Nyfure

Dont. They are notoriously bad at such things. Lack of Hardware acceleration mainly. These old Chips and problems with single-board-complications are just not worth it at such high prices.
An Intel N100 MiniPC will have much more compute with less complications.

Just saw it too :D

Why would you ever need such a feature? Closed.

you want a frontend, not the “service” itself.
Under “service” i usually understand the main logic part of something. In this case the LLM-processing itself.
Thats probably where the confusion is coming from here.

Please include the actual calculations for energy-prices as many, you may not know, live in different locations and pay different prices compared to you.

As far as i understood tailscale funnel its just a TCP-tunnel.
So you handle TLS on your own system, which makes sure tailscale cannot really interfere.

If you already trust them this far, might aswell do the same with a VPS and gain much more flexibility and independence (you can easily switch VPS provider, you cannot really switch tailscale funnel provider, you vendor-locked yourself in that regard)

I’d connect the VPS and your home system via VPN (you can probably also use tailscale for this) and then you can use a tcp-tunnel (e.g. haproxy), or straight up forward the whole traffic via firewall-rules (a bit more tricky, but more flexible… though not that easy with tailscale… probably best to use TCP-tunnel with PROXY-Protocol).
This way you can use all ports, all protocols, incoming and outgoing traffic with the IP-Address of the VPS.

Tailscale might even already have something that can configure this for you… but i dont really know tailscale, so idk…

And as you terminate TLS on your home-system, traffic flowing through the VPS is always encrypted.

If you want to go overboard, you can block attackers on the server before it even hits your home-system (i think crowdsec can do it, the detector runs on your home-system and detects attacks and can issue bans which blocks the attacker on the VPS)

And yes, its a bit paranoid… but its your choice.
My internet connection here isnt good enough to do major stuff like what i am doing (handling media, backups and other data) so i rent some dedicated machines (okay, i guess a bit more secure than a VPS, but in the end its not 100% in your control either)

Many systems dont support subpaths as it can cause some really weird problems.
As you use tailscale funnels, you really want incoming traffic from the internet. I am not sure thats a good idea for e.g. homeassistant that is limited in access anyways.
Might aswell use tailscale and access the system over VPN.

And for anything serious i wouldnt use something like funnel anyways. Rent a VPS and use that as your reverse-proxy, you can then also do some caching or host some services there. Much simpler to deal with and full support for such things as you then have an actual public IPv4/IPv6 address to use.
Heck, dont even have to pay for it with the Oracle Always-Free system.

How much time do you have? Because even small models will take alot of time on that kind of hardware to spit out a long text…
And the small models arent that great. I think the current best and economic model would be a mistral, mixtral or dolphin.
If you got the power, nous-capybara is very good and “only” 34B parameters (loading alone needs like 40GB of memory).

When i was with a customer who was using one of ther VPS offers, performance was unexpectedly low and upon contacting support it was clear the small fish dont get great support answers, but rather pushed to the FAQ.

And i personally find their offerings and marketing scummy. Big promotional prices, but always some small print with a higher price after x Months.
Or just stuff thats not included by default.
I never had that with other (also very cheap) providers.

As long as it works great for you, i wouldnt see a reason to leave.
There arent that many providers offering such small ressources at all or at such a price. To be fair, not much one can do with those specs… 10GB storage is very limited already.
But for those specs… always free oracle tier would work too (though requires a credit card).

Ionos… not a good provider.
Great it works for you, but i wouldnt touch them with a long pole.
Created by an old internet provider (which is also not very good…), pulling every shady marketing trick weird “cloud” providers have…

Contabo is very cheap too, but i wouldnt trust them with critical stuff.
Netcup is next, quite good and still cheap.
Hetzner is very nice, but the cloud offers are expensive. the dedicated server offers though… holy sweetness, specially the auction servers.
Dont forget smaller providers either, they can have some good stuff, but cannot really compete with the big players. (i have one for clean ip space for mail)

Over the years hosting i learned that paying slightly more is often worth it depending on the needs.
And as my requirements went up, i moved up in the tiers. If you have a need for the dedicated servers, gets cheaper for what you get (though you need to manage the hardware side then too…)

Oh and dont forget the Oracle free offers. I dont really trust Oracle, but free compute is free… maybe dont store sensitive stuff though

I dont see how e.g. arch would be super hard to maintain.
There is a nice GUI program for installing programs and updates. (like many modern distros)
If you dont want to set everything up, go with Endeavour or Garuda.

I find rolling release to be easier to maintain and keep up to date than non-rolling.
Specially if you want up to date packages for desktop use.

Windows has a request assistance function? wtf… where is that found?
I only know Remote desktop tools and most of these work perfectly fine on linux as the client or even under Wine.

[Edit: woah, i did some rambling below here… not related to your specific case here, but some nice information maybe]

Linux as host is where it gets funny… bigger ones support X11, pretty much none support Wayland.
To be fair, its impossible to control mouse and keyboard under Wayland without root.
I think we now have some new desktop packages for gnome and kde which can do that, so now they need to be implemented.

But i dont see an effort being made for Wayland by the bigger providers in the near future… the market just isnt there and there is lots of uncertainty with the featureset.

Switched to Rustdesk a while back, works nicely as client, but only picture output with wayland as host.l as of now.
And i cannot copy&paste under wayland as client… even though it worked before…

Hetzner is wild at how cheap you get hardware and included traffic.
German providers in general, everywhere is very expensive compared to these prices.

The question then is, how much can you actually do in the shell.
Good luck setting up many programs that way when they soley rely on the GUI and documentation about configs or their database structure is nonexistant.

WebCord is a beast! Maybe runs better for you.
Basically Discord desktop client experience, but privacy (well… as much as you can have with discord) from the browser-version. (minus discord desktop client exclusive features of course)

To be fair, windows development also included UI changes beneficial for users, so its not necessarily bad to copy those.
Of course there are many which are… questionable, we of course shouldnt copy those :D

Async is good because threads are expensive, might aswell do something else when you need to wait for something anyways.
But only having async and no other thread when you need some computation is obviously awful… (or when starting anothe rthread is not easily manageable)

Thats why i like go, you just tell it you want to run something in parallel and he will manage the rest… computational work, shift current work to new thread… just waiting for IO, async.

Just pay the few dollars per year and have a stable and reputable domain.
Certainly for fediverse i’d want a stable domain, these are usually hard to migrate.

The performance is absolutely abysmal and the error-rates high. For personal use, just have a normal VPN.

Index of repositories is held locally, so if you use the same repository with multiple machines, they have to rebuild their index every time they switch.
I also have family PCs i wanted to backup too, but borg doesnt support windows, so only hacky WSL would have worked.
But the worst might be the speed of borg… idk what it is, but it was incredibly slow when backing up.