Papers › Adapting Pretrained Text-to-Text Models for Long Text Sequences

Adapting Pretrained Text-to-Text Models for Long Text Sequences

21 Sep 2022arXiv:2209.10052archive 2025-07-28

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

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.

PaperPDFCode

In Syntology Open this paper in Syntology's Atlas, the map of the papers in Syntology's graph and their citations.

Code

facebookresearch/bart_ls officialmentioned in papermentioned on GitHubpytorchNOASSERTION report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

Long-range modelingQuestion AnsweringText Summarization

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Long-range modeling SCROLLS BART-LS Avg. 39.76 #4 of 13 Archive leaderboard report
Long-range modeling SCROLLS BART-LS CNLI 87.1 #4 of 13 Archive leaderboard report
Long-range modeling SCROLLS BART-LS GovRep 59.4 / 29.8 / 30.8 #4 of 13 Archive leaderboard report
Long-range modeling SCROLLS BART-LS Nrtv 26.2 #4 of 13 Archive leaderboard report
Long-range modeling SCROLLS BART-LS QALT EM-T/H 37.8 / 34.0 #4 of 13 Archive leaderboard report
Long-range modeling SCROLLS BART-LS QMSum 35.1 / 11.0 / 22.0 #4 of 13 Archive leaderboard report
Long-range modeling SCROLLS BART-LS Qspr 48.7 #4 of 13 Archive leaderboard report
Long-range modeling SCROLLS BART-LS SumScr 37.7 / 10.2 / 21.5 #4 of 13 Archive leaderboard report
Text Summarization Arxiv HEP-TH citation graph BART-LS ROUGE-1 50.2 #2 of 28 Archive leaderboard report
Text Summarization BookSum BART-LS ROUGE 38.5 #2 of 3 Archive leaderboard report
Text Summarization GovReport BART-LS ROUGE-1 62.0 #2 of 2 Archive leaderboard report
Text Summarization Pubmed BART-LS ROUGE-1 50.3 #3 of 29 Archive leaderboard report
Text Summarization QMSum BART-LS ROUGE-1 37.9 #1 of 1 Archive leaderboard report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections