Automation & Ops amazon-sagemakerecrvllmtext-embeddings-inferencedjl-inferencecuda-compatibility

SageMaker Serving Image Selector

Choose a compatible SageMaker serving container and retrieve the current regional image URI for a Hugging Face model.

FollowSkills review · FSRS-2.0
Not recommended
47/ 100 5-point scale 2.4 / 5
Trust11 / 25 · 2.2/5

The skill directs URI selection to AWS's official DLC catalog and discusses tokens, VPCs, regions, and image provenance. However, the mirroring script can create ECR repositories, authenticate to public/private registries, pull images, and push them without explicit user confirmation, rollback, or least-privilege guidance; sensitive-token data flow is also incomplete, so points are deducted.

Reliability8 / 20 · 2.0/5

The documentation provides detailed decision tables, environment variables, AMI rules, regional handling, and troubleshooting, while mirror_image.py includes argument checks and failure messages. However, key claims, versions, and defect reports are not covered by local tests, and operation depends on external catalogs, AWS CLI, Docker, and other skills. Static review cannot establish execution, so the score is capped and reduced.

Adaptability10 / 15 · 3.3/5

Trigger conditions, model categories, fallback boundaries, and coverage of LLMs, embeddings, rerankers, diffusion, and custom code are relatively clear. Non-fit inputs are not fully defined, Chinese-language guidance is absent, and core operation depends on AWS/Hugging Face services whose mainland-China reachability is not addressed, so points are deducted.

Convention8 / 15 · 2.7/5

The skill has clear structure with decision tables, workflow, configuration, troubleshooting, references, a helper script, and Apache-2.0 licensing. It lacks an independent version, changelog, explicit maintenance owner, and update path; several referenced skills and external materials are outside the supplied target evidence, so points are deducted.

Effectiveness6 / 15 · 2.0/5

The skill offers directly actionable container-selection rules, URI lookup steps, AMI mapping, and an image-mirroring script, which can support the core decision. It provides no complete deployment example, representative verified output, or static proof of current catalog compatibility; users must still validate external data, regions, and model support, so only a moderate static score is warranted.

Verifiability4 / 10 · 2.0/5

The material cites the AWS catalog, regional availability page, GitHub source data, an SDK PR, and a security policy, and the repository includes CI for generated artifacts. It lacks dedicated tests covering this skill's key paths, third-party execution evidence, and corroborated verification, so the conclusion remains largely documentation-based and is capped by static calibration.

Evidence confidence:Low Reviewed Jul 20, 2026 Reviewed revision 86cdeee824b7
The upstream repository has new commits since this review. The score still applies to the reviewed revision shown and may not cover the latest changes.
Before you use it
  • Do not treat the documented “current” versions, verified image defects, or AMI conclusions as reproduced by this review; recheck the AWS catalog, target region, model architecture, and task compatibility before deployment.
  • Before using mirror_image.py, confirm least-privilege AWS/ECR permissions, repository scope, Docker credential handling, and external side effects such as repository creation, with explicit approval and rollback procedures.
  • The workflow depends on AWS, ECR, Hugging Face Hub, and public-network access; mainland-China networks or restricted VPCs may be unable to reach catalogs, models, or public ECR.
Review evidence [1][2][3][4][5][6][7]
See the full review method →

What does this skill do, and when should you use it?

This skill helps agents select the correct serving container and current image URI for Amazon SageMaker model deployments. It mandates Hugging Face-curated DLCs when they are compatible and evaluates model architecture, task, instance type, region, and CUDA version. It covers LLMs, multimodal models, embeddings, rerankers, classifiers, diffusion models, and custom inference code. It also checks vLLM AMI requirements, TEI CPU/GPU variants, VPC image pulls, known-broken images, and first-boot timing.

Reads image examples, tags, CUDA/Python versions, and platform information from the official AWS Deep Learning Containers catalog; maps model tasks to huggingface-vllm, huggingface-vllm-omni, TEI, DJL Inference, HF Inference Toolkit, Hugging Face SGLang, AWS vLLM, BYOC, or SageMaker JumpStart; substitutes the target region into image URIs; checks TEI CPU/GPU instance matching and vLLM InferenceAmiVersion requirements; distinguishes ForCausalLM generative rerankers from ForSequenceClassification cross-encoders using config.json; and, when needed, uses curl, the AWS CLI, ECR, or mirror_image.py to retrieve or mirror images. It also supplies deployment image-uri and environment-variable guidance.

  1. An engineer deploying a Llama, Qwen, Mistral, or other Hugging Face text-generation model to a SageMaker endpoint.
  2. A deployment operator selecting Hugging Face vLLM-Omni for a multimodal model.
  3. A team deploying embeddings or BERT-family cross-encoder rerankers and choosing the correct TEI CPU or GPU variant.
  4. A team deploying Qwen3-Reranker or another generative reranker while avoiding an incorrect TEI selection.
  5. An engineer troubleshooting image pulls and startup failures in a VPC, a specific AWS region, or a cu130 GPU environment.
  6. A team replacing known-broken Hugging Face GPU inference images with DJL Inference or BYOC.

What are this skill's strengths and limitations?

Pros
  • Clearly mandates compatible Hugging Face images over generic AWS vLLM, DJL-LMI, and SGLang images.
  • Covers common deployment hazards involving task type, region, instance type, CUDA/AMI compatibility, and VPC networking.
  • Makes an important distinction between TEI cross-encoder rerankers and vLLM generative rerankers.
  • Documents known issues involving Hugging Face GPU inference tags, TGI, and older Hub download paths.
Limitations
  • It depends on AWS catalog and regional availability data that can change as images are released.
  • No standalone test suite or automated validation of every region and image combination is documented.
  • Endpoint creation, production defaults, and IAM workflows are outside this skill's scope.
  • Image mirroring requires Docker, the AWS CLI, and usable AWS/ECR permissions.

How do you install this skill?

Copy or symlink skills/hf-cloud-serving-image-selection/ from the repository's skills/ directory into a standard Codex .agents/skills location, such as $REPO_ROOT/.agents/skills or $HOME/.agents/skills. The README does not document a dedicated standalone installation command for this skill.

How do you use this skill?

In an Agent Skills-compatible coding agent, use a prompt such as: "Deploy this Qwen model to SageMaker; choose the correct serving image URI and check the region, instance type, and vLLM AMI requirements." Before creating resources, provide the model ID, AWS region, instance type, and deployment mode. The skill directs the agent to read the URI from the AWS DLC catalog instead of hardcoding one from memory.

How does this skill compare with similar options?

The skill explicitly compares Hugging Face vLLM with AWS vLLM, DJL Inference, SGLang, DJL-LMI, and TGI. A compatible Hugging Face image must be selected first; AWS vLLM is only a fallback for verified incompatibility, regional unavailability, or a known-broken Hugging Face image. TGI should not be used.

FAQ

Does this skill automatically deploy a SageMaker endpoint?
No. It selects the serving container, retrieves or derives the image URI, and provides configuration checks. Endpoint creation is handled by deployment scripts or other production-deployment skills.
Why not choose an AWS vLLM image simply because its version is newer?
The skill prioritizes model compatibility. A newer version in the generic repository is not sufficient justification; a compatible Hugging Face image remains mandatory.
Can TEI serve every reranker?
No. BERT-family ForSequenceClassification cross-encoders use TEI, while ForCausalLM generative rerankers such as Qwen3-Reranker should use Hugging Face vLLM.
What is required for vLLM images tagged cu130 or later?
Set InferenceAmiVersion=al2-ami-sagemaker-inference-gpu-3-1 on the ProductionVariant; otherwise the container may fail during startup.

More skills from this repository

All from huggingface/skills

Automation & Ops

SageMaker Deployment Planner

Choose a practical Amazon SageMaker deployment path for your model, traffic pattern, and latency needs.

Automation & Ops

SageMaker Production Defaults

Deploy SageMaker real-time or async endpoints with autoscaling, CloudWatch alarms, and consistent tagging built in.

Data & Analysis

Hugging Face Local Evaluator

Run controlled inspect-ai or lighteval evaluations for Hugging Face Hub models on local hardware.

Dev & Engineering

Hugging Face ZeroGPU Builder

Build dependable Gradio AI demos and GPU workloads on Hugging Face ZeroGPU.

Automation & Ops

AWS Context Discovery

Resolve the active AWS environment before any account operation.

Automation & Ops

SageMaker Python Environment Setup

Build an isolated, version-aligned Python environment for SageMaker and AWS automation.

Data & Analysis

Hugging Face Dataset Explorer

Explore, query, and extract Hugging Face datasets through the Dataset Viewer API.

Dev & Engineering

Hugging Face Local Model Runner

Choose compatible GGUF quantizations from Hugging Face and run them locally with llama.cpp.

Automation & Ops

SageMaker IAM Preflight

Find and validate a usable SageMaker execution role before deployment or training.

Dev & Engineering

LoRA Space Demo Builder

Build and publish a tailored Gradio Space for a Hugging Face LoRA.

Dev & Engineering

Hugging Face Spaces Builder

Build, deploy, debug, and maintain machine-learning apps on Hugging Face Spaces.

Dev & Engineering

Hugging Face Paper Research Assistant

Read AI papers and retrieve structured metadata from Hugging Face paper pages.

Data & Analysis

Hugging Face LLM Trainer

Train, fine-tune, and preserve models on Hugging Face cloud GPUs.

Data & Analysis

TRL Language Model Trainer

Train and fine-tune transformer language models with TRL’s CLI for SFT, preference alignment, and reinforcement learning.

Data & Analysis

Sentence Transformers Training Assistant

Train retrieval, similarity, reranking, and sparse-search models with Sentence Transformers.

Data & Analysis

Transformers.js Inference Skill

Run Hugging Face models directly inside JavaScript applications.

Data & Analysis

Hugging Face Model Memory Estimator

Estimate inference memory without downloading model weights.

Data & Analysis

Trackio Experiment Tracking

Log, monitor, and query machine-learning experiments with dashboards, diagnostic alerts, and automation-ready output.

Data & Analysis

Hugging Face Vision Trainer

Train detection, classification, and segmentation models on Hugging Face Jobs GPUs and persist the results to the Hub.

Dev & Engineering

Hugging Face Hub CLI Skill

Give agents command-line control over Hugging Face Hub models, datasets, Spaces, jobs, and infrastructure.

Related skills