---
# vLLM — high-performance LLM inference with OpenAI-compatible API
# 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:
    dest: /etc/systemd/system/vllm.service
    content: |
      [Unit]
      Description=vLLM OpenAI-Compatible Inference Server
      After=docker.service ollama.service
      Requires=docker.service

      [Service]
      Restart=always
      RestartSec=5
      ExecStartPre=-/usr/bin/docker stop vllm
      ExecStartPre=-/usr/bin/docker rm vllm
      ExecStart=/usr/bin/docker run \
        --name vllm \
        --gpus all \
        --ipc=host \
        -p 8000:8000 \
        -v /opt/cezen/models:/root/.cache/huggingface \
        -e HF_HOME=/root/.cache/huggingface \
        vllm/vllm-openai:latest \
        --model meta-llama/Meta-Llama-3.1-8B-Instruct \
        --gpu-memory-utilization 0.7 \
        --max-model-len 8192 \
        --tensor-parallel-size 1
      ExecStop=/usr/bin/docker stop vllm

      [Install]
      WantedBy=multi-user.target
    mode: "0644"

- name: Create vLLM model directory
  file:
    path: /opt/cezen/models/hf_cache
    state: directory
    owner: cezen
    group: cezen