note · growing
Fine-tuning methods
Full SFT, LoRA, QLoRA, DPO, GRPO, distillation — and when each one is the answer.
Training from scratch is almost no product team's job. In practice the choice is which parameter-efficient method.
| Method | What it does | When |
|---|---|---|
| Full SFT | updates every weight | rarely — costly, forgetful |
| LoRA | low-rank adapter on a frozen model | the default starting point |
| QLoRA | LoRA over a 4-bit base | fitting onto one GPU |
| DPO | aligns on preference pairs | you have good/bad answer pairs |
| GRPO | RL against a reward function | correctness is programmatically checkable |
| Distillation | copies a big model into a small one | cost and latency pressure |
Curate200-500 examples
QLoRA SFT
DPOif you have pairs
Evalbefore promoting
An adapter does not replace retrieval. A fine-tuned model still invents the document it never saw — just in a nicer format.