# Amazon SageMaker

Cloud service · AWS. Updated 2026-11-12. Page: https://blowstack.com/tools/cloud-services-hub/amazon-sagemaker

> Amazon SageMaker is a unified platform for data, analytics, and AI, bringing together AWS machine learning and analytics capabilities with unified access to data.

Practice exams covering Amazon SageMaker: [AIB-C01 (beta) AWS Certified AI Business Strategist](https://blowstack.com/practice-exams/aws-certified-ai-business-strategist), [AIF-C01 AWS Certified AI Practitioner](https://blowstack.com/practice-exams/aws-certified-ai-practitioner), [CLF-C02 AWS Certified Cloud Practitioner](https://blowstack.com/practice-exams/aws-certified-cloud-practitioner), [DEA-C01 AWS Certified Data Engineer - Associate](https://blowstack.com/practice-exams/aws-certified-data-engineer-associate), [AIP-C01 AWS Certified Generative AI Developer - Professional](https://blowstack.com/practice-exams/aws-certified-generative-ai-developer-professional), [MLA-C02 (beta) AWS Certified Machine Learning Engineer - Associate](https://blowstack.com/practice-exams/aws-certified-machine-learning-engineer-associate), [SCS-C03 AWS Certified Security - Specialty](https://blowstack.com/practice-exams/aws-certified-security-specialty), [SAP-C02 AWS Certified Solutions Architect - Professional](https://blowstack.com/practice-exams/aws-certified-solutions-architect-professional).

## 10 key features of Amazon SageMaker

Amazon SageMaker simplifies the entire machine learning process by offering a fully managed service with built-in algorithms, one-click training and deployment, automatic model tuning, a comprehensive IDE, elastic inference, multi-model endpoints, robust security, and support for both real-time and batch processing.

1. **Fully Managed Service**: Amazon SageMaker offers a fully managed service, enabling data scientists and developers to build, train, and deploy machine learning models quickly without managing the underlying infrastructure.
2. **Built-In Algorithms and Frameworks**: SageMaker provides several built-in algorithms and supports multiple machine learning frameworks, including TensorFlow, PyTorch, and MXNet, making it easier to start model training without the need to write algorithms from scratch.
3. **One-Click Training and Deployment**: With SageMaker, you can easily train your model with a single click and then directly deploy it to a production-ready hosted environment, simplifying the process from model creation to deployment.
4. **Automatic Model Tuning**: Amazon SageMaker autotunes your models by automatically adjusting hundreds of different combinations of algorithm parameters to arrive at the most accurate predictions.
5. **Integrated Development Environment**: SageMaker Studio provides a fully integrated development environment (IDE) for building, training, and deploying models. It offers a visual interface where you can easily perform all machine learning development steps.
6. **Multi-Model Endpoints**: SageMaker offers support for multi-model endpoints, allowing you to optimize inference costs by adjusting the compute capacity according to the demand and deploying multiple models on a single endpoint.
7. **Secure and Compliant**: Ensuring security and compliance, SageMaker is designed to meet rigorous standards, including HIPAA eligibility and compliance with frameworks like GDPR, helping organizations manage and protect user data effectively.
8. **Real-Time and Batch Processing**: SageMaker supports both real-time and batch processing for inference, enabling you to choose the most suitable option for your application, whether it requires immediate responses or can handle asynchronous batch requests.
9. **Asset Publishing and Subscriptions**: Share and collaborate on data and AI assets through publishing and subscribing workflows.
10. **Generative AI Output Guardrails**: Apply Amazon Bedrock Guardrails in SageMaker to protect and filter model outputs for responsible generative AI application development.

## Where Amazon SageMaker is used

Amazon SageMaker enables the development and deployment of machine learning models, automates data preparation, supports real-time fraud detection, facilitates predictive maintenance in manufacturing, and helps create personalized recommendation systems.

1. **Developing and Deploying Machine Learning Models**: Amazon SageMaker provides a comprehensive platform for developers and data scientists to easily build, train, and deploy machine learning models at scale. The platform simplifies the process of model development by providing pre-built algorithms and support for popular frameworks, enabling rapid experimentation and iteration. Once a model is ready, SageMaker offers tools for easy deployment, allowing models to be quickly integrated into applications and utilized for predictions.
2. **Automating Data Preparation**: With Amazon SageMaker Data Wrangler, users can simplify the process of data preparation for machine learning. This feature provides a visual interface for users to clean, explore, and visualize their data without the need for extensive code, making it faster and more intuitive to prepare datasets for training.
3. **Real-time Fraud Detection**: Amazon SageMaker can be used to develop and deploy machine learning models that help in identifying potentially fraudulent activities in real-time. By leveraging its powerful computing resources and scalable environment, SageMaker enables organizations to process vast amounts of transaction data in real-time, apply sophisticated algorithms to detect anomalies, and flag suspicious transactions, significantly reducing the risk of fraud.
4. **Predictive Maintenance in Manufacturing**: Manufacturing companies can utilize Amazon SageMaker to predict equipment failures before they happen, minimizing downtime and maintenance costs. By analyzing historical operation data, SageMaker can train models to forecast equipment malfunctions and schedule maintenance proactively, ensuring optimal operation and efficiency in manufacturing processes.
5. **Personalized Recommendations**: Amazon SageMaker allows for the creation of personalized recommendation systems that can enhance customer experiences and drive sales. Using machine learning algorithms, companies can analyze customer data and past interactions to generate individualized product or service recommendations, significantly improving customer engagement and loyalty.

## How Amazon SageMaker is priced

Amazon SageMaker pricing includes On-Demand for flexible usage, Savings Plans for committed use with discounts, and Spot Instances offering deep discounts for temporary compute capacity.

- **On-Demand Pricing**: Pay for SageMaker as you use it, based on the resources consumed. This includes instance hours, data processing, storage used, and data transfer. It's suitable for flexible, as-needed usage without upfront commitments.
- **Savings Plans**: Offer a significant discount compared to on-demand pricing, in exchange for a commitment to use a specific amount of resource (measured in $/hour) for a one or three-year period.
- **Spot Instances**: Allows users to take advantage of unused Amazon EC2 capacity at up to a 90% discount compared to On-Demand prices. However, these instances can be terminated by AWS at any moment if the capacity is needed elsewhere.
- **Data Agent Credits**: SageMaker Data Agent, the AI agent within SageMaker notebooks, charges $0.04 per credit consumed by requests. Credit consumption depends on the work performed and is metered to two decimal places, with a minimum of 0.01 credits per request.
- **Amazon Q Developer Subscription**: Amazon Q Developer is available within SageMaker Unified Studio and charges a fixed cost per user per month.
- **SageMaker Catalog Usage**: SageMaker Catalog charges for requests, metadata storage, compute, and AI recommendations for business descriptions, with no upfront commitments or minimum fees. Requests, metadata storage, and compute have monthly free allowances; certain core APIs are always free.
- **Lakehouse Usage**: Lakehouse metadata storage and API requests follow AWS Glue Data Catalog pricing. Data storage and access charges depend on Amazon S3 or Amazon Redshift Managed Storage, while automated statistics collection and Apache Iceberg table maintenance incur additional usage-based fees. Fine-grained permissions powered by AWS Lake Formation have no extra charge.
- **Unified Studio Free Tier**: SageMaker Unified Studio includes always-free core requests for manual IAM Identity Center domain creation, project management, user management, and policy grants. Existing AWS Free Tier allocations or pay-per-use agreements also apply to AWS services used through Unified Studio.
- **Redshift Reserved Instances**: SQL analytics with Amazon Redshift includes reserved instances, which provide cost savings for long-term commitments.

## Services Amazon SageMaker works with

- **[Amazon S3](https://blowstack.com/tools/cloud-services-hub/amazon-s3)**: Used for storing and retrieving large datasets, model artifacts, and other data needed for training and inference.
- **[AWS Identity and Access Management (IAM)](https://blowstack.com/tools/cloud-services-hub/aws-identity-and-access-management-(iam))**: Provides fine-grained access control policies to manage permissions for SageMaker resources and actions.
- **[AWS Lambda](https://blowstack.com/tools/cloud-services-hub/aws-lambda)**: Enables integration with SageMaker endpoints for real-time data processing and event-driven workflows.
- **[Amazon CloudWatch](https://blowstack.com/tools/cloud-services-hub/amazon-cloudwatch)**: Provides monitoring and logging for SageMaker resources, allowing you to track metrics and set alarms.
- **[AWS Glue](https://blowstack.com/tools/cloud-services-hub/aws-glue)**: Offers data preparation and transformation capabilities to preprocess data before feeding it into SageMaker models.
- **[Amazon Redshift](https://blowstack.com/tools/cloud-services-hub/amazon-redshift)**: Allows you to perform advanced analytics and store large amounts of structured data, which can be used in SageMaker.
- **[AWS Data Pipeline](https://blowstack.com/tools/cloud-services-hub/aws-data-pipeline)**: Enables data workflow automation to transport data between various AWS services and SageMaker.
- **[AWS Step Functions](https://blowstack.com/tools/cloud-services-hub/aws-step-functions)**: Facilitates building and orchestrating machine learning workflow pipelines using SageMaker.
- **[Amazon API Gateway](https://blowstack.com/tools/cloud-services-hub/amazon-api-gateway)**: API Gateway can expose SageMaker inference endpoints through a REST API using an AWS service integration. This adds API authentication, throttling, and request mapping in front of model inference.
- **[Amazon Athena](https://blowstack.com/tools/cloud-services-hub/amazon-athena)**: SageMaker notebooks, Data Wrangler, and Unified Studio can query data with Athena for analysis and machine learning preparation. Studio also supports propagating users' identities to Athena for governed data access.
- **[Amazon Augmented AI (Amazon A2I)](https://blowstack.com/tools/cloud-services-hub/amazon-augmented-ai-(amazon-a2i))**: SageMaker inference workflows can use Amazon A2I to send selected predictions to human reviewers. A custom human-review workflow can, for example, route low-confidence model results to a review workforce.
- **[Amazon Aurora](https://blowstack.com/tools/cloud-services-hub/amazon-aurora)**: SageMaker Lakehouse can receive operational data through Aurora zero-ETL integrations. Aurora Machine Learning also lets SQL queries invoke SageMaker endpoints, such as scoring customer records without moving inference logic into an application.
- **[Amazon Aurora Serverless](https://blowstack.com/tools/cloud-services-hub/amazon-aurora-serverless)**: Supported Aurora Serverless v2 configurations can use Aurora Machine Learning to invoke SageMaker endpoints from SQL. Supported Serverless v2 databases can also supply data to SageMaker Lakehouse through Aurora zero-ETL integrations.
- **[Amazon Bedrock](https://blowstack.com/tools/cloud-services-hub/amazon-bedrock)**: SageMaker Unified Studio integrates Bedrock tools for developing generative AI applications alongside analytics and machine learning workflows. Supported models trained or fine-tuned in SageMaker can also be imported into Bedrock through Custom Model Import.
- **[Amazon CodeWhisperer](https://blowstack.com/tools/cloud-services-hub/amazon-codewhisperer)**: SageMaker Studio integrated CodeWhisperer code suggestions into its notebook development environment. These capabilities are now part of Amazon Q Developer.
- **[Amazon Cognito](https://blowstack.com/tools/cloud-services-hub/amazon-cognito)**: SageMaker Ground Truth can use Cognito user pools to authenticate members of private labeling workforces. Workers sign in to the labeling portal with identities managed in the associated user pool.
- **[Amazon Comprehend](https://blowstack.com/tools/cloud-services-hub/amazon-comprehend)**: SageMaker Canvas provides ready-to-use text analysis models powered by Comprehend. Users can perform tasks such as sentiment analysis, entity extraction, and personally identifiable information detection without training a custom model.
- **[Amazon DataZone](https://blowstack.com/tools/cloud-services-hub/amazon-datazone)**: SageMaker Unified Studio uses DataZone capabilities for domains, projects, and governed data collaboration. SageMaker Catalog builds on DataZone to help users discover data assets and request access for analytics and AI work.
- **[Amazon DynamoDB](https://blowstack.com/tools/cloud-services-hub/amazon-dynamodb)**: SageMaker Lakehouse can receive DynamoDB data through zero-ETL integration capabilities in AWS Glue. This makes operational table data available for analytics and machine learning without maintaining a custom extraction pipeline.
- **[Amazon EC2](https://blowstack.com/tools/cloud-services-hub/amazon-ec2)**: SageMaker provisions and manages EC2-backed compute for notebooks, training, processing, and model hosting. Training jobs can also use managed Spot Training to run on spare EC2 capacity with interruption handling.
- **[Amazon Elastic Block Store (Amazon EBS)](https://blowstack.com/tools/cloud-services-hub/amazon-elastic-block-store-(amazon-ebs))**: SageMaker uses EBS volumes for notebook storage and for storage attached to supported training, processing, and inference instances. Volume size and encryption settings can be configured for the relevant SageMaker resources.
- **[Amazon Elastic Container Registry (Amazon ECR)](https://blowstack.com/tools/cloud-services-hub/amazon-elastic-container-registry-(amazon-ecr))**: SageMaker pulls training, processing, and inference container images from ECR. Custom algorithms and model-serving environments can be packaged as images in private ECR repositories.
- **[Amazon Elastic File System (Amazon EFS)](https://blowstack.com/tools/cloud-services-hub/amazon-elastic-file-system-(amazon-efs))**: SageMaker Studio can use EFS for persistent user files, and supported training jobs can read training data directly from EFS. This allows notebooks or training jobs to work with a shared file system instead of copying every dataset into local storage.
- **[Amazon Elastic Kubernetes Service (Amazon EKS)](https://blowstack.com/tools/cloud-services-hub/amazon-elastic-kubernetes-service-(amazon-eks))**: SageMaker HyperPod supports EKS as an orchestrator for resilient machine learning clusters. SageMaker Operators for Kubernetes also let Kubernetes workloads create and manage SageMaker jobs and inference resources.
- **[Amazon EMR](https://blowstack.com/tools/cloud-services-hub/amazon-emr)**: SageMaker Studio notebooks can connect to EMR clusters and EMR Serverless applications to run Spark data processing. Supported connections can propagate the user's identity for governed access to data.
- **[Amazon EventBridge](https://blowstack.com/tools/cloud-services-hub/amazon-eventbridge)**: SageMaker publishes job, pipeline, endpoint, and other resource state-change events to EventBridge. EventBridge rules and schedules can also start SageMaker pipelines, enabling event-driven retraining and deployment workflows.
- **[Amazon Forecast](https://blowstack.com/tools/cloud-services-hub/amazon-forecast)**: SageMaker Canvas has integrated Amazon Forecast technology for building time-series forecasting models through its no-code interface. This supports workflows such as predicting product demand from historical sales data.
- **[Amazon Fraud Detector](https://blowstack.com/tools/cloud-services-hub/amazon-fraud-detector)**: Models hosted on SageMaker endpoints can be imported into Fraud Detector and used in fraud detection workflows. Their prediction scores can be combined with detector rules and other fraud signals.
- **[Amazon FSx](https://blowstack.com/tools/cloud-services-hub/amazon-fsx-(for-all-types))**: SageMaker training jobs can use FSx for Lustre file systems as training data sources. This provides high-throughput shared file access for large datasets, including datasets linked to S3.
- **[Amazon Kinesis Video Streams](https://blowstack.com/tools/cloud-services-hub/amazon-kinesis-video-streams)**: AWS provides an inference integration template that reads video from Kinesis Video Streams and sends frames to a SageMaker endpoint. This supports applications such as detecting objects in live camera footage.
- **[Amazon Managed Workflows for Apache Airflow (Amazon MWAA)](https://blowstack.com/tools/cloud-services-hub/amazon-managed-workflows-for-apache-airflow-(amazon-mwaa))**: MWAA can orchestrate SageMaker through Airflow's SageMaker operators and sensors. A DAG can start training or processing jobs, wait for completion, and deploy the resulting model.
- **[Amazon Neptune](https://blowstack.com/tools/cloud-services-hub/amazon-neptune)**: Neptune ML uses SageMaker to train and host graph machine learning models. Neptune queries can use these models for tasks such as predicting missing relationships or classifying graph nodes.
- **[Amazon OpenSearch Service](https://blowstack.com/tools/cloud-services-hub/amazon-opensearch-service)**: OpenSearch Service supports machine learning connectors that invoke models hosted on SageMaker endpoints. Search and ingest pipelines can use these models to generate embeddings or enrich documents.
- **[Amazon Q](https://blowstack.com/tools/cloud-services-hub/amazon-q)**: Amazon Q Developer is integrated into SageMaker development environments to assist with code, explanations, and troubleshooting. In Unified Studio, it also assists with analytics and data development tasks.
- **[Amazon QuickSight](https://blowstack.com/tools/cloud-services-hub/amazon-quicksight)**: QuickSight can augment datasets with predictions from supported SageMaker models. For example, a dataset can gain a predicted churn score before it is used in dashboards.
- **[Amazon RDS](https://blowstack.com/tools/cloud-services-hub/amazon-rds)**: SageMaker Data Wrangler can import data from supported RDS databases through database connections. Supported RDS for MySQL databases can also provide operational data to SageMaker Lakehouse through zero-ETL integrations.
- **[Amazon Rekognition](https://blowstack.com/tools/cloud-services-hub/amazon-rekognition)**: SageMaker Canvas exposes ready-to-use image analysis models powered by Rekognition. Users can detect objects or extract text from images without building an image model themselves.
- **[Amazon SageMaker Data Wrangler](https://blowstack.com/tools/cloud-services-hub/amazon-sagemaker-data-wrangler)**: Data Wrangler prepares and transforms data for SageMaker machine learning workflows. Its processing flows can be exported into SageMaker jobs and pipelines to repeat the same preparation during training.
- **[Amazon SageMaker Ground Truth](https://blowstack.com/tools/cloud-services-hub/amazon-sagemaker-ground-truth)**: Ground Truth creates labeled datasets for training SageMaker models using human workers and automated labeling. Its output manifests can be consumed by SageMaker training workflows.
- **[Amazon Security Lake](https://blowstack.com/tools/cloud-services-hub/amazon-security-lake)**: SageMaker can consume Security Lake data through a subscriber integration to build and train security-focused machine learning models. This supports tasks such as analyzing normalized security events for anomalous activity.
- **[Amazon Simple Notification Service (Amazon SNS)](https://blowstack.com/tools/cloud-services-hub/amazon-simple-notification-service-(amazon-sns))**: SageMaker asynchronous inference can publish success and failure notifications to SNS topics. Ground Truth also uses SNS topics for streaming labeling inputs and labeling results.
- **[Amazon Textract](https://blowstack.com/tools/cloud-services-hub/amazon-textract)**: SageMaker Canvas provides ready-to-use document analysis capabilities powered by Textract. Users can extract information from documents, receipts, and identity documents without training their own extraction model.
- **[Amazon VPC](https://blowstack.com/tools/cloud-services-hub/amazon-vpc)**: SageMaker resources can connect to customer VPC subnets and use security groups to control network access. Studio can also operate in VPC-only mode so application traffic uses customer-controlled networking.
- **[AWS Auto Scaling](https://blowstack.com/tools/cloud-services-hub/aws-auto-scaling)**: SageMaker inference endpoints integrate with Application Auto Scaling to adjust provisioned inference capacity. Scaling policies can respond to metrics such as invocations per instance or follow a schedule.
- **[AWS Clean Rooms](https://blowstack.com/tools/cloud-services-hub/aws-clean-rooms)**: AWS Clean Rooms ML integrates with SageMaker AI for custom machine learning workflows on collaboration data. Collaborators can train custom models while applying Clean Rooms controls to the underlying data.
- **[AWS CloudFormation](https://blowstack.com/tools/cloud-services-hub/aws-cloudformation)**: CloudFormation can provision SageMaker resources, including domains, notebook instances, models, endpoints, and pipelines. This allows machine learning infrastructure to be deployed and updated as versioned templates.
- **[AWS CloudTrail](https://blowstack.com/tools/cloud-services-hub/aws-cloudtrail)**: SageMaker API activity is recorded in CloudTrail for auditing and investigation. Supported SageMaker data events can also capture activity such as model endpoint invocations when data-event logging is configured.
- **[AWS CodeArtifact](https://blowstack.com/tools/cloud-services-hub/aws-codeartifact)**: SageMaker Studio environments can be configured to install Python packages from CodeArtifact repositories. This provides controlled access to private dependencies used in notebooks, training code, and other development workflows.
- **[AWS CodeBuild](https://blowstack.com/tools/cloud-services-hub/aws-codebuild)**: SageMaker Projects MLOps templates use CodeBuild to build and test machine learning artifacts, including custom container images. These builds can feed model-building and deployment pipelines.
- **[AWS CodeCommit](https://blowstack.com/tools/cloud-services-hub/aws-codecommit)**: SageMaker notebook instances and Studio support Git-based workflows with CodeCommit repositories. SageMaker Projects templates can also use CodeCommit to store model-building and deployment code.
- **[AWS CodeConnections](https://blowstack.com/tools/cloud-services-hub/aws-codeconnections)**: SageMaker Projects templates can use CodeConnections connections to access supported external Git repositories through CodePipeline. Repository changes can then initiate the project's model-building or deployment workflow.
- **[AWS CodePipeline](https://blowstack.com/tools/cloud-services-hub/aws-codepipeline)**: SageMaker Projects uses CodePipeline in MLOps templates to automate model building and deployment. Pipelines can coordinate source changes, builds, approval steps, and deployment to SageMaker endpoints.
- **[AWS Config](https://blowstack.com/tools/cloud-services-hub/aws-config)**: AWS Config records supported SageMaker resource configurations and evaluates them against compliance rules. Managed rules can check notebook settings such as encryption, direct internet access, and root access.
- **[AWS Deep Learning Containers](https://blowstack.com/tools/cloud-services-hub/aws-deep-learning-containers)**: SageMaker supports AWS Deep Learning Containers as managed training and inference environments. These images supply framework runtimes and dependencies for workloads such as PyTorch and TensorFlow training.
- **[AWS DeepRacer](https://blowstack.com/tools/cloud-services-hub/aws-deepracer)**: DeepRacer uses SageMaker infrastructure to train reinforcement learning models. Training produces driving policies that can be evaluated in simulation and deployed to a DeepRacer vehicle.
- **[AWS IAM Identity Center (AWS Single Sign-On)](https://blowstack.com/tools/cloud-services-hub/aws-iam-identity-center-(aws-single-sign-on))**: SageMaker domains can use IAM Identity Center to authenticate users and provide access to Studio. Trusted identity propagation can carry the user's identity into supported analytics services for fine-grained authorization.
- **[AWS IoT Greengrass](https://blowstack.com/tools/cloud-services-hub/aws-iot-greengrass)**: Models developed with SageMaker can be packaged and deployed to Greengrass devices for local inference. SageMaker Neo-compiled models can use the Neo runtime on supported edge devices.
- **[AWS Key Management Service (AWS KMS)](https://blowstack.com/tools/cloud-services-hub/aws-key-management-service-(aws-kms))**: SageMaker uses KMS keys to encrypt supported storage volumes, model artifacts, job outputs, and other protected resources. Customer-managed keys let organizations control access to encrypted machine learning data.
- **[AWS Lake Formation](https://blowstack.com/tools/cloud-services-hub/aws-lake-formation)**: SageMaker integrates with Lake Formation for governed access to lake data. Supported Studio connections can propagate user identities so Lake Formation permissions apply to notebook queries.
- **[AWS Marketplace](https://blowstack.com/tools/cloud-services-hub/aws-marketplace)**: SageMaker can use machine learning algorithms and model packages subscribed to through AWS Marketplace. Compatible products can be used for managed training or deployed to SageMaker inference endpoints.
- **[AWS PrivateLink](https://blowstack.com/tools/cloud-services-hub/aws-privatelink)**: SageMaker supports interface VPC endpoints through PrivateLink for supported APIs, inference, and Studio connectivity. This lets clients access SageMaker privately without sending traffic through the public internet.
- **[AWS Resource Access Manager (AWS RAM)](https://blowstack.com/tools/cloud-services-hub/aws-resource-access-manager-(aws-ram))**: SageMaker Feature Store integrates with AWS RAM to share supported feature resources across accounts. This allows teams to discover and reuse centrally managed machine learning features.
- **[AWS Secrets Manager](https://blowstack.com/tools/cloud-services-hub/aws-secrets-manager)**: SageMaker uses Secrets Manager for supported connection credentials, including database connections used by Data Wrangler and repository credentials used by development workflows. Secrets can be referenced without embedding passwords or tokens in notebooks.
- **[AWS Security Hub](https://blowstack.com/tools/cloud-services-hub/aws-security-hub)**: Security Hub includes security controls for SageMaker notebook configurations. Findings can identify settings such as enabled direct internet access, enabled root access, or missing KMS encryption.
- **[AWS Security Token Service (AWS STS)](https://blowstack.com/tools/cloud-services-hub/aws-security-token-service-(aws-sts))**: SageMaker assumes execution roles through STS to obtain temporary credentials for jobs and hosted workloads. Those credentials authorize access to resources such as training data in S3 and container images in ECR.
- **[AWS Service Catalog](https://blowstack.com/tools/cloud-services-hub/aws-service-catalog)**: SageMaker Projects uses Service Catalog products to provision standardized MLOps environments. Administrators can publish approved project templates that create repositories, pipelines, and associated infrastructure.
- **[AWS Systems Manager](https://blowstack.com/tools/cloud-services-hub/aws-systems-manager)**: SageMaker supports remote debugging of training containers through Systems Manager Session Manager. Authorized developers can open a secure session to inspect a running training environment without configuring inbound SSH access.
- **[AWS Trusted Advisor](https://blowstack.com/tools/cloud-services-hub/aws-trusted-advisor)**: Trusted Advisor includes SageMaker checks that identify potentially idle or underutilized resources, such as notebook instances and inference endpoints. These recommendations help locate machine learning resources that may be generating unnecessary costs.

## What is Amazon SageMaker

Amazon SageMaker is a unified platform for data, analytics, and AI. It brings together AWS capabilities for model development, generative AI, data processing, and SQL analytics in a unified studio, with governed access to data in data lakes, data warehouses, and third-party or federated sources.

SageMaker Lakehouse unifies data across Amazon S3 data lakes and Amazon Redshift data warehouses, allowing Apache Iceberg–compatible tools and engines to query a single copy of analytics data.

SageMaker Unified Studio uses Amazon DataZone domains to organize assets and users, and projects to support collaboration, secure artifact sharing, and work across compute services.

SageMaker Unified Studio also provides a fully managed, serverless notebook with a built-in AI agent for analytics and AI work.

Unified Studio resource quotas apply at the account level, so exhausting a quota in one project can affect other projects in the same account.

Traditionally, the process of developing ML models involves a complex cycle of building, training, and tuning the model, followed by deployment and management of the model in production. SageMaker streamlines and simplifies this process significantly.

The existing machine learning capabilities for data wrangling, building, training, and deploying models are now called Amazon SageMaker AI. SageMaker AI is integrated into the broader SageMaker platform and remains available as a standalone service.

At the core of SageMaker's value proposition is its ability to significantly reduce the time and effort required to get machine learning models from concept to production. It achieves this through a combination of high-level tools and automatic features that handle the more laborious tasks associated with model development. For example, SageMaker automatically tunes your model by adjusting thousands of different combinations of algorithm parameters to arrive at the most effective predictions the model can provide. This process, known as hyperparameter optimization, is both time-consuming and complex but is made significantly more approachable through SageMaker.

Moreover, SageMaker is designed with flexibility in mind, allowing it to support nearly all algorithms and frameworks. This means developers and data scientists are not restricted in their choice of tools and can bring their own or use those pre-built and optimized on SageMaker. Its integration with popular frameworks and interfaces ensures that users can continue working with the tools they are familiar with, reducing the learning curve and accelerating the model development cycle.

For training jobs using an instance preference list, SageMaker AI checks the quota for every listed instance type, not just the selected type. Each type must have sufficient quota for its requested instance count; otherwise, the request is rejected even if an earlier preference could have succeeded.

Deployment and scalability are other areas where SageMaker excels. Once a model is ready, deploying it into production is as simple as a few clicks. SageMaker handles all the underlying infrastructure requirements, automatically scaling resources up or down based on demand to ensure that your application maintains high performance without incurring unnecessary costs. This managed service approach allows developers to focus on the model's performance and impact, rather than on managing servers and infrastructure.

SageMaker AI also offers Serverless Inference, which provisions and scales inference compute automatically without requiring you to select or manage serving instances.

Furthermore, SageMaker offers a secure environment for your machine learning workflow, including encryption and compliance with various standards, ensuring that your data and models are protected. Its integration with Amazon's cloud services ecosystem means that SageMaker can easily access and process large datasets stored in Amazon S3, use AWS Lambda for running serverless functions, or tap into a plethora of other services to enhance the functionality and efficiency of your machine learning projects.

SageMaker Unified Studio supports authentication with IAM user credentials, identity-provider credentials through AWS IAM Identity Center, or SAML credentials.

In essence, Amazon SageMaker democratizes machine learning by making it accessible to developers and data scientists irrespective of their machine learning expertise. It speeds up the experimentation and development phase, simplifies deployment, and manages the lifecycle of machine learning models, allowing teams to focus more on solving the problem at hand than on the underlying infrastructure and mechanics of machine learning.

## Free Amazon SageMaker tools

- [Amazon SageMaker GPU Instance Calculator](https://blowstack.com/tools/aws-sagemaker-gpu-instance-calculator): Find the ml.* GPU instances that hold your model on an endpoint or training job, their price and the quota code to request.
- [Amazon SageMaker Endpoint Error Decoder](https://blowstack.com/tools/aws-sagemaker-endpoint-error-decoder): Paste an endpoint's FailureReason or invocation error and get its causes and the command to check each.
- [Amazon SageMaker Quota Finder](https://blowstack.com/tools/aws-sagemaker-quota-finder): Pick an endpoint, job, cluster, notebook or Studio space and get every quota it counts against, with the value to request.
- [Amazon SageMaker Inference Cost Calculator](https://blowstack.com/tools/aws-sagemaker-inference-cost-calculator): Compare a workload's monthly cost on real-time, serverless and asynchronous endpoints and batch transform, with the break-even volumes.

## Concepts related to Amazon SageMaker

- [A/B Testing](https://blowstack.com/tools/cloud-concepts-hub/ab-testing): A/B testing is a controlled experiment that randomly assigns comparable users or requests to two variants and measures which performs better against predefined criteria.
- [Apache Parquet](https://blowstack.com/tools/cloud-concepts-hub/apache-parquet): Apache Parquet is an open, column-oriented binary file format designed for efficient storage and analytical processing of structured and nested data.
- [Asynchronous Inference](https://blowstack.com/tools/cloud-concepts-hub/asynchronous-inference): Asynchronous inference is a model-serving pattern that accepts prediction requests without keeping the caller waiting for the model’s result.
- [Batch Size](https://blowstack.com/tools/cloud-concepts-hub/batch-size): Batch size is the number of training examples processed together, with its precise meaning determined by whether it refers to a device-level batch or an optimizer update.
- [Bayesian Optimization](https://blowstack.com/tools/cloud-concepts-hub/bayesian-optimization): Bayesian optimization is a strategy for optimizing expensive black-box functions by using a probabilistic model of previous evaluations to choose promising inputs.
- [Blue-Green Deployment](https://blowstack.com/tools/cloud-concepts-hub/blue-green-deployment): Blue-green deployment releases a new application version into a separate environment and then redirects production traffic from the current environment to the new one.
- [Canary Deployment](https://blowstack.com/tools/cloud-concepts-hub/canary-deployment): Canary deployment releases a new application or model version to a small portion of production traffic before expanding exposure after validation.
- [Canary Testing](https://blowstack.com/tools/cloud-concepts-hub/canary-testing): Canary testing validates a candidate configuration by exposing it to a small share of production traffic before deciding whether to expand or reverse the rollout.
- [Champion/Challenger Testing](https://blowstack.com/tools/cloud-concepts-hub/championchallenger-testing): Champion/challenger testing compares a currently accepted model with candidate replacements to determine whether a challenger should become the new production standard.
- [Class Resampling](https://blowstack.com/tools/cloud-concepts-hub/class-resampling): Class resampling changes how frequently examples from each target class appear during training to reduce the effects of class imbalance.
- [Classification Metrics](https://blowstack.com/tools/cloud-concepts-hub/classification-metrics): Classification metrics quantify how well a model assigns examples to discrete categories by comparing its predictions with known labels.
- [Continued Pre-Training](https://blowstack.com/tools/cloud-concepts-hub/continued-pre-training): Continued pre-training further trains an existing foundation model on additional data using a self-supervised objective to adapt its learned representations and knowledge.
- [Cosine Similarity](https://blowstack.com/tools/cloud-concepts-hub/cosine-similarity): Cosine similarity measures how closely two nonzero vectors point in the same direction, regardless of their magnitude.
- [Cross-Validation](https://blowstack.com/tools/cloud-concepts-hub/cross-validation): Cross-validation estimates how well a modeling approach generalizes by repeatedly training and evaluating it on different, appropriately separated portions of a dataset.
- [Data Augmentation](https://blowstack.com/tools/cloud-concepts-hub/data-augmentation): Data augmentation creates additional training examples by transforming existing data or generating controlled variations that preserve task-relevant meaning and valid labels.
- [Data Cataloging](https://blowstack.com/tools/cloud-concepts-hub/data-cataloging): Data cataloging is the practice of maintaining a searchable inventory of data assets and metadata describing their meaning, ownership, origins, and permitted use.
- [Data Drift](https://blowstack.com/tools/cloud-concepts-hub/data-drift): Data drift is a change in the statistical distribution of data over time or relative to a reference dataset that can affect a machine learning model’s performance.
- [Data Provenance](https://blowstack.com/tools/cloud-concepts-hub/data-provenance): Data provenance is the documented origin, ownership, and history of data, including how it was collected, transformed, and used.
- [Data Shuffling](https://blowstack.com/tools/cloud-concepts-hub/data-shuffling): Data shuffling is the random reordering of records to reduce unintended ordering effects during dataset preparation and machine learning training.
- [Data Validation](https://blowstack.com/tools/cloud-concepts-hub/data-validation): Data validation is the process of checking data against defined structural, quality, and business rules before accepting or using it.
- [Deployment Rollback](https://blowstack.com/tools/cloud-concepts-hub/deployment-rollback): Deployment rollback restores a previously validated application, infrastructure, or model release when a new deployment fails or causes unacceptable behavior.
- [Distributed Training](https://blowstack.com/tools/cloud-concepts-hub/distributed-training): Distributed training coordinates multiple compute devices or machines to train a single machine learning model.
- [Early Stopping](https://blowstack.com/tools/cloud-concepts-hub/early-stopping): Early stopping ends model training before its maximum duration when monitored performance indicates that further training is unlikely to provide sufficient benefit.
- [Embeddings](https://blowstack.com/tools/cloud-concepts-hub/embeddings): Embeddings are learned numerical representations that encode characteristics of data as vectors so that meaningful relationships can be compared mathematically.
- [Ensemble Learning](https://blowstack.com/tools/cloud-concepts-hub/ensemble-learning): Ensemble learning combines predictions or outputs from multiple models to produce a result that can outperform an individual model.
- [Environmental Sustainability](https://blowstack.com/tools/cloud-concepts-hub/environmental-sustainability): Environmental Sustainability is the practice of reducing the environmental impact of systems throughout their lifecycle while meeting required business and technical outcomes.
- [Epochs](https://blowstack.com/tools/cloud-concepts-hub/epochs): Epochs are complete passes through a training dataset, used to measure and control how long a machine learning model trains.
- [Experiment Tracking](https://blowstack.com/tools/cloud-concepts-hub/experiment-tracking): Experiment tracking is the systematic recording of machine learning runs, their inputs, configurations, outputs, and evaluation results so teams can compare and reproduce their work.
- [Feature Engineering](https://blowstack.com/tools/cloud-concepts-hub/feature-engineering): Feature engineering is the process of creating, transforming, and selecting input variables so machine learning models can learn useful patterns from data.
- [Fluency](https://blowstack.com/tools/cloud-concepts-hub/fluency): Fluency is the degree to which generated language reads naturally, with correct grammar, clear phrasing, and smooth sentence flow.
- [Graceful Degradation](https://blowstack.com/tools/cloud-concepts-hub/graceful-degradation): Graceful degradation is the deliberate preservation of essential service when failures or resource constraints require reduced functionality, performance, or output quality.
- [Gradient Clipping](https://blowstack.com/tools/cloud-concepts-hub/gradient-clipping): Gradient clipping limits the magnitude of gradients during model training to reduce unstable parameter updates.
- [Hallucinations](https://blowstack.com/tools/cloud-concepts-hub/hallucinations): Hallucinations are AI-generated outputs that present fabricated, incorrect, or unsupported information as though it were reliable.
- [Hypothesis Testing](https://blowstack.com/tools/cloud-concepts-hub/hypothesis-testing): Hypothesis testing uses sample data to evaluate whether evidence supports rejecting a defined claim about a population or system.
- [In-Context Learning](https://blowstack.com/tools/cloud-concepts-hub/in-context-learning): In-context learning is a foundation model’s ability to adapt its responses using instructions, examples, and information supplied at inference time without updating its model weights.
- [Incident Response Lifecycle](https://blowstack.com/tools/cloud-concepts-hub/incident-response-lifecycle): The incident response lifecycle is a repeatable process for preparing for, investigating, containing, resolving, and learning from security incidents.
- [Instruction Tuning](https://blowstack.com/tools/cloud-concepts-hub/instruction-tuning): Instruction tuning is fine-tuning a pretrained model on instructions paired with desired responses so it learns to follow requests more reliably.
- [Interpretability-Performance Tradeoff](https://blowstack.com/tools/cloud-concepts-hub/interpretability-performance-tradeoff): The interpretability-performance tradeoff is the balance between how readily people can understand a model’s decisions and how well the model meets predictive and operational requirements.
- [JSON](https://blowstack.com/tools/cloud-concepts-hub/json): JSON is a text-based data interchange format that represents structured values using objects, arrays, strings, numbers, booleans, and null.
- [Knowledge Distillation](https://blowstack.com/tools/cloud-concepts-hub/knowledge-distillation): Knowledge distillation trains a student model to reproduce a teacher model’s behavior, often preserving useful capabilities in a smaller, less expensive model.
- [KV Caching](https://blowstack.com/tools/cloud-concepts-hub/kv-caching): KV caching retains a transformer's attention key and value tensors for previously processed tokens so autoregressive generation can reuse them instead of recomputing them.
- [Lazy Loading](https://blowstack.com/tools/cloud-concepts-hub/lazy-loading): Lazy loading defers loading data, code, or model resources until they are actually needed.
- [Learning Rate Schedules](https://blowstack.com/tools/cloud-concepts-hub/learning-rate-schedules): Learning rate schedules are rules that adjust an optimizer’s learning rate during training to balance progress, stability, and convergence.
- [Lemmatization](https://blowstack.com/tools/cloud-concepts-hub/lemmatization): Lemmatization maps inflected word forms to their dictionary form, or lemma, using linguistic rules, vocabulary, and sometimes grammatical context.
- [LIME](https://blowstack.com/tools/cloud-concepts-hub/lime): LIME is a model-agnostic explanation technique that approximates a model’s behavior near one prediction with a simpler, interpretable model.
- [LLM-as-a-Judge](https://blowstack.com/tools/cloud-concepts-hub/llm-as-a-judge): LLM-as-a-Judge uses a large language model to assess generated responses against explicit evaluation criteria.
- [Low-Rank Adaptation](https://blowstack.com/tools/cloud-concepts-hub/low-rank-adaptation): Low-Rank Adaptation is a parameter-efficient fine-tuning technique that customizes a pretrained model by training small, low-rank weight updates while keeping the original model weights frozen.
- [Min-Max Scaling](https://blowstack.com/tools/cloud-concepts-hub/min-max-scaling): Min-Max Scaling is a feature transformation that uses minimum and maximum reference values to linearly map numeric data to a chosen range, commonly 0 to 1.
- [Missing Value Imputation](https://blowstack.com/tools/cloud-concepts-hub/missing-value-imputation): Missing value imputation replaces absent data values with justified estimates or explicit placeholders so datasets can be analyzed or used by machine learning systems.
- [Mixed Precision Training](https://blowstack.com/tools/cloud-concepts-hub/mixed-precision-training): Mixed precision training combines lower- and higher-precision numerical formats to accelerate model training while preserving numerical stability and model quality.
- [Model Cards](https://blowstack.com/tools/cloud-concepts-hub/model-cards): Model cards are structured documents that describe an AI model’s intended uses, evaluation results, limitations, and responsible-use considerations.
- [Model Distillation](https://blowstack.com/tools/cloud-concepts-hub/model-distillation): Model distillation transfers learned behavior from a teacher model to a student model, typically a smaller one, to reduce inference cost and resource requirements.
- [Model Drift](https://blowstack.com/tools/cloud-concepts-hub/model-drift): Model drift is a change over time in production data or its relationship to outcomes that can reduce a machine learning model’s predictive reliability.
- [Model Parallelism](https://blowstack.com/tools/cloud-concepts-hub/model-parallelism): Model parallelism divides a single machine learning model across multiple compute devices so they cooperate to train it or serve predictions.
- [Model Robustness](https://blowstack.com/tools/cloud-concepts-hub/model-robustness): Model robustness is an AI model’s ability to maintain acceptable performance under noisy inputs, meaningful operating changes, and deliberate attempts to disrupt its behavior.
- [Model Routing](https://blowstack.com/tools/cloud-concepts-hub/model-routing): Model routing selects the most suitable AI model for each request or task using capability, quality, cost, latency, and operational constraints.
- [Model Transparency](https://blowstack.com/tools/cloud-concepts-hub/model-transparency): Model transparency is the degree to which people can understand and inspect a model’s design, development, intended use, limitations, and decision-making behavior.
- [Multimodal Fusion](https://blowstack.com/tools/cloud-concepts-hub/multimodal-fusion): Multimodal fusion combines information from different data modalities, such as text, images, audio, and tables, to support a shared prediction, interpretation, or generated response.
- [Multimodal Generation](https://blowstack.com/tools/cloud-concepts-hub/multimodal-generation): Multimodal generation is the creation of content using generative AI that connects different data types, such as text, images, audio, and video, across its inputs and outputs.
- [Mutual TLS](https://blowstack.com/tools/cloud-concepts-hub/mutual-tls): Mutual TLS is a TLS configuration in which both communicating peers authenticate each other with digital certificates while encrypting data in transit.
- [N-Gram Overlap](https://blowstack.com/tools/cloud-concepts-hub/n-gram-overlap): N-gram overlap measures how many contiguous sequences of tokens a generated text shares with one or more reference texts.
- [Nondeterminism](https://blowstack.com/tools/cloud-concepts-hub/nondeterminism): Nondeterminism is the property of a system that allows repeated executions with the same apparent inputs to produce different results.
- [Nucleus Sampling](https://blowstack.com/tools/cloud-concepts-hub/nucleus-sampling): Nucleus sampling is a text-generation technique that samples each next token from the smallest set of likely tokens whose cumulative probability reaches a configured threshold, called top-p.
- [Online Inference](https://blowstack.com/tools/cloud-concepts-hub/online-inference): Online inference uses a deployed machine learning model to generate predictions or other outputs on demand as application requests arrive.
- [Open-Source Models](https://blowstack.com/tools/cloud-concepts-hub/open-source-models): Open-source models are AI and machine learning models released with artifacts and licenses that permit their use, study, modification, and redistribution.
- [Parameter-Efficient Fine-Tuning](https://blowstack.com/tools/cloud-concepts-hub/parameter-efficient-fine-tuning): Parameter-efficient fine-tuning adapts a pretrained model to a task by training a small subset of parameters or added trainable components instead of updating all model weights.
- [Precision@k](https://blowstack.com/tools/cloud-concepts-hub/precision@k): Precision@k measures the fraction of the top k retrieved results that are relevant to a query.
- [Price-Performance Ratio](https://blowstack.com/tools/cloud-concepts-hub/price-performance-ratio): Price-Performance Ratio measures the cost of delivering a defined level of useful performance under specified workload conditions.
- [Prompt Compression](https://blowstack.com/tools/cloud-concepts-hub/prompt-compression): Prompt compression reduces the tokens sent to a foundation model while preserving the instructions and information needed for an effective response.
- [Random Search](https://blowstack.com/tools/cloud-concepts-hub/random-search): Random Search is a hyperparameter optimization method that evaluates randomly sampled model configurations within a defined search space.
- [Recall@k](https://blowstack.com/tools/cloud-concepts-hub/recall@k): Recall@k measures the fraction of all known relevant items that a retrieval system returns within its first k results.
- [Reranking](https://blowstack.com/tools/cloud-concepts-hub/reranking): Reranking is a second-stage search process that reorders retrieved candidates using additional relevance signals before results are presented or supplied to a foundation model.
- [Response Streaming](https://blowstack.com/tools/cloud-concepts-hub/response-streaming): Response streaming delivers a response incrementally as output becomes available instead of waiting for the entire response to finish.
- [Selective Disclosure](https://blowstack.com/tools/cloud-concepts-hub/selective-disclosure): Selective disclosure is the deliberate sharing of relevant information with appropriate audiences while withholding details that could compromise privacy, security, or model safety.
- [Semantic Similarity](https://blowstack.com/tools/cloud-concepts-hub/semantic-similarity): Semantic similarity measures how closely two pieces of text match in meaning, even when they use different words.
- [Serialization](https://blowstack.com/tools/cloud-concepts-hub/serialization): Serialization converts in-memory data into a defined text or binary representation for storage, transmission, and later reconstruction.
- [Shadow Testing](https://blowstack.com/tools/cloud-concepts-hub/shadow-testing): Shadow testing evaluates a candidate system using copies of live production requests while only the established production system returns responses to users.
- [Small Language Models](https://blowstack.com/tools/cloud-concepts-hub/small-language-models): Small language models are language models with relatively compact parameter counts and resource requirements, selected to perform suitable tasks efficiently.
- [Source Citation](https://blowstack.com/tools/cloud-concepts-hub/source-citation): Source citation is the practice of identifying and documenting the origins of information so that data, model artifacts, and AI-generated claims can be traced to supporting sources.
- [Stop Sequences](https://blowstack.com/tools/cloud-concepts-hub/stop-sequences): Stop sequences are configured text patterns that cause a language model’s generation process to end when a matching pattern is produced.
- [Temperature Sampling](https://blowstack.com/tools/cloud-concepts-hub/temperature-sampling): Temperature sampling controls the randomness of a generative model’s token selection by reshaping the probabilities used during inference.
- [Tokenization](https://blowstack.com/tools/cloud-concepts-hub/tokenization): Tokenization is the process of converting text into smaller units called tokens and mapping them to numerical identifiers that a language model can process.
- [Top-K Sampling](https://blowstack.com/tools/cloud-concepts-hub/top-k-sampling): Top-K sampling is a text-generation decoding method that selects each next token from only the K highest-probability candidates.
- [Transfer Learning](https://blowstack.com/tools/cloud-concepts-hub/transfer-learning): Transfer learning reuses knowledge learned by a model on one task or dataset to improve learning on a different, related task or domain.