ollama

hosts/nas/apps/ollama/default.nix

@@ -1,4 +1,17 @@
-{ config, ... }:
+{ config, pkgs, ... }:
+let
+  llamaPackage = pkgs.llama-cpp.overrideAttrs (old: {
+    src = pkgs.fetchFromGitHub {
+      owner  = "ggml-org";
+      repo   = "llama.cpp";
+      rev    = "b4920";
+      sha256 = "sha256-SnQIeY74JpAPRMxWcpklDH5D4CQvAgi0GYx5+ECk2J4=";
+    };
+    # Optionally override other attributes if you need to
+    # version = "my-fork-version";
+    # pname = "llama-cpp-custom";
+  });
+in
 {
   services.ollama = {
     enable = true;
@@ -11,12 +24,15 @@
     home = "/media/nas/ssd/nix-app-data/ollama";
   };
 
+  environment.systemPackages = [ llamaPackage ];
+
   services.llama-cpp = {
     enable = true;
     port = 8127;
     host = "0.0.0.0";
     openFirewall = true;
     model = "/media/nas/ssd/nix-app-data/llama-cpp/models/functionary-small-v3.2-GGUF/functionary-small-v3.2.Q4_0.gguf";
+    package = llamaPackage; # pkgs.unstable.llama-cpp;
     extraFlags = [
       "--n_gpu-layers"
       "500"
@@ -24,7 +40,8 @@
       "0"
       "--numa"
       "numactl"
-    ]
+      "--jinja"
+    ];
   };
 
   services.open-webui = {