Amazon SageMaker
A fully managed machine learning platform from Amazon AWS that supports the end-to-end model lifecycle, including training, hyperparameter tuning, and deployment. The platform provides reinforcement learning training pipelines, compute scheduling, and environment orchestration capabilities for building and operating machine learning workloads at scale.
Timeline (last 90 days)
SageMaker separates training jobs from the Notebook, executing them independently using managed compute resources, returning the Notebook to its role as an exploration tool.
SageMaker provides the Data Wrangler tool to assist with data cleaning and feature engineering.
Batch Transform does not run as a persistent service; it completes inference on a batch of data at once and then releases resources, suitable for scenarios such as offline scoring and periodic reporting, and is usually lower in cost.
Real-time Inference provides a persistent HTTPS endpoint, suitable for low-latency, continuously online online-prediction scenarios.
SageMaker enables the interaction in which an application sends a request, the endpoint calls the model and returns the prediction result, through inference endpoints (Endpoint).
Model lineage information in SageMaker can be associated with training jobs and model artifacts through capabilities such as SageMaker Lineage and Experiments.
The SageMaker Model Registry allows teams to annotate, approve and manage the status of model versions.
The model artifacts produced by SageMaker training jobs are stored in S3.
ChurnWatch uses Amazon S3 to store training datasets, Amazon SageMaker to train the churn prediction model, Python scripts to perform drift detection with PSI, and Amazon CloudWatch for metric monitoring.
The SageMaker Model Monitor service is used to continuously detect data drift and model quality drift.
8 more timeline events
All Facts (18)
SageMaker separates training jobs from the Notebook, executing them independently using managed compute resources, returning the Notebook to its role as an exploration tool.
50%UnverifiedSageMaker provides the Data Wrangler tool to assist with data cleaning and feature engineering.
50%UnverifiedBatch Transform does not run as a persistent service; it completes inference on a batch of data at once and then releases resources, suitable for scenarios such as offline scoring and periodic reporting, and is usually lower in cost.
50%UnverifiedReal-time Inference provides a persistent HTTPS endpoint, suitable for low-latency, continuously online online-prediction scenarios.
50%UnverifiedSageMaker enables the interaction in which an application sends a request, the endpoint calls the model and returns the prediction result, through inference endpoints (Endpoint).
50%UnverifiedModel lineage information in SageMaker can be associated with training jobs and model artifacts through capabilities such as SageMaker Lineage and Experiments.
50%UnverifiedThe SageMaker Model Registry allows teams to annotate, approve and manage the status of model versions.
50%UnverifiedThe model artifacts produced by SageMaker training jobs are stored in S3.
50%UnverifiedChurnWatch uses Amazon S3 to store training datasets, Amazon SageMaker to train the churn prediction model, Python scripts to perform drift detection with PSI, and Amazon CloudWatch for metric monitoring.
50%UnverifiedThe SageMaker Model Monitor service is used to continuously detect data drift and model quality drift.
50%UnverifiedAs a service within the AWS ecosystem, SageMaker's advantage lies in its relatively natural integration with storage, compute, and permission systems; if a team is already using AWS, the migration cost is relatively manageable; if the tech stack is not on AWS, the integration cost needs to be evaluated.
50%UnverifiedServerless Inference is suitable for scenarios with sparse or bursty traffic and is billed by number of invocations.
50%UnverifiedSageMaker provides experiment tracking capabilities that associate the hyperparameters, metrics, and artifacts of different training runs.
50%UnverifiedSageMaker provides a managed training environment that supports bringing your own container or built-in algorithms; training jobs can start compute resources on demand and automatically release them after the job ends.
50%UnverifiedAmazon SageMaker covers the complete workflow from data labeling, distributed training, and hyperparameter tuning to model hosting.
50%UnverifiedA typical cloud-based GenAI workflow is: store data in S3 for cleaning and formatting, evaluate ready-made models in Bedrock, submit distributed training in SageMaker, package as an inference endpoint with auto-scaling, and continuously monitor and iterate.
50%UnverifiedFor distributed training tasks that support checkpointing, using Spot Instances in conjunction with SageMaker's checkpoint-saving feature is an effective means of reducing training costs.
50%UnverifiedAmazon SageMaker has introduced a new capability called aws-ai-ml, made available to coding agents through the Agent Toolkit for AWS.
50%