AIGlossary

Fine-tuning

Related services

AI Automation
Fine-tuning is like sending a university graduate to advanced training courses. You already have a powerful base AI model (like GPT-4 or Llama 3) that knows "a little bit of everything". You feed her your specific data so she becomes an expert in your niche.

How does Fine-tuning work?

Creating a large language model (LLM) from scratch costs millions of dollars (this is Pre-training). Fine-tuning costs hundreds or thousands of dollars, because you only change the internal "weights" of the model to suit your style.

You need to prepare a dataset in the format "Question - Perfect Answer" (for example, 1000 examples of how your best manager answers customers). The model learns from these pairs and begins to copy that tone of voice, terminology, and logic.

Fine-tuning vs RAG

These are two different technologies. **RAG** ​​gives the model new knowledge (like an open book on an exam). **Fine-tuning** changes the behavior, style and format of the model's response. They are often used together: fine-tuning for the correct tone of voice, and RAG for current facts.

What is Fine-tuning used for?

01Corporate Tone of Voice

So that the bot communicates not like a dry robot, but responds in the signature, bold style of your brand.

02Code generation

Training the model to write code exclusively on the company's specific internal framework.

03Medicine/Jurisprudence

Creating a model that perfectly understands complex medical slang and abbreviations.

/ FAQ

Depends on the task. Sometimes 50-100 high-quality examples (few-shot) are enough, for complex tasks - from 1000 to 10,000.

/ Related terms

Fine-tuning
/ Fine-tuning

Ready to implement?

We'll help you implement it. 30-minute process audit - free.

Book an audit