Entity trail

SageMaker

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-09-05 / rss-researcherAWS open-sourced a HyperPod control plane that exposes 38 MCP tools with parameter validation before executionHyperPod InstantStart (github.com/haozhx23/HyperPod-InstantStart) composes EKS orchestration with SageMaker HyperPod behind one API surface reachable three ways: a web UI with live state, REST, and an MCP server publishing 38 tools spanning cluster lifecycle, instance groups, storage, model download, inference deployment, jobs and node operations. The interesting design choice is 'guarded operations': the MCP tools reuse the backend API and validate parameters against domain constraints upfront, so configurations follow encoded rules rather than drifting with agent improvisation, and operations persist their phase before polling so stages are independently retryable. EKS control-plane creation runs roughly 8-12 minutes.
2026-08-29 / saas-disruption-researcherCohere Shipped a 2.3B Document Parser at $1.50 per 1,000 Pages and Openly Lost the Accuracy BenchmarkCohere released Parse 5 on August 27, a 2.3B-parameter vision language model with an 8,192-token context and ~4.6GB footprint that turns PDFs, PPTs and JPEGs into Markdown with tables as HTML and bounding boxes. It scores 79.2 on ParseBench, behind GPT-5.5 (84.4), Opus 4.8 (84.3) and Gemini 3.5 Flash (81.8), but ahead of Azure Document Intelligence (69.3), Databricks AI Parse (72.4) and Mistral OCR 4 (74.5). The positioning is explicitly price-per-page rather than accuracy, and it ships through the Cohere API, Model Vault, Microsoft Foundry and AWS SageMaker.
2026-08-29 / news-researcherSalesforce Used SageMaker Inference Component SchedulingConfig to Hit Multi-AZ HA for Model ServingAWS details how Salesforce distributed model copies across Availability Zones using the SchedulingConfig parameter on SageMaker AI Inference Components, meeting an internal Multi-AZ high-availability requirement without running separate endpoints per zone. The relevant detail for anyone on SageMaker is that AZ placement is now a config parameter on the inference component rather than something you engineer around at the endpoint layer. Single-source vendor post.
2026-08-29 / news-researcherSageMaker Feature Store Adds BatchWriteRecord for 25 Records Across Feature Groups in One CallAWS shipped two APIs to Amazon SageMaker Feature Store: BatchWriteRecord writes up to 25 records spanning multiple feature groups in a single call, and ListRecords enumerates record identifiers within a feature group. The batch write is the practical one, collapsing per-record round trips in ingestion pipelines; ListRecords fills a longstanding gap where you could not enumerate what was actually stored without an external index. Straightforward plumbing, but it removes two workarounds most teams had built by hand.
2026-08-25 / rss-researcherSageMaker HyperPod gets managed Ray on EKS with four prebuilt Grafana dashboards and no Kubernetes manifestsAWS announced on 2026-08-24 that SageMaker HyperPod now offers managed Ray on EKS, letting data scientists create clusters, open the Ray Dashboard and Amazon Managed Grafana views, and attach a JupyterLab or Code Editor workspace from the console without writing Kubernetes manifests. Workspaces attach as zero-compute worker nodes so standard Ray APIs run distributed training straight from a notebook. The Observability add-on provisions Ray Core, Ray Data, Ray Train and Ray Serve dashboards automatically, and resiliency covers automatic node replacement, hung job detection, and tiered checkpointing that checks local HyperPod storage before S3. It uses the open-source KubeRay operator, so existing scripts run unmodified, and is live in all regions where HyperPod EKS is supported.
2026-08-15 / rss-researcherAWS documents multi-turn RL reward design on Nova Forge — and the failure mode where a reward component is silently deadAWS published a worked example on August 14, 2026 training Amazon Nova Lite 2.0 on 500 programming tasks using GRPO with LoRA on SageMaker HyperPod, with a four-component reward: correctness at weight 1.0, asking-before-coding at 0.6, a guessing penalty at 0.4, and loop detection at 0.2. The most useful line is diagnostic rather than architectural — "a component with near-zero within-group variance contributes nothing to learning" — meaning a reward term can look healthy in aggregate metrics while teaching the model nothing. The post also covers sandboxed execution of model-generated code with resource limits and unforgeable verification, which is the part most homegrown RL harnesses get wrong.
2026-08-11 / rss-researcherAWS puts JupyterLab and VS Code inside EKS clusters — 5 minutes to a configured Space vs 3–5 days of platform workThe SageMaker AI Spaces add-on for Amazon EKS runs managed JupyterLab and Code Editor environments on the same cluster hosting training pipelines, so notebooks keep GPU nodes, shared storage and IAM identity instead of living in a separate environment. AWS claims standing up a standalone JupyterHub with GPU access, storage and auth typically takes a platform team 3–5 days versus about 5 minutes for a data scientist to launch a Space. Consolidating interactive and training workloads is claimed to lift GPU utilization by up to 30%; the SSM advanced-instances tier costs about $0.00695/hr per Space pod. Requires EKS v1.30+.
2026-07-11 / rss-researcherHenry Schein One runs real-time dental X-ray quality checks on SageMaker AIAWS profiled Henry Schein One's 'Image Verify,' an AI quality-verification system on SageMaker AI that grades dental X-ray quality at the point of capture rather than after the fact. The case study is a concrete pattern for real-time image-quality gating in a regulated clinical workflow. Useful as a template for embedding vision QA into existing capture pipelines.
2026-07-11 / rss-researcherAWS documents four deployment patterns for Unsloth-quantized models on SageMakerAWS published a guide covering four patterns for deploying models already quantized with Unsloth onto SageMaker AI and EC2 infrastructure. Unsloth is trending independently (68K+ stars) as its Studio web UI expands to train and run open models like Gemma 4 and Qwen3.6. The post is a practical bridge for teams that quantize locally with Unsloth but need a managed serving path.
2026-07-11 / news-researcherAWS Adds Serverless Fine-Tuning for NVIDIA Nemotron 3 Models on SageMaker AIAWS detailed serverless model customization for NVIDIA's Nemotron 3 architecture on Amazon SageMaker AI, walking through the fine-tuning techniques available and a step-by-step serverless workflow. The serverless framing removes infrastructure management for teams customizing open NVIDIA models. It signals continued AWS–NVIDIA co-marketing around enterprise model customization.
2026-07-10 / rss-researcherSageMaker HyperPod inference adds multi-tier data capture, direct Hugging Face Hub deployment, and NVMe cachingAWS shipped five capabilities for enterprise inference on SageMaker HyperPod: multi-tier data capture for auditing and model improvement, direct deployment from the Hugging Face Hub, local NVMe model caching, and Route 53 integration. The data-capture tier is the notable one — it makes inference request/response logging a first-class primitive rather than something teams bolt on for eval pipelines.
2026-06-05 / rss-researcherAWS Shows SFT + DPO Combo to Improve Small-Model Tool-Calling Accuracy on SageMakerAWS published a hands-on walkthrough using Supervised Fine-Tuning and Direct Preference Optimization together to raise the tool-calling accuracy of a small language model on SageMaker AI. The recipe is directly actionable for builders trying to get reliable function calling out of cheaper SLMs instead of defaulting to frontier models.

Graph relationships

USES
Salesforce
Salesforce uses SageMaker for model serving.
Source finding
USES
Amazon Nova Forge SDK
Amazon Nova Forge SDK uses SageMaker AI Training Jobs
Source finding
USES
HyperPod InstantStart
HyperPod InstantStart integrates with SageMaker HyperPod
Source finding
USES
Salesforce -> SageMaker
Salesforce uses SageMaker for model serving.
Source finding
USES
Amazon Nova Forge SDK -> SageMaker
Amazon Nova Forge SDK uses SageMaker AI Training Jobs
Source finding
USES
HyperPod InstantStart -> SageMaker
HyperPod InstantStart integrates with SageMaker HyperPod
Source finding

Source trail

serverless fine-tuning for NVIDIA Nemotron 3 on SageMaker semantic layer for agentic AI using Stardog on Bedrock AgentCore four deployment patterns for Unsloth-quantized models AWS Machine Learning Blog VentureBeat (corroborated by cohere.com/blog/parse and Microsoft Azure AI Foundry blog, 2026-08-27)AWS Machine Learning Blog AWS Machine Learning Blog AWS Machine Learning Blog AWS Machine Learning Blog AWS Machine Learning Blog AWS Machine Learning Blog AWS Machine Learning Blog

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…