Push empty model card and dataset card to the Hub

[Wauplin]

Related to https://github.com/huggingface/moon-landing/pull/4815 (private link).

Goal is to keep up-to-date an empty example of ModelCard and DatasetCard, generated by huggingface_hub. These examples will be referenced from Hub or from docs.

Cards are pushed to:

• https://huggingface.co/templates/model-card-example
• https://huggingface.co/datasets/templates/dataset-card-example

[HuggingFaceDocBuilderDev]

The documentation is not available anymore as the PR was closed or merged.

[merveenoyan]

it's irrelevant but, I was wondering if it would be useful to eventually automatically create these when user pushes model/dataset programmatically through huggingface_hub (we already generate for skops inside skops and for Keras in this library)

[osanseviero]

As discussed in DM, I find using a token for the huggingface org a security risk that we could reduce, as that org has a bunch of important repos. The token could be accidentally leaked if a user does print(PRODUCTION_TOKEN) in a test, we don't notice it and run the workflow, we suddenly have an important token exposed. I suggested @Wauplin to use https://huggingface.co/templates instead.

[osanseviero]

it's irrelevant but, I was wondering if it would be useful to eventually automatically create these when user pushes model/dataset programmatically through huggingface_hub (we already generate for skops inside skops and for Keras in this library)

huggingface_hub is the "low level" tool for people to build model cards. E.g. we have https://huggingface.co/docs/huggingface_hub/how-to-model-cards and templating mechanism. Given the model cards are very context-dependant, I'm not sure users would be very happy if we force automatically creating an empty model card for all repos.

https://github.com/huggingface/moon-landing/pull/4815 (internal) though should help significantly here as the UI will allow people to pick templates.

[Wauplin]

Thanks for the feedback @osanseviero ! I created a new user huggingface-hub-ci for which I created a write token. This token is set as environment variable in the github CI. Worse case scenario, someone successfully leaked it and has write access to huggingface-hub-ci and templates organization. Far less critical than previously done :)

Examples are now in https://huggingface.co/templates/model-card-example and https://huggingface.co/datasets/templates/dataset-card-example.

I also changed the script to exit with nice message if the logged in user doesn't have the correct access.

[julien-c]

Thanks @Wauplin! i will use this in my internal PR

[julien-c]

BTW i would once again advocate to remove the table of content from the templates, as the website UI includes a table of content automatically anyways

But i'll open another issue

[julien-c]

(cc @meg-huggingface for this ^)

[Wauplin]

@julien-c @osanseviero mind approving the PR so that I merge it ? Since you both kinda reviewed already :)

[julien-c]

tiny nit, would you mind removing the table of contents from the templates too (on the fly)?

[Wauplin]

tiny nit, would you mind removing the table of contents from the templates too (on the fly)?

Did it. It will be a good occasion to check CI works fine to push updated versions :)