stanford-star/the-join

The Join

A collection of real-world relational databases across many domains (academic, e-commerce, finance, sports, biomedical, government, text2sql and more) in RelBench format, built for pretraining relational foundation models. Databases overlapping RelBench or dbinfer sources are excluded. Tasks are forecast or autocomplete, train split only, with val_timestamp and test_timestamp null. Tasks were generated with agentic-task-gen. Pretraining data of stanford-star/rt-j; the preprocessed form is the-join-preprocessed.

<dataset>/
  manifest.yaml                 # tables, primary keys, foreign-key graph
  schema.svg                    # ER diagram
  db/*.parquet                  # relational tables
  tasks/<task>/manifest.yaml    # task spec (kind: forecast | autocomplete)
  tasks/<task>/train.parquet

STATS/databases.parquet and STATS/tasks.parquet catalog every database (domain, size, license, source) and task; both are shown in the dataset viewer.

import relbench
ds = relbench.load_dataset("stanford-star/the-join/<dataset>")
task = relbench.load_task("stanford-star/the-join/<dataset>", "<task>")

License

Each database keeps the license of its upstream source, listed in the license and source_url columns of STATS/databases.parquet. Our contributions (the conversion to this format, schemas, task definitions, splits and statistics) are CDLA-Permissive-2.0.

Citation

@inproceedings{ranjan2026rtj,
    title={{RT-J}: Large-Scale Pretraining of Relational Transformers for Context-Efficient Predictions},
    author={Rishabh Ranjan and Vignesh Kothapalli and Harshvardhan Agarwal and Charilaos Kanatsoulis and Roshan Upendra and Tom Palczewski and Carlos Guestrin and Jure Leskovec},
    booktitle={The Fortieth Annual Conference on Neural Information Processing Systems},
    year={2026}
}

@inproceedings{relbenchv2,
  title={RelBench v2: A Large-Scale Benchmark and Repository for Relational Data},
  author={Gu, Justin and Ranjan, Rishabh and Kanatsoulis, Charilaos and Tang, Haiming and Jurkovic, Martin and Hudovernik, Valter and Znidar, Mark and Chaturvedi, Pranshu and Shroff, Parth and Li, Fengyu and Leskovec, Jure},
  booktitle={3rd Workshop on Navigating and Addressing Data Problems for Foundation Models (DATA-FM) at ICLR 2026},
  year={2026},
  note={arXiv:2602.12606 [cs.LG]},
  url={https://arxiv.org/abs/2602.12606}
}