local-container-registry

name	local-container-registry
description	Run a local Docker registry, publish reusable CUDA-capable ML base images, and handle ML container/runtime or local ML service setup issues.

Skill: Local Container Registry

Goal

Use a reusable local registry + baked ML image workflow instead of repeatedly rebuilding CUDA/torch stacks.

Use This Skill When

• You need localhost:5000/... images across repos.
• CUDA/torch libraries are missing in containers.
• You need a reusable ML toolbox image.
• You need to stand up a local ML service like embeddings inference or the Open Hax proxy/model gateway.

Steps

1. Start/validate the registry.
2. Prefer official CUDA-capable upstream images for the base.
3. Build/tag/push a baked ML image.
4. Point repo builds at PROMPTBENCH_BASE_IMAGE.
5. Use dedicated service images when the goal is a specific local ML API.

Example

docker compose -f ~/.config/local-container-registry/docker-compose.yml up -d
curl -sSf http://localhost:5000/v2/_catalog

docker build -f Dockerfile.ml-base \
  -t localhost:5000/shibboleth/ml-base:cuda12.4-2026-03-18 \
  .

docker push localhost:5000/shibboleth/ml-base:cuda12.4-2026-03-18

PROMPTBENCH_BASE_IMAGE=localhost:5000/shibboleth/ml-base:cuda12.4-2026-03-18 \
  docker compose build