You are deploying a machine learning model as a real-time inference service for testing. The service has low CPU utilization and requires less than 48 MB of RAM. The compute target must initialize automatically, minimize cost, and reduce administrative overhead.

Which compute target should you use?