Getting TensorFlow running on AGX Xavier running L4T R31.1


I have a Jetson AGX Xavier flashed with an image running L4T R31.1. I do not want to reflash or change this.

I am trying to install CUDA + cuDNN to use tensorflow on the Jetson module. When I use JetPack4.3 for this, it throws errors and terminates without installation (I assume this is due to an older L4T version).

Questions are: (1) How can I get these packages (CUDA + cuDNN) on my module (is there a way to build from source, should I use a different JetPack version? )

(2) Which versions of CUDA, cuDNN, and tensorflow should I be looking to install so that it runs error-free (Will newer versions of these packages run on L4T R31.1?) ?

R31.1 is developer preview. It is to let users try Xavier and not stable. We would suggest you upgrade to r32.2.3 or r32.3.1 for further development.