SKOS concept scheme for the Model Openness Framework (MOF) Specification v1.0 (LF AI & Data Generative AI Commons, December 2024). MOF is a tiered assessment framework: if a model provides a required set of information elements under appropriate open licenses, it qualifies for a given tier (Class III: Open Model, Class II: Open Tooling Model, Class I: Open Science Model). Each tier specifies which of 17 components must be present and under which license category.
| ID | Label | Notation | Type | Definition |
|---|---|---|---|---|
class-i | Class I - Open Science Model | I | The highest class of openness. Adds 4 components (plus optional model metadata) enabling end-to-end analysis, reproduction of a similar model, and data exploration. All Class I + II + III components required. | |
research-paper | Research paper | I.2 | Requirement | A peer-reviewed or preprint research paper detailing model methodology, results and analysis. MUST be available in the distribution or on a permanent open access platform. MUST be released under CC-BY-4.0 or equivalent open content license. MAY replace the technical report (III.3). |
datasets | Datasets (training, validation and testing) | I.3 | Requirement | All data used for training, validation, fine-tuning, alignment (RLHF) and testing. Any media format. Includes tokenized datasets and any metadata (annotations, labels, bounding boxes, etc.). MUST be released under an appropriate open data or content license. |
data-preprocessing-code | Data preprocessing code | I.4 | Requirement | All code for preprocessing, cleaning and formatting training/validation/test data, fine-tuning data transformations, data ingestion, feature engineering, data augmentation and tokenization. MUST be released under an OSI-approved open source software license. |
intermediate-parameters | Intermediate model parameters (checkpoints) | I.5 | Recommendation | Model parameter checkpoints from key intermediate stages of training, in addition to the final parameters required by III.2. SHOULD be included to enable reproduction and analysis of training dynamics. |
model-metadata | Model metadata | I.6 | Permission | Structured metadata about the model distribution (optional component for Class I). Provides machine-readable information about model components, their locations and relationships. |
class-ii | Class II - Open Tooling Model | II | Builds on Class III. Adds 6 components enabling users to understand the training process, validate benchmarks and perform inference optimizations. All Class III + Class II components required. | |
training-code | Training, validation and testing code | II.2 | Requirement | Full code for training, validating and testing the model: model construction, training loop, hyperparameter selection, checkpointing, fine-tuning, RLHF and adapter code. MUST be released under an OSI-approved open source software license. |
inference-code | Inference code | II.3 | Requirement | Code for performing inference including preprocessing, postprocessing, model optimizations and dependencies. MUST include code required to fully replicate benchmark results. MUST be released under an OSI-approved open source software license. |
evaluation-code | Evaluation code | II.4 | Requirement | Code for model evaluation and benchmarking. MUST be released under an OSI-approved open source software license. |
evaluation-data | Evaluation data | II.5 | Requirement | Data used for model evaluation and benchmarking. MUST be included (standard benchmark datasets MAY be omitted if described in the report). MUST be released under a data or content appropriate open license (e.g. CDLA-Permissive-2.0 or CC-BY-4.0). |
supporting-libraries | Supporting libraries and tools | II.6 | Recommendation | Code libraries, utilities and tools developed in the course of the model. SHOULD be released under an OSI-approved open source software license. Includes tokenizers, data loaders, visualisation code, container definitions, hyperparameter search code, monitoring code, frontend code. |
class-iii | Class III - Open Model | III | The base class of openness. Requires 6 components (plus optional sample outputs) to be present and released under type-appropriate open licenses. Enables unrestricted usage, study, modification and redistribution. | |
model-architecture | Model architecture | III.1 | Requirement | The ML algorithms, neural network layout, connectivity patterns, activation functions and other architectural elements. MUST be distributed as open source code under an OSI-approved open source software license. |
model-parameters | Final model parameters (checkpoints and optimizer state) | III.2 | Requirement | Trained model weights and biases. At minimum the final model parameters and optimizer state MUST be distributed, compressed or uncompressed, in a format compatible with popular deep learning frameworks. MUST be released under an open license; SHOULD use an open data license (e.g. CDLA-Permissive-2.0) rather than a code license. |
technical-report | Technical report | III.3 | Requirement | Documentation providing performance, usage and implication information. MUST be included in the distribution or made available on a permanent open access platform (e.g. arXiv). MUST be distributed under an open license (SHOULD be CC-BY-4.0). MAY be omitted if a research paper is provided. |
evaluation-results | Evaluation results | III.4 | Requirement | Detailed quantitative metrics and qualitative results from evaluating the model. MUST include benchmark test results (standard or custom). Raw outputs MUST be distributed under an open license (SHOULD be CC-BY-4.0). |
model-card | Model card | III.5 | Requirement | A document covering model details, intended uses, evaluation factors, risks and mitigations. MUST be distributed under an open license (SHOULD be CC-BY-4.0). |
data-card | Data card | III.6 | Requirement | Summary statistics and details about training datasets: features, instances, intended uses, motivation and collection process. MAY be combined with the model card. MUST be distributed under an open license (SHOULD be CC-BY-4.0). |
sample-outputs | Sample model outputs | III.7 | Recommendation | Text, image, audio, code or other outputs generated by the model. RECOMMENDED but not required. If included MUST be shared publicly without copyright or restrictions where legally permitted. |
mof-config-file | MOF configuration file | Recommendation | A machine-readable configuration file declaring which MOF components are included in the distribution and their locations. SHOULD be included in every MOF-conformant distribution. (No direct bridge concept; closest is doc-identifier or component-external-ref.) |