hendrik @ hendrik @palaver.p3x.de

A software developer and Linux nerd, living in Germany. I'm usually a chill

---

Read more

Posts

8

Comments

1,820

Joined

4 yr. ago

You're a genius. I hadn't even thought about that... 🏆

Most frontends should have you covered and scale down the image appropriately (and automatically). I'm not entirely sure about that. I think working on resolutions higher than supported should either not work due to the image encoder, or lead to degraded performance. Usually they're scaled to what the model needs, somewhere within the pipeline. You can crop them if you like, that sheds off some pixels. Or split them up and feed them in one part after the other, if that somehow makes sense. But I bet with most software you can just upload a random image and it'll do whatever scaling is required for you.

Sure. It could do your summer vacation including those nasty traffic jams without your participation. Send back a few pictures from important landmarks and monuments, all the while you sit in front of your computer in your air conditioned home like the hacker in the video. Dead landscape might be another option. I'm not sure how the processing power of a Nissan compares to a Mars rover 😅 Maybe it needs to drive very slow to be able to keep up with the incoming sensor data, or due to delay...

Uh sorry, no. Since I don't use vllm, I don't know. It certainly depends on the method you choose. The activation-aware ones will use a lot of resources. Just truncating the numbers to 8bit (or whatever) uses very little resources, I did that on my laptop. Also depends on model architecture and the size of the model you feed in. Since you gave an 32b parameter model, I'd expect it to take about 64GB loaded fully into memory using 16bit floating point numbers. (32 million billion times 2 bytes.) But I don't really know whether quantization methods load the full thing, or if they do it in chunks. You'd have to google that or try it.

Wow, super impressive. Now we need a live video feed from some forward facing camera to give some FPV perspective and a gamepad. And with some more clever hacks, the Leaf could become something like a Mars rover.

This would be better suited in some casual ranting community. Or one concerned with tech bros. I think it's completely off topic here.

I think they have you covered with this: https://github.com/vllm-project/llm-compressor

I always find quantizations available since I use gguf models. But I've done it before, it's not hard to do. Pick a method with good performance, it's going to affect your inference speed. I don't know which one is "best" for vllm.

I think MiniCPM advertises with strong OCR capabilities. Maybe try that one. But is that 384 number correct? Seems to me even the common Llama 3.2 models do images up to 1120x1120px for a while and that should be enough. MiniCPM can do even more (1.8 million pixels).

Setup shouldn't be too hard. Lots of interfaces and inference frameworks have vision capabilities these days. It should work pretty much out of the box with something like Ollama, AnythingLLM or llama.cpp based stuff.

Other approaches include doing OCR by more traditional means and then feed the result into an LLM. Just make sure to check that whatever you use can read Chinese.

I'm not entirely sure what your prompt should look like. Obviously attach the image. With ChatGPT I believe I had a bit better results when I told it to first copy and write down the text, and then translate it in a second step. And then do whatever I want after that. I'm not sure what it does if you make it do everything in one go.

Hmm, interesting. But we all have different experiences anyway. I believe my mom's computer "broke" twice in the last two years during some major Windows updates. One time some service pack broke a lot of printers for quite a while and she was affected as well. I don't remember what the second incident was, since I didn't fix it, but it also required manual intervention. And she doesn't even do a lot with the thing except office stuff, documents and mails, so I doubt she was at fault.

I certainly also had stuff break on Linux, but it's been kind of quiet the last years. But I'm kind of the wrong person to judge since I currently don't take part in everyday Windows use. At least not when I get to decide and maintain the computer. But I feel it has improved as well. There has been a time where I had to install my gaming windows several times because the order in which I installed all the drivers mattered for some reason. It got cluttered and slower over time so I had to reinstall it during the lifetime of a computer. And I had friends infected with trojans and cryptojacking malware every other month or so. Back then I had a very comfortable life full of hubris with my Linux on the desktop. Granted, it needed more fiddling at that time, but that was acceptable. But times have changed and everything got better and it's nothing like that anymore. And for a long time now.

Unpopular opinion, but I install Linux on my computers and they just work for like 5 years straight, with me spending exactly 0 hours each day fixing anything. Whereas I fix other peoples stupid printer issues and Windows becoming incompatible with the hardware or some nasty messages from some antivirus or strange software, multiple times a year...

I see however how some disgrunteled people would write something like this.

I'm not sure whose reading skills are not on par... But that's what I get from the article. They'll face consequences for stealing them. Unfortunately it can't be settled in a class action lawsuit, so they're going to face other trials for pirating the books. And they won't get away with this.

That almost sounds right, doesn't it? If you want 5 million books, you can't just steal/pirate them, you need to buy 5 million copies. I'm glad the court ruled that way.

I feel that's a good start. Now we need some more clear regulation on what fair use is and what transformative work is and what isn't. And how that relates to AI. I believe as it's quite a disruptive and profitable business, we should maybe make those companies pay some extra. Not just what I pay for a book. But the first part, that "stealing" can't be "fair" is settled now.

Maybe Discover isn't the best choice. I believe that's made for the KDE desktop and Gnome should come with "Gnome Software" per default?! I'm not entirely sure what kind of concept Fedora has. I usually use the command line or some of the older package managers with more options and settings, so I can't really tell what's best here. These modern and shiny ones also regularly confuse me and I install Flatpaks by accident or whatever. Maybe try something else, maybe the Fedora community has some recommendations for a better one.

Okay, that's weird. It shouldn't try to remove sudo. That's just silly. But I haven't used Fedora, I have no idea what's going on here.

Hmm, that doesn't show a lot of details but it seems correct. Gnome should have an Extension manager. It should show up if you type some first letters of the word "Extensions" into the app overview. Maybe look if shows up there and got activated.

The Gnome-shell extension is supposed to show up as a button in the top right menu. How did you install it? Should be called "gnome-shell-extension-gsconnect".

I don't have good first hand experience, but i know the Awesome Selfhosted list has a plethora of them.

I'm not sure if this is what you're looking for, but for AI generated novels, we have Plot Bunni. That's specifically made to draft, generate an outline and chapters and then the story. Organize ideas... It has a lot of rough edges though. I had some very limited success with it, and it's not an editor. But it's there and caters to storywriting.

Fingers crossed, but we also know Lemmy might not be ready for that type of philosophy. I mean I still don't know what exactly happened, but lemm.ee wasn't successful in the end. And the underlying issues are still there. So the next admin team might face the same dynamics.

Wikipedia also offers some (limited) information: https://en.wikipedia.org/wiki/.zip_(top-level_domain)