TART: An Open-Source Tool-Augmented Framework for Explainable Table-based Reasoning

TL;DR

TART is an open-source framework that enhances large language models' ability to perform accurate and explainable reasoning over tables by integrating specialized tools and a new benchmark dataset.

Contribution

The paper introduces TART, a tool-augmented reasoning framework for tables, and the TOOLTAB dataset for training LLMs in table-tool integration, improving reasoning accuracy and explainability.

Findings

TART significantly outperforms existing methods like Chain-of-Thought.

TART paired with CodeLlama achieves 90% of GPT-3.5-turbo's accuracy.

The framework improves data processing precision and reasoning clarity.

Abstract

Current Large Language Models (LLMs) exhibit limited ability to understand table structures and to apply precise numerical reasoning, which is crucial for tasks such as table question answering (TQA) and table-based fact verification (TFV). To address these challenges, we introduce our Tool-Augmented Reasoning framework for Tables (TART), which integrates LLMs with specialized tools. TART contains three key components: a table formatter to ensure accurate data representation, a tool maker to develop specific computational tools, and an explanation generator to maintain explainability. We also present the TOOLTAB dataset, a new benchmark designed specifically for training LLMs in table-tool integration. Our experiments indicate that TART achieves substantial improvements over existing methods (e.g., Chain-of-Thought) by improving both the precision of data processing and the clarity of the reasoning process. Notably, TART paired with CodeLlama achieves 90.0% of the accuracy of the closed-sourced LLM GPT-3.5-turbo, highlighting its robustness in diverse real-world scenarios. All the code and data are available at https://github.com/XinyuanLu00/TART.