---
schema_version: '1.0'
id: news-20260817-9f2f80
title: Efficient Streaming Language Models with Attention Sinks
url: https://arxiv.org/abs/2309.17453
url_hash: 9f2f8008d0d4fcaf8d85fd3eedead6fd60c1e6e6599eeb22e6f37dedfa3156b9
canonical_url: https://arxiv.org/abs/2309.17453
source: arxiv.org
category: news/tech
category_raw: it_ai
region: null
tags: []
lang: null
published_at: null
fetched_at: '2026-08-17T00:38:47.986086Z'
updated_at: '2026-08-17T00:39:14Z'
status: published
content_hash: null
content_changed_at: null
license_note: link-only
summary: null
summary_source: null
summary_en: null
summary_omitted_reason: license_restricted
redistribution: false
keywords_en:
- artificial intelligence
- Attention
- Efficient
- Efficient Transformers
- Language
- LanguageModel API
- Models
- promptStreaming
- Sinks
- Streaming
- technology
- with
- Work with Pride
entities_en:
- name: Attention
  type: concept
- name: Efficient Transformers
  type: method
- name: LanguageModel API
  type: organization
- name: Models
  type: artifact
- name: promptStreaming
  type: method
- name: Work with Pride
  type: organization
topics_en:
- artificial intelligence
- technology
numeric_facts: []
entities:
- name: LanguageModel API
  type: organization
- name: Models
  type: artifact
- name: Work with Pride
  type: organization
- name: Attention
  type: concept
key_facts: []
related: []
related_auto:
- name: HackerNews
  type: organization
  weight: 2.0
computed_at: null
sanitized_at: '2026-08-20T14:56:50.092484+00:00'
---

# Efficient Streaming Language Models with Attention Sinks

## TL;DR
（要約なし・原文はリンク先を参照 / No summary available; see the source link.）

## Key Points
- (なし)

## Details
(本文なし。リンク先参照)

## Source
元記事: [Efficient Streaming Language Models with Attention Sinks](https://arxiv.org/abs/2309.17453)