Using TensorRT 7.1.3 (Jetpack4.4) with Python3.7 instead of 3.6 - how?

Hi everyone,

for a bigger project with ROS I developed inference code which needs to run under Python3 using the TensorRT (important: IT HAS TO BE TRT7.1.3 or lower) Python API. No problem whatsoever on the Jetson.

Now an other code module forces me to use Python3.7.

I created a Python3.7 virtualenv. But how exactly can I get it to use the Python API correctly. Without doing anything, when trying to import TensorRT, it can’t be found. If I add the dist-packages of Python3.6 to 3.7, I get the error that the python3-libnvinfer package was build with Python3.6 so it can’t be used with Python3.7.

When I try to install via pip (via nvidia-pyindex) I can’t get the correct TRT7.1.3 version - only 7.2 which will not fit my system nor my code.

On my x86 machine/container, I can simply use the .whl file provided in the TAR file (following the installation instructions: Installation Guide :: NVIDIA Deep Learning TensorRT Documentation) and simply use it to install TRT with pip in my virtual env.

But when it comes to ARM / Jetson / Jetpack, the only possibility to install TensorRT is to install the whole SDK. There is no TAR installation method for ARM based systems where I could copy the correct .whl file out of (The whl. files from the x86 TAR file are obviously just for x86 systems …)

Can I somewhere get the .whl file for TensorRT 7.1.3 on ARM/Jetson or does anyone has an other idea how to get TensorRT up and running using a Python3.7 virtual env?


GPU Type: Jetson AGX Xavier
Nvidia Driver Version:
CUDA Version: 10.2
CUDNN Version: 7.6.5
Operating System + Version: Jetpack 4.4
Python Version (if applicable): 3.7 (and 3.6)
TensorFlow Version (if applicable): -
PyTorch Version (if applicable): 1.6.0
Baremetal or Container (if container which image + tag): Container: l4t-base:r32.4.3

This looks like a Jetson issue. We recommend you to raise it to the respective platform from the below link



We only have the python package for v3.6.
For other version, you can build it with pybind11 on your own.

You can find the source in our OSS GitHub below: