Project Lightning Talk: Modelpack: Standardizing The Packaging And Distribution Of... Andrew Block

This talk introduces Model Pack, a project designed to simplify the distribution and management of AI and ML models. The speaker discusses the challenges faced in the AI and ML landscape, such as packaging, storage, and serving models, and highlights how Model Pack addresses these issues through standardized practices. As a vendor-neutral initiative within the Cloud Native Computing Foundation, Model Pack utilizes OCI artifacts for easy model management and integration with existing cloud-native tools. The speaker emphasizes the importance of community collaboration and the benefits of using Model Pack for CI/CD integrations, security features, and interoperability with current infrastructures. Overall, this session provides insights into how Model Pack streamlines the AI model lifecycle from packaging to deployment.

All right, everyone. I'm going to talk to you today about Model Pack and its ability to distribute and package AI and ML models. Now, for those of you who work in the AI and ML space, it's a challenging time. It's a challenging time because you have to manage a lot of the content. And there are three ways that you have to potentially manage this content. One is

you have to package your content. And when you package any content, you're going to have various files, folders, archives, different types of data. You then have proprietary information that might be in those models. You also need to figure out how to store the models. You're storing them in a variety of locations. Is it going to be an S3 bucket? Is it going to be within Git? Is

it going to be in some proprietary model store that your provider or whatever tooling you're using? And finally, how are you serving the content? You know, each framework is going to have its own type of of content ability. You might have like if you're using Python, it may have its own idiosyncrasies on top of that. And if you're using Kubernetes, there's a entire management ecosystem that you

have to potentially deal with. Those are all challenges. So, that's why the community came up with a project called Model Pack. Model Pack is a vendor neutral open standards way for managing AI and ML models. It is a sandbox project in the CNCF since last year in May 2025. how do you go ahead and leverage the power of Model Pack? Well, first of all, it packages AI

and ML models as OCI artifacts using standardized and easy distribution and that can be used for easy distribution and consumption. Uh its benefits. The benefits of Model Pack are easy. First of all, you can apply your open cloud native standards. So, we're all here. We're all part of the cloud native community. We're just reusing a lot of the abilities that are already out there in the community

using the open container initiative. Now, OCI, it is, you may or may not know, is a standard for how you manage container images. Well, recently, you can now go ahead and manage other things aside from container images using a concept called the OCI artifacts. So, you know, Robert mentioned earlier, one way you can also manage use artifacts is through Helm. Helm charts can now be distributed as

Some of the benefits in addition to not only using cloud native standards is that you can use reuse your existing infrastructure because you can use basically everything you use today with containers, you can now use with AI and ML models. And then finally, you have the interoperability between all your existing tools and frameworks. We're not building new things. We're able to leverage what's already out there in

the cloud native community. So, what features and capabilities are part of Model Pack? Well, first of all, as I mentioned earlier, we have that standard packaging for assembling all of your model files, your dependencies, and your metadata. There's an entire CLI that's provided by Model Pack that you can simplify and abstract a lot of the underlying capabilities for you to unpackage and build the AI and ML

models. You can you have CI/CD integration because you can easily integrate with the existing tool chains and frameworks. Jenkins, GitHub Actions, you name it, can all easily be brought into the Model Model Pack ecosystem. You have security built in because guess what? You're using OCI content. We have over a decade's worth of of technologies and and and tooling around OCI for how you potentially sign and distribute

content. You can sign and you can sign your container images. You can sign your artifacts that are part of Model Pack. You can also generate SBOMs and using the same tooling that you already have. Then finally, most importantly, community first. This is another open source community project. You can standardize and work with the community to be more productive and effective with AI and ML content. So, publishing

and building AI and ML content is easy. You have a model repository. You then go ahead and use the tooling that's provided by Model Pack to create a model artifact, and you publish it to a container registry. How many of you have a container registry? Raise your hand. You can go ahead and use Model Pack then. And you get all the benefits of version management, distribution, access

control, and security and compliance out of the box. No additional capabilities cuz it's already there. Just go ahead and use it. And using it at runtimes is simple. Use Model Pack use the model CTL to pull down it locally to your machine, or you can use a container runtime using Docker, Podman, or in Kubernetes, you can then leverage the CSI driver or the the out-of-the-box image volume

source. All different ways that you can consume the Model Pack content as you need to. So, if you want to go ahead and learn more about Model Pack, while you're here, you're number one, so that's one way to learn. If you want to go interact with the community, go ahead and do so at our Model Pack kiosk tomorrow from 10:00 from 10:00 to 2:40. You can also

go ahead and go to our website modelpack.org, go to the github.com/modelpack, or go to the Model Pack Slack channel on CNCF Slack. Thanks, everyone. Wow, that was dead ass on