Lol the header image is pretty good.
this post was submitted on 23 Nov 2023
1 points (100.0% liked)
LocalLLaMA
1 readers
1 users here now
Community to discuss about Llama, the family of large language models created by Meta AI.
founded 10 months ago
MODERATORS
Another freindly reminder to not just host you model on single host, this bit worrying but nevertless backup your favorite model if you model have to much sensitive stuff.
note : HF reportedly block in china
gotta start torrenting models
You could write a magnet link into the HF description, would make for an easy switch if it gets erased, people back things up.
I'm surprised model torrenting isn't more of a thing. The most legit use-cases possible for torrents are to pass around Linux ISOs and AI models.
You would think HF would realize they can save a ton of bandwidth by posting a magnet link alongside the download button.
Funny. I can/will seed the hell out of some models when this happens.
note : HF reportedly block in china
iirc China has their own version of huggingface. Modelscope?
If HF is blocked there, how are we getting models like Yi and QWEN?
Disappointing
I am predicting someone using an AI model to have Xi-Pooh singing, and spilling a pot called China in the thumbnail.
I think so-vits-SVC is also Chinese made. HF for some reason was scared of gpt-4chan before this. I'm surprised as much stays up on it as it does.
404 media has also posted some scaremongering AI articles before. Their aim is to probably discredit HF like they discredit funders for several OSS projects here: https://www.404media.co/404-media-podcast-week-13-a16z-funds-civitai/
dont comply, then get banned by the ccp in china its really that simple.
They're already banned in China.
Sad, and we thought HF was the harbor of unaligned models, but maybe im missing the whole story. Hopefully they dont kill models for saying taiwan good or something
Well genocide Joe deserves all the hate possible for his support for terrorism and crimes against humanity
I guess will have to make a singing Winnie the Pooh model now, same thing.
The whole GPT-4chan fiasco should have already alerted people about who these people are.
I’m out of the loop. What was this fiasco?
Back in the early days, they deleted a model because it didn't suit their delicate sensibilities.
Looks like I'm backing up the models I like to use. Call me paranoid but this is what happens when you put all your eggs in one basket.
That isn't paranoid at all. I backup everything to two 18TB HDDs everytime I make a new SD LoRA or train an SVC/RVC voice clone.
I appreciate people like you
As an aside, these huge HDD sizes are cool and all, but it's soooooo much data to trust on one or two drives, it's wild
I'm not alone. 🫠
Generally the larger disks have similar failure rates to smaller onces, which means that, if you're running a storage array, increasing the number of drives actually increases your chances of experiencing a failure, even assuming the best case that drive failures are statistically independent events, which they are not.
Larger drives also make keeping full backups much easier (i.e. entire backup fits on a single disk, which makes it easier to make and store).
I'm guessing they received a complaint about the Xi Jinping one, but not the Biden one?
You see that's the interesting thing about China, and in fact literally all "conspiracy theories" around the world.
People think of conspiring as sat around a table and talked, or had a phone call together, or e-mailed with each other, etc. That a group of people agreed on something together, and then went forward with their evil plots.
But in reality, almost everything that happens like this is because the expectation is assumed.
ie: China doesn't need to say anything to HuggingFace to get them to do this. HuggingFace already knows what China expects of them. They don't need to ask. They don't need to be told. They already know.
NYT have written plenty of stories on that, it's just nobody gives a shit
https://www.google.com/search?q=hunter%20Biden%20laptop%20new%20York%20times
That scans. I think Disney owns the copyright for Winnie the Pooh.
By that logic every LLM put there will engage in talk about Xi
Someone should merge the two models.
It would be like Winne the Pooh but with ice cream instead of honey.
"In the USA you can go to Central Park in New York and shout that you really hate president Biden, and you are free to do that. They call that free speech."
"Well, you also have free speech in China. You are also free to do that in Chaoyang Park in Beijing. You can shout you hate Biden there as well."
Hugging face makes me nervous. There’s no way they’re making money now and at some point they’re going to lock down all their goodies behind a paywall
IPFS HF competitor when?
Isn’t this called democracy … we should remember this when you see any politician in any country trying to silence press, or comedians. Cause they’ll definitely remove LLMs or models that make them look or sound bad.
Is china a democracy?