Zubnet AILearnWiki › Instruction Tuning
Training

Instruction Tuning

Instruction Fine-Tuning, IFT, SFT
Fine-tuning a pre-trained language model on a dataset of (instruction, response) pairs to teach it to follow instructions. A base model that just predicts text becomes a model that answers questions, follows directions, and behaves like an assistant. This is the step that turns GPT into ChatGPT, or a base Llama into Llama-Chat.

Why it matters

Instruction tuning is the bridge between a raw language model (which can only complete text) and a useful assistant (which can follow instructions). Without it, even the most capable base model just generates plausible-sounding text rather than actually doing what you ask. It's arguably the most important post-training step.

Deep Dive

The process: collect thousands to millions of (instruction, ideal response) pairs covering diverse tasks — Q&A, summarization, coding, creative writing, math, conversation. Fine-tune the base model on these pairs using standard supervised learning (minimize the loss on the response tokens given the instruction). The model learns the meta-pattern: "when given an instruction, produce a helpful response."

SFT vs. RLHF vs. DPO

Instruction tuning (Supervised Fine-Tuning / SFT) is typically the first post-training step, followed by alignment via RLHF or DPO. SFT teaches the model the format and basic helpfulness. RLHF/DPO then refines the behavior — making responses more helpful, less harmful, and better calibrated. Some approaches (like ORPO) combine SFT and preference alignment into a single step.

Data Quality Over Quantity

Research consistently shows that a small set of high-quality instruction-response pairs outperforms a large set of low-quality ones. The LIMA paper (Zhou et al., 2023) showed that fine-tuning with just 1,000 carefully curated examples could produce surprisingly good results. The key is diversity (covering many task types) and quality (responses that are genuinely excellent, not just adequate). This is why instruction data curation has become a specialized discipline.

Related Concepts

← All Terms
← Instruction Following Jailbreak →