Model Lineage in AI

Model lineage

The chain of provenance for a model recording its base model, the fine-tuning datasets and procedures applied, and any further derivatives, such that any deployed model can be traced back to its constituent training operations.

ByCASRAI Editorial Board

· Last updated 22 Aug 2026

Why it matters

Licence terms, security patches, and known biases propagate along the lineage graph. If a base model carries a non-commercial restriction, every fine-tune inherits it; if a base model has a documented vulnerability or bias, every downstream derivative should be flagged until independently re-evaluated. For a research office reviewing a proposal to deploy or publish results from a fine-tuned model, lineage is the first thing to ask for: without it, there is no reliable way to know what the model actually is.

How this differs from its close siblings

Model lineage is the broadest of four closely related terms on this site. Model fine-tune lineage is one specific record type within it — the sequence of fine-tuning operations (dataset, method, hyperparameters) applied to a base model, rather than the full derivation graph across all operation types. Model versioning is a different axis entirely: it identifies which specific revision you are looking at, not how that revision was derived. A model checkpoint is narrower still — a single saved parameter snapshot at one point in training, the kind of artefact a lineage record points to rather than describes. In practice these overlap heavily, and a mature registry usually tracks all four as facets of one underlying record rather than as four separate documents.

Emerging mechanisms

Hugging Face’s model-card finetuned_from field and draft SPDX AI/ML bill-of-materials profiles are the two most concrete mechanisms currently in use for recording lineage in a machine-readable way; neither is yet a mandatory or universally adopted standard.

References

Hugging Face Model Card metadata schema (finetuned_from field).
SPDX AI/ML profile (draft), Linux Foundation.
See also: Training data provenance, Persistent Identifiers and Citation for AI Models and Training Datasets.

Use in your systems

JATS XML <role> element

xml

<role vocab="credit"
      vocab-identifier="https://casrai.org/dictionary/"
      vocab-term="Model lineage"
      vocab-term-identifier="https://casrai.org/dictionary/term/model-lineage" />

Schema.org DefinedTerm (JSON-LD)

json

{
  "@context": "https://schema.org",
  "@type": "DefinedTerm",
  "@id": "https://casrai.org/dictionary/term/model-lineage",
  "name": "Model lineage",
  "identifier": "https://casrai.org/dictionary/term/model-lineage",
  "description": "The chain of provenance for a model recording its base model, the fine-tuning datasets and procedures applied, and any further derivatives, such that any deployed model can be traced back to its constituent training operations.",
  "inDefinedTermSet": "https://casrai.org/dictionary/domain/ai-ml-research-outputs#set",
  "url": "https://casrai.org/dictionary/term/model-lineage",
  "sameAs": [
    "model provenance",
    "model ancestry"
  ],
  "license": "https://creativecommons.org/licenses/by/4.0/",
  "publisher": {
    "@id": "https://casrai.org/#organization"
  },
  "author": {
    "@id": "https://casrai.org/#editorial-team"
  },
  "datePublished": "2026-05-21T02:22:51",
  "dateModified": "2026-08-22T15:43:49",
  "inLanguage": "en-GB",
  "isAccessibleForFree": true
}