  warn(f"Failed to load image Python extension: {e}")
track: yolo_model=/home/jk/Desktop/work/python_project/yolov8_tracking/yolov8/ultralytics/weights/yolov8s.pt, reid_model=/home/jk/Desktop/work/python_project/yolov8_tracking/examples/weights/lmbn_n_cuhk03_d.pt, tracking_method=deepocsort, source=/home/jk/Desktop/work/python_project/yolov8_tracking/test_1.mp4, imgsz=[640], conf=0.5, device=, show=False, save=False, classes=None, project=/home/jk/Desktop/work/python_project/yolov8_tracking/runs/track, name=exp, exist_ok=False, half=False, vid_stride=1, hide_label=False, hide_conf=False, save_txt=False
/home/jk/.local/lib/python3.8/site-packages/examples/weights
/home/jk/.local/lib/python3.8/site-packages/examples
add by xxx = osnet_x1_0_imagenet.pth
add by xxxx cached_file= /home/jk/.cache/torch/checkpoints/osnet_x1_0_imagenet.pth
Successfully loaded imagenet pretrained weights from "/home/jk/.cache/torch/checkpoints/osnet_x1_0_imagenet.pth"
Traceback (most recent call last):
  File "examples/track.py", line 246, in <module>
    main(opt)
  File "examples/track.py", line 241, in main
    run(vars(opt))
  File "/home/jk/.local/lib/python3.8/site-packages/torch/autograd/grad_mode.py", line 27, in decorate_context
    return func(*args, **kwargs)
  File "examples/track.py", line 114, in run
    predictor.results = predictor.postprocess(preds, im, im0s)
  File "/home/jk/.local/lib/python3.8/site-packages/ultralytics/yolo/v8/detect/predict.py", line 14, in postprocess
    preds = ops.non_max_suppression(preds,
  File "/home/jk/.local/lib/python3.8/site-packages/ultralytics/yolo/utils/ops.py", line 246, in non_max_suppression
    i = torchvision.ops.nms(boxes, scores, iou_thres)  # NMS
  File "/home/jk/.local/lib/python3.8/site-packages/torchvision/ops/boxes.py", line 40, in nms
    _assert_has_ops()
  File "/home/jk/.local/lib/python3.8/site-packages/torchvision/extension.py", line 48, in _assert_has_ops
    raise RuntimeError(
RuntimeError: Couldn't load custom C++ ops. This can happen if your PyTorch and torchvision versions are incompatible, or if you had errors while compiling torchvision from source. For further information on the compatible versions, check https://github.com/pytorch/vision#installation for the compatibility matrix. Please check your PyTorch version with torch.__version__ and your torchvision version with torchvision.__version__ and verify if they are compatible, and if not please reinstall torchvision so that it matches your PyTorch install.

PyTorch and torchvision versions do not match,
Insert image description here
but I printed it out and there is no problem with the version.

Finally uninstalled and reinstalled.

pip uninstall torch 
pip uninstall torchvision
pip uninstall torchaudio

Uninstall several times

pip uninstall torch 
pip uninstall torchvision
pip uninstall torchaudio

In fact, during the uninstallation process, I found that there are multiple versions of torchvision, which is probably caused by this reason.

Then follow step 3 and try again (reinstallation is quick)

Run successfully.