Pytorch Incompatibility Issue on Jetson Orin

Possible Problem 1:

GPU is unavailable

Install the Jetson version of PyTorch

Tutorial: https://docs.nvidia.com/deeplearning/frameworks/install-pytorch-jetson-platform/index.html

You can choose the version yourself: https://developer.download.nvidia.cn/compute/redist/jp/

The version I am using:

torch-2.5.0a0+872d972e41.nv24.08.17622132-cp310-cp310-linux_aarch64.whl

After downloading, run:

export TORCH_INSTALL=torch-2.5.0a0+872d972e41.nv24.08.17622132-cp310-cp310-linux_aarch64.whl
pip install --no-cache $TORCH_INSTALL

Test:

import torch

torch.cuda.is_available()

Possible Problem 2:

ImportError: libcusparseLt.so.0: cannot open shared object file: No such file or directory

Solution:

Install the CUDA version corresponding to the cuSPARSELt library compatible with the Jetson Orin platform

Download Link:

https://developer.nvidia.com/cuda-12-6-0-download-archive?target_os=Linux&target_arch=aarch64-jetson&Compilation=Native&Distribution=Ubuntu&target_version=22.04&target_type=deb_local

https://developer.nvidia.com/cusparselt-downloads

Possible Problem 3:

torchvision is not available

Solution:

Manually install the matching vision version, torch 2.5 -> torchvision 0.20.0

git clone --branch v0.20.0 [https://github.com/pytorch/vision.git](https://link.zhihu.com/?target=https%3A//github.com/pytorch/vision.git)

Compile after downloading:

export BUILD_VERSION=0.20.0
python3 setup.py install --user

Reference Link: https://zhuanlan.zhihu.com/p/1933164131969659101