Prepare a model’s prediction endpoint — SageMaker AI

Your goal

Create a model record and configure the resources for its prediction endpoint.

Why this matters

A model deployment needs an explicit model reference and compute configuration. You will create those records and inspect the selected model and instance count. No model server or prediction request is started.

Words you will use

Inference means getting a prediction from a model. An endpoint serves requests. An endpoint configuration describes the model and resources used to host it.

How to proceed

Start a fresh emulator, read each numbered step, and enter the practice settings shown. Run the action, inspect its API receipt, then continue. Finish lab checks the emulator state again before saving your progress.

1. Create a practice model record

Create a model record with a fixed example container image reference and an execution role. The emulator does not download that image, train a model, or provide real predictions.

2. Choose the endpoint’s initial size

Choose an initial instance count and save an endpoint configuration for the model. This records a deployment setting; it does not allocate paid prediction servers.

Initial instance count: Choose 1–3 instances for the emulated endpoint configuration. Suggested practice value: 1.

3. Inspect the model reference and capacity

Describe the endpoint configuration. Verify the model name and initial count in its production variant. A saved deployment plan is separate from a tested prediction service.

Practice, then check your understanding

Perform the actions yourself and inspect the returned state. Errors keep your settings so you can retry. Starting another lab clears the previous live environment. Earned XP is kept, and signed-in achievements sync to your account. Replay does not award duplicate XP.

Read the official AWS documentation for service behavior and production details.