Models·
NVIDIA Kumo Tabular Sets New Benchmark for Zero‑Shot Tabular Prediction
NVIDIA releases Kumo Tabular, an open foundation model that delivers zero‑shot predictions on tabular data with state‑of‑the‑art accuracy and no training required.
Overview
NVIDIA Kumo Tabular is a transformer‑based foundation model that can predict labels for new rows of a table in a single forward pass, without any fine‑tuning or feature engineering.
How it works
The model treats each cell as a token, applying Fourier embeddings to numerical and categorical values while handling missing entries natively. Column, row and in‑context attention let it capture intra‑row relationships and leverage labeled context rows for inference.
Available in three sizes ranging from 28 million to 215 million parameters, Kumo Tabular was pretrained exclusively on synthetic tabular data and released under the OpenMDW‑1.1 license for commercial use.
It tops the leaderboards on TabArena, BeyondArena, TALENT and ScoringBench, demonstrating a new accuracy‑efficiency frontier for zero‑shot tabular prediction.
- aigpu
- ai gpu
- ai gpu cloud
- aigpu dubai
- kumotabular
- tabular
- foundation model
- machine learning
By AiGpu Editorial · Editorial rewrite based on public reporting (Hugging Face Blog)
← All articles