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

650 relational databases spanning academic, e-commerce, finance, sports, biomedical, and government domains are ported to a self-describing manifest format. The collection is built by star-project for large-scale pretraining of relational and tabular foundation models, with tasks shipping labels as-is. The dataset was last updated on June 12,我们发现了一个问题，输入中的最后更新日期是2026年，这是一个未来的日期，这可能是一个错误或占位符。根据事实性协议，我们直接陈述这个数字，但不在推断中使用它来暗示新鲜度。

Use Cases

• Pretrain relational foundation models based on the collection's 650 diverse databases.
• Benchmark tabular model generalization across domains like e-commerce and biomedical data mentioned in the description.
• Develop self-describing data processing pipelines based on the RelBench manifest format.

Strengths

• Contains 650 distinct relational databases, indicating substantial scale.
• Spans multiple domains including academic, e-commerce, finance, sports, biomedical, and government, suggesting diversity.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and file size are unknown, which may limit suitability assessment.
• The last updated date is 2026-06-12, a future date, which may indicate metadata inaccuracies.

Provenance

Source

star-project on Hugging Face

Collection Method

Ported from various sources to the RelBench manifest format.