services:
  orpheus-fastapi:
    container_name: orpheus-fastapi
    build:
      context: .
      dockerfile: Dockerfile.cpu  
    ports:
      - "5005:5005"
    env_file:
      - .env
    environment:
      - ORPHEUS_API_URL=http://llama-cpp-server:5006/v1/completions
    restart: unless-stopped
    depends_on:
      llama-cpp-server:
        condition: service_started

  llama-cpp-server:
    image: ghcr.io/ggml-org/llama.cpp:server
    ports:
      - "5006:5006"
    volumes:
      - ./models:/models
    env_file:
      - .env
    depends_on:
      model-init:
        condition: service_completed_successfully
    restart: unless-stopped
    command: >
      -m /models/${ORPHEUS_MODEL_NAME}
      --host 0.0.0.0
      --port 5006
      --ctx-size ${ORPHEUS_MAX_TOKENS}
      --n-predict ${ORPHEUS_MAX_TOKENS}
      --threads ${LLAMA_CPU_THREADS:-6}
      --threads-batch ${LLAMA_CPU_THREADS:-6}
      --rope-scaling linear
      --no-mmap
      --no-slots
      --no-webui
  model-init:
    image: curlimages/curl:latest
    user: ${UID}:${GID}
    volumes:
      - ./models:/app/models
    working_dir: /app
    command: >
      sh -c '
      if [ ! -f /app/models/${ORPHEUS_MODEL_NAME} ]; then
        echo "Downloading model file..."
        wget -P /app/models https://huggingface.co/lex-au/${ORPHEUS_MODEL_NAME}/resolve/main/${ORPHEUS_MODEL_NAME}
      else
        echo "Model file already exists"
      fi'
    restart: "no"