I'm also wondering how much of the git lfs part we should leave in here. It seems the cases where this is best is fairly limited, so maybe the test is, can we clearly articulate the nature of those cases (e.g., want full version control access?), and if we're having trouble with that, cut that section down to merely mentioning it exists and linking to the docs.

I'm also wondering how much of the git lfs part we should leave in here. It seems the cases where this is best is fairly limited, so maybe the test is, can we clearly articulate the nature of those cases (e.g., want full version control access?), and if we're having trouble with that, cut that section down to merely mentioning it exists and linking to the docs.

I opened the draft PR so you could see what I had so far (since you had this issue in mind). I hadn't removed all the old content yet, instead moving most of it further down the page in case I wanted to pull more for the main content. This is also why (as noted in your first comment) there are placeholders.

I will re-read tomorrow, but am happy for any feedback. My current plan:

I think lines 89-114 (git lfs and gitattributes content) should probably be deleted, while lines 116-126 should fall under an "other topics of note" style heading or be worked in as notes at relevant locations further up. HF is quite picky with merge conflicts, so I want to make sure to include a warning about that.

This sounds good to me. The other remaining point I noticed when reviewing this is that it only talks about dataset repos on HF. But the guidance (or at least most of it?) applies just as much to model repos.

Presumably we don't want to have near-duplicates of this for dataset and model repo guidance. Would the idea be to create a separate model repo guidance that by and large refers to this page asking the reader to replace the concept of "dataset" with "model"? Or would it be better to have a single page that in the (hopefully very few) places where it matters distinguishes between dataset repo and model repo type? (One pending project candidate that needs this guidance in fact needs it for a model repo.)

This sounds good to me. The other remaining point I noticed when reviewing this is that it only talks about dataset repos on HF. But the guidance (or at least most of it?) applies just as much to model repos.

Presumably we don't want to have near-duplicates of this for dataset and model repo guidance. Would the idea be to create a separate model repo guidance that by and large refers to this page asking the reader to replace the concept of "dataset" with "model"? Or would it be better to have a single page that in the (hopefully very few) places where it matters distinguishes between dataset repo and model repo type? (One pending project candidate that needs this guidance in fact needs it for a model repo.)

It could probably be refactored to a general HF upload guide, especially since I already referenced a model vs dataset vs space distinction. I think models are generally more standardized, so it should be simple enough to mostly point to the docs. The key points to note there are about different checkpoints getting their own repositories and then generating a collection.

The checkpoints sound a little less straightforward, so maybe best to defer refactoring to a subsequent PR and get the CLI guidance online now, even if in phrasing only for datasets?