AI Researchers from UC Berkeley and Meta Present AST-T5: A Novel Pretraining Paradigm that Harnesses the Power of Abstract Syntax Trees (ASTs) to Boost the Performance of Code-Centric Language Models
