Research computing and AI infrastructure leaders with proprietary data and a mix of HPC and AI demand can run the whole model lifecycle without sending that data to an external platform. Fuzzball is a container-first workflow orchestration platform from CIQ that runs training, fine-tuning, validation, and inference on infrastructure the organization controls. A workflow is defined once and runs on premises or on AWS, Google Cloud, Oracle Cloud, CoreWeave, or Azure, with one IAM, RBAC, and secrets posture that travels with it. Fuzzball runs on top of an existing Slurm or PBS deployment and uses it as a resource manager, so nothing has to move on day one. Service Endpoints turn a completed workflow into an inference service that runs as an autoscaled pool behind one stable endpoint, and validated model presets start from the workflow catalog by name. Fuzzball provides a web interface and an API, with container-native scheduling and GPU-aware fairshare.
Arcee AI, Lawrence Berkeley National Laboratory, Sandia National Laboratories,
We have not yet collected reviews for CIQ Fuzzball. Share your experience with PeerSpot's community.
Share a review