Three million models: should sharing terms be machine-readable? #81
francescobianco
started this conversation in
Open Source AI
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
At the Raleigh Salon, one topic was machine-readable information about how a model is shared: use restrictions, provenance, sharing terms. That means more than the license, and it should be usable for decisions "at governance-relevant speed and scale".
The scale is already there: Hugging Face went from about one million models in 2024 to more than three million today. Nobody reads three million model cards.
APIs went through the same thing: OpenAPI turned prose documentation into something machines can check. What would a model manifest need to contain to be useful, and who would you trust to verify it: the publisher, the hub, or a third party?
All reactions