I just flashed my Xavier with Jetpack 4.2, and am trying to install Tensorflow for use with ROS. Unfortunately, that requires the use of python 2. With the previous version of Jetpack (4.1.1), I was able to install a prebuilt package for Tensorflow 1.12 with python2.7 support, but the repositories don’t seem to contain any Tensorflow packages (1.12 or 1.13) for python 2.7.
Are there plans to provide prebuilt wheels for this configuration?
These are the repositories I’m looking at:
Old (Jetpack 4.1.1): Index of /compute/redist/jp/v411/tensorflow-gpu
Current (Jetpack 4.2): https://developer.download.nvidia.com/compute/redist/jp/v42/tensorflow-gpu/
Hi,
From JetPack4.2, we will only provide the python3 package.
Maybe you can try to build it from source for the python2 support.
Thanks.
Hi AastaLLL,
I can look into doing a source build. Is there a reason python2 packages aren’t being provided? I’m worried that if Nvidia isn’t providing them, then building from source may prove difficult, and I don’t want to end up duplicating Nvidia’s work.
Hi,
Just because that the EOL of python2 is coming.
@AastaLLL, while python 2 is nearing end of life, many of us still depend on it for daily use in the ROS robotics community. Spending 24 hours compiling Tensorflow on the tiny processor of the Jetson Nano ruins my workflow. Is it really that big of an inconvenience for you to run another build for 2.7? Also, official EOL is not for another 6 months.
I too need these, what did you end up doing?