Amazon S3

Headlines · 6

1. 2d
   AWS details a nine-stage AI agent evaluation framework on Bedrock AgentCore

   An AWS Machine Learning Blog post describes how AWS's Sales, Marketing, and Global Services (SMGS) Operations team built a continuous quality-evaluation framework for a production multi-agent system on Amazon Bedrock AgentCore Evaluations. The framework spans nine lifecycle stages—defining quality, experimentation, CI merge checks, CD production gates, canary scoring, adversarial testing, live drift detection, a real-time pulse, and an automated feedback loop—and runs at three levels at once: tools, individual sub-agents, and multi-agent orchestration, each with its own golden set and promotion criteria. AWS reports that its internal sales agents produced fluent but wrong answers such as labeling $410.08 of three-month revenue as "$410.08 Million" and presenting estimates as sourced facts while conventional monitoring stayed green; in one pre-production test, mismatched data, tool, and prompt changes dropped the pass rate from 78 percent to 38 percent.

   AWS Machine Learning Blog · 🔥 0
2. 6d
   LiveKit publishes Egress troubleshooting guide

   LiveKit published a "Troubleshooting Egress" guide on its blog that catalogs the common error codes and statuses its Egress recording component can return across startup, running and upload stages. It explains the causes and fixes for errors such as 412 start signal not received, 400 page load or S3/GCP/Azure upload failures, 404 participant or track not found, 413 duration limits, 429 concurrency and usage limits, and 403 AssumeRole not enabled. The guide also covers debugging custom templates locally, keeping storage credentials valid for the length of a recording, differences between the Egress recording and the audio an agent hears, and missing audio on SIP calls that use silence suppression.

   LiveKit Blog · 🔥 0
3. 7d
   AWS details llama.cpp decision-model serving on SageMaker AI with SemIf

   AWS published a walkthrough for serving decision models with llama.cpp on Amazon SageMaker AI using specialized Deep Learning Containers. The standard path deploys a GGUF model from Amazon S3 and maps InvokeEndpoint to llama-server's OpenAI-compatible chat completions; an optional SemIf child image (SemIf was formerly OpenJev) returns probabilities over caller-declared options by reading final-position logits instead of decoding an answer token. On 231 public JevBench tasks with Qwen3.5-4B Q4_K_M, a 32-thread CPU endpoint (ml.c6i.8xlarge) was correct on 179/231 (77.5%, p50 1.715s, p95 19.291s), while full GPU offload on ml.g6e.xlarge scored 180/231 (77.9%, p50 0.352s, p95 0.599s) — 4.9x faster at p50 and 32.2x at p95. AWS says the two images used different llama.cpp builds and disagreed on six hard tasks, so it calls the accuracy gap runtime variation rather than proof that GPU offload improves quality, and cites Jev, Kev and Strands Decider 2B as related decision-model work.

   AWS Machine Learning Blog · 🔥 0
4. Oct 12
   Apple's Minions turns idle Macs into an internal virtual GPU cloud

   Apple researchers published a paper describing Minions, an internal "virtual GPU cloud" that aggregates idle Apple Silicon Macs across the network to run machine learning inference. A Minion Server splits submitted jobs into subtasks stored in PostgreSQL, a Minion Executor on each Mac pulls subtasks over REST APIs and runs them on the local GPU with inputs and outputs in Amazon S3, and a Minion Watchdog requeues unfinished work when a machine sleeps, goes offline, powers off or its user stops the executor, using five-minute heartbeats. The paper says Minions only handles inference — embeddings with Hugging Face transformer models and text generation through Ollama installed on each Mac — and Apple reports that some jobs that once faced 10-plus-hour centralized GPU queues could start within minutes, figures that come from Apple's own testing and leave cross-team data confidentiality as an open concern.

   Apple Machine Learning Research · 🔥 0
5. Oct 8
   AWS details multi-tenant GPU cluster sharing on SageMaker HyperPod

   An AWS Machine Learning Blog post lays out a reference architecture for letting multiple teams share a single Amazon SageMaker HyperPod (EKS) GPU cluster while preserving isolation and fairness. It uses AWS IAM Identity Center for centralized authentication (federating with Microsoft Entra ID via SCIM provisioning and SAML 2.0), per-team SageMaker AI domains and execution roles, and Kubernetes namespaces for workload isolation. HyperPod Task Governance handles fair resource allocation, namespace-level cost allocation provides per-team chargeback visibility, and storage is split across FSx for Lustre or OpenZFS directories and S3 buckets.

   AWS Machine Learning Blog · 🔥 0
6. Oct 5
   AWS ships MCP server to promote Amazon Quick resources across accounts

   An AWS blog post details the Amazon Quick Resource Migrator, a sample Model Context Protocol server hosted on Amazon Bedrock AgentCore runtime that promotes Amazon Quick agents, action connectors, knowledge bases, flows and spaces from a development AWS account to production. It reads and recreates resources through the Amazon Quick (Quick Sight) API, replays source permissions in the target account, offers a read-only preview, is idempotent across re-runs, and writes versioned S3 backups with a restore tool. S3 documents behind knowledge bases are not copied, and connector credentials are created as placeholders and must be re-authenticated in the target account; full source is in an aws-samples repository.

   AWS Machine Learning Blog · 🔥 0

Experience and discussion from the community

Share my Amazon S3 experienceAsk about Amazon S3