Launches
Stop flattening your database: NVIDIA pitches Kumo Relational without task-specific training
Developing 3 sources
A pretrained relational foundation model for classification and regression across linked tables using historical labeled examples as inference context.
What happened lately
Kumo Relational combines row representations and message passing across primary-key and foreign-key relationships. Its October 21 announcement describes availability through NVIDIA’s library and NIM, building on earlier KumoRFM-2 research. No task-specific training does not mean no labeled data: historical examples supply context at inference time. NVIDIA’s benchmark and database-scale figures remain developer-reported results.