Inference on Jetson nano but training on x86 machine with Nvidia GPU. Tensorflow version to use ?

Robotics & Edge Computing Jetson & Embedded Systems Jetson Nano
1

Tensor flow version to use for training on x86 desktop with Nvidia GPU ?

I am installing tensorflow-gpu==1.13.1+nv19.4 on my Jetson Nano (via the pip command mentioned in another thread). I intend to use Nano for inference on a robot .

For training I have an x86 desktop machine with an Nvidia GPU. Previously I had encountered issues when I used different tensorflow versions for training and inference. So I would like to avoid this pitfall.

To use the same version of tensor flow for training is it enough that I install via pip install tensorflow-gpu==1.13 on the desktop machine ? or should I worry about nv19.4 too ?

2

I’ve successfully transferred models to my Nano that were trained on TF 1.12 from my workstation and on TF 1.13 using Google Colab, so TF 1.13 on your desktop should be fine.

I think the ‘+nv19.4’ in the package name reflects the fact that you’re running an NVIDIA-enhanced version of TensorFlow on the Nano, so you can ignore it when training on other platforms.