Installing PyTorch from wheel fails.
Followed instructions in
for both 1.11 and 1.12
Also did:
sudo apt-get install python3-pip libopenblas-base libopenmpi-dev libomp-dev
(1) cannot install torchvision :
python3 setup.py (after export BUILD_VERSION=0.12.0)
OSError: libmpi.so.20: cannot open shared object file: No such file or directory
Below exists in the system:
Command:
I am using Orin . Linux 20.04. I downloaded the torch for 5.0.2 as recommend in the link mentioned for this version. Below are the details of my jetson box. I am using Python 3.8
I think, it does look to me that I am using Torch lib that is older. unfortunately, I would not like to use the container deployment.
Is there a torch wheel that uses libmpi.so.40 ?? Shouldnt 1.11 and 1.12 be created with libmpi.so.40!!! Is there a patch you can suggest?
NVIDIA Jetson AGX Orin
L4T 35.1.0 [ JetPack 5.0.2 ]
Ubuntu 20.04.4 LTS
Kernel Version: 5.10.104-tegra
CUDA 11.4.239
CUDA Architecture: 8.7
OpenCV version: 4.5.4
OpenCV Cuda: NO
CUDNN: 8.4.1.50
TensorRT: 8.4.1.5
Vision Works: NOT_INSTALLED
VPI: 2.1.6
Vulcan: 1.3.203
Yes, they were and those are the wheels that I use on JetPack 5 as well (with libmpi.so.40). You could try making a symbolic link from libmpi.so.40 → libmpi.so.20. Or you can use these newer PyTorch wheels for JetPack 5:
(2) Is there a problem with the part of the command in the document:
export “LD_LIBRARY_PATH=/usr/lib/llvm-8/lib:$LD_LIBRARY_PATH”;
(3) I did put a link . It got stuck in OSError: libcublas.so.10: cannot open shared object file: No such file or directory
I unzipped the wheel and the version file (version.py) inside the zip points to Cuda 10 whereas in Orin we have 11.4.
Can this be a problem that all libraries are messed up?