Skip vLLM image pull when no GPU present

ansible/roles/vllm/tasks/main.yml

@@ -1,10 +1,23 @@
 ---
 # vLLM — high-performance LLM inference with OpenAI-compatible API
-# Entry tier: runs as a Docker container (easier to manage than pip install)
+# Skipped automatically if no GPU is present.
+
+- name: Check for NVIDIA GPU
+  shell: nvidia-smi --query-gpu=name --format=csv,noheader 2>/dev/null | head -1
+  register: gpu_check
+  ignore_errors: true
+  changed_when: false
+
+- name: Skip vLLM if no GPU detected
+  debug:
+    msg: "No GPU detected — skipping vLLM image pull. Run manually when GPU is available: docker pull vllm/vllm-openai:latest"
+  when: gpu_check.stdout == "" or gpu_check.rc != 0
+
 - name: Pull vLLM Docker image
   shell: docker pull vllm/vllm-openai:latest
   retries: 3
   delay: 15
+  when: gpu_check.stdout != "" and gpu_check.rc == 0
 
 - name: Create vLLM systemd service
   copy:
@@ -38,16 +51,6 @@
       WantedBy=multi-user.target
     mode: "0644"
 
-- name: Enable vLLM (but don't start yet — model selection needed first)
-  systemd:
-    name: vllm
-    enabled: yes
-    daemon_reload: yes
-  # Note: vLLM service is enabled but not started by default.
-  # Start manually after choosing a model:
-  #   sudo systemctl start vllm
-  # Or change the --model flag in /etc/systemd/system/vllm.service first.
-
 - name: Create vLLM model directory
   file:
     path: /opt/cezen/models/hf_cache