AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Unlimiformer: Long-Range Transformers with Unlimited Length Input

Amanda Bertsch, Uri Alon, Graham Neubig, Matthew R. Gormley

NeurIPS 2023 · Architectures and efficiency › Attention and transformers · approach: Sparse efficient transformers for long context · 23 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Apply Unlimiformer at test time without additional trainingBART-base · GovReportROUGE-1: 48.7 -> 49.7Improved
Use Unlimiformer for validation and early stopping (ablation)BART-base · GovReportROUGE-1: 48.7 -> 51.0Improved
Train on non-overlapping chunks, then add test-time Unlimiformer (ablation)BART-base · GovReportROUGE-1: 46.2 -> 53.4Improved
Train on non-overlapping chunks without test-time Unlimiformer (ablation)BART-base · GovReportROUGE-1: 48.7 -> 46.2Hurt
Apply test-time SLED instead of standard finetuningBART-base · GovReportROUGE-1: 48.7 -> 45.8Hurt
Add Unlimiformer with alternating trainingBART-base · GovReportROUGE-1: 55.2 -> 56.6Improved
Add Unlimiformer with alternating trainingPRIMERA · SummScreenROUGE-1: 33.0 -> 33.3Improved
Apply Unlimiformer with retrieval trainingPRIMERA · BookSumEntMent: 11.6 -> 25.5Improved
Apply Unlimiformer only at test timeBART-base · BookSumEntMent: 10.0 -> 21.9Improved
Use random-encoded rather than retrieval training (ablation)BART + Unlimiformer · BookSumROUGE-1: 36.8 -> 37.3Mixed
Use random-encoded rather than retrieval training (ablation)BART + Unlimiformer · BookSumROUGE-2: 8.3 -> 6.7Hurt
Increase the length of input available to Unlimiformer (ablation)Unlimiformer · BookSumEntMentImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy