Self-Instruct

A training method that uses an existing language model to automatically generate instruction-following data. The model generates task descriptions and corresponding responses to expand training datasets without requiring extensive manual annotation. It is an important technical paradigm in synthetic data generation.