IngesTables: Scalable and Efficient Training of LLM-Enabled Tabular Foundation Models

This repository contains the implementation of IngesTables, a tabular foundation model. An earlier version of the model was accepted at the NeurIPS'23 Tabular Representation Learning Workshop. Stay tuned for updates!

It contains library code that defines the data preprocessing, training, and evaluation. It also contains scripts for running it locally or on Google Cloud Platform (GCP). Do note that using the GCP scripts may incur costs and would transmit data to GCP and be accessible to those who can access your GCP project.

🧑‍🏫 Tutorials

Here is the list of tutorials and reproducible experiments to get started with IngesTables for various tasks:

• Regression
• Classification

BibTeX

If you found any part of this codebase to be useful, please consider citing our work:

@inproceedings{yak2023ingestables,
  title={IngesTables: Scalable and Efficient Training of LLM-Enabled Tabular Foundation Models},
  author={Scott Yak and Yihe Dong and Javier Gonzalvo and Sercan Arik},
  booktitle={NeurIPS'23 Table Representation Learning Workshop},
  year={2023}
}

This is not an officially supported Google product.