Krea shares GPU clusters between research and customer-facing image generation, giving research priority when training needs the machines. Its engineering account describes a bridge that lets inferenceRunning a trained model to get answers — the phase where AI is actually used, as opposed to trained.Full definition → workloads run with external computing providers while remaining managed through Kubernetes, its system for organizing workloads. The bridge presents a virtual node to Kubernetes, translates assigned jobs into provider operations and reports their status back. This keeps the management interface consistent even when execution moves elsewhere. The article explains that portability mechanism; the scheduling policy that decides when to move work is reserved for a later account.
Terms in this piece · Glossary
inference — Running a trained model to get answers — the phase where AI is actually used, as opposed to trained.
