Finetuned

Fine-tuning is a machine learning technique in which a model that has been pretrained on a large dataset is further trained on a smaller, task-specific dataset in order to adapt it to a particular task or domain. The term is commonly written as fine-tuned or finetuned. The pretrained model provides general representations that can be specialized for the target task, reducing the need for large labeled datasets and lowering training costs compared with training from scratch.

Typical workflow involves starting with a pretrained model, assembling a target dataset, and choosing a fine-tuning strategy. In full fine-tuning, all model parameters are updated during training on the target data. In cheaper approaches, such as feature extraction, only the final layers or a small number of adapters are trained while earlier layers are frozen. Other parameter-efficient methods include LoRA and adapter modules that inject trainable parameters without updating the entire network.

Key considerations include the size and quality of the target data, domain shift between pretraining and target tasks, and the risk of overfitting on small datasets. Regularization, early stopping, and learning-rate scheduling are commonly used. Data leakage and mismatched labels should be avoided. Evaluation should compare to baselines trained from scratch and to other transfer-learning approaches.

Applications span natural language processing, computer vision, and audio or multimodal tasks. In NLP, models like BERT or GPT are often fine-tuned for classification or question-answering; in vision, convolutional networks pretrained on ImageNet are fine-tuned for specialized tasks such as medical imaging or remote sensing.

Limitations include dependence on the quality of the pretraining data, potential for overfitting on small domains, and computational requirements. Fine-tuning remains a core technique in modern machine learning, enabling rapid adaptation of large, general-purpose models to a wide range of tasks.