Recently I have been working on Nvidia Jetson Nano. I have been trying to convert tensorflow model to tensorrt on Ubuntu to run the customized model on Jetson Nano.
I have been facing a challenge with version compatibility of tensorrt with tensorflow, cuda and cudnn.
The software versions I have are:
OS: Ubuntu desktop - 18.04 64bit
GPU: GeForce rtx 2060 super
Below I have attached an image that comprises the error I have got when I tried to verification of tensorrt.
Would any of you let me know what am I missing here? Is it tensorrt and tensorflow version compatibility problem?
According to tensorrt documentation tensorrt 6.0.1 has been tested with tensorflow 1.14.0 with cudnn 7.6.5.
But tensorflow documents says that tf 1.14 supports cudnn 7.4.
Am I missing on something?
Are there any other ways to convert the model?