level: technical
nvidia introduced kumo relational, a pre-trained foundation model that makes predictions directly on relational databases. given a predictive query, it gathers historical examples as context and predicts outcomes like churn, demand, or fraud in a single forward pass. it needs no task-specific training, no feature engineering, and no flattening of tables. the model is the first few-shot foundation model to outperform supervised machine learning on common relational benchmarks.
across 41 tasks on 15 databases, kumo relational beats supervised relational deep learning models and expert-engineered gradient-boosted tree pipelines. on the sap salt enterprise benchmark, it surpasses autogluon ensembles by 8% in mean reciprocal rank. on relbench v1 classification tasks, it outperforms gradient-boosted trees on expert-engineered features by nearly 4 auroc points on average. it scales to databases with 500b+ rows and works through nvidia's open-source structured-data-models and kumo-engine libraries.
the model is pre-trained on synthetic relational data generated by plurel, which simulates real-world schemas, foreign-key links, and column values. it uses a predictive query language to specify tasks and prevents data leakage by replaying historical database states. limitations include best performance on data with meaningful relationships and temporal dynamics, and task-specific models may edge out on very large training sets. the code is apache 2.0 licensed and weights are under the openmdw license.
why it matters: data science teams can replace weeks of feature engineering and model training with a single pre-trained model that queries relational databases directly.