On my custom Xavier AGX board, I have managed to succesfully install Pytorch v. 1.9.0 using the instructions found here.
wget https://nvidia.box.com/shared/static/h1z9sw4bb1ybi0rm3tu8qdj8hs05ljbm.whl -O torch-1.9.0-cp36-cp36m-linux_aarch64.whl
sudo apt-get install python3-pip libopenblas-base libopenmpi-dev
sudo pip3 install Cython
sudo pip3 install numpy torch-1.9.0-cp36-cp36m-linux_aarch64.whl
The following lines:
import torch
print (torch.__version__)
torch.cuda.is_available()
torch.cuda.get_device_name(0)
produce this output:
1.9.0
True
‘Xavier’
So, I guess, PyTorch has been successfully installed.
I then went on to install the corresponding TorchVision version, i.e. 0.10.0:
sudo apt-get install libjpeg-dev zlib1g-dev libpython3-dev libavcodec-dev libavformat-dev libswscale-dev
git clone --branch release/0.10 https://github.com/pytorch/vision torchvision
cd torchvision
export BUILD_VERSION=0.10.0
sudo python3 setup.py install --user
Below is part of the output messages I got during installation:
Installed /home/nvidia/.local/lib/python3.6/site-packages/torchvision-0.10.0a0+ca1a620-py3.6-linux-aarch64.egg
Processing dependencies for torchvision==0.10.0a0+ca1a620
Searching for pillow>=5.3.0
Reading Links for pillow
Downloading https://files.pythonhosted.org/packages/d0/57/4b81f4fb0f0f93425488fc4a4e77e34c946ff76a65722739cbad35ddfe34/Pillow-8.4.0-cp36-cp36m-manylinux_2_17_aarch64.manylinux2014_aarch64.whl#sha256=72cbcfd54df6caf85cc35264c77ede902452d6df41166010262374155947460c
Best match: Pillow 8.4.0
Processing Pillow-8.4.0-cp36-cp36m-manylinux_2_17_aarch64.manylinux2014_aarch64.whl
Installing Pillow-8.4.0-cp36-cp36m-manylinux_2_17_aarch64.manylinux2014_aarch64.whl to /home/nvidia/.local/lib/python3.6/site-packages
Adding Pillow 8.4.0 to easy-install.pth file
Installed /home/nvidia/.local/lib/python3.6/site-packages/Pillow-8.4.0-py3.6-linux-aarch64.eggSearching for torch==1.9.0
Best match: torch 1.9.0
Adding torch 1.9.0 to easy-install.pth file
Installing convert-caffe2-to-onnx script to /home/nvidia/.local/bin
Installing convert-onnx-to-caffe2 script to /home/nvidia/.local/bin
So, I guess that TorchVision installation is ok as well. But unfortunately this isn’t the case:
import torchvision
gives me the error:
Traceback (most recent call last):
File “”, line 1, in
ModuleNotFoundError: No module named ‘torchvision’
Does anyone can provide some hint on what could be the problem here?