Foundation models for tabular data, like TabPFN, achieve strong performance on small datasets when pre-trained solely on synthetic data. We show that this performance can be significantly boosted by a targeted continued pre-training phase. Specifically, we demonstrate that leveraging a small, curated collection of large, real-world datasets for continued pre-training yields superior downstream predictive accuracy compared to using broader, potentially noisier corpora like CommonCrawl or GitTables. Our resulting model, Real-TabPFN, achieves substantial performance gains on 29 datasets from the OpenML AutoML Benchmark.
@article{arxiv.2507.03971,
title = {Real-TabPFN: Improving Tabular Foundation Models via Continued Pre-training With Real-World Data},
author = {Anurag Garg and Muhammad Ali and Noah Hollmann and Lennart Purucker and Samuel Müller and Frank Hutter},
journal= {arXiv preprint arXiv:2507.03971},
year = {2025}
}