Training Open-Source AI with Volunteer Computing Project?

[revenge of rebecca]

Training AI is very expensive, and many of the most powerful models are likely to be closed-source proprietary. I am wondering if there are any volunteer distributed computing projects out there for training a LLM or text-to-image model? By volunteer computing projects, I am thinking about projects like Folding@home and SETI@home, in which volunteers donate computer resources to advance the project. In this case, I am thinking about a project specifically to train AI models.

The only such project that I know of is Leela Chess Zero, which uses volunteer computing power to train its AI to play chess better. However, I am not particularly interested in chess, and I am more interested in the LLMs and the text-to-image models. Those models have some economic use, and I think it would be a shame for them to be non-libre and monopolized by sociopathic megacorporations like Google and Microsoft. Therefore, I would like to figure out how to donate some spare computing time to open-source LLM and text-to-image AI projects.

Does anyone know of any such projects? If not, are there any particular barriers or limitations that prevent such a project from being feasible?

 

[Justtocheck]

Yes there was, but it ended because of the lack of computing capacity. That's the thing that takes lots of money. The training set is out there. It was called open-assistant.

https://www.youtube.com/@YannicKilcher

 

[revenge of rebecca]

I will take a look at this later. Thank you.