Compatible jetpack, libnvinfer, cuda, cudnn and tensorflow versions

I am trying to install the right version of tensorflow for the given cuda, cudnn and libnvinfer libraries flashed onto my jetson xavier agx. I am not able to find anywhere how to relate these different libraries. I tried to install tensorflow 2.3.1 because thats the only latest version which supports cuda 10.2 which is flashed onto my jetson xavier. But I encounter “Could not load dynamic library 'libnvinfer.so.7” while trying to convert to a tensorrt model. I know that the libnvifer v8 has been flashed onto my jetson device but I am not sure which tensorflow version supports this libnvifer version. I dont know whether it is the jetpack version which I chose incorrectly or the tensorflow version itself, while installing tensorflow on the jetson. Apparently, I dont think there is any “concrete way” to determine this. I just need a comprehensive list about which of these library versions are compatible with each other. e.g like this “從原始碼開始建構  |  TensorFlow”. Along with this is there a way to know additional info like jetpack version, libnvifer versions for particular tensorflow/tensorrt versions?

Hi,

Please make sure you are using our prebuilt package.
You can find the installation steps in the below document:

All the available versions can be found via the corresponding URL.
For example JetPack 4.6

$ sudo pip3 install --pre --extra-index-url https://developer.download.nvidia.com/compute/redist/jp/**v46** tensorflow==

More, please noted that our latest v2.6.0+nv21.9 has some issues with cuTensor library.
It’s recommended to use other releases instead.

Thanks.

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.