{ pkgs, service_configs, config, inputs, optimizePackage, ... }: { services.llama-cpp = { enable = true; model = builtins.toString ( pkgs.fetchurl { url = "https://huggingface.co/bartowski/google_gemma-3-27b-it-GGUF/resolve/main/google_gemma-3-27b-it-IQ2_XS.gguf"; sha256 = "080150a6758afefb40950f2753ea9f113b3cdc2b8ad4243a3711959748cd4df0"; } ); port = service_configs.ports.llama_cpp; host = "0.0.0.0"; package = ( optimizePackage ( inputs.llamacpp.packages.${pkgs.system}.default.overrideAttrs (old: { cmakeFlags = old.cmakeFlags ++ [ "-DGGML_AVX2=ON" ]; }) ) ); extraFlags = [ "--flash-attn" ]; }; services.caddy.virtualHosts."llm.${service_configs.https.domain}".extraConfig = '' ${builtins.readFile ../secrets/caddy_auth} reverse_proxy :${builtins.toString config.services.llama-cpp.port} ''; }