One Model to Rule Them All? SAP-RPT-1 and the Future of Tabular Foundation Models

Tabular foundation models are being developed by companies like SAP, which recently released its Relational Pretrained Transformer (RPT) models for enterprise data. These models are based on the ConTextTab and TabPFN architectures, which adapt the transformer model for use with relational, or tabular, data. Unlike earlier models trained only on synthetic data, ConTextTab incorporates semantic embeddings from real-world datasets to improve performance. The piece provides a hands-on Python demonstration showing how to use the SAP-RPT-1 API for a classification task using few-shot in-context learning. This serves as a case study to evaluate the practical implications and future of large-scale models for enterprise use cases.