Closed lfdominguez closed 9 months ago
It sounds to me like you are using ollama-flake as a flake input and overriding nixpkgs with an older version than what's expected. Like this:
ollama = {
url = "github:abysssol/ollama-flake";
inputs.nixpkgs.follows = "nixpkgs";
};
If that is the case, you have a few options:
nix flake lock --update-input nixpkgs
ollama = {
url = "github:abysssol/ollama-flake";
#inputs.nixpkgs.follows = "nixpkgs"; # remove this part
};
nix flake lock --update-input ollama
# at use of ollama-flake
(ollama.packages.${system}.cuda.override { cudaGcc = pkgs.gcc11; }) # use older version of gcc
Hopefully this is helpful and one of these options is relevant, but if not just leave some more details/context and I'll see how I can help.
Well that's the problem, inputs.nixpkgs.follows = "nixpkgs";
but I'm using a more recent nixpkgs, that's why ollama fail to build, I'm using gcc > 11 on my nixpkgs... Thanks
I believe that you are using an older version of nixpkgs because it contains a version of nvcc (nvidia cuda compiler) which requires a version of gcc <= 11. On the latest nixpkgs-unstable, nvcc works with gcc12, though not gcc13 (I had to override it to gcc12 in the flake). So, it seems that your version of nvcc hasn't been updated to support gcc12, which implies an older nvcc and nixpkgs.
Ahhh you mean nixpkgs unstable... I understand now, nop I'm on stable branch only hehehe thanks
Hi, I'm trying to compila the gpu version, but fail with this:
I have an NVidia 3060 Laptop.