CodeBPE: Investigating Subtokenization Options for Large Language Model Pretraining on Source Code.

Nadezhda Chirkova, Sergey Troshin

Browse the full ICLR paper archive.