qwen my beloved <3, now declarative!

machines/homepc/configuration.nix

@@ -39,6 +39,9 @@
   # };
 
   # This is here because I don't have another computer that could run local AI, and regardless the packages would be different on every one.
+  # TODO: honestly, while I currently only have one pc that can run local AI, might change in the future.
+  # And this config is getting a bit complicated for a single pc config
+  # Should be moved to it's own shit
   environment.systemPackages = with pkgs; [
     ollama-cuda
     opencode
@@ -80,6 +83,20 @@
     };
   };
 
+  # As long as this is here the models are declarative. llama-server will grab them if not downloaded already.
+  environment.etc."llama-swap/config.yaml".text = ''
+    models:
+      "Qwen3.5-35B-A3B-GGUF":
+        cmd: llama-server --port ''${PORT} -hf unsloth/Qwen3.5-35B-A3B-GGUF:UD-Q4_K_XL --ctx-size 128000 --temp 0.6 --top-p 0.95 --top-k 20 --min-p 0.00 --cache-type-k f16 --cache-type-v f16 -np 1  --cpu-moe -fa on  --slots --slot-save-path /home/laythe/llamapcache  --jinja -kvu --no-mmproj --ubatch-size 16 --batch-size 16
+        ttl: 1200
+      "Qwen3-1.7B-GGUF":
+        cmd: llama-server --port ''${PORT} -hf unsloth/Qwen3-1.7B-GGUF:Q4_K_S --ctx-size 32768 --temp 0.6 --top-p 0.95 --top-k 20 --min-p 0.00 -fa on --jinja -kvu
+        ttl: 120
+      "Qwen3-8B-GGUF":
+        cmd: llama-server --port ''${PORT} -hf unsloth/Qwen3-8B-GGUF:Q4_K_S --ctx-size 32768 --temp 0.6 --top-p 0.95 --top-k 20 --min-p 0.00 -fa on --jinja -kvu -np 1
+        ttl: 120
+  '';
+
   # Set your time zone.
   # time.timeZone = "Europe/Amsterdam";