https://derekwang2002.github.io/ai-intelligence/en/projects/kumo-tabular/ Material changes in AI models, agents, engineering and infrastructure, with primary sources and recommendations. en Mon, 02 Nov 2026 16:56:00 GMT https://derekwang2002.github.io/ai-intelligence/en/events/ev-20260929-01/ ev-20260929-01-published Tue, 29 Sep 2026 16:01:24 GMT New TRIAL <p>NVIDIA released Kumo Tabular, an open-weight tabular foundation model that predicts classification or regression targets from labeled rows in one forward pass without task-specific training or feature engineering. Three models span 28M to 215M parameters and were pretrained entirely on synthetic tables. NVIDIA reports first-place aggregate results on TabArena, BeyondArena, TALENT and ScoringBench. The weights are ungated under OpenMDW-1.1 and inference code is Apache-2.0, but the training recipe and synthetic-data generator are promised rather than available.</p><p><strong>Why read: </strong>Open weights turn tabular cold starts into one forward pass; compare against tree baselines.</p><p>Recommendation: TRIAL</p><p>Sources: <a href="https://huggingface.co/blog/nvidia/kumo-tabular">NVIDIA Kumo Tabular release article</a> · <a href="https://huggingface.co/nvidia/Kumo-Tabular">NVIDIA Kumo Tabular model card and weights</a> · <a href="https://github.com/NVIDIA/structured-data-models">NVIDIA structured-data-models repository</a></p>