MLA-C01 Question 198
Select 2You have trained a machine learning model using a custom framework on your local machine. Now, you want to deploy this model to Amazon SageMaker for inference. Which of the following methods can be used to integrate the locally trained model into SageMaker?
- A
Save the model as a serialized artifact (e.g., .pkl file), upload it to an S3 bucket, and reference it when creating a SageMaker model.
- B
Wrap the locally trained model in a Docker container and push the container to Amazon ECR, then deploy it using a SageMaker endpoint.
- C
Convert the model into a SageMaker-specific model format using the SageMaker SDK, then deploy it to SageMaker.
- D
Use the SageMaker Neo service to compile the model code directly from your local machine and deploy it to SageMaker.
- E
Upload the model file directly to SageMaker Studio and deploy it as a SageMaker endpoint.
Show answer and explanation
Correct answers: A, B
Explanation
To integrate a model trained outside of SageMaker, you can either save the model artifact (e.g., .pkl file) to S3 and reference it during model creation or package the model into a custom Docker container and deploy it using a SageMaker endpoint. These methods allow you to incorporate externally built models into the SageMaker environment effectively. Other options, such as converting the model into a SageMaker-specific format or using SageMaker Neo, are not applicable for this scenario.
- A. Correct.
Correct: You can save the model artifact (e.g., a .pkl file) on your local machine, upload it to an S3 bucket, and reference it when creating a SageMaker model. SageMaker supports this workflow as long as the model artifact is accessible and compatible with the inference environment.
- B. Correct.
Correct: You can package your locally trained model into a Docker container and push it to Amazon ECR. SageMaker supports deploying custom containers for model inference, making this a valid approach.
- C. Incorrect.
Incorrect: There is no requirement to convert a model into a SageMaker-specific format. SageMaker supports various model artifacts and custom containers without the need for conversion.
- D. Incorrect.
Incorrect: SageMaker Neo is used to optimize models for specific hardware platforms, not for directly integrating locally trained models into SageMaker.
- E. Incorrect.
Incorrect: You cannot directly upload model files to SageMaker Studio for deployment. SageMaker Studio is a development environment, but model artifacts need to be stored in S3 or used within a custom container for deployment.