Fine-Tuning

In short

Taking a pre-trained model and further training it on specific data to specialize it for a particular use case.

Like training a general physician to become a cardiologist. They already know medicine broadly — now you’re sharpening their expertise in one specific area.

After Pre-Training gives a model broad language understanding, fine-tuning narrows and sharpens it. You take the pre-trained model and continue Training it on a much smaller, carefully curated dataset specific to your needs — your company’s support tickets, legal documents, medical records, internal policies, whatever. The model adjusts its parameters slightly so it becomes better at your particular task.

While pre-training costs millions, fine-tuning can range from a few hundred to a few thousand dollars depending on the model size and dataset. Modern techniques like LoRA make this even cheaper by only updating a small fraction of the parameters rather than all of them.

The decision of whether to fine-tune, use RAG, or rely on Prompt Engineering is actually more of a business decision than a technical one. Fine-tuning works best when you need consistent behavior and you have high-quality domain-specific Data. RAG is often better when your data changes frequently.

Related

• Pre-Training - what happens before fine-tuning
• Training - fine-tuning is a type of training
• RAG - an alternative approach to specialization
• Prompt Engineering - another alternative, no training needed
• Data - you need good domain-specific data