TableBench: A Comprehensive and Complex Benchmark for Table Question Answering

TL;DR

This paper introduces TableBench, a comprehensive benchmark for table question answering that highlights the gap between current LLM capabilities and real-world industrial requirements, and presents TableLLM trained on a new dataset.

Contribution

The paper proposes a new complex benchmark, TableBench, and a specialized model, TableLLM, to better evaluate and improve LLM performance on real-world table QA tasks.

Findings

GPT-4 outperforms other models but still lags behind humans.

Current LLMs show significant room for improvement on real-world table QA.

TableLLM achieves comparable performance to GPT-3.5 on TableBench.

Abstract

Recent advancements in Large Language Models (LLMs) have markedly enhanced the interpretation and processing of tabular data, introducing previously unimaginable capabilities. Despite these achievements, LLMs still encounter significant challenges when applied in industrial scenarios, particularly due to the increased complexity of reasoning required with real-world tabular data, underscoring a notable disparity between academic benchmarks and practical applications. To address this discrepancy, we conduct a detailed investigation into the application of tabular data in industrial scenarios and propose a comprehensive and complex benchmark TableBench, including 18 fields within four major categories of table question answering (TableQA) capabilities. Furthermore, we introduce TableLLM, trained on our meticulously constructed training set TableInstruct, achieving comparable performance with GPT-3.5. Massive experiments conducted on TableBench indicate that both open-source and proprietary LLMs still have significant room for improvement to meet real-world demands, where the most advanced model, GPT-4, achieves only a modest score compared to humans.