Instruction Tuning
Supervised fine-tuning on diverse instruction-response pairs to improve a model's ability to follow commands.
Full Definition
Instruction tuning is a form of supervised fine-tuning where the training data consists of natural-language instructions paired with high-quality responses across a wide variety of tasks — summarisation, translation, Q&A, coding, brainstorming, and more. The model learns to interpret what a human wants from a directive and generate an appropriate response, even for task types not explicitly seen during tuning. This generalisation to new instructions is the defining characteristic of instruction-tuned models and is what makes models like GPT-4, Claude, and Llama Instruct so much more useful than their base counterparts for everyday tasks.
Examples
Training on 52,000 instruction-output pairs from the Stanford Alpaca dataset to transform Llama 7B into a capable instruction-following assistant.
Google's FLAN instruction-tuning dataset containing over 1,800 tasks expressed as natural-language instructions.
Apply this in your prompts
PromptITIN automatically uses techniques like Instruction Tuning to build better prompts for you.
Related Terms
Fine-Tuning
Continuing training of a pretrained model on a smaller, task-specific dataset to…
View →Reinforcement Learning from Human Feedback
A training technique that uses human preference ratings to align model outputs w…
View →Supervised Learning
Training a model on input-output pairs where the correct output is provided as a…
View →