Jozu drives open source ModelPack & KitOps Modelkit
DevSecOps company Jozu is taking a prominent role in two open source projects, KitOps and ModelPack.
KitOps is an open source DevOps tool that packages (and version tracks) an AI or ML model, a dataset, a codebase and a corresponding configuration into a reproducible artefact known as ModelKit (see below).
ModelPack establishes open standards for packaging, distributing and running “AI artefacts” in cloud-native environments… artefacts of this kind could include functional outputs of AI models (sometimes described as tangible identifiable outputs), which can range from simple text and code to complex applications and visualisations.
ModelKits are built on existing standards to ensure compatibility.
Container-like portability
Backed by the CNCF, these projects are said to be the first to bring container-like portability to AI/ML workflows and make model delivery reproducible and secure.
“The AI/ML ecosystem has lacked open standards for packaging, versioning and sharing projects,” said Mohamed Nanabhay, managing partner at Mozilla Ventures. “The ModelPack specification, by extending the existing OCI standard for the large artifacts common in AI, has taken a huge step toward giving AI/ML teams the same flexibility as containers brought to application teams.”
Advocates in this space say that AI/ML teams have struggled to package, version and move models, datasets (and code) between development and production environments.
Kubernetes infrastructure
ModelPack extends the OCI (1.1) standard to support the large artifacts common in AI, while KitOps assembles them into versioned ModelKits that work with existing container registries and Kubernetes infrastructure so no new tooling is required.
“ModelKits make AI deployments cryptographically verifiable, compliant and reproducible across any Kubernetes environment,” said Görkem Ercan, CTO and co-founder of Jozu. “Teams can now move models through Dev, Test and Prod with the same confidence and ease that they move containers.”
“Given that the AI/ML ecosystem is rapidly evolving, the ability to bring interested parties together within a shared community provides the opportunity to drive innovation through shared collaboration,” said Andrew Block, ModelPack maintainer at Red Hat.
Jozu now offers enterprise support for both KitOps and ModelPack through its Jozu Hub platform, which provides full DevSecOps management for ModelKit projects across connected, on-premises and air-gapped environments.