Machine Learning Operations Flashcards
What are the three deployment types for Blue/Green deployments?
All at once
Canary
Linear
Can you test multiple models in SageMaker on live traffic?
Yes, using production variants.
How are production variants splitting trafic?
A/B testing let’s say 90 percent to one model and 10 to another.
What AWS Service allows you o deploy your model to an actual Edge device?
Neo and IoT Green Grass
Can you use spot instances for Sagemaker Model training?
Yes, but you need to use checkpoints to S3.
What is the easiest way to deploy a model in SageMaker?
SageMaker jump start.
What is the the ModelBuilder in the SageMaker Python SDK?
It allows you to deploy models for inference with more control.
Can AWS CloudFormation be used for deployments?
Yes this is for advanced users.
What is the Cloudformation notation for a model?
AWS::SageMaker::Model
What is a real-time inference?
Low latency requirements.
What is SageMaker Serverless Inference?
No management of infrastructure.
Ideal for a workload that has idle periods and uneven traffic over time.
Must tolerate cold starts.
What is Asynchroneous Inference?
Queues and processes requests asynchronously.
Good for inferences that may take some time to process.
What is autoscaling inference?
Dynamically adjusts compute resources for endpoints based on traffic.
What is SageMaker Inference Recommender?
It recommends the best instance type and configuration for your model.
What is an inference pipeline?
A linear sequence of 2 -15 containers that handle both real-time inference and batch transforms.
