Discussion: download and upload models have a somewhat inconsistent API

[osanseviero]

Hi all. Just a small-ish thing to discuss.

Methods that get information about/from repos take the full model id (e.g. hf_api.model_info("osanseviero/my-model") or snapshot_download("osanseviero/my-model)).

On the other hand, create_repo takes as two separate arguments the model name and the organization (if any). I wonder if this could lead to confusions in the future, and, if so, we should check if the name param has a / and determine if there's an org in there.

[nreimers]

I think this makes a lot of sense. Actually implemented it in the save_to_hub from sentence-transformers

Users are used to pull images like {org}/{name}, so it makes sense to have a push_to_hub("{org}/{name}")

With the current setup, with two parameters, users need to read the docs in order to know what there is a second parameter for the organization. And as we know: Users are not good in reading the docs ;)

[osanseviero]

@LysandreJik @sgugger @julien-c in case you have any suggestions/ideas :)

We now have better error messages when this error happens, so it might not be such a big issue.

[LysandreJik]

I wouldn't mind having org/name be interpreted as name, organization=org if specified, I think it would be coherent with the downloading methods.

[julien-c]

Yes I'd be supportive of this as well

[severo]

Related: while working on the models' data in https://observablehq.com/@huggingface/kaggle-dataset-huggingface-modelhub, I have been surprised that some of the "modelId"s are not "user/model", but just "model" (I understand all of them are models provided by Hugging Face as the organization). eg: https://huggingface.co/bert-base-uncased

[LysandreJik]

Yes, this is for historic purposes. The first models had no organization and were therefore at the root.

Unfortunately, for backwards compatibility purposes, it is now complicated to change the behavior.