22 February 2024

Implement MLOps practices with

Amazon SageMaker
Gaurav Singh Smiti Guru
Senior Solutions Architect Senior Solutions Architect
AWS India AWS India

© 2024, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Agenda

• What is MLOps

• What and why MLOps on AWS

Amazon SageMaker Studio, Amazon SageMaker Processing, Amazon SageMaker Pipelines, Amazon
SageMaker Model Registry, Amazon SageMaker projects, Amazon SageMaker ML Lineage Tracking,
Amazon SageMaker Feature Store

• Demo: Data scientist and MLOps on AWS

• Recap

© 2024, Amazon Web Services, Inc. or its affiliates. All rights reserved.
ML and path to production
Machine learning code and data science notebooks…

ML
code

© 2024, Amazon Web Services, Inc. or its affiliates. All rights reserved.
ML and path to production
…is one small part of the overall deployment picture

Machine Monitoring
Data verification resource
management

Configuration
Data collection
ML Serving
code Analysis tool infrastructure

Feature Process
extraction management tools

© 2024, Amazon Web Services, Inc. or its affiliates. All rights reserved.
ML and path to production
Considerations and challenges that lead to MLOps

• Culture
• Lack of cross-functional teams
• Priorities and needs (personas)
• Organizational structure
• Skillsets
• Unique aspects of ML lifecycle

© 2024, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Amazon SageMaker for MLOps

© 2024, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Amazon SageMaker Studio
ML experimentation on AWS with Amazon SageMaker Studio notebooks

© 2024, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Amazon SageMaker Processing
Automate pre-processing data and offload data scientists from managing containers

Container

Code
Inputs
Outputs

Supported pre-built processors: +

Custom containers
Framework processors

© 2024, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Amazon SageMaker Pipelines
Automated workflows to pre-process, train, evaluate, and register models

Step types:
• Process
• Train
• Tune
• Create model
• Register model
• Transform
• Condition
• Call back
• AWS Lambda function
• Clarify check
• Quality check
• Amazon EMR
• Fail

© 2024, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Amazon SageMaker Model Registry

Create model groups in Benchmark and observe the model Track the activities, metrics, and
your model registry versions in your model group and settings per model version
promote versions of the model by
changing their status

© 2024, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Amazon SageMaker projects
Example repositories and CI/CD pipelines per Amazon SageMaker project
Feature branch
Feature and development branches can be used
Feature branch for the development of new features/versions
of the Amazon Sagemaker Pipelines
Model registry
Development branch
Main model group
Main branches merge requests deploy the
Amazon Sagemaker Pipelines which will
Model build repository be used to (re-)train the production ready v1 v2 v3

AWS CodeCommit models ml-training-

Production ready
models
pipeline
Main branch

1. Validate 2. Code level 4. Build 5. Execute

3. Build
repo structure test e.g. unit Amazon Amazon
custom
meet static code Sagemaker Sagemaker
Model build CI/CD containers
requirement analysis Pipelines Pipelines
AWS CodePipeline

Main branch
Main branches contains endpoint
Model registry configuration
New model Changes configuration and testing code
Main model group
version event committed to
Model deploy repository in registry the main branch Pre-production environment Production environment

AWS CodeCommit v1 v2
Production ready
v3 testing

models Amazon
EventBridge 1. Manual 2. Model 3. Testing 4. Manual 5. Deploy
approval endpoint approval to
deployment production

Model Deploy CI/CD

AWS CodePipeline Endpoint Endpoint

© 2024, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Amazon SageMaker ML Lineage Tracking
Track workflow steps, model and dataset lineage, and establish model governance and audit

The query parses the

graph and returns the
lineage results

© 2024, Amazon Web Services, Inc. or its affiliates. All rights reserved.
 Feature drift

Challenges
of separate feature Feature duplication
stores

Slow model development/

deployment

© 2024, Amazon Web Services, Inc. or its affiliates. All rights reserved.
How it works: Amazon SageMaker Feature Store

Real-time
inference
Online
Streaming feature
store
Batch
inference
Offline
Batch feature
Amazon store
SageMaker Model
training
Feature Store
Store, discover, and
Raw data Feature processing share features for Ingest data Store Serve
Data in its original Transform raw data into machine learning Move streaming features Online and offline feature Features for real-time and
form that has not meaningful features for or batch features to a central stores maintaining batch applications for model
been processed better models repository consistency and accuracy training

© 2024, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Integrating Amazon SageMaker Feature Store
Feature authoring Feature discovery Online inference

Amazon SageMaker Amazon SageMaker Amazon SageMaker

Data Wrangler Studio hosting

AWS Glue DataBrew AWS Lambda

Feature processing

Amazon EMR

Training, batch scoring

AWS Glue
Amazon SageMaker
Amazon SageMaker Feature Store Amazon SageMaker
Processing training

Amazon SageMaker
Streaming
batch transform

Amazon Kinesis
Amazon Athena

Amazon Managed
Streaming for Apache Kafka

Feature AWS Step Amazon SageMaker Apache

pipelines Functions Pipelines Airflow

© 2024, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Demo:
Data scientist and MLOps on AWS

© 2024, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Recap
• What is MLOps and why it matters
• Challenges in ML lifecycle
• Amazon SageMaker tools to address these challenges:
• Amazon SageMaker Studio
• Amazon SageMaker Pipelines
• Amazon SageMaker Model Registry
• Amazon SageMaker projects
• Amazon SageMaker ML Lineage Tracking
• Amazon SageMaker Feature Store

• Demo - automating workflows and CI/CD pipelines with Amazon SageMaker

© 2024, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Additional resources
Getting started with Amazon SageMaker Pipelines:

• Amazon SageMaker Pipelines brings DevOps capabilities to your machine learning projects

• Building, automating, managing, and scaling ML workflows using Amazon SageMaker Pipelines

Custom Amazon SageMaker projects:

• Build custom Amazon SageMaker project templates – best practices

• Create Amazon SageMaker projects using third-party source control and Jenkins

• Build MLOps workflows with Amazon SageMaker projects, GitLab, and GitLab pipelines

• Multi-account model deployment with Amazon SageMaker Pipelines

Demo resources and code repositories:

• Amazon SageMaker custom project templates

• Amazon SageMaker MLOps with Amazon SageMaker Feature Store and Data Wrangler

© 2024, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Visit the AI and machine learning resource hub

Dive deeper into these resources, get inspired and learn how you can use
AI and machine learning to accelerate your business outcomes.

• 7 leading machine learning use cases

• Achieving transformative business results with artificial intelligence
• 6 steps to success with generative AI
• Maximize the business value of generative AI
• Answering your 4 biggest questions about generative AI security
• Automate document data extraction and analysis
• Drive business growth with personalization
• … and more!
Visit resource hub

© 2024, Amazon Web Services, Inc. or its affiliates. All rights reserved.
AWS Training and Certification
Access 600+ free digital courses with AWS Skill Builder

Focus on the cloud skills and services that are most relevant to you
across 30+ AWS solutions, including digital self-paced learning
plans and ramp up guides.

• Build your future in the AWS Cloud at your own pace

• Advance your skills and knowledge with learning plans
• Validate your cloud expertise with AWS Certification

Learn your way. Visit skillbuilder.aws »

© 2024, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Why work with an AWS Partner?
AWS Partners are uniquely positioned to help your organization at any
stage of your cloud adoption journey, providing
• Innovation – innovative and cost-effective scalable cloud solutions and
capabilities for your organization — helping you keep pace with cutting edge
technology changes
• Expertise – strategic experts and experienced builders, providing
groundbreaking, relevant and reliable solutions to help your business grow.
• Global Reach – Choose from a global community of trusted AWS Partners
across software, hardware, and services.

Find and connect with AWS Partner

© 2024, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Thank you for attending AWS Innovate - AI/ML and Data Edition
We hope you found it interesting! A kind reminder to complete the survey.
Let us know what you thought of today’s event and how we can improve the event
experience for you in the future.

[email protected]

twitter.com/AWSCloud

facebook.com/AmazonWebServices

youtube.com/user/AmazonWebServices

slideshare.net/AmazonWebServices

twitch.tv/aws

© 2024, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Thank you!

© 2024, Amazon Web Services, Inc. or its affiliates. All rights reserved.