signalAI热榜2026-09-30
NVIDIA Releases Open-Source Tabular Foundation Model Kumo Tabular, Ranking First on Four Benchmarks Including TabArena
NVIDIA has released Kumo Tabular, an open-source tabular foundation model that performs classification and regression with a single forward pass, requiring no training or feature engineering. It ranks first on four benchmarks: TabArena, BeyondArena, TALENT, and ScoringBench. The model comes in three sizes (28M to 215M parameters), is pretrained entirely on synthetic data, and is released under the OpenMDW-1.1 license.
- for who
- Enterprise machine learning practitioners working with tabular data
- why now
- NVIDIA's new open-source Kumo Tabular tops four benchmarks, now commercially usable.
- what changes
- They can replace gradient boosting workflows with in-context learning that requires no training, tuning, or feature engineering, achieving state-of-the-art accuracy and efficiency
- to do
- Download Kumo Tabular from Hugging Face and test it with the structured-data-models library on your own tabular datasets
key points
- Kumo Tabular enables zero-training tabular prediction with a single forward pass
- Ranked first on TabArena, BeyondArena, TALENT, and ScoringBench with ELO 1950 on TabArena
- Three sizes from 28M to 215M parameters, pretrained on synthetic data, open-source license
#tabular foundation model#nvidia#open source#tabarena#gradient boosting trees
score
score 9 out of 10. 0-10: how dense the facts are, multiplied by how much you can do with them after reading. 8+ means the topic's evidence bar is met: benchmarks and availability for a new model, amount and investors for a funding round, revenue figures for a solo-money story. Below 5 an item does not enter the digest. A press release scores 3 or less, a reprint loses 2, anything older than 14 days loses 1, a headline that misleads loses 3.
read the source