distribution-spec VS oras

One of the main reasons teams struggle to build and maintain their MLOps pipelines are vendor specific packaging. As a model is handed off between data science teams, app development teams, and SRE/DevOps teams, the teams are required to repackage the model to work with their unique toolset. This is tedious, and stands in contrast to well adopted development processes where teams have standardized on the use of containers to ensure that project definitions, dependencies, and artifacts are shared in a consistent format. KitOps is a robust and flexible tool that addresses these exact shortcomings in the MLOps pipeline. It packages the entire ML project in an OCI-compliant artifact called a ModelKit. It is uniquely designed with flexible development attributes to accommodate ML workflows. They present more convenient processes for ML development than DevOps pipelines. Some of these benefits include:

Internally, Google used a platform called Borg which is still used by Google to this day. It also served as the basis for Kubernetes. Borg is a container-based platform whose goal was to allow developers to focus on code, not infrastructure. Google has an entire infrastructure team to manage the datacenters. This system came out circa 2004. This predates the advent of modern OCI Containers by about a decade.

Back in 2013, a little company called Docker made it really easy to start using containers to package up applications. A big key to their success was the OCI (you can learn about that here), an industry wide initiative to have standards around how we package up our applications. Because of OCI standards, we have hundreds (maybe thousands?) of tools that can be combined to manage and deploy applications. So why aren’t we using this for packaging up Notebooks and AI models as well? It would make deploying, sharing, and managing our models easier for everyone involved.

Kubernetes on the backend used to utilize docker for much of its container runtime solutions. One of the modular features of Kubernetes is the ability to utilize a Container Runtime Interface or CRI. The problem was that Docker didn't really meet the spec properly and they had to maintain a shim to translate properly. Instead users could utilize the popular containerd or cri-o runtimes. These follow the Open Container Initiative or OCI's guidelines on container formats.

OCI stands for Open Container Initiative, and its goal as an organization is to define a specification for container formats and runtime.

https://opencontainers.org/

Here is Containerfile from the repo: https://github.com/ublue-os/bazzite/blob/main/Containerfile

apko allows us to build OCI container images from .apk packages.

Flagger's load testing service can be installed via a Kustomization resource based on manifests packaged as an artifact in an Open Container Initiative (OCI) registry

Interesting, I'd not known of this RFC before.

Another example of a content-addressed data store could be OCI registries (more commonly known as container image registries). Using them to store arbitrary artefacts is quite well supported now: https://oras.land/

See https://oras.land/

> ... but still only supports one niche operating system.

1. Linux is not a niche in the target market for this project.

2. The project is written in Common Lisp with hard dependencies on SBCL-provided libraries[1], so there's reason to suspect it should work on other OSes supported by SBCL.

3. Sure, the presence of Makefile and sb-posix imply it requires a POSIX compliant OS, but Linux is not the only one that fits the bill.

4. The included Linux-only binary 'oras' is clearly a vendored artifact, not part of this project, and clearly an OCI client. A simple search shows it is indeed cross-platform[2].

Perhaps you should try what almost every Linux user has had to do when encountering software actually built for only one "niche" operating system that they want to use on their OS: look.

1. https://github.com/ocicl/ocicl/blob/170aff0/ocicl.asd#L34

2. https://github.com/oras-project/oras/releases

oras isn't that unpopular, Helm is using it as an SDK for example. Here are other projects who are using it. https://github.com/oras-project/oras/network/dependents