An end-to-end open-domain TableQA agent with reinforcement learning, achieving ~88% accuracy on the OpenWikitable dataset.

Usage

• Clone this repository.

git clone https://github.com/TabibitoQZP/OpenTableR1
cd OpenTableR1

• Prepare the data directory.

mkdir -p data/openWTQ

• Download trained models.

cd data
git clone https://huggingface.co/TabibitoQZP/Qwen3-4B-Fncall-SFT
git clone https://huggingface.co/TabibitoQZP/Fncall-Adapter
cd ..

• Download the OpenWikiTable dataset.

cd data/openWTQ
wget https://github.com/sean0042/Open_WikiTable/raw/main/data/data.tar.gz
tar -xf data.tar.gz
cd ../../

• Run the inference code.

python -m otqa.inference

You can check the source code otqa/inference.py and change the argument to configure your own test.

Paper

This work is part of our ongoing research. A manuscript describing OpenTable-R1 is in preparation—stay tuned for updates!