Join: 650 Multi-Domain Relational Databases for Foundation Model Pretraining

The Join is a collection of 650 relational databases spanning domains like academia, e-commerce, finance, sports, and biomedicine. It is ported to the RelBench manifest format and built for pretraining relational and tabular foundation models. The dataset was created by relbench and was last updated on 2026-06-12.

Use Cases

• Pretraining relational foundation models based on the collection's 650 self-describing databases.
• Benchmarking tabular model generalization across domains based on the described academic, e-commerce, and biomedical datasets.
• Developing text-to-SQL systems based on the included text2sql domain data.
• Studying cross-domain relational patterns based on the multi-spanning collection.

Strengths

• Contains 650 distinct relational databases.
• Spans multiple domains including academic, e-commerce, finance, sports, and biomedical.
• Databases are self-describing and formatted for the RelBench manifest.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Last updated 2026-06-12; freshness should be verified.

Provenance

Source

relbench on Hugging Face

Collection Method

Ported from various sources to the RelBench manifest format.

Freshness

Last updated 2026-06-12.