Fine-Tuning — Technology Wiki

Overview

Direct Answer

Fine-tuning is the process of taking a pre-trained neural network model and retraining its weights on a smaller, task-specific dataset to adapt its learned representations to a new domain or objective. This approach leverages existing feature knowledge whilst specialising the model for particular downstream tasks.

How It Works

The process begins with a model already trained on large-scale data, which has developed generalised feature detectors across its layers. Training resumes on the task-specific dataset, typically with a reduced learning rate to preserve earlier learned representations whilst allowing subtle weight adjustments. Some layers may be frozen to maintain their feature extractors, whilst deeper or output layers are trained more aggressively.

Why It Matters

Fine-tuning dramatically reduces training time and data requirements compared to training from scratch, lowering computational costs and enabling rapid deployment in resource-constrained settings. It achieves superior accuracy on specialised tasks where collecting large labelled datasets is prohibitively expensive, making advanced AI accessible to organisations without massive data resources.

Common Applications

Practical applications include adapting large language models to domain-specific language (legal contracts, medical notes), customising vision models for medical imaging or defect detection, and personalising recommendation systems. Named applications span natural language processing, computer vision in manufacturing, and financial fraud detection systems.

Key Considerations

Practitioners must balance learning rate selection to avoid catastrophic forgetting, where the model loses previously learned features, and avoid overfitting on small task datasets. Dataset quality and representativeness are critical, and the choice of which layers to freeze involves careful tradeoffs between computational efficiency and task performance.

Related in Language Models

Word Embedding

Dense vector representations of words where semantically similar words are mapped to nearby points in vector space.

LoRA

Low-Rank Adaptation — a parameter-efficient fine-tuning technique that adds trainable low-rank matrices to frozen pretrained weights.

Parameter-Efficient Fine-Tuning

Methods for adapting large pretrained models to new tasks by only updating a small fraction of their parameters.

Adapter Layers

Small trainable modules inserted between frozen transformer layers that enable task-specific adaptation without modifying the original model weights.

Prefix Tuning

A parameter-efficient method that prepends trainable continuous vectors to the input of each transformer layer, guiding model behaviour without altering base parameters.

Pre-Training

The initial phase of training a deep learning model on a large unlabelled corpus using self-supervised objectives, establishing general-purpose representations for downstream adaptation.

More in Deep Learning

Mixture of Experts

Architectures

An architecture where different specialised sub-networks (experts) are selectively activated based on the input.

Gradient Checkpointing

Architectures

A memory optimisation that trades computation for memory by recomputing intermediate activations during the backward pass instead of storing them all during the forward pass.

State Space Model

Architectures

A sequence modelling architecture based on continuous-time dynamical systems that processes long sequences with linear complexity, offering an alternative to attention-based transformers.

Deep Learning

Architectures

A subset of machine learning using neural networks with multiple layers to learn hierarchical representations of data.

Pipeline Parallelism

Architectures

A form of model parallelism that splits neural network layers across devices and pipelines micro-batches through stages, maximising hardware utilisation during training.

Neural Network

Architectures

A computing system inspired by biological neural networks, consisting of interconnected nodes that process information in layers.

Dropout

Training & Optimisation

A regularisation technique that randomly deactivates neurons during training to prevent co-adaptation and reduce overfitting.

Key-Value Cache

Architectures

An optimisation in autoregressive transformer inference that stores previously computed key and value tensors to avoid redundant computation during sequential token generation.