High-performance & cost-effective model deployment with Amazon SageMaker
summary
Overview of concrete SageMaker deployment choices and measured limits: real-time endpoints require always-on instances, serverless (GA today) uses 1–6 GB memory and 1–50 concurrency and charges only during invocation, async supports up to 1 GB payloads and 15 minute runtimes with queueing and can scale to zero (demo shows 0–5 instances), batch handles multi‑GB jobs over days. Inference Recommender runs default jobs in ~1 hour and suggests instance types; SageMaker Savings Plan claims ~64% potential savings for 1–3 year commitments.
AI-written summary · about
chapters (7)
- 0:00Introduction
- 1:05Sagemaker deployment options
- 4:25Deep dive into inference modes
- 13:07Inference selection flowchart
- 15:29Programmatic access to SageMaker
- 18:26Cost optimization techniques
- 32:36Hands-on demo
related talks
48mAmazon SageMaker — Survey of Ways to (Re)deploy Models on AWS
33mBuild, train, and deploy ML models using Amazon SageMaker
27mScaling ML to the next level: Hosting thousands of models on SageMaker
21mAWS On Air ft. Amazon SageMaker Inference Recommender
41mDeploy ML models for inference at high performance & low cost
45mDeploying ML Models to Production with Amazon SageMaker
20mIndustrial-Scale Machine Learning
19mDeveloping large-scale machine learning algorithms on Amazon SageMaker