MathSolver

MathSolver is an advanced problem-solving tool that leverages Large Language Models (LLMs) with multiple strategies to solve mathematical problems. It supports various solving methods and integrates with popular LLM providers.

Paper

COMING SOON

Features

• Multiple Solving Strategies:

  • Direct solving
  • Divide-and-conquer approach
  • Graph-based problem decomposition

• LLM Provider Support:

  • OpenAI
  • Replicate
  • Ollama (local models)

• Evaluation Framework:

  • Benchmarking system
  • Accuracy metrics
  • Solution iteration tracking

• Flexible Integration:

  • Command-line interface (CLI)
  • Python API
  • Configurable providers

Installation

git clone https://github.com/JasonHonKL/mathsolver
cd mathsolver
pip install -e .

Currently it is not availbe in pypi but will be available soon

Usage

CLI

1. Configure your LLM provide

    mathsolver config --provider openai --api-key YOUR_API_KEY --model your-model

2. Solve a single problem

    mathsolver solve --problem "What is the integral of x^2 from 0 to 3?" --method divide-conquer

Run benchmarks

    mathsolver benchmark --benchmark-file problems.json --output-file results.json

Python API

from mathsolver import OpenAIProvider, DivideConquerSolver

# Initialize provider and solver
provider = OpenAIProvider(api_key="your_api_key")
solver = DivideConquerSolver(provider, max_iterations=5)

# Solve a problem
solution = solver.solve("Solve 3x + 2 = 14")
print(f"Answer: {solution['final_answer']}")
print(f"Iterations: {solution['iterations']}")
print(f"Subquestions: {solution['subquestions']}")

Configuration

mathsolver config \
  --provider ollama \
  --base-url http://localhost:11434 \
  --model llama2

Supported configuration options: - Provider type (openai|replicate|ollama) - API keys - Model selection - Base URLs for custom deployments

Run benchamrks with

mathsolver benchmark --benchmark-file math_problems.json --output-file results.json

Contributing

Contributions are welcome! Please follow these steps:

1. Fork the repository
2. Create a feature branch
3. Submit a Pull Request

LICEENSE

This project is licensed under the MIT License - see the LICENSE file for details.