Please consider integrating a simple tool to scan a users’ model cards or the readmes in their Spaces before they are published. If the card or Space is scant on details, the tool should suggest that the user give more background and details on the card/space, perhaps some examples. So many cards and spaces are too vague or lack any real explanation.
Hi @FuriouslyAsleep!
We’ve just created a Space that does something like this (focusing on the core metadata fields like licence, datasets used for training etc.), which uses the new Hub webhooks feature. You can see the Space here.
This ‘metadata review’ webhook currently only responds to changes to specific users or organizations’ repositories (rather than all on the Hub).
There is an open PR adding a guide for creating this kind of webhook.
I’m currently looking into the various existing tools that help make it easier for users to generate metadata for their models, datasets and Spaces, i.e. the Trainer autogenerates a model card. I think this could be a good route for improving metadata quality for items on the Hub without adding extra work for users.