Fine-tuning adapts a pre-trained model to a more specific task or domain by continuing its training on a smaller, specialized dataset, rather than training a new model from scratch. Because the model already has broad, general knowledge from its original training, fine-tuning typically only needs to nudge its existing weights toward the target task, which makes it far faster and cheaper than a full training run.
This makes fine-tuning practical for a wide range of applied uses: a company might fine-tune a general-purpose language model on its own support documentation to build a more accurate customer service assistant, or fine-tune it on legal or medical text to sharpen its performance in that specific field. The tradeoff is that fine-tuning can narrow a model's behavior toward the fine-tuning data, which is why teams typically evaluate a fine-tuned model carefully to confirm it hasn't lost general capabilities it still needs.
