Linux Mint 22.3 (Zena); Python 3.12
I cant seem to get tensorflow[and-cuda] to work in Python - especially in a virtual environment.
Following the steps from https://www.tensorflow.org/install/pip
import tensorflow as tf
always ends with
Skipping registering GPU devices…
Lots of online posts say this is because the Nvidia libraries that have been installed are not in the PATH or maybe LD_LIBRARY_PATH. And (at least all of Google’s ) hints to fix this include code like
import nvidia.cuda_nvcc
print(nvidia.cuda_nvcc.__file__)
but this always prints ‘None’.
Again, according to Google, that is bcause “the nvidia-cuda-nvcc Python package is a namespace package used to distribute CUDA binaries via PyPI. It does not contain traditional Python scripts or an init.py file, meaning it lacks a file path”
and the hint to fix that takes you back to cudnn.__file__
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$(python -c 'import os; import nvidia.cudnn; print(os.path.dirname(nvidia.cudnn.__file__) + "/lib")')
Can anybody tell me exactly one file it might actually be looking for so I can find that file and manually set the PATH to that, or those, places?
Thanks
John C