Back to Archive

AWS News - 2026-09-26

2026-09-26
最終更新: 2026-09-26 12:17:41 JST

AI による概要

12 記事

この日は 12 件で、大規模な強化学習の実行基盤と、生成 AI の本番運用にまつわる解説が中心でした。機械学習ブログでは、Mixture-of-Experts (MoE) の強化学習を Amazon EKS 上でスケールさせる構成が公開されました。Amazon EKS と Elastic Fabric Adapter (EFA)、Amazon S3 を組み合わせたアーキテクチャにより、大規模な RLHF や GRPO のトレーニングにおいて強化学習のロールアウト全体のスループットを 40% 向上させています。あわせて、オープンソースの強化学習フレームワーク SkyRL を Amazon SageMaker HyperPod 上で動かし、Qwen3-VL-8B のビジョン言語モデルを GRPO でポストトレーニングする手順が紹介されました。コンテナイメージの構築、SageMaker Studio からの Ray クラスターの起動、ジョブの投入と監視、学習済み LoRA アダプターの推論用ホスティングまでを扱っています。品質管理では、NarrateAI が Amazon Bedrock 上で構築した本番向けの LLM 品質保証の仕組みが解説されました。適応的なパイプラインのオーケストレーション、クロスアカウントでのマルチモデルフェイルオーバー、リアルタイムのストリーミング評価、複合的な評価、データ正確性の検証という 5 つの手法により、応答をリアルタイムにストリーミングしながら数値の正確性で約 99% に到達しています。音声では、公開されている Qwen3-TTS-12Hz-1.7B-Base を Amazon SageMaker JumpStart からフルマネージドのリアルタイムエンドポイントにデプロイし、短い参照音声から声を複製する方法が紹介されました。言語をまたいでもクローン元の話者の特徴が保たれます。トレーニングの配置では、Amazon SageMaker HyperPod と Cloud Native Qumulo を組み合わせ、学習用のコンピュートをあるリージョンに置きながらデータセットを別のリージョンに置く構成が検証されました。リージョンをまたいだ学習でも、NeuralCache のウォームアップを短時間挟めば同一リージョン構成と同等のスループットが出たと報告されています。事例では、Datacor がガスと溶接の販売業者向けに、Amazon Quick Sight のダッシュボードと自然言語クエリを自社の TrackAbout プラットフォームに組み込み、クラウドをまたぐ自動データパイプラインとマルチテナントの行レベルセキュリティを備えたセルフサービスのレンタル分析を実現した取り組みが紹介されました。サービス更新では、AWS DataSync のコンソールに監視ダッシュボードが追加され、アカウント全体のデータ転送を俯瞰できるようになりました。タスク実行ごとにステータス、データとファイルの転送レート、所要時間、転送量を確認でき、ステータス・タスク・タスクモード・実行 ID・開始時刻で絞り込み、失敗した実行を選んでエラーを確認できます。Amazon Transcribe は、カスタム語彙・カスタム語彙フィルター・カスタム言語モデルを、自分で管理する AWS KMS キーで保存時に暗号化できるようになりました (これまでは常に AWS 所有のキーでの暗号化。キーを指定しなければ従来どおりで対応は不要)。キーの権限を自分で制御でき、使用はすべて AWS CloudTrail に記録されます。また、AWS IAM のアウトバウンド ID フェデレーションが OIDC ディスカバリー API のインターフェイス VPC エンドポイントに対応し、OIDC のディスカバリーメタデータと JWKS の検証キーのエンドポイントに、AWS PrivateLink 経由で VPC 内からアクセスできるようになりました。これまでこれらのエンドポイントはパブリックインターネット経由でしか到達できませんでした。コンピュートでは、Amazon EC2 の M8i / M8i-flex、R8i / R8i-flex、C8i / C8i-flex の各インスタンスが AWS European Sovereign Cloud (ドイツ) リージョンでも利用できるようになりました。いずれも AWS だけで使えるカスタムの Intel Xeon 6 プロセッサを搭載し、同等のクラウド向け Intel プロセッサの中で最高の性能と最速のメモリ帯域を備えます。前世代の Intel ベースのインスタンスと比べて価格性能が最大 15% 向上し、メモリ帯域は 2.5 倍です。汎用の M8i / M8i-flex は M7i / M7i-flex 比で最大 20% 高性能で、PostgreSQL で最大 30%、NGINX の Web アプリケーションで最大 60%、AI のディープラーニング推薦モデルで最大 40% 高速です。メモリ最適化の R8i / R8i-flex も R7i 比で 20% 高性能で、同様に PostgreSQL 最大 30%、NGINX 最大 60%、推薦モデル最大 40% の改善が示されています (R8i-flex は初のメモリ最適化 Flex インスタンスです)。コンピュート最適化の C8i / C8i-flex は C7i / C7i-flex 比で最大 20% 高性能で、NGINX で最大 60%、推薦モデルで最大 40%、Memcached で 35% 高速です。

主要トピック

  • 強化学習: MoE の強化学習を EKS + EFA + DeepEP でスケールし、ロールアウトのスループットを 40% 向上

  • 強化学習: SkyRL を SageMaker HyperPod で動かし Qwen3-VL-8B を GRPO でポストトレーニング

  • 品質保証: NarrateAI が Bedrock 上の 5 つの手法で、ストリーミングしながら数値の正確性 約 99% を達成

  • 音声合成: Qwen3-TTS を SageMaker JumpStart からデプロイ、短い参照音声で言語をまたぐ声のクローン

  • トレーニング: SageMaker HyperPod と Qumulo でリージョンをまたぐ学習、同一リージョン構成と同等のスループット

  • 分析事例: Datacor が Quick Sight の埋め込みと自然言語クエリでセルフサービスのレンタル分析を実現

  • 転送: AWS DataSync にアカウント横断の監視ダッシュボードを追加、失敗した実行のエラーも確認可能

  • セキュリティ: Amazon Transcribe のカスタムリソースを顧客管理の KMS キーで暗号化可能に

  • セキュリティ: IAM アウトバウンド ID フェデレーションの OIDC ディスカバリーが PrivateLink 経由で VPC 内から到達可能に

  • コンピュート: EC2 の M8i / R8i / C8i と各 flex が European Sovereign Cloud (ドイツ) でも利用可能に、前世代比で価格性能 最大 15% 向上

AI (Claude Opus 5) が生成 · 2026-09-27 08:18:43 JST

AWS What's New

AWS DataSync launches a monitoring dashboard for tracking task executions across your account

AWS DataSync now provides a monitoring dashboard in the DataSync console that gives you visibility into your data transfers across your account. For each task execution, you can see its status, data and file transfer rates, duration, and the total data and files transferred. You can filter executions by status, task, task mode, execution ID, or start time to focus on a specific set of runs. The dashboard summarizes the filtered results, including the number of successful and failed executions, and the cumulative data and files transferred, grouped by task. To troubleshoot issues, you can select a failed execution to view its errors. The dashboard also displays counts of your configured tasks, locations, and agents, along with real-time total transfer rates.

Previously, monitoring multiple transfers required reviewing each execution individually or building custom Amazon CloudWatch dashboards. Now you can monitor the health, performance, and progress of concurrent transfers from a single view. This is useful for large migrations and recurring transfers that run many tasks at once.

The AWS DataSync dashboard is available at no additional cost in all commercial AWS Regions and the AWS GovCloud (US) Regions where AWS DataSync is available. To view your dashboard, go to the AWS DataSync console. To learn more about DataSync, visit the AWS DataSync User Guide.

AWS IAM outbound identity federation now supports interface VPC endpoints for OIDC discovery

AWS Identity and Access Management (IAM) outbound identity federation now supports Amazon Virtual Private Cloud (VPC) endpoints for the OpenID Connect (OIDC) discovery APIs. You can now access the OIDC discovery metadata and JSON Web Key Set (JWKS) verification key endpoints from within your VPC using AWS PrivateLink, without requiring traffic to traverse the public internet.

IAM outbound identity federation eliminates the need to use long-lived credentials when your AWS workloads access external services. Instead, your workloads request short-lived JSON Web Tokens (JWTs) from AWS Security Token Service (AWS STS). External services verify these tokens using public verification keys and metadata available at OIDC discovery endpoints. Previously, the OIDC discovery endpoints were only reachable over the public internet, so a verifying workload running in a VPC without internet access could not retrieve them. With this launch, you can create an interface VPC endpoint to reach these endpoints privately, keeping the verification key retrieval traffic within the AWS network. This capability helps you meet network security requirements for workloads that operate in VPCs with restricted internet access, while still enabling external services to verify JWTs.

This feature is available in all commercial AWS Regions, the AWS GovCloud (US) Regions, and China Regions. There is no additional charge for this feature beyond standard AWS PrivateLink pricing. To learn more, see the IAM User Guide.

Amazon EC2 C8i and C8i-flex instances are now available in additional regions

詳細を表示

Starting today, Amazon Elastic Compute Cloud (Amazon EC2) C8i and C8i-flex instances are available in the AWS European Sovereign Cloud (Germany) region. These instances are powered by custom Intel Xeon 6 processors, available only on AWS, delivering the highest performance and fastest memory bandwidth among comparable Intel processors in the cloud. These C8i and C8i-flex instances offer up to 15% better price-performance, and 2.5x more memory bandwidth compared to previous generation Intel-based instances. They deliver up to 20% higher performance than C7i and C7i-flex instances, with even higher gains for specific workloads. The C8i and C8i-flex are up to 60% faster for NGINX web applications, up to 40% faster for AI deep learning recommendation models, and 35% faster for Memcached stores compared to C7i and C7i-flex.

C8i-flex are the easiest way to get price performance benefits for a majority of compute intensive workloads like web and application servers, databases, caches, Apache Kafka, Elasticsearch, and enterprise applications. They offer the most common sizes, from large to 16xlarge, and are a great first choice for applications that don't fully utilize all compute resources.

C8i instances are a great choice for all memory-intensive workloads, especially for workloads that need the largest instance sizes or continuous high CPU usage. C8i instances offer 13 sizes including 2 bare metal sizes and the new 96xlarge size for the largest applications.

To get started, sign in to the AWS Management Console. Customers can purchase these instances via Savings Plans, On-Demand instances, and Spot instances. For more information about the new C8i and C8i-flex instances visit the AWS News blog.

Amazon EC2 R8i and R8i-flex instances are now available in additional regions

詳細を表示

Starting today, Amazon Elastic Compute Cloud (Amazon EC2) R8i and R8i-flex instances are available in the AWS European Sovereign Cloud (Germany) region. These instances are powered by custom Intel Xeon 6 processors, available only on AWS, delivering the highest performance and fastest memory bandwidth among comparable Intel processors in the cloud. The R8i and R8i-flex instances offer up to 15% better price-performance, and 2.5x more memory bandwidth compared to previous generation Intel-based instances. They deliver 20% higher performance than R7i instances, with even higher gains for specific workloads. They are up to 30% faster for PostgreSQL databases, up to 60% faster for NGINX web applications, and up to 40% faster for AI deep learning recommendation models compared to R7i.

R8i-flex, our first memory-optimized Flex instances, are the easiest way to get price performance benefits for a majority of memory-intensive workloads. They offer the most common sizes, from large to 16xlarge, and are a great first choice for applications that don't fully utilize all compute resources.

R8i instances are a great choice for all memory-intensive workloads, especially for workloads that need the largest instance sizes or continuous high CPU usage. R8i instances offer 13 sizes including 2 bare metal sizes and the new 96xlarge size for the largest applications. R8i instances are SAP-certified and deliver 142,100 aSAPS, the highest among all comparable machines in on-premises and cloud environments, delivering exceptional performance for mission-critical SAP workloads.

To get started, sign in to the AWS Management Console. Customers can purchase these instances via Savings Plans, On-Demand instances, and Spot instances. For more information about the new R8i and R8i-flex instances visit the AWS News blog.

Amazon EC2 M8i and M8i-flex instances are now available in additional regions

詳細を表示

Starting today, Amazon EC2 M8i and M8i-flex instances are now available in the AWS European Sovereign Cloud (Germany) region. These instances are powered by custom Intel Xeon 6 processors, available only on AWS, delivering the highest performance and fastest memory bandwidth among comparable Intel processors in the cloud. The M8i and M8i-flex instances offer up to 15% better price-performance, and 2.5x more memory bandwidth compared to previous generation Intel-based instances. They deliver up to 20% better performance than M7i and M7i-flex instances, with even higher gains for specific workloads. The M8i and M8i-flex instances are up to 30% faster for PostgreSQL databases, up to 60% faster for NGINX web applications, and up to 40% faster for AI deep learning recommendation models compared to M7i and M7i-flex instances.

M8i-flex are the easiest way to get price performance benefits for a majority of general-purpose workloads like web and application servers, microservices, small and medium data stores, virtual desktops, and enterprise applications. They offer the most common sizes, from large to 16xlarge, and are a great first choice for applications that don't fully utilize all compute resources.

M8i instances are a great choice for all general purpose workloads, especially for workloads that need the largest instance sizes or continuous high CPU usage. The SAP-certified M8i instances offer 13 sizes including 2 bare metal sizes and the new 96xlarge size for the largest applications.

To get started, sign in to the AWS Management Console. For more information about the new instances, visit the M8i and M8i-flex instance page or visit the AWS News blog.

Amazon Transcribe adds customer-managed KMS keys for custom resources

Amazon Transcribe now lets you encrypt your custom vocabularies, custom vocabulary filters, and custom language models at rest with a customer-managed AWS KMS key that you own and control.

Previously, these custom resources were always encrypted with an AWS owned key. Now you can supply your own symmetric AWS KMS key when you create or update these resources, so the artifacts Transcribe stores on your behalf are encrypted under a key in your account. If you do not provide a key, your resources continue to be encrypted with an AWS owned key—no action is required unless you opt in.

With a customer-managed key, you control key permissions and decide exactly which principals and services can encrypt and decrypt your custom resources. Every use of your key is logged in AWS CloudTrail, giving you a full audit trail for compliance and monitoring. You can also disable the key or transition resources to a different key to revoke access on your own schedule.

This feature is available in all AWS Regions where Amazon Transcribe is offered. To learn more, see Amazon Transcribe Developer Guide.

AWS Machine Learning Blog

Multi-Region training with Amazon SageMaker HyperPod and Qumulo

Amazon SageMaker HyperPod and Cloud Native Qumulo let you place training compute in one AWS Region while keeping your dataset in another. This post shares the architecture and validation results from a cross-Region training run, where a remote cluster matched a co-located cluster's throughput after a brief NeuralCache warmup.

How Datacor built self-service rental analytics with Amazon Quick Sight

Learn how Datacor built a self-service rental analytics experience for gas and welding distributors by embedding Amazon Quick Sight dashboards and natural language querying into its TrackAbout platform, powered by an automated cross-cloud data pipeline and multi-tenant row-level security.

Deploying real-time personalized speech with Qwen3-TTS on Amazon SageMaker AI

Deploy the publicly available Qwen3-TTS-12Hz-1.7B-Base text-to-speech model from Amazon SageMaker JumpStart to a fully managed, real-time endpoint, and clone a voice from a short reference clip. Cross-lingual cloning preserves the speaker's identity across languages.

NarrateAI: production-ready LLM quality assurance on Amazon Bedrock

NarrateAI delivers production-ready LLM quality assurance on Amazon Bedrock. This post details five techniques—adaptive pipeline orchestration, cross-account multi-model failover, real-time streaming evaluation, composite evaluation, and data accuracy verification—that reach about 99% numerical accuracy while streaming responses in real time.

Accelerate multimodal RL training with SkyRL on Amazon SageMaker HyperPod

Learn how to run SkyRL, an open-source reinforcement learning framework, on Amazon SageMaker HyperPod to post-train a Qwen3-VL-8B vision-language model with GRPO. This walkthrough covers building the container image, launching a Ray cluster from SageMaker Studio, submitting and monitoring the job, and hosting the trained LoRA adapter for inference.

Scaling MoE reinforcement learning on Amazon EKS with EFA and DeepEP with 40% more throughput

Learn how to scale Mixture-of-Experts (MoE) reinforcement learning on Amazon EKS using Elastic Fabric Adapter (EFA) and DeepEP. This post presents an architecture that combines Amazon EKS, EFA, and Amazon S3 and increased aggregate reinforcement learning rollout throughput by 40% for large-scale RLHF and GRPO training.
前日
Archive
翌日