Instruction tuning (SFT): teaching a model to answer

Instruction tuning, or supervised fine-tuning (SFT), trains a pretrained model on instruction–response pairs, scoring only the response tokens.

Part of Post-training.

Read first

Read next

Related