This page may contain stale information. Last updated: 2026-07-01
Definition
Tabular foundation models apply foundation-models / in-context learning paradigms to structured row-column data — the most common format in enterprise ML (banking, e-commerce, insurance).
Key Approach (TabFM)
- Reframe prediction as in-context learning: labeled rows as context, predict new rows in one forward pass
- Train on massive synthetic SCM-generated datasets (only viable path to scale)
- Eliminates manual feature engineering and per-dataset hyperparameter search
Ecosystem
| Component | Provider |
|---|---|
| TabFM | google / tabfm |
| TabPFN | Prior art (row/column attention) |
| TabICL | Prior art (compression) |
| BigQuery AI.PREDICT | Planned SQL integration |