You work for a social media company.

AutoML Edge  is a service that allows you to train and deploy custom image classification models for mobile devices 1 2 .  It supports exporting models as  Core ML  files, which are compatible with iOS applications 3 .

Using a Core ML model directly on the device eliminates the need for network requests and reduces prediction latency. It also minimizes the cost of serving predictions, as there is no need to pay for cloud resources or network bandwidth.

Option A is incorrect because sending batch requests during prediction does not reduce latency, as the requests still need to be processed by the cloud service. It also incurs more cost than using a local model on the device.

Option C is incorrect because TFLite models are not compatible with iOS applications.  TFLite models are designed for Android and other platforms that support TensorFlow Lite 4 .

Option D is incorrect because exposing the model as a Vertex AI endpoint requires network requests and cloud resources, which increase latency and cost. It also does not leverage the benefits of AutoML Edge, which is optimized for mobile devices.