Amazon SageMaker AI
Amazon SageMaker AI is a managed ML platform that lets developers build, train, and host models as SageMaker AI endpoints. It supports Meta’s Muse Spark Open (BF16 and NVFP4 variants) on multi‑GPU instances (ml.p6‑b300.48xlarge or ml.p5e.48xlarge), Hugging Face models, and includes training sessions that support Low‑Rank Adaptation (LoRA). Endpoints accept multi‑modal input, return text, and can be VPC‑attached and KMS‑encrypted. New updates: a reusable CDK construct automatically deregisters and re‑registers Application Auto Scaling during endpoint updates, eliminating CloudFormation failure errors. Amazon SageMaker AI also hosts a Multi‑Turn Reinforcement Learning (MTRL) pipeline that trains agents via Bedrock AgentCore. In a Wordle benchmark, a gpt‑oss‑20b model improved solve rate from 61% to 84 % (and from 15% to 69 % at low effort) for about $339 in training tokens. Amazon SageMaker Edge Manager REPLACED_BY Amazon SageMaker Neo Exowatt used Amazon SageMaker AI Amazon SageMaker AI customized Qwen3-1.7B model Amazon SageMaker AI customized Qwen3-14B model· Entity summary generated by Graphiti
Relation timeline
BY VALID-FROM DATE- ExowattUSEDAmazon SageMaker AI
Exowatt used Amazon SageMaker AI
- Amazon SageMaker AICUSTOMIZEDQwen3-1.7B model
Amazon SageMaker AI customized Qwen3-1.7B model
- Amazon SageMaker AICUSTOMIZEDQwen3-14B model
Amazon SageMaker AI customized Qwen3-14B model
- Amazon SageMaker AIREPLACED BYAmazon SageMaker Neo
Amazon SageMaker Edge Manager REPLACED_BY Amazon SageMaker Neo
- Amazon SageMaker AISENDS PROMPTS TOAmazon Bedrock AgentCore
Amazon SageMaker AI sends prompts to Amazon Bedrock AgentCore
- Multi-Turn Reinforcement LearningRUNS ONAmazon SageMaker AI
Multi-Turn Reinforcement Learning runs on Amazon SageMaker AI
- Amazon SageMaker AIALLOWSLow‑Rank Adaptation
Amazon SageMaker AI allows Low-Rank Adaptation (LoRA)
- Amazon SageMaker AI training session APIsWORKS INAmazon SageMaker AI
Amazon SageMaker AI training sessions works in GitHub Copilot
- Amazon SageMaker AIINCLUDESAmazon SageMaker AI training session APIs
Amazon SageMaker AI includes Amazon SageMaker AI training sessions
- AWS Cloud Development KitDEPLOYED ONAmazon SageMaker AI
AWS Cloud Development Kit stacks handle the deployment of Amazon SageMaker AI real‑time endpoints
- hexgrad/Kokoro-82MRUNS ONAmazon SageMaker AI
Kokoro-82M runs on Amazon SageMaker AI real‑time endpoint as a custom container
- Amazon SageMaker AIPUBLISHES TOAmazon CloudWatch
Amazon SageMaker AI publishes the ConcurrentRequestsPerModel metric to Amazon CloudWatch
- Amazon SageMaker AISIGN UP WITHGoogle
Amazon SageMaker AI customers can now sign up using a Google identity
- Amazon SageMaker AISIGN UP WITHApple
Amazon SageMaker AI customers can now sign up using an Apple identity
- SageMaker StudioPART OFAmazon SageMaker AI
Amazon SageMaker Studio is part of Amazon SageMaker AI
- Amazon SageMaker AIRUN ONAmazon Web Services
Amazon SageMaker AI runs on AWS
- Amazon SageMaker AISIGN UP WITHAmazon
Amazon SageMaker AI customers can now sign up using an Amazon identity
- Amazon SageMaker AIINTEGRATES WITHHugging Face
Amazon SageMaker AI supports deployment of Hugging Face models and integrates with Hugging Face
- Amazon SageMaker AISIGN UP WITHGitHub
Amazon SageMaker AI customers can now sign up using a GitHub identity
- Amazon SageMaker JumpStartPART OFAmazon SageMaker AI
Amazon SageMaker JumpStart is a machine learning hub within Amazon SageMaker AI
- Key Management SystemBUILT FROMAmazon SageMaker AI
Key Management System is built from code in the Confidential Federated Compute Github repository.
Relations are facts Graphiti extracted from reports, not verbatim quotes. An end date means later reporting showed the relation no longer holds.
Events 7
- Exowatt beats frontier models on accuracy with SageMaker serverless model customization1 cited report · 2026-10-28
- Fleet-wide computer vision lifecycle at the edge with SageMaker SDK v31 cited report · 2026-10-27
- Multi-Turn Reinforcement Learning on Amazon SageMaker AI: A hands-on guide1 cited report · 2026-10-24
- Announcing new ways to customize models with SageMaker AI, without managing GPUs1 cited report · 2026-10-23
- Stop guessing auto scaling targets for TTS on Amazon SageMaker AI1 cited report · 2026-10-23
- Developers can now access SageMaker Studio through the new AWS experience1 cited report · 2026-10-20
- Amazon SageMaker JumpStart: 2026 year-to-date in review1 cited report · 2026-10-20
Source reports 8
- Exowatt beats frontier models on accuracy with SageMaker serverless model customizationAWS Machine Learning Blog · first-party · 2026-10-28
- Fleet-wide computer vision lifecycle at the edge with SageMaker SDK v3AWS Machine Learning Blog · first-party · 2026-10-27
- Multi-Turn Reinforcement Learning on Amazon SageMaker AI: A hands-on guideAWS Machine Learning Blog · first-party · 2026-10-24
- Announcing new ways to customize models with SageMaker AI, without managing GPUsAWS Machine Learning Blog · first-party · 2026-10-23
- Stop guessing auto scaling targets for TTS on Amazon SageMaker AIAWS Machine Learning Blog · first-party · 2026-10-23
- Developers can now access SageMaker Studio through the new AWS experienceAWS Machine Learning Blog · first-party · 2026-10-20
- Amazon SageMaker JumpStart: 2026 year-to-date in reviewAWS Machine Learning Blog · first-party · 2026-10-20
- Toward provably private learning from federated dataGoogle Research · first-party · 2026-10-02