docker-compose.gpu.yml

# Docker Compose配置文件 - GPU部署
version: '3.8'

services:
  adaptive-rag:
    build:
      context: .
      dockerfile: Dockerfile.gpu
    container_name: adaptive-rag-gpu
    restart: unless-stopped
    environment:
      - CUDA_VISIBLE_DEVICES=0
      - PYTORCH_CUDA_ALLOC_CONF=max_split_size_mb:512
      - TOKENIZERS_PARALLELISM=false
      - HF_HOME=/app/models
      - TRANSFORMERS_CACHE=/app/models
    env_file:
      - .env
    ports:
      - "8000:8000"
      - "8001:8001"  # 可选：监控端口
    volumes:
      - ./data:/app/data
      - ./models:/app/models
      - ./logs:/app/logs
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu]
    depends_on:
      - ollama

  ollama:
    image: ollama/ollama:latest
    container_name: ollama-gpu
    restart: unless-stopped
    ports:
      - "11434:11434"
    volumes:
      - ollama-data:/root/.ollama
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu]
    command: ["ollama", "serve"]

  # 可选：监控服务
  nvidia-smi-exporter:
    image: mindprince/nvidia_gpu_prometheus_exporter:0.1
    container_name: gpu-monitor
    restart: unless-stopped
    ports:
      - "9445:9445"
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu]

volumes:
  ollama-data:
    driver: local