Exam AWS Certified Machine Learning - Specialty topic 1 question 139 discussion

https://docs.aws.amazon.com/sagemaker/latest/dg/model-ab-testing.html

Answer is C, hosting two models under single endpoint has less operational overheads than two hosting endpoints

Need to Create the ProductionVariant before you update the weight

Need to create the variant before you update the weight

in option A it's mentioned that we set initial_weight to 0 which isn't true as the value should be 1 -> C

While Option C is a viable method, Option A is generally more straightforward and aligns well with common practices for deploying and managing model versions in SageMaker. Supported by Copilot

The Answer is A. The question says "Developers want to introduce a new version of the model for a limited number of users who subscribed to a..." In order to introduce a new production version with least overhead you have to create a production variant by using CreateEndpointConfig operation and set the InitialVariantWeight to 0. You then specify the TargetVariant parameter for InvokeEndpoint calls for users who subscribed to the preview feature and gradually update the weight. https://docs.aws.amazon.com/sagemaker/latest/dg/model-ab-testing.html preview feature of the app

-CreateEndPointConfig with initial weight to 0 prohibits any traffic to new variant - TargetVariant Parameter in the endpoint calls made by selected users ensures new variant be used - Change of InitialWeight causes gradual release of new variant

Obviously C

https://docs.aws.amazon.com/sagemaker/latest/dg/deployment-best-practices.html You can modify an endpoint without taking models that are already deployed into production out of service. For example, you can add new model variants, update the ML Compute instance configurations of existing model variants, or change the distribution of traffic among model variants. To modify an endpoint, you provide a new endpoint configuration. SageMaker implements the changes without any downtime. For more information see, UpdateEndpoint and UpdateEndpointWeightsAndCapacities. According to this doc, new variants can be deployed with UpdateEndpoint, and weights can be updated with UpdateEndpointWeightsAndCapacities. Though for using UpdateEndpoint we need to create an endpoint config. I will go with C

The company can implement the testing model with the least amount of operational overhead by using Option A. The developers can update the ProductionVariant data type with the new version of the model by using the CreateEndpointConfig operation with the InitialVariantWeight parameter set to 0. They can specify the TargetVariant parameter for InvokeEndpoint calls for users who subscribed to the preview feature. When the new version of the model is ready for release, they can gradually increase InitialVariantWeight until all users have the updated version

The best option for the company to implement the testing model with the least amount of operational overhead is option C. Option C uses the SageMaker feature of production variants, which allows the company to test multiple models on a single endpoint and control the traffic distribution between them. By setting the DesiredWeight parameter to 0 for the new version of the model, the company can ensure that only users who subscribed to the preview feature will invoke the new version by specifying the TargetVariant parameter. When the new version of the model is ready for release, the company can gradually increase the DesiredWeight parameter until all users have the updated version. This option minimizes the operational overhead by avoiding the need to create and manage additional endpoints, load balancers, or DNS records.

C is correct. The existing model will be updated using parameter DesiredWeightAndCapacity for new production variant and lead to less operational effort.

This one is tricky, but I think it is testing the difference between UpdateEndpointWeightsAndCapacities and ProductionVariant UpdateEndpointWeightsAndCapacities: Updates variant weight of one or more variants associated with an existing endpoint, or capacity of one variant associated with an existing endpoint https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_UpdateEndpointWeightsAndCapacities.html ProductionVariant: Identifies a model that you want to host and the resources chosen to deploy for hosting it. If you are deploying multiple models, tell SageMaker how to distribute traffic among the models by specifying variant weights. https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_ProductionVariant.html So it must be A, because the variant must exist before it is updated This link gave me confidence to choose A https://docs.aws.amazon.com/sagemaker/latest/dg/model-ab-testing.html

I agree with C.

Please see step 4: https://docs.aws.amazon.com/sagemaker/latest/dg/model-ab-testing.html & in option A it's mentioned that we set initial_weight to 0 which isn't true as the value should be 1.

I did not found the InitialVariantWeight, only DesiredWeight, therefore is C: https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_DesiredWeightAndCapacity.html