Supervised Fine-Tuning (SFT)
Definition
Supervised fine-tuning adapts a pretrained model based on given input and output examples.
The further optimization of a pretrained model using labeled input-output examples and a supervised learning objective.
Simply explained
The model practices with tasks and corresponding pattern answers so that it can handle similar tasks better later on.
Why it matters in practice
The term helps to select, develop, evaluate AI systems in a well-founded manner and realistically assess their technical risks.
Example from business
A model learns the desired support style from verified question-answer examples.
Limits and pitfalls
- Ensure SFT with knowledge updating, mix test data, or multiply problematic examples.
Origin & history
SFT originated from transfer learning and often forms the first adaptation stage of modern language models.
Use in mid-sized companies
- Possible example: A model learns the desired support style from verified question-answer examples.
Frequently asked questions
What does Supervised Fine-Tuning (SFT) mean?
The further optimization of a pretrained model using labeled input-output examples and a supervised learning objective. What mistakes should be avoided? Ensure SFT with knowledge updating, mix test data, or multiply problematic examples.
Sources
Source checked on 10.10.2026. Editorial application examples are hypothetical; no proof of real customer results.
Editorially reviewed on: