On Github and Gitlab users can “archive” projects to make them read-only and clearly indicate they are not being actively maintained.
I think it would be useful for model and dataset repositories on Huggingface Hub to have a similar type of mode for repositories. This would help model creators and organizations make it abundantly clear to users that a particular model/dataset is old and outdated. The best thing would also be if repository owners in this process of archiving could recommend options to users (link to models/datasets that are recommended to be used instead).
This has not been a huge issue thus far on HF Hub. But in my opinion it’s becoming more and more of an issue. As the number of models on organizations’ pages grow, it becomes increasingly difficult for users to sort through all of the options and find the most relevant one. Especially when there are multiple different versions of the same-ish model with slight variations.
I know you have introduced “Collections” as a way to help users find relevant models and datasets. I still however think something like an “archive this model/dataset” would be incredibly useful for model/dataset uploaders.
As an example from the orgs I belong to:
- Users frequently end up using suboptimal versions of wav2vec2 models that were finetuned for monolingual use based off of multilingual checkpoints as opposed to from checkpoints that were pretrained from scratch on the specific language.
- Users end up using NER models that are old/inferior.
Removing the models is not always an option, as they may be actively used and/or have been referenced in research.