version: "3.8"

services:
  llm-asyncio:
    build:
      context: .
      dockerfile: Dockerfile
    shm_size: "1000gb"
    volumes:
      - ./workspace:/opt/workspace/
      - ./cache:/root/.cache/
      - ../model:/opt/model/
    environment:
      PYTORCH_CUDA_ALLOC_CONF: expandable_segments:True
      PYTHONPATH: /opt/workspace/
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: "all"
              capabilities: [gpu]
    container_name: llm-asyncio
    ports:
      - "8000:8000"
    networks:
      - llm-network
    entrypoint: >
      /bin/bash -c "
      uvicorn main:app --reload --host 0.0.0.0 --port 8000
      "
    tty: true

  redis:
    image: redis:latest
    container_name: redis-server
    ports:
      - "6380:6380"
    restart: always
    networks:
      - llm-network

  worker:
    build:
      context: .
      dockerfile: Dockerfile
    shm_size: "1000gb"
    volumes:
      - ./workspace:/opt/workspace/
      - ./cache:/root/.cache/
      - ../model:/opt/model/
    environment:
      PYTORCH_CUDA_ALLOC_CONF: expandable_segments:True
      PYTHONPATH: /opt/workspace/
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: "all"
              capabilities: [gpu]
    networks:
      - llm-network
    entrypoint: >
      /bin/bash -c "
      python /opt/workspace/worker.py
      "
    restart: always
    tty: true
    scale: 2

networks:
  llm-network:
    driver: bridge