No description
Find a file
Ignasi 6c08602e27
add multimodal projector and chat template media types to spec (#1)
* docs: add multimodal projector and chat template media types to spec

Add two new media types to the AI model specification:
- `application/vnd.docker.ai.gguf.v3.mmproj` for multimodal projector weights
- `application/vnd.docker.ai.chat.template.jinja` for Jinja prompt templates

Also fix the media type in the example manifest to use the correct format
(`application/vnd.docker.ai.gguf.v3` instead of the non-existent
`application/vnd.docker.ai.gguf.model.v3`).

* Update spec.md

Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>

---------

Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
2025-09-25 17:47:17 +02:00
config.md emphasize NOT wrt the container runtime spec 2025-03-24 20:19:14 -04:00
LICENSE Add LICENSE file 2025-03-24 20:20:37 -04:00
README.md emphasize NOT wrt the container runtime spec 2025-03-24 20:19:14 -04:00
spec.md add multimodal projector and chat template media types to spec (#1) 2025-09-25 17:47:17 +02:00

OCI Model Artifact Specification

This repository contain a draft specification for packaging large language models as OCI artifacts.

The specification defines:

  • Guidelines for packaging an LLM packaging as an OCI image manifest following the artifact guideline.
  • Media types and packaging guidelines for the files that together comprise the artifact.
  • The schema for a config JSON file that describes the model and it's files.

Distribution

Artifacts adhering to this specification are distributable by any registry that supports the OCI distribution specification.

Runtime

Models packaged according to this specification may be run by any model runtime that supports this specification. Please note that these artifacts are NOT intended to be run directly by a general-purpose container runtime.