ModelPack: Packaging ML models as OCI artifacts made easy
summary
ModelPack packages real ML models as OCI artifacts, using a modl CLI to generate a model manifest (similar to a lightweight Dockerfile), build and push layered artifacts to standard OCI registries, and mount them into Kubernetes via a CSI volume mount so nodes reuse container storage instead of downloading huge files. The project provides signatures and SBOMs, integrates into CI/CD, runs a demo loading a 6B model served by VLM that answers queries, and exists as an open source CNCF sandbox.
AI-written summary · about
chapters (7)
- 0:00ModelPack and OCI artifacts
- 5:41Container wars and OCI standards
- 10:40OCI artifacts for model packaging
- 19:47Getting started with ModelPack
- 21:10Demo: building and pushing model
- 25:07Deploying models on Kubernetes
- 29:23Q&A on ModelPack
related talks
05mModelpack: Standardizing The Packaging And Distribution Of...
05mProject Lightning Talk: ModelPack: Bringing the OCI Standard to AI/ML
18mBaby Sitting Your AI Models in Production? STOP!
23mBringing Container-Native Simplicity to AI/ML
20mBuilding Governable ML Models with R
27mBuilding Secure MLOps Pipelines With KitOps + Argo Workflows
06mEvolving with Kubernetes: Embracing Model Ops
34mOpen Standards For Machine Learning Model Deployment