search

Qengineering / Jetson-Nano-Ubuntu-20-image

Jetson Nano with Ubuntu 20.04 image
https://qengineering.eu/install-ubuntu-20.04-on-jetson-nano.html
BSD 3-Clause "New" or "Revised" License
646 stars 70 forks source link

Upgrading Tensorflow #52

Closed benjmerales closed 9 months ago

benjmerales commented 9 months ago

I have a saved H5 model that is built on Tensorflow 2.12. I successfully converted it to a compatible TensorRT version. But it seems there are some layers that 2.4 is yet to discover (Keras's keepdims keyword).

Is there right way to install the latest version of Tensorflow? One that can determine the Nano's GPU device. Since normally using pip install -U tensorflow will cause TF to restrict the GPU.

Qengineering commented 9 months ago

@benjmerales,

pip install will not give you CUDA acceleration.

I'm afraid, it's even worse.

Jetson Nano can't support TensorFlow > 2.4.1. with CUDA, as these TF versions need CUDA 11, which can't be installed on the nano due to low-level incompatibility.

benjmerales commented 9 months ago

Bummer. Guess my only solution now is to create a new model that can support TensorFlow 2.4.1, huh.

Thanks. Closing this issue now.