Error during optimisation of the model using TF-TRT

Autonomous Machines Jetson & Embedded Systems Jetson Nano
, , ,
1

HI NVIDIA Developers,
I installed the Jetpack 4.5.1 package. Thanks to this, it seems to me that TensorRT, Cudnn and cuda are already installed.
First, I used tensorflow1.15.5 +nv21.5 to do inference. It works despite the fact that some libraries seem to be missing.

screen code_jessy errors
screen code_jessy errors1850×920 390 KB

Is there a way to download them all?

Secondly, I then want to do some optimization of my starting model using TF-TRT. To do this, I use the code below:

optimisation_code
optimisation_code891×180 20.7 KB

Unfortunately, I have a problem that I can’t solve:

error_optimisation
error_optimisation1838×213 89.8 KB

I tried to install libnvinfer packages with:
sudo apt-get install python3 -libnvinfer-dev
But this did not solve my problem.

Would you also have a solution here on what might be missing because I am a bit lost.

Thanks in advance.

Paul Griffoul

3

Hi,

The CUDA related library (ex. cufft) is in /usr/local/cuda-10.2/lib64.
Could you try to export the environment variable to see if it works?

$  export LD_LIBRARY_PATH=/usr/local/cuda-10.2/lib64:$LD_LIBRARY_PATH

The 2nd TensorRT error is caused by the missing library.
Please try the command first to load the required cu* libraries.

Thanks.