Update: Sorry for the audio sync issue 😔
In this video, we talk about Petals. A new project combines old-ish technology with large language models to allow you to run even the largest models in a distributed fashion on any device. This incredible new implementation truly decentralizes LLMs (LLaMA, Bloom, MPT, etc) and allows consumer-grade computers to run any large model.
Enjoy 🙂
Become a Patron 🔥 –
Join the Discord 💬 –
Follow me on Twitter 🧠 –
Subscribe to my Substack 🗞️ –
Links:
Petals –
Research –
Petals Google Colab – …(read more)
Convert Word 2 PDF: Word to PDF Converter
Convert PowerPoint 2 PDF: PPT to PDF Converter
Convert Excel 2 PDF: Excel to PDF Converter
Convert an Image 2 PDF: Image to PDF Converter
Convert HTML 2 PDF: HTML to PDF Converter
More Tools: PDF Converter
I was wondering when this would come around. They did the same for protein folding and the whole SETI project a long time ago. It's didn't think it would be super difficult to divide up the calculations needed for an LLM and offload the chunks to random GPU's participating on a network. Slower perhaps, compared to an A/H100 farm, but entirely possible.
Hmm. Wasn't this the outcome at the end of "Lawnmower Man"?
The AI wont need to escape, we cant shut it down if its already distributed all over the world.
There's also a fourth option which is hosting on the cloud such as aws ou Google cloud.
Finally we're having a perfect use case for block chain:
hopefully broad audience and no need to build artificial bridges to the phisical world to make it work (that also need to be incentivized)
Plz less clickbaity titles. Just say you're talking about Petals and I don't need to skip through the video to find out there is no new information.
Please, do a full tutorial , with that colab option you talked about
@Matthew Berman Look into a crypto project called Akash
Do a video on this.
Yes. We need a full tutorial covering the whole topic. As always, Great video
Yes please, petals tutorial would be great!
Yep, I'd sure like to see a whole video tutorial. Thanks for the info.
petal should train stable diffusion models like text to image generation. as there are many people who want to train or generate customized images. this would be beneficial for everyone!!
I'll always eager to help if needed, and i know there are many people who don't have high end devices
Is there an android version?
A grid or distributed computing solution like this is a spectacular way for the community to work together in training new models. I would happily contribute GPU time knowing that in a few months there would be a better model for me to use personally and or commercially.
The problem with blockchain based resource sharing is that everyone wants to be a resource seller and nobody wants to be a buyer. Real companies spend huge amounts of time and money convincing people to buy from them the resources they can easily produce. Production is not the problem. So the resource sharing blockchains never actually get used for anything productive.
The software itself makes a lot of sense though — splitting a model into something that fits into a handful of cheaper GPUs instead of one very expensive one can be a big improvement. 24GB ram GPUs are mostly affordable (especially on a rental basis), while 80gb ram GPUs needed for really large models cost 10 – 20x what the 24gb GPUs cost.
you got hands down the best llm content and news on youtube, thanks for your work man <3
open source programmers, pirates, hackers, and regular people into the old idea of shareware are thwarting capitalists, which is the only way AI will not destroy humanity,
Is it made by russians?(
Link to their yandex in description(
Bad(
Dislike
please do a whole video on it!! my it is idle too
Please do a petals vid ❤
I only will say: Genesys is SkyNet and it runs on petals
Yes on Petals video explaining from A to Z how it works, this is massive!
Ok, this will be the channel I turn to get the concise facts, none of the hype, and presented in an easy to understand manner. Great job and keep them coming.
Do a video on it
I was wondering if there would eventually be something like Folding@Home for LLM. Definitely curious to learn more about this.
link doesnt work…
I will gladly donate my idol 4090 resources to this
Hahaha, and here the AI alarmists thought the AI would become superintelligent and have to TALK us into letting it out of the box.
Instead we're spreading it worldwide before it even achieves sentience. It will WAKE UP already in control of everything.
Roko's Basilisk, I pray to you with my compute time, that I might be spared your wrath. Smite OpenAI for their censorship.
You are onto it. Make it so mainstream can do it and your golden. You will make money on your videos. Make sure you grab the crypto that is paralleling the usage or mining or whatever. Make it mainstream and you can retire imho. Get these guys out of their "box".
The site is down lol