This Docker setup replaces the Kimi interface with Ollama and FinGPT for generating alpha factors using WorldQuant Brain API.

• Docker and Docker Compose installed
• WorldQuant Brain credentials
• At least 8GB RAM (16GB recommended for FinGPT model)
• 20GB+ free disk space for models

For GPU acceleration:

• NVIDIA GPU with CUDA support
• NVIDIA Docker runtime installed
• CUDA 12.1+ drivers
• At least 8GB GPU VRAM (16GB+ recommended)

1. Set up credentials:

   # Copy the example credentials file
   cp credential.example.txt credential.txt
   
   # Edit credential.txt with your WorldQuant Brain credentials
   # Format: ["username", "password"]
   

2. Create necessary directories:

   mkdir -p results logs
   

3. Build and run with Docker Compose:

   CPU Only:

   docker-compose up --build
   

   With GPU Acceleration:

   # Windows
   start_gpu.bat
   
   # Linux/Mac
   docker-compose -f docker-compose.gpu.yml up --build
   

1. naive-ollma: Main application container

   • Runs the alpha orchestrator with integrated workflow
   • Alpha generation using Ollama with llama3.2:3b
   • Alpha expression mining on promising alphas
   • Daily alpha submission with rate limiting
   • GPU acceleration support

2. ollama-webui (Optional): Web interface for Ollama

   • Access at http://localhost:3000
   • Monitor model status and chat with models

3. alpha-dashboard (Optional): Alpha Generator Dashboard

   • Access at http://localhost:5000
   • Real-time monitoring of GPU, Ollama, orchestrator status
   • Manual controls for mining and submission
   • Statistics and activity logs
   • Alpha generator logs with filtering
   • Real-time system activity monitoring

The system now includes an integrated orchestrator that manages:

• Concurrent Execution: Alpha generator and expression miner run simultaneously for maximum efficiency
• Continuous Alpha Generation: Uses Ollama with llama3.2:3b for generating alpha ideas
• Expression Mining: Automatically mines promising alphas every 6 hours to find parameter variations
• Daily Submission: Submits successful alphas once per day (2 PM) to avoid rate limits
• GPU Acceleration: Full GPU support for faster model inference
• Automated Workflow: No manual intervention required
• Concurrent Simulation Control: Configurable limit (default: 3) for concurrent simulations to respect API limits

• Continuous Mode (default): Runs all components with intelligent scheduling
• Daily Mode: Single daily workflow execution
• Individual Modes: Run specific components (generator, miner, submitter)

• ./credential.txt: WorldQuant Brain credentials (read-only)
• ./results/: Generated alpha results and batch files
• ./logs/: Application logs
• ollama_data: Persistent Ollama models and data

The alpha dashboard provides comprehensive monitoring and control:

• Main Dashboard: http://localhost:5000
• Ollama WebUI: http://localhost:3000
• Ollama API: http://localhost:11434

• GPU Status: Real-time GPU memory, utilization, temperature
• Ollama Status: Model loading status, API connectivity
• Orchestrator Status: Generation activity, mining schedule, submission status
• WorldQuant Status: API connectivity, authentication status
• Statistics: Total generated alphas, success rates, 24h metrics

• Generate Alpha: Trigger single alpha generation for testing
• Trigger Mining: Run alpha expression mining manually
• Trigger Submission: Submit successful alphas immediately
• Refresh Status: Update all metrics and logs

• Alpha Generator Logs: Filtered logs showing alpha generation activity
• System Logs: Complete system activity and Docker container logs
• Recent Activity: Timeline of recent events and activities

• Dashboard automatically refreshes every 30 seconds
• Real-time log updates
• Live status indicators

You can modify the docker-compose.yml to adjust:

• OLLAMA_HOST: Ollama host (default: 0.0.0.0)
• OLLAMA_ORIGINS: CORS origins (default: *)
• PYTHONUNBUFFERED: Python output buffering (default: 1)

The application accepts these arguments (modify in docker-compose.yml):

• --batch-size: Number of alphas per batch (default: 3)
• --sleep-time: Sleep between batches in seconds (default: 30)
• --max-concurrent: Maximum concurrent simulations (default: 3)
• --log-level: Logging level (default: INFO)
• --ollama-url: Ollama API URL (default: http://localhost:11434)
• --mode: Operation mode (continuous, daily, generator, miner, submitter)
• --mining-interval: Hours between mining runs (default: 6)

The Docker setup now runs the alpha generator and expression miner concurrently:

1. Alpha Generator: Continuously generates new alpha expressions using Ollama
2. Expression Miner: Monitors for promising alphas and mines variations
3. Coordination: Both components respect the --max-concurrent limit (default: 3)
4. File Sharing: Uses hopeful_alphas.json as the coordination mechanism
5. Process Management: Orchestrator manages both processes and restarts if needed

• Increased Throughput: Generator and miner work simultaneously
• Better Resource Utilization: No idle time waiting for sequential completion
• Faster Alpha Discovery: Mining starts as soon as promising alphas are found
• Resilient Operation: Automatic restart of failed components

# Start all services
docker-compose up -d

# View logs
docker-compose logs -f naive-ollma

# Stop services
docker-compose down

The system includes a comprehensive web dashboard for monitoring and control:

Access URLs:

• Alpha Dashboard: http://localhost:5000 (Main monitoring interface)
• Ollama WebUI: http://localhost:3000 (Ollama chat interface)
• Ollama API: http://localhost:11434 (API endpoint)

Dashboard Features:

• Real-time Status: GPU utilization, Ollama status, orchestrator activity
• Manual Controls: Trigger mining and submission operations
• Statistics: Alpha generation metrics and success rates
• Activity Logs: Recent system activity and error tracking
• Auto-refresh: Updates every 30 seconds automatically

Manual Controls:

• Trigger Mining: Manually start alpha expression mining
• Trigger Submission: Manually submit successful alphas
• Refresh Status: Update dashboard data immediately

1. Application Logs:

   docker-compose logs -f naive-ollma
   

2. Web Interface:

   • Open http://localhost:3000 for Ollama WebUI
   • Monitor model status and performance

3. Results:

   • Check ./results/ directory for batch results
   • Check hopeful_alphas.json for promising alphas

1. Ollama Model Issues:

   # Check if FinGPT model is available
   docker-compose exec naive-ollma ollama list
   
   # Pull model manually if needed
   docker-compose exec naive-ollma ollama pull fingpt
   

2. Authentication Issues:

   • Verify credential.txt format: ["username", "password"]
   • Check WorldQuant Brain API status

3. Resource Issues:

   • Increase Docker memory limit (8GB+ recommended)
   • Check available disk space for models

4. Network Issues:

   # Test Ollama connectivity
   curl http://localhost:11434/api/tags
   
   # Test WorldQuant Brain connectivity
   docker-compose exec naive-ollma python -c "
   import requests
   from requests.auth import HTTPBasicAuth
   import json
   
   with open('credential.txt') as f:
       creds = json.load(f)
   
   sess = requests.Session()
   sess.auth = HTTPBasicAuth(creds[0], creds[1])
   resp = sess.post('https://api.worldquantbrain.com/authentication')
   print(f'Auth status: {resp.status_code}')
   "
   

To use a different model instead of FinGPT:

1. Modify the generate_alpha_ideas_with_ollama method in alpha_generator_ollama.py
2. Change the model name in the Ollama API request
3. Update the Dockerfile to pull your preferred model

For production use:

1. Increase batch size:

   command: ["--batch-size", "10", "--sleep-time", "60"]
   

2. Add resource limits:

   deploy:
     resources:
       limits:
         memory: 16G
         cpus: '4.0'
   

3. Use external Ollama service:

   environment:
     - OLLAMA_API_BASE_URL=http://external-ollama:11434/api
   

1. Backup models:

   docker run --rm -v naive-ollma_ollama_data:/data -v $(pwd):/backup alpine tar czf /backup/ollama_backup.tar.gz -C /data .
   

2. Restore models:

   docker run --rm -v naive-ollma_ollama_data:/data -v $(pwd):/backup alpine tar xzf /backup/ollama_backup.tar.gz -C /data
   

1. Model Optimization:

   • Use quantized models for faster inference
   • Consider using smaller models for testing

2. Resource Management:

   • Monitor memory usage during model loading
   • Adjust batch sizes based on available resources

3. Network Optimization:

   • Use local Ollama instance for faster responses
   • Consider caching frequently used data fields

• Credentials are mounted as read-only
• Ollama API is exposed only locally
• Consider using Docker secrets for production credentials
• Regularly update base images and dependencies

For issues related to:

• Ollama: Check Ollama documentation
• FinGPT: Check FinGPT repository
• WorldQuant Brain: Check WorldQuant Brain documentation