Examples
Worked examples
- Is an instance
A model card recording 'SFT on 200k instruction pairs, then DPO on 50k preference pairs, LoRA rank 64'.
- Is an instance
A fine-tune chain documented as Mistral-7B -> instruction-SFT -> domain-SFT -> safety-DPO.
Counter-examples
Looks similar, but isn't
- Not an instance
A base-only model with no fine-tuning history.
- Not an instance
A statement 'fine-tuned for our use case' without procedure details.
Editorial commentary
Model fine-tune lineage is the specific record of the fine-tuning operations applied to a base model to produce a derivative: which dataset was used, which method (supervised fine-tuning, DPO, RLHF, LoRA, or another parameter-efficient technique), what hyperparameters were set, and which resulting checkpoint identifier the run produced. It is deliberately narrower than the general concept of model lineage: it documents one class of derivation event, not the full ancestry graph of a model.
How this differs from model lineage
Model lineage is the umbrella term covering every kind of derivation relationship a model can have — fine-tunes, merges, distillations, quantisations, and forks. Fine-tune lineage is the record you look up specifically when the derivation in question was a fine-tuning run. If an institution only ever fine-tunes (and never merges or distils models), its fine-tune-lineage records and its model-lineage records will, in practice, be the same document under two names — the distinction only earns its keep once an organisation also tracks non-fine-tuning derivations. Treat this term as the fine-tuning-specific slice of the broader lineage record, not as a competing concept; the two pages should be read together.
What a complete record captures
Dataset identifier and version, fine-tuning method, key hyperparameters (learning rate, epochs, LoRA rank where applicable), base model identifier and version, and the resulting checkpoint identifier. Without all five, a downstream reviewer cannot reproduce or meaningfully audit the fine-tuning step.
Why it matters for research administration
A researcher publishing results from a fine-tuned model, or a lab procuring one from a vendor, needs this record to assess whether the training data included licensed, personal, or otherwise restricted material — the same due-diligence question a data-management plan answers for a raw dataset.
References
- See also: Model lineage, Fine-tuning, Model checkpoint.
Also known as
fine-tuning provenance
Machine-readable encodings
Use in your systems
<role vocab="credit"
vocab-identifier="https://casrai.org/dictionary/"
vocab-term="Model fine-tune lineage"
vocab-term-identifier="https://casrai.org/dictionary/term/model-fine-tune-lineage" />{
"@context": "https://schema.org",
"@type": "DefinedTerm",
"@id": "https://casrai.org/dictionary/term/model-fine-tune-lineage",
"name": "Model fine-tune lineage",
"identifier": "https://casrai.org/dictionary/term/model-fine-tune-lineage",
"description": "The specific portion of model lineage that records the sequence of fine-tuning operations applied to a base model: dataset, method (SFT, DPO, RLHF, LoRA), hyperparameters, and resulting checkpoint identifier.",
"inDefinedTermSet": "https://casrai.org/dictionary/domain/ai-ml-research-outputs#set",
"url": "https://casrai.org/dictionary/term/model-fine-tune-lineage",
"alternateName": [
"fine-tuning provenance"
],
"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": "2026-09-05T17:31:45",
"inLanguage": "en-GB",
"isAccessibleForFree": true
}
