Google AI Introduces TabFM: A Hybrid-Attention Tabular Foundation Model for Zero-Shot Classification and Regression

TabFM is a foundation model designed to make predictions on tabular data without requiring dataset-specific training, hyperparameter tuning, or feature engineering. It works by reading an entire dataset as a single prompt and generating predictions in one forward pass using a technique called in-context learning, which combines alternating row-and-column attention with a transformer-based approach. This matters because tabular data forms the backbone of enterprise operations for tasks like customer churn and fraud detection, yet traditional methods like XGBoost require extensive manual preparation work before they can produce reliable results. TabFM was trained on hundreds of millions of synthetically generated datasets using structural causal models and is available on Hugging Face and GitHub, with plans for integration into BigQuery through a SQL command.