← Back to Quiz Home
This quiz covers operationalizing ML models, including monitoring for drift, feature management, and deployment strategies.
What does "Data Drift" mean in the context of SageMaker Model Monitor?
Detection of drift early allows you to retrain the model before accuracy degrades significantly.
How can you serve two versions of a model (A and B) on a single SageMaker Endpoint to test performance?
A/B testing via production variants is a native feature that allows safe rollout of new models.
What is the primary purpose of the SageMaker Feature Store?
It solves the problem of "skew" where the features calculated during offline training differ from those calculated during real-time inference.
Which SageMaker Feature Store component provides low-latency access for real-time inference?
The Online Store is optimized for single-record retrieval to feed the model at runtime.
How do you secure a SageMaker Notebook to prevent data exfiltration to the public internet?
Removing internet access ensures that users cannot upload sensitive data to public repositories or Dropbox.
What is "SageMaker Pipelines"?
Pipelines allow you to automate the end-to-end ML workflow as code (Python SDK).
How do you deploy a custom SciKit-Learn model trained on your laptop to SageMaker?
SageMaker supports "Bring Your Own Container" (BYOC) for any custom framework.
What happens if you enable "Inter-Container Traffic Encryption" for a training job?
This is critical for compliance when training on distributed sensitive data.
Which service orchestrates the "Human-in-the-loop" workflow for labeling training data?
Ground Truth manages the labeling workforce (private, vendor, or public) and assists with automated labeling.
What is "Model Quality Drift"?
Unlike Data Drift (inputs), Quality Drift measures the actual performance (outputs).
How do you optimize inference latency for a deep learning model on SageMaker?
Compilation optimizes the graph execution specifically for the chip (Intel, Nvidia, Inferentia).
What is the "SageMaker Model Registry"?
The Registry is the central integration point between the Data Scientist (Training) and the MLOps Engineer (Deployment).
Which deployment option allows you to test a new model in production without showing predictions to users (Shadow Mode)?
Shadow variants receive a copy of the traffic, generate predictions (which are logged but discarded), allowing you to verify performance safely.
How can you run a script automatically every time a Notebook Instance starts (e.g., to install a specific library)?
Lifecycle configs allow admins to ensure consistent environments and security agents are installed.
What is "Bias Drift" in Model Monitor?
Clarify helps detect pre-training bias and post-training bias drift.
Which IAM permission is required for a SageMaker Role to write artifacts to S3?
Least privilege dictates scoping permissions to only the buckets used for the job.
How does SageMaker "Data Parallel" distributed training work?
Data Parallel is the most common way to speed up training by throwing more compute at the dataset.
What is the "Offline Store" in Feature Store backed by?
The Offline Store is an append-only log in S3, ideal for generating historical training datasets with point-in-time correctness.
Which SageMaker tool helps you debug training jobs by capturing tensors?
Debugger can catch issues like vanishing gradients or loss not decreasing.
What is "Managed Spot Training" checkpoints?
Checkpoints are critical for Spot training to ensure you don't lose days of progress upon interruption.
Quiz Progress
0 / 0 questions answered
(0%)
0 correct
Quiz Complete!
0%
📚 Study Guides
📬 Weekly DevOps, Cloud & Gen AI quizzes & guides