# Four services across two axes: cpu/cuda backend, prod/dev deployment. The dev
# services bind-mount the source over /app; the cuda services need the
# nvidia-container-toolkit on the host.
services:
  parakeet-asr-cpu-prod:
    image: ghcr.io/taketwo/parakeet-asr:latest
    build:
      context: ..
      dockerfile: docker/Dockerfile
      args:
        ONNX_VARIANT: cpu
    pull_policy: always
    restart: unless-stopped
    ports:
      - "8000:8000"
    volumes:
      # Persist downloaded model weights across restarts.
      - ../data:/app/data
    profiles:
      - cpu-prod

  parakeet-asr-cuda-prod:
    image: ghcr.io/taketwo/parakeet-asr:cuda
    build:
      context: ..
      dockerfile: docker/Dockerfile
      args:
        ONNX_VARIANT: cuda
    pull_policy: always
    restart: unless-stopped
    ports:
      - "8000:8000"
    volumes:
      - ../data:/app/data
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]
    profiles:
      - cuda-prod

  parakeet-asr-cpu-dev:
    build:
      context: ..
      dockerfile: docker/Dockerfile
      args:
        ONNX_VARIANT: cpu
    ports:
      - "8000:8000"
    volumes:
      - ../data:/app/data
      - ..:/app
    profiles:
      - cpu-dev

  parakeet-asr-cuda-dev:
    build:
      context: ..
      dockerfile: docker/Dockerfile
      args:
        ONNX_VARIANT: cuda
    ports:
      - "8000:8000"
    volumes:
      - ../data:/app/data
      - ..:/app
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]
    profiles:
      - cuda-dev