Fine-Tuning

Fine-tuning refers to adapting an already pre-trained AI model for specific tasks. It's like retraining an experienced chef from French to Italian cuisine — the fundamental skills are there, but the details are adjusted. Instead of training a model from scratch (which can take months and cost millions), you take an existing model and train it further with new, task-specific data. In full fine-tuning, all weights of the network are updated. Today, however, parameter-efficient methods (PEFT, such as LoRA) dominate: they freeze the base model and train only small, additional adapters across all layers. This saves compute time and data, and reduces the risk of catastrophic forgetting — the model overwriting its existing knowledge. Fine-tuning is the standard method for adapting large language models to specialized applications.