Note

This site contains only the example notebooks built on SageMaker Python SDK v3, which is what pip install sagemaker installs. Notebooks built on the deprecated SDK 2.x are archived in the v2 example notebooks documentation.

Amazon SageMaker Example Notebooks

Welcome to Amazon SageMaker. This site highlights example Jupyter notebooks for a variety of machine learning use cases that you can run in SageMaker.

This site is based on the SageMaker Examples repository on GitHub. To run these notebooks, you will need a SageMaker Notebook Instance or SageMaker Studio. Refer to the SageMaker developer guide’s Get Started page to get one of these set up.

On a Notebook Instance, the examples are pre-installed and available from the examples menu item in JupyterLab. On SageMaker Studio, you will need to open a terminal, go to your home folder, then clone the repo with the following:

git clone https://github.com/aws/amazon-sagemaker-examples.git

---

Introduction

• Introduction to Amazon SageMaker

We recommend the following notebooks as a broad introduction to the capabilities that SageMaker offers. To explore in even more depth, we provide additional notebooks covering even more use cases and frameworks.

Training

• SageMaker V3 Distributed Local Training Example
• SageMaker V3 JumpStart E2E Training Example
• PyTorch’s example to demonstrate Amazon SageMaker Heterogeneous Cluster for model training
• Managed Spot Training for XGBoost (V3)
• Regression with Amazon SageMaker XGBoost algorithm (V3 - sagemaker-core)
• Train a PyTorch model with MNIST dataset
• Building your own algorithm container
• Horovod Distributed Training with SageMaker TensorFlow script mode (SageMaker Python SDK V3)
• Getting Started with AWS Batch for SageMaker Training jobs
• Optional: Delete AWS Batch Resources

Model Customization

• CPT Training on HyperPod
• SageMaker V3 Custom Distributed Training Example
• Direct Preference Optimization (DPO) Training with SageMaker
• Fine-tune LLM with PyTorch FSDP and QLora on Amazon SageMaker AI using ModelTrainer
• Multi-Turn RL (Agentic RFT) Example - Fine-tuning with SageMaker
• Nova Data Mixing
• Recipe Override with ModelTrainer.from_recipe
• Recipe Override with SFTTrainer
• RLAIF Example - Finetuning with Sagemaker
• RLVR Example - Finetuning with Sagemaker
• Continued Finetuning (or) Finetuning on Model Artifacts
• Iterative Training (Resume from Checkpoint)
• Serverless Fine-Tuning and Deployment (End-to-End)
• Part 2: Model Evaluation
• SFTTrainer Example - Finetuning with Sagemaker
• Finetuning with Jumpstart base model
• Continued Finetuning (or) Finetuning on Model Artifacts
• Finetuning with Serverful Compute (TrainingJobCompute)
• Training Monitoring
• Fine-Tuning with HyperPod
• Fine-Tuning with Serverful Training Jobs (SMTJ)
• Import Fine-tuned LLaMA 3 models on SageMaker JumpStart to private model hub (SageMaker Python SDK v3)
• Appendix

Evaluation

• SageMaker Benchmark Evaluation - Basic Usage
• SageMaker Custom Scorer Evaluation - Demo
• SageMaker InspectAI Evaluation
• SageMaker LLM-as-Judge Evaluation
• Part 2: LLM-as-Judge with Custom Models

Inference

• Deploy Models to Bedrock with BedrockModelBuilder
• SageMaker V3 In-Process Mode Example
• SageMaker V3 Inference Pipeline - ModelBuilder vs Core
• SageMaker V3 Local Container Mode Example
• SageMaker V3 Model Optimization Example
• A/B Testing with Amazon SageMaker
• Faster autoscaling on Amazon SageMaker realtime endpoints with inference components (Application Autoscaling)
• Amazon SageMaker Asynchronous Inference using the SageMaker Python SDK
• Leverage deployment guardrails to update a SageMaker Inference endpoint using linear traffic shifting
• Contents
• Introduction
• Setup
• Step 1: Create and deploy the models
• Step 2: Invoke Endpoint
• Step 3: Create CloudWatch alarms to monitor Endpoint performance
• Step 4: Update Endpoint with deployment configurations
• Cleanup
• Amazon Augmented AI(A2I) Integrated with AWS Marketplace ML Models
• Package a machine learning model for listing on the AWS Marketplace
• Notebook CI Test Results
• Amazon SageMaker Multi-Model Endpoints using TorchServe
• Running multi-container endpoints on Amazon SageMaker
• SageMaker Serverless Inference
• Development Environment and Permissions
• Shadow Variant Experiments via API
• References

MLOps

• Amazon SageMaker Feature Store: Introduction to Feature Store
• SageMaker Pipelines with MLflow
• Track an experiment while training a Pytorch model locally or in your notebook
• EMR Serverless Step in SageMaker Pipelines
• SageMaker Lineage Tracking - V3 SDK Example
• E2E ML Pipeline with Model Registry
• SageMaker V3 PyTorch Processing
• SageMaker Clarify E2E Test