I’ve built tensorflow from source on a drive AGX machine (call it system-A) and generated a whl file. I’ve verified that it works correctly on system-A. However, when I want to use the whl file to install tensorflow on an identical machine (system-B), I get an error message that
pip install tensorflow-1.11.0-cp27-cp27mu-linux_aarch64.whl
ERROR: tensorflow-1.11.0-cp27-cp27mu-linux_aarch64.whl is not a supported wheel on this platform.
I’ve verified that the OS, driveworks, python, pip, cuda, cudnn, tensorRT versions and other libs are identical on the 2 machines. Is there some reason why the tensorflow whl install is not transferable?
What’s even more puzzling is that if I reflash system-A, the same whl file no longer works for tensorflow installation, so I’m having to build from source each time, leading to a lot of overhead. Any help is appreciated