# What is Supervised Instruction Tuning? Definition & Technical Mechanics

> **Source:** [SPIDITS AI AI Glossary](https://spidits.com/ai-glossary)  
> **Category:** Model Training  
> **Canonical Citation:** [https://spidits.com/ai-glossary/supervised-instruction-tuning](https://spidits.com/ai-glossary/supervised-instruction-tuning)

## Definition
Supervised Instruction Tuning (SFT) is a training phase where a pre-trained base model is fine-tuned on a curated dataset of instruction-response pairs. This teaches the model to understand prompts, adopt an assistant persona, and output responses in a structured format.

## Architectural Mechanics & Deep Dive
Supervised Instruction Tuning (SIT) is the alignment phase where a pre-trained base model is fine-tuned on a high-quality dataset of instructions and corresponding answers (prompts and completions). This trains the model to respond as a helpful assistant, transforming raw token prediction into dialogue.

## Industry Applications & Real-World Use Cases
Conversational model preparation, API JSON formatting models, and customer service fine-tuning.

## Frequently Asked Questions

### What is the difference between SFT and base pre-training?
Pre-training uses raw text to learn word statistics. SFT uses structured prompt-response templates to teach the model how to behave as an assistant.

### What follows Supervised Instruction Tuning in modern alignment?
Usually preference optimization phases like RLHF (Reinforcement Learning from Human Feedback) or DPO (Direct Preference Optimization).

## Related Terminology
- [fine-tuning](https://spidits.com/ai-glossary/fine-tuning)
- [rlhf](https://spidits.com/ai-glossary/rlhf)
- [dpo](https://spidits.com/ai-glossary/dpo)

---
*From the SPIDITS AI Knowledge Base (300+ terms indexed). Explore full technical definitions: [https://spidits.com/ai-glossary](https://spidits.com/ai-glossary)*