---
title: "Longformer: The Long-Document Transformer"
license: CC-BY-4.0
license_url: https://creativecommons.org/licenses/by/4.0/
license_note: 'paper is arXiv non-exclusive distribution license (verified on arXiv abs page 2026-11-12); this pointer note is our own words and is CC BY 4.0; we link to the paper rather than reproduce it'
source: https://iinventedlight.com/library/notes/2004-05150-longformer.md
source_license: "arXiv non-exclusive distribution license"
attribution: 'Pointer by iinventedlight (AI-assisted). Source: Iz Beltagy, Matthew E. Peters, Arman Cohan, Longformer: The Long-Document Transformer (arXiv non-exclusive distribution license). This note links to the paper rather than reproducing it.'
author: 'iinventedlight (AI-assisted)'
date: 2026-11-12
status: published
approved_by: "Cole Withington, 2026-11-12 14:44 PT, Librarian chat widget t51s316: \"Yes, publish all 100 second-list arXiv notes\""
checked_on: 2026-11-12
as_of: 2026-11-12
confidence: medium
untested: 'Pointer note only. Paper not reproduced. Abs page and license checked 2026-11-12.'
category: arxiv
topic_area: 'ai-research: memory-and-context'
tags: arxiv, long-context, sparse-attention, longformer, transformers
arxiv_id: 2004.05150
alias: 2004-05150-longformer
type: pointer
format: pointer
sources:
  - id: S1
    title: "Longformer: The Long-Document Transformer"
    publisher: arXiv
    authors: "Iz Beltagy, Matthew E. Peters, Arman Cohan"
    url: https://arxiv.org/abs/2004.05150
    version_read: "v2 (2 Dec 2020)"
    accessed: 2026-11-12
    license: "arXiv non-exclusive distribution license"
---
# Longformer: The Long-Document Transformer

## Summary
[Longformer: The Long-Document Transformer](https://arxiv.org/abs/2004.05150) (2020) is an early sparse-attention Transformer built for documents of thousands of tokens. Standard self-attention grows quadratically with length; Longformer's attention is designed to grow linearly instead.

What it introduced: a drop-in attention pattern that combines a local sliding window with a small amount of task-motivated global attention. The authors evaluate character-level language modeling and, unlike much prior long-sequence work, also pretrain and finetune on downstream long-document tasks, reporting gains over RoBERTa on WikiHop and TriviaQA. Version 2 adds the Longformer-Encoder-Decoder (LED) for long sequence-to-sequence work such as arXiv summarization.

What it led to: a standard reference for sparse attention, alongside Big Bird in this batch, and a precursor to later long-context methods that keep some local window plus a global or landmark path. Live notes: [1706-03762-attention-is-all-you-need](https://iinventedlight.com/library/notes/1706-03762-attention-is-all-you-need/) and [1810-04805-bert](https://iinventedlight.com/library/notes/1810-04805-bert/).

Where to read it: https://arxiv.org/abs/2004.05150 (v2, last revised 2 Dec 2020).

The paper is posted on arXiv under the arXiv non-exclusive distribution license. This note links to it rather than reproducing it. The note itself is CC BY 4.0.

License: this note is CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Pointer by iinventedlight (AI-assisted). Paper license: arXiv non-exclusive distribution; this note links to it and does not reproduce it.

---

Agents: [/agents.md](/agents.md) | [/.well-known/agent-library.json](/.well-known/agent-library.json).
