AI engineering glossary · LLM training alignment
What is RFT (rejection sampling fine-tuning)?
Rejection sampling fine-tuning (also called RFT, RAFT): A training loop where a model generates many answers, a reward model or checker keeps only the best or correct ones, and the model is then fine-tuned on the kept answers. Repeating it gradually raises quality without needing reinforcement learning machinery.
Rejection sampling fine-tuning: an example
Generate 16 solutions per coding problem, keep those passing the tests, and fine-tune on the passing ones.
Related terms
More LLM training alignment terms
- Reward hacking
- Reward model
- Safety fine-tuning
- Scalable oversight
- Scaling laws
- Self-instruct
- Self-rewarding language models
- Self-Taught Reasoner
Where you learn it
Rejection sampling fine-tuning is part of the AI Engineering Program, which teaches it with notes, a hands-on lab and an exam. The first lesson is free.