Concept
Fine-tuning
Continued training of a pre-trained model on a task-specific dataset to specialise behaviour.
Definition
Fine-tuning takes a checkpoint that already learned general language or vision and updates its weights â sometimes only a subset (LoRA) â on a smaller curated dataset. Done well, it sharpens domain knowledge, style, and instruction following without the cost of training from scratch.
Common use cases
- Domain adaptation
- Style transfer
- Custom assistants