llama swap

modules/hosts/desktop/configuration.nix

@@ -2,9 +2,8 @@
 # your system. Help is available in the configuration.nix(5) man page, on
 # https://search.nixos.org/options and in the NixOS manual (`nixos-help`).
 
-{ inputs, lib, pkgs, self, ... }:
-{
-flake.nixosModules.lily-desktop = { pkgs, ... }: {
+{ inputs, lib, pkgs, self, ... }:{
+flake.nixosModules.lily-desktop = { pkgs, ... }:{
   # Use the systemd-boot EFI boot loader.
   boot.loader = {
 	systemd-boot.enable = true;
@@ -73,7 +72,8 @@ flake.nixosModules.lily-desktop = { pkgs, ... }: {
 	vim
 	wget
 	home-manager
-   ] ++ [ self.packages.${pkgs.stdenv.hostPlatform.system}.nh ];
+	self.packages.${pkgs.stdenv.hostPlatform.system}.nh
+];
 
   # Some programs need SUID wrappers, can be configured further or are
   # started in user sessions.

modules/hosts/desktop/desktop.nix

@@ -4,6 +4,7 @@
 		modules = [ 
 			self.nixosModules.lily-desktop
 			self.nixosModules.lily
+			self.nixosModules.llama
 		];
 	};
 }

modules/nixosModules/ai/llama.nix

@@ -0,0 +1,29 @@
+{self, inputs, ...}: {
+	flake.nixosModules.llama = { lib, pkgs, ... }: {
+		nixpkgs.overlays = [
+			(final: prev: { 
+				llama-cpp-cuda = prev.llama-cpp.override {
+					cudaSupport = true;
+					blasSupport = true;
+				};
+			})
+		];
+		services.llama-swap = {
+			enable = true;
+			port = 9001;
+			openFirewall = true;
+			listenAddress = "0.0.0.0";
+			settings = {
+				models = {
+					"magidonia" = {
+						ttl = 3600;
+						cmd = "${pkgs.llama-cpp-cuda}/bin/llama-server --port \${PORT} -m /var/AI/Models/Chat/ggufs/Magidonia-24B-v4.3-Q4_K_M.gguf";
+					};
+				};
+			};
+		};
+		environment.systemPackages = with pkgs; [
+			llama-cpp-cuda
+		];
+	};
+}