# Fine-Tuning

> Continuing a pretrained model's training on a smaller, task-specific dataset to specialize its behavior for a particular domain or format.

_The Agentic Web Lexicon · /glossary/fine-tuning · [JSON](/api/glossary/fine-tuning) · [all The Agentic Web Lexicon](/glossary)_

- **term:** Fine-Tuning
- **category:** knowledge-memory
- **short_def:** Continuing a pretrained model's training on a smaller, task-specific dataset to specialize its behavior for a particular domain or format.
- **long_def:** Fine-tuning adapts an already-trained model by updating its weights on curated examples, so it learns a style, a domain vocabulary or a structured-output format the base model does not reliably produce. It is a form of transfer learning, distinct from retrieval, which adds knowledge at inference time without changing weights.
- **see_also:** rlhf, rag, embeddings
- **etymology_origin:** — (verify-against-primary-at-build)
- **related_to:** rlhf, rag, embeddings, transformer
- **contrast_with:** Unlike retrieval-augmented generation, which supplies knowledge at query time and changes no weights, fine-tuning bakes behavior into the model by training on examples.
- **example:** A team might fine-tune a model on their support transcripts so it adopts their tone and product vocabulary without needing those examples in every prompt.
- **source:** https://en.wikipedia.org/wiki/Fine-tuning_(deep_learning)
- **status:** active
- **why_it_matters:** Fine-tuning and retrieval are the two levers for specializing an agent; the agentic web leans on retrieval, since machine-readable content updates a model's knowledge without retraining it.
- **sameAs:** https://en.wikipedia.org/wiki/Fine-tuning_(deep_learning)
- **bridge_entity:** rag
- **last_verified:** 2026-07-06
- **md_twin:** /glossary/fine-tuning.md