add ollama back to artemision

systems/artemision/configuration.nix

@@ -1,6 +1,7 @@
 {
   lib,
   pkgs,
+  config,
   ...
 }:
 {
@@ -17,6 +18,7 @@
     ./stylix.nix
     ./wifi.nix
     ./zerotier.nix
+    ../palatine-hill/ollama.nix
   ];
 
   time.timeZone = "America/New_York";
@@ -39,6 +41,20 @@
   sops.age.sshKeyPaths = [ "/etc/ssh/ssh_host_ed25519_key" ];
 
   services = {
+    ollama = {
+      package = lib.mkForce pkgs.ollama-rocm;
+      models = lib.mkForce "${config.services.ollama.home}/models";
+      loadModels = lib.mkForce [
+        "deepseek-r1:1.5b"
+        "qwen3-coder-next"
+        "lennyerik/zeta"
+        "nomic-embed-text:latest"
+        "glm-4.7-flash"
+        "magistral"
+        "devstral-small-2"
+        "starcoder2:7b"
+      ];
+    };
     flatpak.enable = true;
     calibre-web = {
       # temp disable this

systems/palatine-hill/ollama.nix

@@ -25,11 +25,12 @@ in
         "nemotron-cascade-2:30b"
         "magistral"
         "devstral-small-2"
+        "starcoder2:15b"
       ];
       models = vars.primary_ollama;
       environmentVariables = {
         FLASH_ATTENTION = "1";
-        OLLAMA_KV_CACHE_TYPE = "q8_0";
+        OLLAMA_KV_CACHE_TYPE = "q4_0";
         # Ollama memory configuration
         OLLAMA_MAX_LOADED_MODELS = "3";
         OLLAMA_MAX_QUEUE = "512";