docker-compose.yml

version: "3.8"

# 霜云 AI 大模型 - Docker 编排配置
# Shimokumo AI Model - Docker Compose

services:
  shimokumo:
    build:
      context: .
      dockerfile: Dockerfile
    container_name: shimokumo-ai
    restart: unless-stopped
    ports:
      - "${SHIMOKUMO_PORT:-8080}:8080"
    volumes:
      - ./models:/app/models          # 模型权重目录
      - ./characters:/app/characters  # 角色配置目录
      - ./data:/app/data              # 数据目录
      - ./output:/app/output          # 输出目录
      - ./logs:/app/logs              # 日志目录
      - ./config.yaml:/app/config.yaml:ro  # 配置文件（只读）
    env_file:
      - .env
    environment:
      - SHIMOKUMO_MODEL_PATH=/app/models/shimokumo
      - SHIMOKUMO_CHARACTER_FILE=/app/characters/shimokumo.json
      - SHIMOKUMO_LOG_LEVEL=${SHIMOKUMO_LOG_LEVEL:-INFO}
      - NVIDIA_VISIBLE_DEVICES=all
    healthcheck:
      test: ["CMD", "python", "-c", "import requests; requests.get('http://localhost:8080/api/v1/health')"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 60s
    deploy:
      resources:
        limits:
          memory: 32G
        reservations:
          memory: 8G
    profiles:
      - cpu
      - gpu
    command: ["python", "main.py", "--host", "0.0.0.0", "--port", "8080"]

  # GPU 模式专用配置
  shimokumo-gpu:
    build:
      context: .
      dockerfile: Dockerfile.gpu
    container_name: shimokumo-ai-gpu
    restart: unless-stopped
    ports:
      - "${SHIMOKUMO_PORT:-8080}:8080"
    volumes:
      - ./models:/app/models
      - ./characters:/app/characters
      - ./data:/app/data
      - ./output:/app/output
      - ./logs:/app/logs
      - ./config.yaml:/app/config.yaml:ro
    env_file:
      - .env
    environment:
      - SHIMOKUMO_MODEL_PATH=/app/models/shimokumo
      - SHIMOKUMO_CHARACTER_FILE=/app/characters/shimokumo.json
      - SHIMOKUMO_LOG_LEVEL=${SHIMOKUMO_LOG_LEVEL:-INFO}
      - NVIDIA_VISIBLE_DEVICES=all
      - NVIDIA_DRIVER_CAPABILITIES=compute,utility
    healthcheck:
      test: ["CMD", "python", "-c", "import requests; requests.get('http://localhost:8080/api/v1/health')"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 120s
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]
    profiles:
      - gpu
    command: ["python", "main.py", "--host", "0.0.0.0", "--port", "8080"]