Skip to content
Get packs

SageMaker Serving Image Selection

Picks the correct HuggingFace serving container and finds its current image URI for SageMaker deployments

Source Download skill
Repository
huggingface/skills
Stars
11,253
Licence
Apache-2.0
Files
3 files

What it does

When an agent is about to deploy a model to an AWS SageMaker endpoint, this skill makes it choose the right prebuilt serving container (for example HuggingFace vLLM for most text LLMs, TEI for embedding and encoder reranker models) and look up its current address from AWS's official catalog instead of guessing from memory. It steers away from the archived TGI server and from generic images unless a HuggingFace one cannot run the model. A bundled script can copy a public image into a private AWS registry for VPC setups without internet egress.

Great for

  • Engineers deploying open-weight LLMs, embedding or reranker models to SageMaker
  • Teams hitting opaque 'Failed to pass health check' errors caused by a wrong container image
  • Developers in VPCs without a NAT gateway who need images mirrored into private ECR

Skip it if

  • Deployments outside AWS SageMaker, where the container choice rules do not apply
  • Users who need a specific generic vLLM or DJL version for a stated reason, since the skill deliberately resists version-based picks
  • Anyone expecting the skill to verify live AWS catalog contents on its own, because it instructs the agent to read the catalog page rather than bundling current URIs

Try asking

  • “Deploy this Qwen3 LLM to a SageMaker endpoint in us-west-2 and pick the right container”
  • “Host this sentence-transformers embedding model on SageMaker”
  • “Which image should I use to serve this Qwen3-Reranker model on SageMaker?”
  • “Mirror this public ECR image into my private ECR repo so my VPC SageMaker endpoint can pull it”
Step-by-step guide: how to use SageMaker Serving Image SelectionInstall, first run, examples, tips and FAQs

Install in 3 steps

  1. 1

    Download the skill

    The ZIP contains the 3 original files, the Apache-2.0 licence and an attribution note.

    hf-cloud-serving-image-selection.zip
  2. 2

    Add it to your agent

    Claude Code (all projects)

    mkdir -p ~/.claude/skills && unzip hf-cloud-serving-image-selection.zip -d ~/.claude/skills/

    Claude Code / Agent SDK (this project only)

    mkdir -p .claude/skills && unzip hf-cloud-serving-image-selection.zip -d .claude/skills/

    Claude apps (web/desktop): Settings → Capabilities → Skills → Upload skill, and choose the ZIP.

  3. 3

    Ask for the task

    No special command needed — the agent reads the skill's description and loads it when your request matches. Try one of the prompts above.

Get better results

  • Give the agent your target AWS region up front so it substitutes the correct region into the image URI.
  • Review the generated image URI and the deployment log before deploying, since a wrong tag costs about 20 minutes per failed endpoint creation.
  • Before running mirror_image.py, confirm the public image source and the private repo name, since it pushes images into your AWS account.

What's inside

  • references/model-to-image.md
  • scripts/mirror_image.py
  • SKILL.md

Redistributed unmodified from huggingface/skills under the Apache-2.0 licence. The assessment was drafted with AI from the skill's files on 2026-10-29. Skills can instruct your agent to run code — read SKILL.md and any scripts before installing.