dotfiles/nixos/modules/ollama.nix

{
  flake,
  pkgs,
  ...
}:
let
  inherit (flake.config.services.instances)
    ollama
    web
    ;
  service = ollama;
  localhost = web.localhost.address0;
in
{
  services = {
    ollama = {
      acceleration = "rocm";
      package = pkgs.ollama.override {
        config = {
          rocmSupport = true;
          cudaSupport = false;
        };
      };
      enable = true;
      group = service.name;
      host = "http://${localhost}";
      # models = service.paths.path1;
      user = service.name;
    };
    # llama-cpp = {
    #   enable = true;
    #   port = 8080;
    #   host = localhost;
    #   model = "/models/qwen2.5-coder-32b-instruct-q8_0-00004-of-00005.gguf";
    # };
    open-webui = {
      enable = true;
      host = localhost;
      port = service.ports.port0;
      environment = {
        ENABLE_OLLAMA_API = "True";
        ANONYMIZED_TELEMETRY = "False";
        DO_NOT_TRACK = "True";
        SCARF_NO_ANALYTICS = "True";
        OLLAMA_BASE_URL = "http://${localhost}:${toString service.ports.port1}";
        WEBUI_AUTH = "True";
      };
    };
  };

  systemd.tmpfiles.rules = [
    "Z ${service.paths.path1} 0755 ${service.name} ${service.name} -"
  ];

  networking = {
    firewall = {
      allowedTCPPorts = [
        # 8080
        service.ports.port0
        service.ports.port1
      ];
    };
  };
}