In reply to @yorickvp:matrix.org

bisecting the following:

let
  pkgs = import ./. {
    config = {
      allowUnfree = true;
      cudaCapabilities = [ "8.6" ];
      cudaSupport = true;
    };
  };
in
{
  torchtest = (pkgs.python3.pkgs.torch.override { openai-triton = null; }).overridePythonAttrs (o: {
    disallowedReferences = [ pkgs.python3.pkgs.torch.cudaPackages.cuda_nvcc.stdenv.cc.cc.lib ];
    USE_CUDNN = 0;
    USE_KINETO = 0;
    USE_QNNPACK = 0;
    USE_PYTORCH_QNNPACK = 0;
    USE_XNNPACK = 0;
    INTERN_DISABLE_ONNX = 1;
    ONNX_ML = 0;
    USE_ITT = 0;
    USE_FLASH_ATTENTION = 0;
    USE_MEM_EFF_ATTENTION = 0;
    USE_FBGEMM = 0;
    USE_MKLDNN = 0;
  });
}

In reply to @glepage:matrix.org
https://github.com/mikex86/LibreCuda

In reply to @ss:someonex.net
yorickvp can you summarize your findings in a github issue, and link the pr?

In reply to @yorickvp:matrix.org
My bisect findings is that this has been happening ever since the mixed gcc was introduced

Hey connor (he/him) (UTC-5),
I have a quick question regarding the torch update.
The CUDA build of pytorch now fails because of the addition of a new cuda capability 9.0a.

Someone suggested the following solutions:
https://github.com/NixOS/nixpkgs/pull/329836#issuecomment-2279508588

What is your opinion ?

In reply to @glepage:matrix.org
Ok, thank you very much for sharing your patch !