docker-compose.yml

# Single-instance F5-TTS server.
# For multi-instance scale-out (4× on a 16-core box, agoracosmica production
# pattern), see docker-compose.scale.yml.

services:
  f5-server:
    build:
      context: .
      dockerfile: Dockerfile
    image: chipmates/f5-server:latest
    ports:
      - "8000:8000"
    environment:
      # Checkpoint — auto-downloads from HuggingFace on first start.
      # Known options: "hvoss" (default), "aihpi". Both CC-BY-NC-4.0.
      # Pass an absolute path (e.g. /models/my_checkpoint.safetensors) to use
      # a custom commercially-licensed checkpoint.
      F5_DEFAULT_CKPT: "hvoss"
      F5_DEFAULT_LANGUAGE: "de"

      # Speed: 1.0 = engine-native (~180 WPM DE). Lower = slower. Apply per-request
      # via the `speed` param of /v1/audio/speech, or set a default here.
      F5_SPEED: "1.0"

      # CORS — "*" for hobbyist single-machine use, scope down for production.
      # Example: "https://yourdomain.com,http://localhost:5173"
      CORS_ALLOW_ORIGINS: "*"

      # CPU thread limits per instance — keep low if running multiple instances.
      OMP_NUM_THREADS: "4"
      MKL_NUM_THREADS: "4"
    volumes:
      # Persist downloaded checkpoint across `docker compose down`.
      - f5_models:/models
    # GPU — uncomment if you have nvidia-container-toolkit installed:
    # deploy:
    #   resources:
    #     reservations:
    #       devices:
    #         - driver: nvidia
    #           count: 1
    #           capabilities: [gpu]
    restart: unless-stopped
    healthcheck:
      test: ["CMD", "curl", "-sf", "http://localhost:8000/health"]
      interval: 30s
      timeout: 10s
      start_period: 120s   # Cold start downloads ~1.35 GB checkpoint
      retries: 3

volumes:
  f5_models: