Google Research Launches TabFM: A Zero-Shot Tabular Foundation Model That Predicts Without Training on User Data

Google Research launches TabFM, a groundbreaking zero-shot tabular foundation model that makes instant predictions on tabular data without ever training on user data, now publicly available on GitHub with full scikit-learn compatibility.

• Google Research releases TabFM, a pretrained tabular foundation model that enables zero-shot classification and regression on tabular datasets with mixed column types, requiring no training on user data at inference time.
• TabFM leverages in-context learning by reading training data as context to make instant predictions, and is fully compatible with scikit-learn, supporting both JAX and PyTorch backends for CPU and GPU environments.
• The model is publicly available on GitHub with pre-trained weights for version 1.0.0, complete with quick-start examples, evaluation results, and unit tests for both classification and regression tasks.