Google Research · 2026-06-30 · notable
TabFM — Google's zero-shot foundation model for tabular data
TabFM is a Google Research foundation model that classifies and predicts on tabular data in a single forward pass, no per-task training. Weights are on Hugging Face, Apache-2.0 code is on GitHub, BigQuery hook next.
A pretrained-once foundation model that skips XGBoost's tuning ritual for tabular classification and regression.
Key specs
| License | Apache-2.0 |
|---|---|
| GitHub stars | 117 |
| Availability | HF + GitHub |
What is it?
TabFM is an open foundation model for tables — you point it at rows of features and a label column and it predicts new rows in a single forward pass, with no per-task training or hyperparameter sweep. Google Research shipped weights on Hugging Face and Apache-2.0 code on GitHub.
How does it work?
TabFM alternates row and column attention over the raw table, then compresses each row into a dense embedding that a small in-context Transformer reads to make predictions. It was pretrained on hundreds of millions of synthetic tables generated from structural causal models so it sees a wide range of feature relationships before ever touching real data.
Why does it matter?
Tabular data still runs banks, ads, health, and search, and today the default is XGBoost with hand-tuned features. On the TabArena benchmark (38 classification and 13 regression sets) TabFM beats heavily-tuned XGBoost with no per-task work, which cuts model iteration from days to seconds.
Who is it for?
Data scientists and ML engineers who ship classification and regression on structured data.
Try it
pip install and load google/tabfm-1.0.0-pytorch from Hugging Face; BigQuery AI.PREDICT integration lands in coming weeks.