Cuda enabled Tensorflow and/or Pytorch shell

I have tried everything to get a working python setup with Cuda Supported Pytorch and Tensorflow. It seems that the pytorch-bin package is broken. (Torch-bin depends on triton). My PC has not enough resources to build it From scratch so i need to rely on a Binary Cache.

Could anyone Post their working Pytorch/Tensorflow Cuda shell.nix, their Cache/Cachix Setup and the important parts of their system Configuration. And the Branch that they are on?

Thank you in advance.

I tried to get a working tensorflow shell (without cuda) today and it seems the packaged version is currently broken as well. So the best solution is to bump those packages or get someone to do so (especially as you lack the resources to compile them). The only current workarounds i can think of are using a docker-container as a build environment or installing via pip in a venv.

1 Like

Do you have a working pip setup? because when i tried to use Pip my GPU wasn’t recognised

I did go the container route and had to find out that tensorflow depends on AVX which my Intel P8600 doesn’t support.