NVIDIA Releases Kumo Tabular: Open Tabular Foundation Models That Predict New Rows in a Single Forward Pass
NVIDIA released Kumo Tabular, open tabular models that predict new rows in one forward pass.
NVIDIA released Kumo Tabular, tabular foundation models in Small, Medium, and Large sizes from about 28 million to 215 million parameters. They classify or regress new rows in one forward pass from labeled context, with no per-dataset training or feature engineering. Weights ship under OpenMDW-1.1, which permits commercial use, and run through the Apache-2.0 structured-data-models library. NVIDIA reports a TabArena Elo of 1950 and 17x faster inference than LimiX-2 on one RTX 6000 Pro; training used only synthetic structural causal model tables.
- Three sizes span about 28M to 215M parameters.
- OpenMDW-1.1 weights allow commercial use, unlike several rivals.
- Pretraining used only synthetic tables, up to 60,000 rows.
- NVIDIA reports first place on TabArena with Elo 1950.
- Regression outputs 999 quantiles as an uncertainty estimate.
- Vendors
- NVIDIA
- Products
- structured-data-models
- Organizations
- NVIDIAPrior LabsGoogle Research
From the original article
4 min read at marktechpost.com
NVIDIA has released Kumo Tabular, a new family of tabular foundation models (TFMs) for classification and regression. If you have followed TabPFN or TabICL, the setup will look familiar. The model takes labeled rows as context and predicts new rows in one forward pass.
By Asif Razzaq · marktechpost.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.