systems/palatine-hill/ollama.nix

@@ -12,19 +12,18 @@ in
       package = pkgs.ollama;
       syncModels = true;
       loadModels = [
-        "deepseek-r1:latest"
         "deepseek-r1:1.5b"
         "deepseek-r1:32b"
+        "deepseek-r1:70b"
         "qwen3"
         #"qwen3-coder-next"
-        "qwen3-coder"
+        "qwen3-coder-next"
         "lennyerik/zeta"
-        "qwen2.5-coder:1.5b-base"
         "qwen2.5-coder:14b"
+        "qwen2.5-coder:32b"
         "nomic-embed-text:latest"
         "llama4:scout"
         "mistral:7b"
-        "deepseek-v3.2:cloud"
         "minimax-m2.7:cloud"
         "lfm2:24b"
         "glm-4.7-flash"
@@ -34,9 +33,9 @@ in
         FLASH_ATTENTION = "1";
         OLLAMA_KV_CACHE_TYPE = "q8_0";
         # Ollama memory configuration
-        OLLAMA_MAX_LOADED_MODELS = "2";
-        OLLAMA_MAX_QUEUE = "4";
-        OLLAMA_NUM_PARALLEL = "2";
+        OLLAMA_MAX_LOADED_MODELS = "3";
+        OLLAMA_MAX_QUEUE = "512";
+        OLLAMA_NUM_PARALLEL = "3";
 
         # ROCm memory optimization
         #HIP_VISIBLE_DEVICES = "0";