GPUStack Operator

Model Delivery

ModelArtifact gives a model version a stable identity. Its delivery can be handled by the engine, a node cache, a PVC or an image.

Contents

Sources and delivery

Model Artifact covers resolution, validation and the delivery choices. Model Image Source covers image-backed weights.

Cache operations

Model Store Operations covers configuration, watermarks and removal. Model Prefetch covers warming weights before a Pod needs them. Node Model Store describes the per-node record and plugin.

Observing and syncing

The Model Artifact API covers reading artifact progress and node caches from the API. Node-to-Node Sync covers how a node pulls weights from another node’s cache.