Nvidia releases Kumo Tabular, training-free tabular foundation models

Nvidia has released Kumo Tabular, a family of tabular foundation models for classification and regression. The Small, Medium and Large versions span 28M to 215M parameters and predict new rows in a single forward pass with no training or hyperparameter tuning, with weights under the commercial-friendly OpenMDW-1.1 license.

• Three sizes: Small, Medium and Large, from 28M to 215M parameters
• Predicts new rows in one forward pass with no training or feature engineering
• OpenMDW-1.1 weights allow commercial use; SDM code is Apache-2.0
• Ranks first on TabArena with Elo 1950, 17x faster than LimiX-2