TabICL: A Tabular Foundation Model for In-Context Learning on Large Data

TabICL is a scalable tabular foundation model designed for classification tasks. Pre-trained on synthetic datasets with up to 60K samples, it can handle even larger datasets thanks to its memory-efficient inference.

Installation

pip install tabicl

The source code is available at GitHub - soda-inria/tabicl.

Citation

If you use TabICL for research purposes, please cite our paper:

@article{qu2025tabicl,
  title={TabICL: A Tabular Foundation Model for In-Context Learning on Large Data},
  author={Qu, Jingang and Holzm{\"u}ller, David and Varoquaux, Ga{\"e}l and Morvan, Marine Le},
  journal={arXiv preprint arXiv:2502.05564},
  year={2025}
}