Trial Jan 15: @dusty_nv , thank you for the torchvision pip wheel. It installed successfully.
Installation output:
pip3 install torchvision-0.16.0+fbb4cc5-cp310-cp310-linux_aarch64.whl
Defaulting to user installation because normal site-packages is not writeable
Processing ./torchvision-0.16.0+fbb4cc5-cp310-cp310-linux_aarch64.whl
Requirement already satisfied: numpy in /usr/lib/python3/dist-packages (from torchvision==0.16.0+fbb4cc5) (1.21.5)
Requirement already satisfied: torch in /home/jet/.local/lib/python3.10/site-packages (from torchvision==0.16.0+fbb4cc5) (2.1.0)
Requirement already satisfied: pillow!=8.3.*,>=5.3.0 in /usr/lib/python3/dist-packages (from torchvision==0.16.0+fbb4cc5) (9.0.1)
Requirement already satisfied: requests in /usr/lib/python3/dist-packages (from torchvision==0.16.0+fbb4cc5) (2.25.1)
Requirement already satisfied: typing-extensions in /home/jet/.local/lib/python3.10/site-packages (from torch->torchvision==0.16.0+fbb4cc5) (4.9.0)
Requirement already satisfied: networkx in /home/jet/.local/lib/python3.10/site-packages (from torch->torchvision==0.16.0+fbb4cc5) (3.2.1)
Requirement already satisfied: sympy in /usr/lib/python3/dist-packages (from torch->torchvision==0.16.0+fbb4cc5) (1.9)
Requirement already satisfied: fsspec in /home/jet/.local/lib/python3.10/site-packages (from torch->torchvision==0.16.0+fbb4cc5) (2023.12.2)
Requirement already satisfied: jinja2 in /home/jet/.local/lib/python3.10/site-packages (from torch->torchvision==0.16.0+fbb4cc5) (3.1.3)
Requirement already satisfied: filelock in /home/jet/.local/lib/python3.10/site-packages (from torch->torchvision==0.16.0+fbb4cc5) (3.13.1)
Requirement already satisfied: MarkupSafe>=2.0 in /usr/lib/python3/dist-packages (from jinja2->torch->torchvision==0.16.0+fbb4cc5) (2.0.1)
Installing collected packages: torchvision
Successfully installed torchvision-0.16.0+fbb4cc5
Verification:
python3
Python 3.10.12 (main, Nov 20 2023, 15:14:05) [GCC 11.4.0] on linux
Type “help”, “copyright”, “credits” or “license” for more information.import torch
torch.version
‘2.1.0’
torch.cuda.is_available()
True
torch.backends.cudnn.version()
8904
import torchvision
torchvision.version
‘0.16.0+fbb4cc5’
Summary
For others experiencing the same issues in this thread, summary of Jan 15 Trial:
- Began with re-imaged SD card with Jetpack 6.0 DP
sudo apt upgdateandsudo apt upgrade- Installed Pytorch v2.1.0 via Pip Wheel method. I did not download v2.2.0 because a compatible version of torchvision was not available at the time of this writing
- I attempted torchvision v0.16.1 installation with Pip Wheel method after modifying
cpp_extension.py(see 2 replies above), but errors occurred and torchvision did not fully install. - Changed
cpp_extension.pyback to its original state - Installed Torchvision v0.16.0 with @dusty_nv Pip Wheel pulled from l4t-pytorch:r36.2.0 container (see previous reply to download wheel)
- Success!
The main goal was to install Hello AI World by Building the Project from Source, including the Transfer Learning with Pytorch. I went through the process this morning and it worked! Just make sure NOT TO RE-INSTALL PYTORCH/TORCHVISION on pain of breaking the installation summarized above. So there is no need to run install-pytorch.sh.