Endpoint Auto Scaling

Automatically increasing or decreasing inference endpoint instances by configuring minimum and maximum instance counts and trigger metric thresholds. It controls cost while maintaining response times, but introduces cold start latency.