What strategies would you use to manage GPU/CPU resource allocation at Luma AI when handling sudden spikes in API requests?
