Already? I’m still using Fedora 39 since that’s the only version supported by CUDA Toolkit :S

Only 3.8B parameters according to the paper, so it ought to be quite easy on the hardware as well if they do.

Easiest GUI toolkit I’ve used was NiceGUI. The end result is a web app but the python code you write is extremely simple, and it felt very logical to me.

For LLMs it entirely depends on what size models you want to use and how fast you want it to run. Since there’s diminishing returns to increasing model sizes, i.e. a 14B model isn’t twice as good as a 7B model, the best bang for the buck will be achieved with the smallest model you think has acceptable quality. And if you think generation speeds of around 1 token/second are acceptable, you’ll probably get more value for money using partial offloading.

If your answer is “I don’t know what models I want to run” then a second-hand RTX3090 is probably your best bet. If you want to run larger models, building a rig with multiple (used) RTX3090 is probably still the cheapest way to do it.

What kind of issues do they have? I’ve used gtx970, 1080, rtx3080 and now 3090 and I’ve never had any issues worth mentioning.

It ought to be mandatory to write this out whenever talking about Linux. I’ve seen more than one person bash Linux in a public forum “because it has digital rights management built into the kernel” after they’ve misinterpreted some news headline.

If someone is careless, they should create a wrapper around rm, or just use a FM.

I think that’s the situation OP is in… They don’t trust themself with these kinds of commands, while other commenters here are trying to convince them that they should just use rm -rf anyway

Maybe they’re afraid of accidentally writing rm -rf folder/.git /* or something

Did you try running xev and pressing the key to see if it shows up as something?

If you’re using btrfs then you might need to rebalance it. I had the same problem, i.e. “no free space” while tools like df reporting that there should be available disk space, and it confused the hell out of me until I found the solution.

See manual: https://btrfs.readthedocs.io/en/latest/Balance.html

This are the commands I run every now and then, especially if my drive has been close to full and I delete a bunch of files to make more space:

sudo btrfs balance start -dusage=10 /
sudo btrfs balance start -dusage=20 /
sudo btrfs balance start -dusage=30 /

The / at the end is the path, since it’s my root mount which uses btrfs. The example in the manual does 40 and 50 too, but higher numbers take longer time, even on an nvme ssd.

My wife and I gave a Linux computer to a friend’s kid. I think I used something called Grapejuice to install Roblox, which ran perfectly for about a year. Then it broke because they wrote a new game client or something, but the kid just said “it’s ok, I’ll play other games instead.” Best Linux gamer attitude :)

I have some hobby projects in Python but I’ve never needed the pro features, I do pay for Clion though

It looks pretty, but IMO one of the selling points of zsh is that it allows async updating of the prompt, allowing you to use slow commands like “git status” without adding a delay every time the prompt needs to be printed.

E.g. the default prompt from prezto is quite light and responsive, but when inside a git repo adds the info on the right side (shows when you have commits ahead/behind the remote branch, stashes, modified/deleted/added/staged files, etc) when that becomes available.

Didn’t look like any of the example themes on ohmyposh.dev had the $RPROMPT stuff, which I guess would be difficult support for a cross-shell theming engine.

In the defence of client side AC; if the entire game runs on the server, then network delay makes FPS:es awful to play. Being able to trust clients and let them do hit detection is quite important in making online FPS:es responsive. In addition, cheats that remove walls/grass, highlight players or even autoaim are near impossible to detect server side. One could try to use heuristics and statistics but it would be difficult to tell the difference between cheaters and players who are just good at aiming and map awareness.

That’s been the primary reason why I’ve kept a Windows dual boot, though when I tried Steam VR on Linux a month ago it mostly worked well. Still some features that are unavailable, and a couple of bugs, but usable.

I also use KDE because I like customizing my DE, but I’m not sure I agree that it’s hard to break. When I just switched from Xfce to KDE I downloaded several global themes using the built-it theme browser, and a few of those definitely messed things up. It’s also happened more than once that I boot my computer and end up with only the desktop background (i.e. no panels or context menu) because KDE thought there was some wrong with the theme, which can be difficult to recover from for someone who doesn’t know how to ctrl-alt-F3 and edit settings manually. Though it’s ofc. more stable when not testing global themes, and only changing other appearance settings.

Definitely not cheap, but at least not as bad as having to buy an A100 for €7000 to get 40GB VRAM. I’m hoping second hand GPU prices will plummet after Christmas

You can buy 2x second hand RTX3090 for about the same price of one new RTX4090, though you’ll probably need to get a new PSU as well. Or rent the hardware through runpod.io or similar for around $1/hour. Still a lot of money for most people but it’s not completely unachievable… Spend some time in the local LLM community and 48GB VRAM will start to feel like the bare minimum if you want to use any of the better models :S

There are tons of options for running LLMs locally nowadays, though none come close to GPT4 or Claude 2 etc. One place to start is /c/localllama@sh.itjust.works

Looks really cool, but a video of more than 3 seconds would’ve been nice :)