ONNX Runtime Installation Issue on Jetson Orin Nano with JetPack 5.1.2

Issue Overview

Users are experiencing difficulties installing ONNX Runtime on the Nvidia Jetson Orin Nano development board running JetPack 5.1.2. The primary symptom is an error message indicating that the ONNX Runtime wheel file is not supported on the platform. This issue occurs during the installation process using pip or pip3 commands. The problem affects multiple versions of ONNX Runtime, including 1.15.1 and 1.16.0, and appears to be related to Python version compatibility and wheel file naming conventions.

Possible Causes

1. Incorrect wheel file name: The downloaded wheel file may have an extra ‘m’ in its name, causing compatibility issues with newer Python versions.

2. Python version mismatch: The installed Python version may not match the version for which the wheel file was built.

3. JetPack version compatibility: There might be compatibility issues between the ONNX Runtime version and the installed JetPack version.

4. Pip version outdated: An outdated pip version could potentially cause installation problems.

5. Virtual environment conflicts: The use of virtual environments may lead to discrepancies in Python versions and package installations.

Troubleshooting Steps, Solutions & Fixes

1. Correct wheel file naming:

   • Ensure the wheel file is named correctly without the extra ‘m’ in the filename.
   • Rename the file from onnxruntime_gpu-1.16.0-cp38-cp38m-linux_aarch64.whl to onnxruntime_gpu-1.16.0-cp38-cp38-linux_aarch64.whl.

2. Verify Python version:

   • Check your Python version using:

     python3 --version
     

   • Ensure it matches the version specified in the wheel filename (e.g., cp38 for Python 3.8).

3. Use the correct pip version:

   • Verify pip version with:

     pip3 --version
     

   • If using a virtual environment, ensure you’re using the correct pip:

     which pip3
     

4. Install ONNX Runtime:

   • Download the correct wheel file:

     wget https://nvidia.box.com/shared/static/iizg3ggrtdkqawkmebbfixo7sce6j365.whl -O onnxruntime_gpu-1.16.0-cp38-cp38-linux_aarch64.whl
     

   • Install using pip3:

     pip3 install onnxruntime_gpu-1.16.0-cp38-cp38-linux_aarch64.whl
     

5. Verify installation:

   • After installation, check if ONNX Runtime can be imported:

     python3
     >>> import onnxruntime
     

6. Use virtual environments cautiously:

   • If using virtual environments, ensure you’re activating the correct one before installation.
   • Create a new virtual environment with the matching Python version if needed:

     python3.8 -m venv py38-venv
     source py38-venv/bin/activate
     

7. Update pip:

   • If you encounter any issues, try updating pip:

     pip install --upgrade pip
     

8. Check file integrity:

   • Verify the downloaded wheel file’s integrity using the provided checksum:

     md5sum onnxruntime_gpu-1.16.0-cp38-cp38-linux_aarch64.whl
     

   • Compare the output with the provided checksum: a50941a48fbd216da67be82a5314ee5f.

9. JetPack compatibility:

   • Confirm your JetPack version:

     sudo apt-cache show nvidia-jetpack
     

   • Ensure you’re using the ONNX Runtime version compatible with JetPack 5.1.2.

10. Alternative Python versions:

    • If issues persist with Python 3.8, consider trying Python 3.10 if available on your system, as some users reported success with this version.

By following these steps, users should be able to successfully install ONNX Runtime on their Jetson Orin Nano with JetPack 5.1.2. If problems persist, it may be necessary to consult NVIDIA’s official documentation or seek support from the Jetson community forums.