About the Data Card Generator
We read data documentation standards and the laws that ask for dataset summaries, field by field, and turn them into an outline anyone can use.
The reason for this site
A card a buyer asks for, the training data summary California's AB 2013 demands and the public summary of training content the EU AI Act asks of general-purpose model providers overlap, but they are not the same document. Seeing what each asks for, in its own words, saves writing three.
Who uses it
Teams preparing a dataset for an AI buyer or for publication, and developers who must summarise their training data in California or the EU.
What the site covers
Start with the calculator: it answers one question, "data card generator", on the page, with no account needed. Next come 4 guides on nearby questions. Then the data (figures, and how they were made) and the sources, every one linked.
- Datasheet for datasets template: what the original paper asks: The Datasheets for Datasets proposal in its own words: motivation, composition, collection process and recommended uses, beside the other standards.
- Who requires a data card? The two laws on our board: California AB 2013 and the EU AI Act's training-content summary are the two laws on our board that ask for dataset documentation. What each says.
- Dataset card template: the Hugging Face card as its docs describe it: How a Hugging Face dataset card is built, in the Hub's own docs, and how many standards on our board ask about provenance and personal data, which those docs do not.
- Data card: what one holds, from the texts that define it: What a data card documents, read from the Data Cards paper and Playbook and from Croissant's own working group page, each quoted in its own words with its source.
Limits
- We do not tell anyone their card satisfies a law.
- We store nothing typed into the generator.
- We never paraphrase a standard into a requirement its text does not make.
Accountability
Joseph Ellul-Turner, Director.
Talk to us
Reach us through Contact or at enquiries@datacardgenerator.com; the rules that apply are on Terms and Privacy.
This site summarises published standards and laws on dataset documentation; it is not legal advice, and no standards body endorses it. If a company licenses a dataset through an introduction we make, the company pays our fee of 12.5% of the deal with a minimum as its exclusive agent, or about 10% where the dataset joins a bundle or the pool, agreed in writing first; any fee a buyer pays us is credited to the company.