[QUANTING UPDATE]
Hey everybody, recently I had to pause the quanting work due to the (somewhat new) imposed HF public storage limits. Hopefully, the pay-as-you-go public storage option will see the light of day soon.
Meanwhile, I will clean up some old, obsolete quants in the repo and return to addressing your requests in a couple of days.
@ArtusDev Can you share what size your total public repos were previously?
Did you just get blocked from uploading anything new until you go under 11.2TB limit or they force you to remove them by x date? I am curious to know the enforcement mechanism, since I think it will shape how this platform will look like in a few months.
Meanwhile, I will clean up some old, obsolete quants
One man's trash is another man's treasure.
I feel like this could make it so that in a few years when I'll want to try out an old open model (since open weights tend to stay available whenever you want it), the accounts hosting them now will do similar kind of cleaning and old models may get hard or impossible to find.
Starting this month, I was billed for grandfathered private storage limits, I posted about this here. Are you also hit by something that seemed to have started on January 1st 2026?
@ubergarm Can you share how you're dealing with max public repo size? Were you granted any special storage pool by HF?
@adamo1139 We are currently sitting at about 50TB across all the quants. Most of the space from that is coming from previous HF storage grant for which I'm very grateful.
Regarding the "cleanup" thing - yes it's not ideal, especially considering that EXL3 especially doesn't have a lot of quanters so most models don't have alternatives at all (compared to gguf). But reality is also that nothing can be free or infinite forever so it's understandable.
I was hit with the grandfathered private storage limits in another org where I'm an admin.
Can you share how you're dealing with max public repo size? Were you granted any special storage pool by HF?
I ran into the issue when hf originally started enforcing public quotas a couple/few months ago. I got a monthly subscription and also applied for a grant. The hf team graciously gave me enough space to keep releasing in the near term for which I am grateful.
Knowing it will eventually become a challenge, I've been releasing less middle sized quant options and focusing on a couple smaller sized, one good middle size, and one larger (but not full) size in a collection.
I too need to go back and clean up some old stuff, especially the older recipes of my big Kimi quants. I will likely need to carefully rewrite the git history and force push to remove the old versions and keep only the updated recipes.
I've wondered how xet effects things, given if a number of people have quants including data chunks with the same hash then in theory it is only saved "once" though I suppose it still "shows up" in the usage quota despite "deduplication" under the hood?
Anyway, cheers and may the quants keep flowing! 🙏
I will likely need to carefully rewrite the git history and force push to remove the old versions and keep only the updated recipes.
I am doing super-squashing to repos to remove git history, I recommend doing that since it's easy. I use a few vibe coded scripts that I can share if you're interested.
I've wondered how xet effects things, given if a number of people have quants including data chunks with the same hash then in theory it is only saved "once" though I suppose it still "shows up" in the usage quota despite "deduplication" under the hood?
I've not seen any effect of xet on the quota numbers. I think they're keeping the difference for themselves. But it does make uploads faster when you're just reuploading things.