x-gpu: &gpu
  deploy:
    resources:
      reservations:
        devices:
          - driver: nvidia
            device_ids:
              - ${HORSEFLOW_GPU}
            capabilities:
              - gpu

services:
  ollama:
    <<: *gpu
    image: ollama/ollama:0.34.1
    environment:
      OLLAMA_KEEP_ALIVE: -1
    volumes:
      - ${HORSEFLOW_OLLAMA_MODELS}:/root/.ollama
    restart: unless-stopped

  api:
    <<: *gpu
    build:
      context: ..
      dockerfile: server/Dockerfile
    environment:
      OLLAMA_URL: http://ollama:11434
      HORSEFLOW_LLM_MODEL: ${HORSEFLOW_LLM_MODEL}
      HORSEFLOW_ASR_MODEL: ${HORSEFLOW_ASR_MODEL}
      HORSEFLOW_LANGUAGE: ${HORSEFLOW_LANGUAGE}
      HORSEFLOW_LANGUAGE_NAME: ${HORSEFLOW_LANGUAGE_NAME}
      HORSEFLOW_COMPUTE_TYPE: ${HORSEFLOW_COMPUTE_TYPE}
      HORSEFLOW_DICTIONARY: ${HORSEFLOW_DICTIONARY}
      HORSEFLOW_ASR_PROMPT: ${HORSEFLOW_ASR_PROMPT}
    volumes:
      - ${HORSEFLOW_ASR_MODELS}:/models
    ports:
      - ${HORSEFLOW_BIND_ADDRESS}:8100:8100
    depends_on:
      - ollama
    restart: unless-stopped
    healthcheck:
      test:
        - CMD
        - python3
        - -c
        - >-
          import urllib.request;
          urllib.request.urlopen("http://127.0.0.1:8100/health", timeout=2)
      interval: 30s
      timeout: 5s
      retries: 3
      start_period: 120s