Advancing Open and Reproducible Relational Learning: RelArena-α, TabPFN-Rel and RPI
Prior Labs is open-sourcing a benchmark, a TabPFN-based relational baseline, and an interface for applying relational models to new databases.
RelArena-α standardizes how methods are run and compared on RelBench v1, including data loading, evaluation, tuning, and support for custom systems. TabPFN-Rel is reported as the current top-ranked model on RelArena-α and is positioned as a strong baseline built from flattening relational data into a single table. The release also includes RPI, a model-agnostic interface for defining problems on new databases and applying models implemented in RelArena-α. HF Daily Papers' note
RelArena-α standardizes how methods are run and compared on RelBench v1, including data loading, evaluation, tuning, and support for custom systems. TabPFN-Rel is reported as the current top-ranked model on RelArena-α and is positioned as a strong baseline built from flattening relational data into a single table. The release also includes RPI, a model-agnostic interface for defining problems on new databases and applying models implemented in RelArena-α. HF Daily Papers' note
score 5