TRU4Zero: Unravel, Uncover, Unite! On Unified Zero-Shot Table Reasoning with LLM Agents

TRU4Zero is a powerful framework for zero-shot table reasoning using Large Language Model (LLM) agents. It unifies multiple table reasoning tasks including table question answering, fact verification, and table-to-text generation into a single framework.

🌟 Key Features

• Zero-Shot Learning: No label required!
• Unified Framework: Handles multiple table reasoning tasks
• LLM Agent Integration: Leverages powerful language models for reasoning
• Parallel Processing: Efficient batch processing with parallel inference

🚀 Quick Start

Installation

1. Create and activate conda environment:

conda create -n tru4zero python=3.10
conda activate tru4zero

2. Install dependencies:

cd tru4zero
pip install -r requirements.txt
python -m spacy download en-core-web-sm
cd tru4zero/feverous && python3 -m pip install -e .

Configuration

You need to set your OpenAI API key in the .env file. Additionally, as for our single agent and multi-agent settings, you can customize client, MODEL, REFLECTION_TURNS, and TEMPERATURE in tru4zero_module/config/tru4zero_global_config.py.

📊 Supported Tasks & Usage

The framework supports three main tasks:

• Table Question Answering
• Fact Verification
• Query-to-SQL Generation

For detailed usage instructions and examples, please refer to the README in each task's directory:

• WikiSQL: See wikisql/README.md
• TAT-QA: See tat-qa/README.md
• FEVEROUS: See feverous/README.md
• SEM-TAB-FACTS: See sem-tab-facts/README.md

📚 Citation (Coming Soon)

If you use TRU4Zero in your research, please cite our work:

# Note that this is a toy version:

@article{tru4zero2024,
  title={TRU4Zero: Unravel, Uncover, Unite! On Unified Zero-Shot Table Reasoning with LLM Agents},
  author={Yiran Rex Ma},
  year={2024}
}

📄 Contact

For any questions or feedback, please feel free to open an issue or contact me at [email protected].