We’re releasing Paris - the world’s first decentralized trained open-weight diffusion model. The model is open for research and commercial use under the MIT license.
We named it Paris after the city that has always been a refuge for those creating without permission. Two remarkable facts that makes Paris first of it’s kind,
It’s a combination of smaller expert diffusion models pre-trained from scratch across different continents in complete isolation. The experts required zero gradient, parameter, or intermediate activation synchronization among each other during training.
This zero communication protocol achieves comparable quality to SOTA distributed approaches using 14× less data and 16× less compute.
How? Full technical report and model weights below.
Full Technical Report : https://github.com/bageldotcom/paris/blob/main/paper.pdf
Model Weights : https://huggingface.co/bageldotcom/paris
We believe we can scale this approach to global state-of-the-art results. But that requires solving some more really, really hard problems. If you’re an ML researcher or engineer interested in helping us achieve this while doing the best open-source work of your career, come work with us: jobs.bagel.com.