add ollama back to artemision

2026-03-26 21:40:12 -04:00
parent 9bd6525dd0
commit cbc1a210dc
4 changed files with 45 additions and 27 deletions
--- a/systems/palatine-hill/ollama.nix
+++ b/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";