services:
  qwen-tts:
    build:
      context: .
      dockerfile: Dockerfile
    image: qwen3-tts-rocm:latest
    container_name: qwen3_tts_app

    # Required for AMD ROCm hardware acceleration
    devices:
      - "/dev/kfd:/dev/kfd"
      - "/dev/dri:/dev/dri"

    # Adds the container to the host's video group to access the GPU render nodes
    group_add:
      - video

    # Solves potential shared memory issues during model inferencing
    ipc: host
    shm_size: "8gb"

    # Mounts current directory code into the container for easy development
    volumes:
      - .:/workspace
      # Optional: Persist Hugging Face model downloads so they don't re-download on restart
      - hf_cache:/root/.cache/huggingface

    # Ports for web UIs (like Gradio) or API servers (FastAPI)
    ports:
      - "7860:7860" # Gradio UI
      - "8000:8000" # FastAPI / Uvicorn Server

    # Keeps the container running interactively so you can execute scripts
    tty: true
    stdin_open: true

    # Change this command if you want to automatically start a specific server script
    #    command: /bin/bash

    environment:
      - MODEL_ID=Qwen/Qwen3-TTS-12Hz-1.7B-VoiceDesign
      - PORT=8000

    command: python3 app/server.py

  qwen-tts-06b:
    # Reuses the same image built by the `qwen-tts` service above (qwen3-tts-rocm:latest).
    # Loads the smaller 0.6B CustomVoice model (9 preset speakers + style control)
    # and serves it on a different port.
    image: qwen3-tts-rocm:latest
    container_name: qwen3_tts_app_06b

    devices:
      - "/dev/kfd:/dev/kfd"
      - "/dev/dri:/dev/dri"

    group_add:
      - video

    ipc: host
    shm_size: "8gb"

    # Shares the HF cache volume so the 0.6B model is downloaded only once
    # and reuses the 1.7B's cached files if/when they overlap.
    volumes:
      - .:/workspace
      - hf_cache:/root/.cache/huggingface

    ports:
      - "8001:8000" # FastAPI / Uvicorn Server (0.6B model; container listens on PORT=8000)

    tty: true
    stdin_open: true

    environment:
      - MODEL_ID=Qwen/Qwen3-TTS-12Hz-0.6B-CustomVoice
      - PORT=8000

    command: python3 app/server.py

volumes:
  hf_cache: