This is important. I dunno about scale, but backups. I started out hosting a chat room on a raspberry pi. It was a fun side project. But then, that became where my friends all hung out. That was the place, so it became important to me. And then the SD card got corrupted. I then moved on to a consumer laptop. It was way more stable, much faster. But if I messed up anything about the installation, I was hosed.

I very highly suggest using Proxmox, like you say, and setting up automatic backups. And occasionally transfer them to a hard drive. It doesn’t matter what kind of virtual CPUs or services you install, gedaliyah@lemmy.world, as long as you have a plan for when something you host becomes important to you and you lose it.

The problems I’ve had with my RPis have all revolved around the fragility of their SD storage. I got burned one too many times trying to host something important in my house with these things, just for them to get corrupted and lose everything. Backing up these systems was its own nightmare, which failed as much as it succeeded.

I’m really interested in this. Is Exllama2 a separately trained variant of Llama2? The use restrictions of Llama2 have always irked me and a similarly performing open variant of that architecture is very intriguing.

Nevermind. This is a processor that runs the model, not the model itself. My bad.

I do, but brain farted and just kept looking on Pygmalion’s site.

Perhaps I’m blind, but I didn’t see any downloads on that page.

I found the download links.

• Pygmalion 2 7B
• Pygmalion 2 13B

Oh, and of course The Bloke quantized the models already:

• Pygmalion 2 7B GGUF
• Pygmalion 2 7B GPTQ
• Pygmalion 2 13B GGUF
• Pygmalion 2 13B GPTQ

From all the war stories you see online, they don’t have that much to lose.

Doesn’t seem like he was implying anything. Just stating the fact that part of the burden of citizenship is sometimes having to interact with law enforcement, maybe even go to trial, even if you’ve done nothing wrong.

I’m no closer to understanding how the original meme is so out of date.

1. Happy to help. Not sure what the year has to do with the meme, tho.

Fair, but the more people you have, with more diverse viewpoints, the harder it will be to get people to agree on what is hateful. And the more nuanced your laws, the harder it will be to agree on what is reasonable or even clear.

If the populace has a bigoted plurality, then they get to declare what is officially hateful. So yes, you’re right.

I put the onus on the collective citizenry, but there is no perfect solution in reality. There is a role for the state to play in protecting people, I just don’t think they should dip much into what speech is or isn’t allowed. The majority should rule in my opinion, but we have the job of maintaining a majority that isn’t regressive bigoted shitheads. It’s an eternal struggle, unfortunately.

You’re partly right. But it’s the job of the citizenry to stand up to this stuff, not the state. We can’t keep our heads down and hope it goes away on its own. We shouldn’t allow the state, with its monopoly on violence, to fight our social battles for us.

I dislike the idea of the state getting to start making decisions on what is “hateful”. And I’m disgusted we don’t have more people standing up and loudly declaring how wrong the hateful viewpoints are. It is our responsibility and we are failing.

It is a tempting proposition to let the state handle hateful speech, but we don’t have to look much further than Florida to see what happens when the shit side is in power and starts redefining what is “hateful”.

Personally I don’t think copyright holders really have a leg to stand on as far as that goes. Simply having and using a copyrighted work isn’t a violation, and the work that is produced in the form of a trained neural network is the very definition of transformative. I also think Meta would have the same issue with trying to use a copyright claim for someone using their llama output to improve other non-llama models. That’s why they had to slip it into a terms of service.

I guess what you might see going forward is every book that’s published comes with a user agreement you agree to by opening the book… But that doesn’t sound practical in any sense.

That wouldn’t even bother me so much if they weren’t trying to lock people into the “llama ecosystem” by restricting its use to only improving other llama models.

Good point. I think I’ll do whatever I want with it and just keep my trap shut. Good luck proving anything Zuck.

I have a feeling that this is going to go similarly to Stable Diffusion’s big 2.0 flop. SD put its limits in through training data. Meta put in its limits via terms and conditions. The end result for both will still be that the community gravitates toward what is usable with the most freedom attached to it. The most annoying part of the TOS is that you can’t use the output to improve other models.

Fuck you Meta, I wanna make a zillion baby specialist models.

The fact that Elon is about to go bankrupt.

Is this a verifiable fact? No. But this fella makes a convincing argument.

Honestly, yes I am.

Thanks for the input. I live in a cold state and all the machines are in the basement, so even in summer it’s pretty cool in there. This gives me hope!

XMPP. It’s an old standard, there are servers you can get an account with or you can host your own. And with OMEMO encryption everything is end to end encrypted.