ha

modules/nixos/services/ai/default.nix

@@ -7,12 +7,11 @@
 }:
 with lib;
 let
-  name = "ai";
-  cfg = config.${namespace}.services.${name};
+  cfg = config.${namespace}.services.ai;
 
   aiConfig = lib.${namespace}.mkModule {
-    inherit config name;
-    serviceName = "open-webui"; # todo multiple?
+    inherit config;
+    name = "ai";
     description = "AI Services";
     options = { };
     moduleConfig = {
@@ -43,14 +42,25 @@ let
             "--seed"
             "3407"
             "--temp"
-            "1.0"
+            "0.7"
             "--top-p"
-            "0.95"
+            "0.9"
             "--min-p"
-            "0.01"
+            "0.05"
             "--top-k"
-            "40"
+            "30"
             "--jinja"
+            "--ctx-size"
+            "4096"
+            "--threads"
+            "8"
+            "--batch-size"
+            "512"
+            "--gpu-layers"
+            "999"
+            "--flash-attn"
+            "auto"
+            "--mlock"
           ];
         };
 
@@ -79,16 +89,52 @@ let
           };
         };
       };
+
+      # Model update script using HuggingFace Hub
+      environment.systemPackages = with pkgs; [
+        amdgpu_top
+        python3Packages.huggingface-hub
+      ];
+
+      # Systemd service for automatic model updates
+      systemd.services.update-qwen-model = {
+        description = "Update Qwen3-Coder-Next model from HuggingFace";
+        serviceConfig = {
+          Type = "oneshot";
+          ExecStart = "${pkgs.writeShellScript "update-qwen-model" ''
+            set -euo pipefail
+
+            MODEL_DIR="${cfg.configDir}/llama-cpp/models"
+            MODEL_NAME="Qwen3-Coder-Next-Q4_0.gguf"
+            REPO_ID="unsloth/Qwen3-Coder-Next-GGUF"
+
+            # Create model directory if it doesn't exist
+            mkdir -p "$MODEL_DIR"
+
+            # Download the latest version of the model
+            echo "Updating $MODEL_NAME from HuggingFace..."
+            ${pkgs.python3Packages.huggingface-hub}/bin/huggingface-cli download \
+              "$REPO_ID" \
+              "$MODEL_NAME" \
+              --local-dir "$MODEL_DIR"
+
+            echo "Model updated successfully"
+          ''}";
+          User = "nix-apps";
+          Group = "jallen-nas";
+        };
+        # Run daily at 3 AM
+        startAt = "*-*-* 03:00:00";
+      };
+
+      # Ensure model is available before llama-cpp starts
+      systemd.services.llama-cpp = {
+        after = [ "update-qwen-model.service" ];
+        wants = [ "update-qwen-model.service" ];
+      };
     };
   };
 in
 {
   imports = [ aiConfig ];
-
-  config = lib.mkIf cfg.enable {
-    environment.systemPackages = with pkgs; [
-      amdgpu_top
-      python3Packages.huggingface-hub
-    ];
-  };
 }