Method UnloadModelOnNode

core/services/nodes/unloader.go:295–307

UnloadModelOnNode sends a model.unload request to a specific node. The worker calls gRPC Free() to release GPU memory.

(nodeID, modelName string)

// UnloadModelOnNode sends a model.unload request to a specific node.
// The worker calls gRPC Free() to release GPU memory.
func (a *RemoteUnloaderAdapter) UnloadModelOnNode(nodeID, modelName string) error {
	subject := messaging.SubjectNodeModelUnload(nodeID)
	xlog.Info("Sending NATS model.unload", "nodeID", nodeID, "model", modelName)

	reply, err := messaging.RequestJSON[messaging.ModelUnloadRequest, messaging.ModelUnloadReply](a.nats, subject, messaging.ModelUnloadRequest{ModelName: modelName}, 30*time.Second)
	if err != nil {
		return err
	}
	if !reply.Success {
		return fmt.Errorf("model.unload on node %s: %s", nodeID, reply.Error)
	}
	return nil
}

// DeleteModelFiles sends model.delete to all nodes that have the model cached.
// This removes model files from worker disks.