With the SDK manager I installed this week Jetpack 6.1 and all additional runtime software proposed by the SDK Manager. I have a 8GB Jetson Orin Nano Developer Kit with 512GB M.2 SSD.
Now I have trouble installing the Tensorflow or Pytorch libraries. I have installed already Python 3.12 and via miniconda I created an environment as well installed some basic libraries (numpy, pandas, matplotlib …). The $PATH includes also the “conda directories”.
I wanted to use the pre-built TensorFlow wheel for Jetson. I downloaded from NVIDIA the tensorflow-2.16.1+nv24.08-cp310-cp310-linux_aarch64.whl, but the installation process stopped with the error “this wheel is not valid for the given system” (or something similar). Is there something you could recommend ?
Hi @joerg.hoenig, you mentioned creating an anaconda environment for Python 3.12, but those wheels are for Python 3.10 (in fact you will find all of our prebuilt wheels for JetPack 6 are for Python 3.10 because this is the default version of Python that comes with Ubuntu 22.04, which is what JetPack 6 currently ships with)
So I would recommend changing your Python environment to 3.10 and trying to reinstall them again. Or you can rebuilt PyTorch/ect for Python 3.12. You can find a bunch of containerized builders and wheels through GitHub - dusty-nv/jetson-containers: Machine Learning Containers for NVIDIA Jetson and JetPack-L4T
You have been correct, with a Python 3.10 environment both wheels could be used for installation without error messages.
In a test program running in Spyder, the tensorflow library could be imported.
Nevertheless, I could not run a simple shallow network for the MNIST data set. It seems, that there is at least a problem with CUDA.
But for this, I will rise another issue.
thank you very much for the quick reply what solved the installation issues.
I guess, that the CUDA installation should have been done during the initial software installation via the SDK Manager.
During the installation process it was a little bit confusing, that the Jetson started to boot the Linux even the Linux image transfer to the Jetson was not finished yet (at about 70%) maybe this prevents the complete software installation ? Should I disconnect from the DevKit all other connections beside the USB-C cable to the control PC ?