Fine-tuning

Taking a pre-trained model and training it further on a specific dataset to adapt it to a specialized task.

How it works

A pre-trained model already encodes vast general knowledge in its weights. Fine-tuning continues training this model on a smaller, task-specific dataset with a low learning rate to update the weights without overwriting the pre-trained knowledge. Full fine-tuning updates all parameters. Parameter-efficient methods like LoRA only update a small set of additional matrices, drastically reducing compute and memory requirements. The result is a model that retains general capabilities but excels at the target task.

Why it matters

Pre-training a frontier model from scratch costs tens of millions of dollars and requires specialised infrastructure. Fine-tuning allows any organisation to adapt a capable base model to their specific domain — medical records, legal documents, customer service — for a fraction of the cost. It is how virtually all production AI deployments are built today, turning general-purpose models into specialised, reliable tools.