StructBERT is a new model that modifies BERT's pretraining objective to enhance its performance in NLP tasks. It introduces two new objectives: word sentence objective and sentence structural objective. By incorporating these objectives, StructBERT consistently outperforms BERT in various downstream tasks.

4m read time From towardsdatascience.com
Post cover image
Table of contents
Large Language Models, StructBERT — Incorporating Language Structures into PretrainingIntroductionPretraining1. Word sentence objective2. Sentence structural objectiveFinal objectiveStructBERT settingsStructBERT versionsConclusionResources

Sort: