Adapting Pretrained Text-to-text Models For Long Text Sequences

Xiong Wenhan, Gupta Anchit, Toshniwal Shubham, Mehdad Yashar, Yih Wen-tau. Arxiv 2022

[Paper] [Code]    
Applications Attention Mechanism Efficiency And Optimization Has Code Model Architecture Pretraining Methods RAG Survey Paper Training Techniques Transformer

We present an empirical study of adapting an existing pretrained text-to-text model for long-sequence inputs. Through a comprehensive study along three axes of the pretraining pipeline – model architecture, optimization objective, and pretraining corpus, we propose an effective recipe to build long-context models from existing short-context models. Specifically, we replace the full attention in transformers with pooling-augmented blockwise attention, and pretrain the model with a masked-span prediction task with spans of varying length. In terms of the pretraining corpus, we find that using randomly concatenated short-documents from a large open-domain corpus results in better performance than using existing long document corpora which are typically limited in their domain coverage. With these findings, we build a long-context model that achieves competitive performance on long-text QA tasks and establishes the new state of the art on five long-text summarization datasets, often outperforming previous methods with larger model sizes. Our code has been released at https://github.com/facebookresearch/bart_ls.

Similar Work