training
Fine-tuning
Model Fine-tuning
Teaching a general AI to specialize in your domain
Reading level
Fine-tuning continues training a pre-trained model on a smaller, task-specific dataset to adapt its behavior. Full fine-tuning updates all parameters (expensive). Parameter-efficient methods (PEFT) like LoRA add low-rank adapters to attention matrices — training <1% of parameters with comparable performance. Instruction fine-tuning (IFT) teaches models to follow instructions. SFT is typically followed by RLHF/DPO for alignment.
Real-world example
A legal firm fine-tunes a general LLM on thousands of their past contracts and case notes. The resulting model becomes much better at analyzing legal documents than the base model — without needing a 100,000-word prompt each time.