Why would you even bother trying to run this all through a VM when you can just run it directly? If you’re to the point of using VMs, you don’t need this tutorial anyways.

Are you seriously telling me you’re jumping through all the hoops to spin up a VM on Linux, and then doing all the configuration for GPU passthrough, because you can’t just figure out how to run it locally?

If your “FIRST STEP” is to choose an OS: Fuck that.

You should never have to change your OS just to use this crap. It’s all written in Python. It should work on every OS available. Your first step is installing the prerequisites.

If you’re using something like Continue for local coding tasks, CodeQwen is awesome, and you’ll generally want a context window of 120k or so because for coding, you want all the code context - or else the LLM starts spitting out repetitious stuff, or can’t ingest all of your context so it’ll rewrite stuff that’s already there.

That way, if the VPN goes down, your torrent client isn’t just downloading stuff nakedly.

You always just bind the torrent client to the VPN adapter so this doesn’t happen. Most modern clients have this (qBittorrent certainly does)

1. • Learning. If you ever found yourself tired of learning new things, your life is basically done.
   • Cost. You already have an internet connection at home. It’s practically a necessity these days. The connection is likely fast enough for most things. Renting even the most piddly of VPS is wildly expensive. Just throw a spare machine at it and go wild.
   • Freedom. Your own data is constantly being collected, regurgitated, and sold back to you. More people need to care about this incessant invasion of our lives.
2. • Backups. 3 copies, on different forms of storage, in multiple PHYSICALLY distinct locations. Just when you have that teeny little imp in the back of your mind say “hmm, I should probably back up soon” – stop everything you’re doing and run a backup.
   • Test your recovery! Backups are only good if you can recover from them. Many have lost data because they failed to ever fail-test their backups.
3. • Google. Legitimately the best skill you can ever attain is simply being able to search effectively and be able to learn jargon quickly. Once you have the lingo down, searches become clearer, quicker, more precise.

Is the router flashable with OpenWRT? :D

jkjk – most modern routers can be turned into just flat access points, ganged with another router.

The router is going to give you more control.

A typical refrigerator is like 40dbA – 25dbA is ABSURDLY quiet. You’re not gonna hit that without a completely fanless system. If 25dbA is his hard cap, he can’t even be breathing in the same area as the computer, because that’s something like 28dbA…

I mean we just had https://nvd.nist.gov/vuln/detail/CVE-2024-6387 – so my guess is that you’re updating quite often to be so confident in your unattended upgrades.

Those were statements. Statements of fact.

Once the models are already trained, it takes almost no power to use them.

Yes, TRAINING the models uses an immense amount of power - but utilizing the training datasets locally consumes almost nothing. I can run the llama 7b set on a 15w Raspberry Pi for example. Just leaving my PC on uses 400w. This is all local – Nothing entering or leaving the Pi. No communication to an external server, nothing being done on anybody else’s server or any AWS instances, etc.

imho - never expose that shit anyways, and VPN into your local network first. Only thing I ever expose to the internet is 80/443.

At the very least, if you’re going to expose an SSH session to the internet, set up some sort of port-knocking. It’s security by obscurity, sure - but it will keep all but the most ardent intruders out.

Once the model is trained, the electricity that it uses is trivial. LLMs can run on a local GPU. So you’re completely wrong.

Absolutely. But I think it might be more advanced than that. They might have some sort of analytics that measures how long people stay on the page, etc to inform their purchasing decisions.

DDNS with Namecheap is as simple as hitting a URL with a /GET request from the IP you want it to point to. No limitations. No special requirements.

Luckily for me I don’t need many email addresses and zoho will do something like 5 for free on your domain. Do you dislike running the email server? I don’t mind all the normal day-to-day upkeep of things, but is email some special kind of hell or something?

Just because you don’t have a website up at [XYZ].com doesn’t mean you’re not using it. You could have a domain controller on the back end doing file services, or you could be using it for network auth, etc. Not all .coms exist for the purpose of putting up a website.

Sorry, but chalk this up to lesson learned. It’s almost always been this way. Domain squatters will do this all the time. In fact, some domain registrars will use you searching their site for an ‘available’ domain, and if you don’t buy it up right away – will buy it and hike the price and sit on it for years in order to lock it down, knowing you wanted it.

btw, Namecheap says Sunglocto dot com is like $10 - so just register a .com. Not through that Epik piece of shit that you used before. Legit, use Namecheap; they’ve never done me wrong and have been my registrar for more than a decade now.

You’re causing your own issues here because you’re wanting to name it all the Japanese way. I use AniDB and everything too - and it works just fine as a single series, with a single year, with all 25 episodes in a Season 1 folder with proper metadata download and everything.

Jellyfin doesn’t see it because it doesn’t know what the hell an “Unlimited Blade Works” released in 2015 is…because it wasn’t released in 2015. You need to use AniDB as a secondary provider for Metadata, not a primary provider, because it doesn’t match up with how Jellyfin and other English-made programs work.

In instances like this, you simply use a single Season 1 folder, with absolute naming for episodes. It’s really that simple.

I don’t understand why you’re naming stuff that way. You use a single FSN:UWB(2014) folder, because the year represents the date the show started. There is no FSN:UBW(2015). It all goes under UBW(2014), even if the episode was released decades later.

Am I the only one here successfully using Sonarr to take care of Anime? Sonarr has the ability to sort by absolute/relative episode you just need a profile for it.

If I really need to bother with any renaming, I’ll use “RenameMyTVSeries” to mass-rename things, and drop them in the folder where Sonarr wants; or usually just have Sonarr grab the anime itself and apply its renaming rules.

Jellyfin is going to want:

• Show Title (YEAR)
• – Season 1
• ---- Episode Name - S01E1234
• – Season 2
• ---- Episode Name - S02E1234

I used to use MythTV back in the analog TV days. It’s much easier to use when you have proper cable channels. I couldn’t be bothered to pay >$140/mo for Cable TV any longer.

So now I just pay $60 for internet, and pirate everything I wanna watch with Sonarr/Radarr/Jellyfin/Jackett/Qbittorrent and a $2/mo VPN from Windscribe.

Honestly, with YouTube experimenting with ‘inline’ commercials, I think MythTV is going to make a comeback; because the big thing MythTV had going for it, was detecting commercials and removing them from the recordings.

Does Incus support things like Kernel Samepage Merging? How does it handle Windows VMs? Does the WebUI give a nice and easy novnc window that just works?