InstructLab

EmbGen pipeline generates synthetic data for specialized language models

Researchers have developed EmbGen, a new pipeline for generating synthetic training data to adapt smaller language models to specialized domains. This method decomposes corpora into entity-description pairs, reassembles…