AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Training-free Context-adaptive Attention for Efficient Long Context Modeling

Zeng You, Yaofo Chen, Shuhai Zhang, Zhijie Qiu, Tingyu Wu, Yingjian Li, Yaowei Wang, Mingkui Tan

TPAMI 2026 · Architectures and efficiency › Attention and transformers · approach: Sparse attention for long-context LLMs · 0 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
use TCA-Attention instead of full attentionLLaMA3.1-8B-Instruct · LongBench-Elatency (ms): 316.14 -> 120.96Improved
use TCA-Attention instead of full attentionLLaMA3.1-8B-Instruct · RULERlatency (ms): 1282.05 -> 453.22Improved
use TCA-Attention instead of XAttentionLLaMA3.1-8B-Instruct · LongBench-Eaverage score: 50.91 -> 53.51Improved
use TCA-Attention instead of MInferenceLLaMA3.1-8B-Instruct · RULER (128K)accuracy: 54.13 -> 77.46Improved
use TCA-Attention instead of full attentionLLaMA3.1-8B-Instruct · short-context GSM8Kaccuracy: 83.85 -> 84.23No clear effect
use TCA-Attention instead of full attentionQwen2.5-7B-Instruct · MT-Bench-101average score: 8.90 -> 8.97Improved
use TCA-Attention instead of full attentionLLaMA3.1-8B-Instruct · OlympiadBenchaverage score: 11.84 -> 12.15No clear effect
use TCA-Attention instead of full attentionLLaMA3.1-8B-Instruct · decoding KV cacheKV cache memory: 8.00GB -> 3.12GBImproved
vary balance parameter alpha from 0.5 to 0.1 (ablation)TCA-Attention · LongBench-Eaverage score: 51.51 -> 51.14Hurt
vary balance parameter alpha from 0.5 to 0.9 (ablation)TCA-Attention · LongBench-Eaverage score: 51.51 -> 51.13Hurt
change calibration dataset to GovReport (ablation)TCA-Attention · LongBench-Eaverage score: 51.54 -> 51.56No clear effect
increase local window size to 4096 (ablation)TCA-Attention · LongBench-Eaverage score: 38.8ms -> 41.5msImproved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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