version: '3.8'
services:
  rvllm:
    build:
      context: .
      dockerfile: Dockerfile
    ports:
      - "8000:8000"
      - "9090:9090"
    volumes:
      - ${MODEL_PATH:-./models}:/models
      - huggingface-cache:/root/.cache/huggingface
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]
    environment:
      - RUST_LOG=info
      - CUDA_VISIBLE_DEVICES=0
    command: ["serve", "--model", "${MODEL_NAME:-meta-llama/Llama-3.2-1B}", "--gpu-memory-utilization", "0.90"]

  # Python vLLM for comparison benchmarks
  rvllm-python:
    image: vllm/vllm-openai:latest
    ports:
      - "8001:8000"
    volumes:
      - ${MODEL_PATH:-./models}:/models
      - huggingface-cache:/root/.cache/huggingface
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]
    environment:
      - CUDA_VISIBLE_DEVICES=0
    command: ["--model", "${MODEL_NAME:-meta-llama/Llama-3.2-1B}", "--gpu-memory-utilization", "0.90"]

volumes:
  huggingface-cache: