MLA-C01 Question 267
Select 2You are deploying a machine learning model into a production environment on Amazon SageMaker. The model has high computational requirements, including complex deep learning inference tasks. For testing, you use a small dataset to validate the model's behavior. Which approach should you take to provision compute resources for the production and test environments?
- A
Use an instance with GPUs for production and a CPU-based instance for testing to save costs.
- B
Use the same GPU-based instance type for both production and testing to ensure consistency.
- C
Use an Amazon SageMaker endpoint with auto-scaling enabled for production and a single-instance endpoint for testing.
- D
Use multiple instances with GPUs for testing to simulate production traffic.
- E
Use Amazon SageMaker Processing Jobs for testing and hosting services for production.
Show answer and explanation
Correct answers: A, C
Explanation
In a production environment, it is important to use GPU instances when models require high computational capabilities, such as deep learning inference, and to enable auto-scaling to handle varying traffic loads. For testing, it is more cost-effective to use CPU-based instances or single-instance endpoints when testing on a smaller dataset. This balances cost savings with functional validation of the model.
- A. Correct.
This is correct. GPU instances are optimal for high-performance production environments, while CPU instances can be sufficient for lightweight testing scenarios, minimizing costs.
- B. Incorrect.
This approach ensures consistency but can be unnecessarily expensive for testing low-resource scenarios where GPUs are not required. It does not optimize costs.
- C. Correct.
This is correct. Auto-scaling ensures that production environments can handle varying traffic, while a single-instance endpoint is cost-effective for testing.
- D. Incorrect.
Testing typically does not require multiple instances unless you are performing load testing. This option would increase costs unnecessarily for most use cases.
- E. Incorrect.
Amazon SageMaker Processing Jobs are suited for data preprocessing, not for testing the inference behavior of models. This option is not suitable for the given scenario.