Examples
Worked examples
- Is an instance
A fine-tuned Llama 3 model deposited to Hugging Face with a Zenodo DOI, model card, and licence
Counter-examples
Looks similar, but isn't
- Not an instance
A model checkpoint stored on a personal laptop with no repository or DOI is not a model-as-output in the citable sense
Editorial commentary
A model-as-output is a trained machine-learning model — its weights, architecture, training metadata, and an accompanying model card — deposited in a model repository (Hugging Face and Zenodo are the most common) with a persistent identifier, and treated as a citable scholarly output independent of any paper that may describe it. This mirrors the logic that made datasets independently citable: citation conventions typically expect the model’s name, version, repository DOI, and its model card, in the way a dataset citation expects a DOI and its metadata record.
A research office is increasingly likely to see model outputs listed in AI/ML researchers’ CVs and grant reports alongside, or instead of, a conventional publication, particularly where the model itself — not a paper describing it — is the project’s primary deliverable. The model card, following the framework Mitchell et al. proposed in 2019, is the structured documentation that makes a deposited model auditable: intended use, training data, evaluation results, and known limitations.
How it relates to sibling output types
- vs. Dataset (as output) — a model is a trained function that maps inputs to outputs; a dataset is the data itself, which may or may not have been used to train any model.
- vs. Container image (as output) — a container image packages the environment a model was trained or run in; the model weights and architecture are a separate, though often co-deposited, artefact.
- vs. Workflow (as output) — a workflow may describe the training pipeline that produced a model, but the trained model itself is the separate citable object.
References
- Hugging Face Model Hub citation conventions.
- Mitchell et al., “Model Cards for Model Reporting,” FAT* (2019).
Also known as
Trained model (output) · ML model output
Machine-readable encodings
Use in your systems
<role vocab="credit"
vocab-identifier="https://casrai.org/dictionary/"
vocab-term="Model (as output)"
vocab-term-identifier="https://casrai.org/dictionary/term/model-as-output" />{
"@context": "https://schema.org",
"@type": "DefinedTerm",
"@id": "https://casrai.org/dictionary/term/model-as-output",
"name": "Model (as output)",
"identifier": "https://casrai.org/dictionary/term/model-as-output",
"description": "A trained machine-learning model — including its weights, architecture, training metadata, and accompanying model card — deposited in a model repository (Hugging Face, Zenodo) with a persistent identifier, treated as a citable scholarly output independent of any paper that may describe it.",
"inDefinedTermSet": "https://casrai.org/dictionary/domain/research-outputs#set",
"url": "https://casrai.org/dictionary/term/model-as-output",
"alternateName": [
"Trained model (output)",
"ML model output"
],
"license": "https://creativecommons.org/licenses/by/4.0/",
"publisher": {
"@id": "https://casrai.org/#organization"
},
"author": {
"@id": "https://casrai.org/#editorial-team"
},
"datePublished": "2026-05-21T02:07:47",
"dateModified": "2026-08-22T14:11:50",
"inLanguage": "en-GB",
"isAccessibleForFree": true
}
