For Heads-of · Practitioner
Fine-Tuning
The process of adapting a pre-trained model to a specific task or domain by training it further on a smaller, task-specific dataset.
- fine-tuning
- model adaptation
- training
- customization
Process
Start with a model already trained on broad data, then train it further on domain-specific examples. It needs far less data and compute than training from scratch.
Use cases
Adapting a general language model to legal document analysis, medical coding, a house writing style or an organisation's own terminology.
Governance implications
Fine-tuning creates a new model with its own risks: the fine-tuning data can introduce bias or be poisoned, and research shows that even benign fine-tuning can weaken safety behaviour learned in base training. Under the EU AI Act, modifying a model substantially can also change who counts as its provider. The quality and provenance of the fine-tuning data are critical; see training data provenance checks.
Related Terms
Data Poisoning
An attack in which an adversary deliberately corrupts data an AI model learns from, to degrade it or plant hidden behaviour. It now reaches beyond training data to fine-tuning sets, retrieval corpora and agent memory, and a poisoned model can pass ordinary evaluation.
Large Language Model (LLM)
A deep learning model trained on very large amounts of text, able to generate and interpret human-like language across a wide range of topics and tasks.
Model Versioning
The practice of recording which model, trained how and on what, was deployed when, enabling comparison, rollback and reproducibility. For systems built on models the organisation did not train, it extends to versioning the surrounding assembly, since that is the only part it controls.
Reinforcement Learning from Human Feedback (RLHF)
A training method in which people rate a model's outputs, a reward model learns to predict those ratings, and the model is then fine-tuned to produce outputs the reward model scores highly. It is the main technique that turned raw language models into usable assistants.
Training Data
The dataset used to train a machine learning model, from which the model learns patterns and relationships.