Error while installing torch2trt on Xavier NX

hi,
I’m using Xavier NX with jetpack 5.1 and
I was trying to clone and use “torch2trt” for real-time pose estimation on NVIDIA Jetson from link: “GitHub - NVIDIA-AI-IOT/trt_pose: Real-time pose estimation accelerated with NVIDIA TensorRT”.

But I got an error says which says

Traceback (most recent call last):  File "setup.py", line 2, in <module>    import torchImportError: No module named torch

But I have install torch and torchvision by following the nvidia forum “PyTorch for Jetson (forum 72048 )”

Hi,

Could you check if you can import torch with python3 directly?

$ python3
# import torch

Thanks.

yeah it shows

  import torch
  print(torch.__version__)
  print(torch.cuda.is_available())
  import torchvision
  print(torchvision.__version__)

also i have check there version and cuda enabled or not !

Hi,

Do you meet this error when running “sudo python3 setup.py install --plugins”?

Thanks.

yups…,but didn’t get error if I use Jetpack 4.6

Hi,

Could you check where the PyTorch is installed?

Torch2TRT is expected PyTorch is installed in /usr/local/lib/python3.8/dist-packages/torch.
But in some cases, it might be installed under $/home/.local/.

Thanks.

Is this still an issue to support? Any result can be shared? Thanks

its still an issue

i found torch on /home/nvidia/.local/lib/python3.8/site-packages

but when i run setup.py file it unable to find torch.
also when i normally use it other code, it runs properly

Hi,

also when i normally use it other code, it runs properly

Does this mean the issue is fixed?
Thanks.

yes…I have resolved this issue