Hmm… I don’t know whether 8.3 TB would be approved, but I think making the request by email (datasets@huggingface.co
) would be the more reliable route:
If this is still unresolved, the current Hugging Face [Storage Limits documentation](https://huggingface.co/docs/hub/storage-limits) explicitly points dataset storage-grant requests from research/non-profit projects to ** datasets@huggingface.co**, with a detailed proposal.
So I would probably use that as the primary route rather than relying on the forum post alone.
I would include at least:
The main thing I would be cautious about is the grant itself. The public documentation describes these grants as case-by-case, rather than as an automatic academic entitlement. So I don’t think we can infer from the public docs that an academic project will necessarily receive ~8.3 TB, or that a Free account will automatically be allowed that amount under a grant.
In other words, I think the part we can answer fairly confidently is where to ask; the part only HF can really answer is what storage arrangement they would approve for this project.
There is also a small wrinkle: the dataset page currently appears to report a total file size of about 8.61 TB, even though the original post says 8.3TB needed (7.83TB quota reached)
. So it is quite possible that the situation has already changed since this post was made — perhaps the upload continued, the quota changed, or some other arrangement was made. I wouldn’t try to infer which from the public page alone.
So if you have already solved the immediate problem, this may be moot. But if you still need a formal quota/grant arrangement — especially if model weights or additional artifacts are still to come — the email route still seems like the clearest documented next step.
How I would interpret Free vs. academic vs. storage grantSo, short version: if you still need the increase, I would email datasets@huggingface.co with the repository, current/expected total size, academic/public-release context, and whatever concrete impact/reuse information you have. The public docs give a fairly clear route for making the request; they just don’t give us enough information to predict whether ~8.3 TB will actually be granted.