Stage 2 — Instruction Fine-Tuning: Teaching the Model to Follow Prompts

~15 min read

Instruction fine-tuning (IFT) turns a text-completing pre-trained model into a conversational assistant by training it on instruction-response pairs, teaching it to answer questions, follow formatting, and summarize rather than just continue text.

Stage 2 — Instruction Fine-Tuning: Teaching the Model to Follow Prompts is a Pro topic

Sign in, then upgrade to Pro or Power to unlock this topic and the full AI Engineering curriculum.

Key points

  • IFT trains on instruction-response pairs, not raw text — teaching the model to follow prompts, not just continue them
  • Before IFT, a pre-trained model may just continue a question rather than answering it
  • After IFT, the model can reliably answer questions, summarize, and write code on request
  • IFT's data comes from curated instruction/response examples, which is far more expensive to scale than raw pre-training text
  • Diminishing returns from more instruction data motivate the next stage: preference fine-tuning via reinforcement learning