Inference endpoint - When to scale up or down?

Inference endpoint scaling needs more capacity when requests wait in a queue and response time climbs. It needs less capacity when GPUs sit unused, and the cost per request rises without any traffic g