AttributeError: module ‘torch._ C‘ has no attribute ‘_ cuda_ setDevice‘

stay Jetson Run on one PyTorch Reasoning program , After installing the basic environment , according to CUDA Version installed PyTorch.

CUDA Version is 11.4：

  And then through PyTorch Official website Get installation command line ：

  After successful installation , When running the inference program, there is Error：

test CUDA, Found not available ：

At first I suspected PyTorch Of CUDA Version of the problem , So we upgraded to 11.6, The problem remains .

The later positioning process ：

(1) JetPack And PyTorch Compatibility problem , Reference link PyTorch for Jetson - version 1.11 now available - Jetson Nano - NVIDIA Developer Forums

I tried to download the following file , But the prompt during installation is similar to Platform Mismatch .

torch-1.12.0a0+2c916ef.nv22.3-cp38-cp38-linux_aarch64.whl

(2) Python Version of the problem ： All materials mentioned that at present Jetson Only supported on Python3.6, And what we pretend is Python3.8. However, this step has not been verified , Later, it was located in the (3) After three reasons, the problem was solved directly .

(3) Ultimate solution

Finally, I found the web page ：PyTorch for Jetson Platform

This page gives JetPack Version and PyTorch Version correspondence ：

our JetPack It uses 5.0.1, Follow the table above , You should use 1.13 Of PyTorch edition , What we have been installing before is 1.12.

adopt This link Found a suitable for Jetson Of PyTorch1.13 Version download address ：https://developer.download.nvidia.cn/compute/redist/jp/v50/pytorch/

Download the required version and install , then , The problem was solved ！