Deployment options of SageMaker
Real time
Serverless
Serverless
Batch
Describe SageMaker Real-Time Deployments (max size, max time, purpose, where response stored)
one predication at time. More configuration.
Up to 6 MB
Max 60 sec
Fast near instant predicitions
Describe SageMaker Serverless Deployments (max size, max time, purpose, where response stored)
Idle period between traffic spikes, can tolerate more latency (cold starts). Less config but slower on cold start
Up to 4MB
Max 60 sec
Describe SageMaker Async Deployments (max size, max time, purpose, where response stored)
Large range payload sizes up to 1 GB for long processing times. Near real time latency requirement.
Request and responses are in S3
Max 1 hr
Describe SageMaker Batch Deployments (max size, max time, purpose, where response stored)
Prediction for entire dataset (multiple predications)
Request and Response in S3
Up to 100MB per invocation
Max 1 Hr
Bulk processing for large datasets
SageMaker feature that can automatically choose hyperparameter ranges, search strategy, and when and early stop for tuning (can save $$)
AMT (automatic model tuning)
SageMaker feature to prepare tabular and image data for machine learning with preparation, transformation and feature engineering. Single interface for data selection, exploration, visualization and processing
SageMaker Data Wrangler
What SageMaker feature can I use to generate new instances of data for underrepresented groups
SageMaker Data Wrangler (Augmented Data)
SageMaker Service that allows for the Ingest features from a variety of sources and transform the data into a feature
SageMaker Feature Store
Benefits of SageMaker Feature Store
Can publish directly from SageMaker Data Wranger into Feature store
Features are discoverable within SageMaker Studio
SageMaker Service that allows you to evaulate Foundation Model performance (model A vs model B)
SageMaker Clarify
Which SageMaker feature allows you to understand how/why ML model is making certain prediction and detect and explain bias in your dataset and models
SageMaker Clarify
Understand how/why ML model is making certain predictions. Great for debugging predictions even after deployed. Helps increase the trust and understanding of the model “why did the model make an incorrect prediction?”
SageMaker Clarify (Model Explainability)
SageMaker feature that uses human feedback to implement Reinforcement Learning.
SageMaker Ground Truth
SageMaker feature that has Essential model information in once place. Used often for document risk and rating.
SageMaker Model Cards
SageMaker feature that is a ceneralized repo for all machine learning models to help track which models are deployed and more
SageMaker - Model Dashboard
SageMaker feature that allows you to define role for personals (Data scientists, MLOps Engineer). Considered a Data Governance Service.
SageMaker Role Managers
SageMaker feature that allow you to monitor quality of models in production by setting up alerts for deviations in the model quality
SageMaker Model Monitor
SageMaker feature that is a central repo allow you to track mange and version ML models. Catalog models, manage model version, associate metadata with model
SageMaker Model Registry
SageMaker CI/CD feature for machine learning. Workflow that automates process of building, training and deploying ML model
SageMaker Pipeline
What are the steps in the SageMaker Pipeline
Processing - feature engineering
Training - for training model
Tuning - hyperparameter Tuning
AutoML - auto train model
Model - create or register a SageMaker model
ClarifyCheck - drift check against baselines (data bias, modal bias, model explainability)
Quality Check - perform drift checks against baselines (data quality, model quality)
SageMaker ML Hub to find pre-trained FM computer vision models, or NLP models
SageMaker JumpStart
Difference between SageMaker JumpStart and Bedrock
JumpStart has a LOT more FM to chose from
SageMaker feature that allows you to build ML models using a visual interface with no coding required.
SageMaker Canvas
