Awesome AI Agents › AI Assistants

microsoft/ToRA

⭐ 1124 Python added to this list on 2025-03-10 repository created 2023-09-27

ToRA is a cutting-edge series of Tool-integrated Reasoning Agents developed by Microsoft, designed specifically to tackle challenging mathematical reasoning problems. These agents leverage the power of large language models (LLMs) combined with external computational tools such as symbolic solvers and computation libraries. This integration allows ToRA to seamlessly blend natural language reasoning with precise computational capabilities, enhancing the problem-solving process for complex math tasks. The project is notable for its performance on various benchmark datasets including GSM8k, MATH, and a suite of 10 math tasks that cover a broad spectrum of mathematical problem types. ToRA models range in size from 7 billion to 70 billion parameters, with the ToRA-Code-34B model being the first open-source model to surpass 50% accuracy on the MATH dataset, outperforming even GPT-4's chain-of-thought results. This demonstrates the effectiveness of combining language models with tool-based reasoning. ToRA's training pipeline involves imitation learning and output space shaping, which help the model learn to interact effectively with external tools during problem solving. The repository provides comprehensive resources including pre-trained models, training scripts, evaluation tools, and example datasets, facilitating both research and practical application. The project is well-documented with a detailed README, providing setup instructions, inference and evaluation commands, and links to the official paper, website, and Hugging Face model hub. It is licensed under MIT, encouraging open collaboration and further development. Overall, ToRA represents a significant advancement in AI-driven mathematical reasoning by integrating language understanding with external computational tools, enabling more accurate and interpretable solutions to complex math problems.

https://github.com/microsoft/ToRA

ai-researchautonomous-agentscomputation-librariesevaluationgsm8khugging-face-modelsimitation-learninglanguage-modellarge-language-modelsllmmath-datasetmathematical-problem-solvingmathematical-reasoningmicrosoftmit-licenseopen-source-modelsoutput-space-shapingsymbolic-solverstool-integrated-reasoning-agentstool-learningtora-70btora-code-34btraining-scripts