NVIDIA Kumo Tabular, an open foundation model for tabular classification and regression, is now available on Hugging Face under the OpenMDW-1.1 license. The model predicts labels of new rows in a single forward pass without training, tuning, or feature engineering, and comes in three sizes from 28M to 215M parameters. It was pretrained entirely on artificial tables generated from Structural Causal Models and ranks first on the TabArena, BeyondArena, TALENT, and ScoringBench benchmarks.
No score is assigned. Sources and their independence are shown in the citation chain below.