docker-compose.yml

version: '3.8'

services:
  # MLflow Tracking Server
  mlflow:
    image: python:3.10-slim
    container_name: llm-mlflow
    ports:
      - "5000:5000"
    volumes:
      - mlflow-data:/mlflow
      - ./mlruns:/app/mlruns
    environment:
      - MLFLOW_BACKEND_STORE_URI=sqlite:///mlflow/mlflow.db
      - MLFLOW_DEFAULT_ARTIFACT_ROOT=/mlflow/artifacts
    command: >
      sh -c "pip install mlflow && 
             mlflow server 
             --host 0.0.0.0 
             --port 5000 
             --backend-store-uri sqlite:///mlflow/mlflow.db 
             --default-artifact-root /mlflow/artifacts"
    networks:
      - llm-network
    restart: unless-stopped

  # Main Application (CPU version)
  app-cpu:
    build:
      context: .
      dockerfile: Dockerfile
      target: cpu
    container_name: llm-app-cpu
    ports:
      - "8501:8501"  # Streamlit
      - "8080:8080"  # Flask API
    volumes:
      - ./data:/app/data
      - ./cache:/app/cache
      - ./logs:/app/logs
      - ./outputs:/app/outputs
      - ./configs:/app/configs
    environment:
      - MLFLOW_TRACKING_URI=http://mlflow:5000
      - CUDA_VISIBLE_DEVICES=""  # Disable CUDA for CPU version
      - ENVIRONMENT=docker
    depends_on:
      - mlflow
    networks:
      - llm-network
    restart: unless-stopped
    profiles:
      - cpu

  # Main Application (GPU version)
  app-gpu:
    build:
      context: .
      dockerfile: Dockerfile
      target: gpu
    container_name: llm-app-gpu
    ports:
      - "8501:8501"  # Streamlit
      - "8080:8080"  # Flask API
    volumes:
      - ./data:/app/data
      - ./cache:/app/cache
      - ./logs:/app/logs
      - ./outputs:/app/outputs
      - ./configs:/app/configs
    environment:
      - MLFLOW_TRACKING_URI=http://mlflow:5000
      - NVIDIA_VISIBLE_DEVICES=all
      - ENVIRONMENT=docker
    depends_on:
      - mlflow
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]
    networks:
      - llm-network
    restart: unless-stopped
    profiles:
      - gpu

  # Redis for caching (optional)
  redis:
    image: redis:7-alpine
    container_name: llm-redis
    ports:
      - "6379:6379"
    volumes:
      - redis-data:/data
    command: redis-server --appendonly yes
    networks:
      - llm-network
    restart: unless-stopped
    profiles:
      - cache

  # PostgreSQL for advanced MLflow backend (optional)
  postgres:
    image: postgres:15-alpine
    container_name: llm-postgres
    ports:
      - "5432:5432"
    volumes:
      - postgres-data:/var/lib/postgresql/data
    environment:
      - POSTGRES_DB=mlflow
      - POSTGRES_USER=mlflow
      - POSTGRES_PASSWORD=mlflow123
    networks:
      - llm-network
    restart: unless-stopped
    profiles:
      - postgres

  # MLflow with PostgreSQL backend
  mlflow-postgres:
    image: python:3.10-slim
    container_name: llm-mlflow-postgres
    ports:
      - "5000:5000"
    volumes:
      - mlflow-artifacts:/mlflow/artifacts
    environment:
      - MLFLOW_BACKEND_STORE_URI=postgresql://mlflow:mlflow123@postgres:5432/mlflow
      - MLFLOW_DEFAULT_ARTIFACT_ROOT=/mlflow/artifacts
    command: >
      sh -c "pip install mlflow psycopg2-binary && 
             mlflow server 
             --host 0.0.0.0 
             --port 5000 
             --backend-store-uri postgresql://mlflow:mlflow123@postgres:5432/mlflow 
             --default-artifact-root /mlflow/artifacts"
    depends_on:
      - postgres
    networks:
      - llm-network
    restart: unless-stopped
    profiles:
      - postgres

  # Jupyter Lab for development (optional)
  jupyter:
    build:
      context: .
      dockerfile: Dockerfile
      target: development
    container_name: llm-jupyter
    ports:
      - "8888:8888"
    volumes:
      - ./:/app
      - ./notebooks:/app/notebooks
    environment:
      - JUPYTER_ENABLE_LAB=yes
      - MLFLOW_TRACKING_URI=http://mlflow:5000
    command: >
      sh -c "jupyter lab 
             --ip=0.0.0.0 
             --port=8888 
             --no-browser 
             --allow-root 
             --NotebookApp.token='' 
             --NotebookApp.password=''"
    depends_on:
      - mlflow
    networks:
      - llm-network
    restart: unless-stopped
    profiles:
      - dev

volumes:
  mlflow-data:
  mlflow-artifacts:
  postgres-data:
  redis-data:

networks:
  llm-network:
    driver: bridge

# Usage Examples:
# 
# CPU-only deployment:
# docker-compose --profile cpu up -d
#
# GPU deployment:
# docker-compose --profile gpu up -d
#
# Full development setup:
# docker-compose --profile gpu --profile dev --profile cache up -d
#
# Production with PostgreSQL:
# docker-compose --profile gpu --profile postgres up -d