Problems with install torch2trt on Jetson Orin NX

Robotics & Edge Computing Jetson Systems Jetson Orin NX
,
1

I want to install the torch2trt Python module in a virtual environment on my Jetson Orin NX (JetPack 6.2.1+b38). I’m following the instructions from the official repo (GitHub - NVIDIA-AI-IOT/torch2trt: An easy to use PyTorch to TensorRT converter).

Before installation, I installed TensorRT and its development packages:

sudo apt-get update
sudo apt-get install tensorrt nvidia-tensorrt-dev python3-libnvinfer-dev

However, when I run:

python3 setup.py install

I get this error:

Traceback (most recent call last):
  File "/home/jetson/torch2trt/setup.py", line 2, in <module>
    import tensorrt
  File "/usr/lib/python3.10/dist-packages/tensorrt/__init__.py", line 76, in <module>
    from .tensorrt import *
ImportError: libcudla.so.1: cannot open shared object file: No such file or directory

I’ve searched online without success. I initially suspected a typo (libcuda.so vs. libcudla.so), and even added a symlink, but the error persists.
How can I fix this and install torch2trt successfully on my Jetson?

4

Hi,

Please also install the CUDA library.

$ sudo apt show nvidia-cuda*

Then the libcudla.so can be found in the location below:

$ sudo find / -iname libcudla.so
/usr/local/cuda-12.6/targets/aarch64-linux/lib/libcudla.so

Thanks.