An AI Agent that demonstrates the principles and performance of the rStar-Math framework, with capabilities to generate integration code for other chatbots and AI agents.

The development of this GitHub Repository was inspired by the "rStar-Math: Small LLMs Can Master Math Reasoning with Self-Evolved Deep Thinking" paper. To read the full paper, visit https://arxiv.org/pdf/2501.04519

• Core Components

  • Monte Carlo Tree Search (MCTS) for step-by-step reasoning
  • Process Preference Model (PPM) for evaluating solution quality
  • Flexible model interface supporting multiple LLMs

• Model Support

  • OpenAI (GPT-4, GPT-3.5)
  • Anthropic (Claude)
  • Mistral AI
  • Groq
  • Google Gemini
  • Local models via llama.cpp

• Integration Templates

  • Rasa chatbot framework
  • LangChain
  • Azure Bot Framework
  • Streamlit
  • Gradio

• Example Notebooks

  • Calculus with visualizations
  • Geometry and proofs
  • Linear algebra operations
  • Statistics and probability
  • Model comparison studies

• Development Tools

  • Comprehensive test suite
  • Performance benchmarking
  • Visualization components
  • API documentation

pip install rstar-math

1. Clone the repository:

git clone https://github.com/yourusername/rStar-Math.git cd rStar-Math

2. Create a virtual environment:

python -m venv venv # Windows venv\Scripts\activate # Unix/MacOS source venv/bin/activate

3. Install dependencies:

pip install -r requirements.txt

4. Install in development mode:

pip install -e .

Create a .env file in the project root:

OPENAI_API_KEY=your_openai_key ANTHROPIC_API_KEY=your_anthropic_key MISTRAL_API_KEY=your_mistral_key GROQ_API_KEY=your_groq_key GEMINI_API_KEY=your_gemini_key

python examples/gradio_integration.py

streamlit run examples/streamlit_integration.py

# Start Jupyter server jupyter lab # Navigate to examples/notebooks/ # Open any of: # - calculus_examples.ipynb # - geometry_examples.ipynb # - linear_algebra_examples.ipynb # - statistics_examples.ipynb

# Run all tests pytest tests/ # Run specific test suite pytest tests/test_new_models.py # Run with coverage report pytest --cov=src tests/

# Run full benchmark suite python tools/benchmark.py # View results in browser python -m http.server 8000 # Open http://localhost:8000/benchmark_results/

# In your Rasa project pip install rstar-math cp examples/rasa_integration.py actions/

from examples.langchain_integration import RStarMathChain chain = RStarMathChain()

# In your Azure Bot project pip install rstar-math cp examples/azure_bot_integration.py bot/

1. Download a compatible model:

# Example: Download LLaMA model wget https://huggingface.co/models/llama-7b/resolve/main/model.bin -O models/llama-7b.bin

2. Run with local model:

from examples.llama_cpp_integration import LlamaCppModel model = LlamaCppModel("models/llama-7b.bin")

from rstar_math.core import MCTS, PPM from rstar_math.models import ModelFactory # Initialize components mcts = MCTS.from_config_file('config/default.json') ppm = ProcessPreferenceModel.from_config_file('config/default.json') model = ModelFactory.create_model('openai', 'YOUR_API_KEY', 'config/default.json') # Solve a problem problem = "What is the derivative of f(x) = x^2 + 3x?" action, trajectory = mcts.search(problem) # Print solution steps with confidence scores for step in trajectory: confidence = ppm.evaluate_step(step['state'], model) print(f"Step: {step['state']}") print(f"Confidence: {confidence:.2f}\n")

from examples.gradio_integration import RStarMathGradio # Launch Gradio interface demo = RStarMathGradio() demo.launch()

from examples.rasa_integration import RStarMathAction # Use in Rasa custom action action = RStarMathAction() await action.run(dispatcher, tracker, domain)

from examples.llama_cpp_integration import LlamaCppModel # Initialize local model model = LlamaCppModel("path/to/model.bin") response = model.generate_response("What is 2 + 2?")

• API Reference
• Model Integration Guide
• Example Notebooks
• Benchmark Results

Run performance benchmarks:

python tools/benchmark.py

This will generate:

• Execution time comparisons
• Memory usage analysis
• Token count statistics
• Confidence score trends

1. Fork the repository
2. Create your feature branch
3. Run tests: pytest tests/
4. Submit a pull request

MIT License - see LICENSE file for details

If you use rStar-Math in your research, please cite:

@article{rstar2024, title={rStar-Math: Small LLMs Can Master Math Reasoning with Self-Evolved Deep Thinking}, author={Original Authors}, journal={arXiv preprint}, year={2024} }

This project is inspired by the paper "rStar-Math: Small LLMs Can Master Math Reasoning with Self-Evolved Deep Thinking" (https://arxiv.org/pdf/2501.04519).

rStar-Math/ ├── src/ # Source code │ ├── core/ # Core components (MCTS, PPM) │ ├── models/ # Model implementations │ └── utils/ # Utility functions ├── tests/ # Test suites ├── examples/ # Example integrations │ ├── notebooks/ # Jupyter notebooks │ └── frameworks/ # Framework integrations ├── docs/ # Documentation ├── tools/ # Development tools └── config/ # Configuration files 

1. Create a new feature branch:

git checkout -b feature/your-feature-name

2. Make changes and run tests:

# Format code black src/ tests/ # Run linter flake8 src/ tests/ # Run tests pytest tests/

3. Submit a pull request:

git add . git commit -m "feat: your feature description" git push origin feature/your-feature-name

1. API Key Issues:

# Check if keys are loaded python -c "import os; print(os.getenv('OPENAI_API_KEY'))"

2. Model Loading Issues:

# Verify model files ls models/

3. CUDA Issues:

# Check CUDA availability python -c "import torch; print(torch.cuda.is_available())"

For more issues, check the troubleshooting guide.