Comet
Experiment tracking, model registry & production monitoring for ML teams
ORMB
Version, share, and serve ML models via OCI image registry
ORMB lets you store, version, and distribute machine learning models as OCI artifacts, using Docker or Harbor registries, with simple CLI commands for save, push, pull, and export.
Overview
Overview
ORMB is a lightweight Go‑based tool that treats machine‑learning models as OCI artifacts. By leveraging existing container registries such as Harbor or Docker Registry, it provides a unified place to store, version, and retrieve models without requiring a separate database.
How it works
Developers place a model directory alongside an ormbfile.yaml specification, then use the CLI to save the model to a local cache, push it to a remote registry, pull it back when needed, and export it to a local filesystem. All operations report digests, sizes, and format metadata, making reproducibility straightforward. ORMB integrates with CI/CD pipelines and serving frameworks like Seldon Core, enabling automated model promotion from training to production.
Deployment
Install the pre‑compiled binary or build from source, configure your preferred OCI‑compatible registry, and start managing models with a single command line. No additional UI is required, and the tool works across Linux, macOS, and Windows environments.
Highlights
OCI‑compatible storage of ML/DL models as image artifacts
Simple CLI for save, push, pull, and export operations
Works with any OCI registry (Harbor, Docker Registry, etc.)
Built‑in versioning and metadata tracking
Pros
- Leverages existing container registry infrastructure
- Lightweight Go binary with minimal dependencies
- Follows the OCI model specification for portability
- Straightforward versioning via image tags
Considerations
- No graphical user interface; CLI‑only interaction
- Requires an external OCI registry to be set up
- Model directories must conform to the ORMB spec
- Community and ecosystem are smaller than some commercial registries
Managed products teams compare with
When teams consider ORMB, these hosted platforms usually appear on the same shortlist.
Looking for a hosted option? These are the services engineering teams benchmark against before choosing open source.
Fit guide
Great for
- Teams already using Docker/Harbor for CI‑CD pipelines
- MLOps workflows needing reproducible model artifacts
- Organizations that prefer CLI‑driven tooling
- Projects that want to store models alongside container images
Not ideal when
- Users requiring a full web‑based model catalog UI
- Small experiments without an existing registry
- Environments lacking Docker or OCI registry access
- Teams needing advanced governance features out of the box
How teams use it
CI/CD pipeline integration
Automatically push trained models to a Harbor registry after each training run.
Model serving with Seldon Core
Pull versioned models from the registry for seamless deployment in production.
Cross‑team model sharing
Provide a single source of truth for model versions accessible to multiple data science teams.
Backup and archival
Store immutable snapshots of models in an OCI registry for compliance and disaster recovery.
Tech snapshot
Go83%
Python9%
Makefile7%
Shell1%
Dockerfile1%
Scheme1%
Frequently asked questions
Which container registries does ORMB support?
Any OCI‑compatible registry, such as Harbor, Docker Registry, or other OCI artifact stores.
How does versioning work?
Each model is tagged like a Docker image (e.g., `my_model:v1`), and the registry tracks digests for immutability.
Can I store models other than TensorFlow SavedModel?
Yes, any model format that can be packaged and described in `ormbfile.yaml` is supported.
How can I view model metadata in the registry?
Metadata is displayed in the registry UI (e.g., Harbor) and also printed by the CLI during push/pull.
Is authentication handled by ORMB?
ORMB relies on the underlying registry’s authentication mechanisms; configure credentials as you would for Docker.
Project at a glance
Dormant- Stars
- 471
- Watchers
- 471
- Forks
- 62
LicenseApache-2.0
Repo age5 years old
Last commit2 years ago
Primary languageGo
Last synced 4 hours ago