BAGEL

BAGEL is a distributed diffusion training infrastructure designed for frontier diffusion workloads on commodity and heterogeneous GPU fleets. It is built for artificial intelligence research labs and teams that require training of state-of-the-art generative models for robotics, video, and world modeling. BAGEL's technology enables the training of these models across heterogeneous hardware, unlocking compute capacity that current training architectures cannot touch.

BAGEL's key capabilities include its Distributed Diffusion Models (DDM) which replace a single large diffusion model with an ensemble of smaller expert models, each trained independently on a partition of the dataset with no gradient synchronization between nodes. At inference, a lightweight router ensembles their outputs, removing the tight coupling that forces conventional training onto homogeneous GPU superclusters. BAGEL also features the Paris Inference Engine, a publicly released DDM that outperforms models trained on traditional monolithic clusters.

The teams that get the most value from BAGEL are those working on frontier diffusion models, particularly in areas such as robotics, video, and world modeling. These teams can leverage BAGEL's distributed training infrastructure to unlock compute capacity and train state-of-the-art generative models more efficiently. By using BAGEL, these teams can focus on developing novel methods for distributed training, enabling them to push the boundaries of what is possible in AI research.