SageMaker V3 Learning Kit

Start with the lifecycle. Then map each stage to the SDK package, object and working example. This is a reference system, not an API memorization exercise.

SDK package map

sagemaker-core

building blocks / plumbing

Foundational resources, sessions, shapes/config objects and lower-level resource abstractions.

sagemaker.core.helper.session_helper.Session
sagemaker.core.resources.Endpoint
sagemaker.core.shapes.Channel
sagemaker.core.shapes.DataSource

sagemaker-train

build and train

High-level model training experience centered on ModelTrainer and training configuration.

sagemaker.train.ModelTrainer
sagemaker.train.configs.Compute
sagemaker.train.configs.InputData
sagemaker.train.configs.OutputDataConfig

sagemaker-serve

deploy and serve

Build deployable models and host them for inference.

sagemaker.serve.ModelBuilder
sagemaker.serve.builder.schema_builder.SchemaBuilder

sagemaker-mlops

operationalize and govern

Lifecycle, registry, pipeline, experiment, monitoring and governance-oriented capabilities.

Model Registry
Pipelines
Experiments
Monitoring

Lifecycle

0. Mental model

SDK V3 package map Prevents confusion between lifecycle concepts and Python package boundaries. SageMaker service vs Python SDK The SDK is the programming interface; SageMaker is the managed AWS service executing the jobs and resources.

1. Data & preparation

Data preparation Models need repeatable preparation, quality checks and feature engineering before training. Feature Store Centralizes reusable model features and can support both offline training and low-latency online retrieval.

2. Build & train

ModelTrainer This is the central high-level V3 training abstraction. Channels & InputData Keeps logically different training inputs separate and named. File, Pipe & FastFile Controls how S3 training data reaches the training container. Training container anatomy Explains where your code, input channels and model output physically appear. scikit-learn training pattern Canonical classical-ML example for understanding the entire lifecycle. PyTorch training pattern Shows that the SageMaker lifecycle stays stable while model code changes. Hugging Face training pattern Connects NLP model catalog knowledge to SageMaker-managed training.

3. Evaluate & tune

Evaluation Separates model fitting from the evidence used to accept or reject a model. Tuning & experiments Makes model iteration traceable and comparable.

4. Model artifact

Model artifact Bridges training and inference.

5. Register & govern

Model Registry Adds versioning, approval and promotion controls around model artifacts. Pipelines / MLOps orchestration Turns manually repeated notebook steps into a reproducible workflow.

6. Deploy & inference

ModelBuilder Central V3 abstraction for creating a deployable model configuration. Real-time endpoint For low-latency request/response inference. Serverless, async & batch inference Inference architecture should match latency, payload size and traffic pattern. Feature Store during inference Keeps inference features consistent with those used during training.

7. Monitor & operate

Monitoring Production correctness includes data behavior, model behavior and operational health. Retraining loop Closes the lifecycle rather than treating deployment as the end.

8. Interview reference

Canonical end-to-end project One project gives you a stable story for interviews and learning. V2 -> V3 mental mapping Many examples and interviewers still use V2 terminology. Where does this object belong? Fast lookup when names blur together.