Chapter 3: Instruction Tuning
What is Instruction Tuning?
Training models to follow instructions in natural language.
Dataset Format
{
"instruction": "Summarize the following text",
"input": "Long text to summarize...",
"output": "Concise summary..."
}
Popular Datasets
| Dataset | Size | Source |
|---|---|---|
| Alpaca | 52K | GPT-4 generated |
| Dolly | 15K | Human written |
| OpenAssistant | 161K | Crowdsourced |
| FLAN | 1.8M | Academic tasks |
Training Template
def format_instruction(example):
return f"""### Instruction:
{example['instruction']}
### Input:
{example['input']}
### Response:
{example['output']}"""
Quality Impact
Better instruction data → Better instruction following:
- Diverse instructions
- High-quality responses
- Consistent formatting
- Task variety