This project reviews the use of augmentations techniques in NLP to optimise GPT-2 text generation performance. Several samples have been tested, examining different augmentation procedure such as corpus replication, random word shuffling and synonym substitution using GloVe as static non-contextual linear embedding model. The result shows whenever it is effective to augment a text corpus before GPT-2 fine-tuning processing, and how this augmentation influences the resulting output.

You may also like

Back to Top