30 lines
755 B
Nix
30 lines
755 B
Nix
{
|
|
pkgs,
|
|
service_configs,
|
|
config,
|
|
inputs,
|
|
...
|
|
}:
|
|
{
|
|
services.llama-cpp = {
|
|
enable = true;
|
|
model = builtins.toString (
|
|
pkgs.fetchurl {
|
|
url = "https://huggingface.co/bartowski/google_gemma-3-12b-it-GGUF/resolve/main/google_gemma-3-12b-it-IQ4_XS.gguf";
|
|
sha256 = "aa7b7ae0b17931c379ede82da59b01f246046925aeb752af1ab4285a3b0d69db";
|
|
}
|
|
);
|
|
port = service_configs.ports.llama_cpp;
|
|
host = "0.0.0.0";
|
|
package = inputs.llamacpp.packages.${pkgs.system}.default;
|
|
extraFlags = [
|
|
|
|
];
|
|
};
|
|
|
|
services.caddy.virtualHosts."llm.${service_configs.https.domain}".extraConfig = ''
|
|
${builtins.readFile ../secrets/caddy_auth}
|
|
reverse_proxy :${builtins.toString config.services.llama-cpp.port}
|
|
'';
|
|
}
|