Refactor llama_server role for Windows support and EL-specific setup

- Added Windows-specific tasks for llama_server role, including installation of dependencies, setup of services using NSSM, and configuration of firewall rules.
- Introduced a new task file for Windows Exporter setup.
- Consolidated EL-specific setup tasks into a separate include file.
- Updated inventory to include additional host for llama server.
- Adjusted error handling in Prometheus service removal task to prevent failures.

ansible/host_vars/ci-01.lab.alexpires.me.yml

@@ -29,3 +29,58 @@ windows_exporter_collectors:
   - tcp
   - time
   - update
+
+llama_server_windows_install_dir: "D:\\llama_server"
+llama_server_windows_user: "llama"
+llama_server_windows_working_dir: "D:\\llama_server"
+llama_server_cuda_version: "13.3"
+
+llama_server_models:
+  - name: "luau-qwen3-4b-fim-v0.1"
+    model_id: "mradermacher/Luau-Qwen3-4B-FIM-v0.1-GGUF"
+    quant: "IQ4_XS"
+    revision: "main"
+  - name: "ministral-3-3b-instruct"
+    model_id: "unsloth/Ministral-3-3B-Instruct-2512-GGUF"
+    quant: "IQ4_XS"
+    revision: "main"
+  - name: "qwen2.5-coder-1.5b-instruct"
+    model_id: "Qwen/Qwen2.5-Coder-1.5B-Instruct-GGUF"
+    quant: "Q4_K_M"
+    revision: "main"
+  - name: "qwen2.5-coder-3b-instruct"
+    model_id: "unsloth/Qwen2.5-Coder-3B-Instruct-128K-GGUF"
+    quant: "Q4_K_M"
+    revision: "main"
+
+llama_server_host: "0.0.0.0"
+
+llama_server_windows_sources:
+  - 10.19.4.136
+
+llama_server_args_extra:
+  [
+    "-t",
+    8,
+    "-np",
+    1,
+    "-c",
+    8192,
+    "-b",
+    1024,
+    "-ub",
+    512,
+    "-fa",
+    "on",
+    "-ctk",
+    "q4_0",
+    "-ctv",
+    "q4_0",
+    "-cram",
+    0,
+    "-fit",
+    "off",
+    "--mmap",
+  ]
+
+llama_server_windows_backend: "vulkan"

ansible/host_vars/gpu-01.lab.alexpires.me.yml

@@ -168,6 +168,37 @@ llama_server_models:
     quant: "IQ4_XS"
     revision: "main"
 
+llama_server_args_extra:
+  [
+    "-t",
+    10,
+    "-np",
+    1,
+    "-b",
+    1024,
+    "-ub",
+    512,
+    "-fa",
+    "on",
+    "-ctk",
+    "q4_0",
+    "-ctv",
+    "q8_0",
+    "-cram",
+    -1,
+    "-sm",
+    "layer",
+    "-dev",
+    "Vulkan1,Vulkan2",
+    "-ts",
+    "8,12",
+    "-fit",
+    "off",
+    "-mg",
+    1,
+    "--mmap",
+  ]
+
 # # Optional global model preset options applied under [*] in models.ini.
 # # Keys map to llama.cpp CLI args without leading dashes, short args, or LLAMA_ARG_* env names.
 # llama_preset_global: