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
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 changed | Where | Measured | Effect |
|---|---|---|---|
| use TCA-Attention instead of full attention | LLaMA3.1-8B-Instruct · LongBench-E | latency (ms): 316.14 -> 120.96 | Improved |
| use TCA-Attention instead of full attention | LLaMA3.1-8B-Instruct · RULER | latency (ms): 1282.05 -> 453.22 | Improved |
| use TCA-Attention instead of XAttention | LLaMA3.1-8B-Instruct · LongBench-E | average score: 50.91 -> 53.51 | Improved |
| use TCA-Attention instead of MInference | LLaMA3.1-8B-Instruct · RULER (128K) | accuracy: 54.13 -> 77.46 | Improved |
| use TCA-Attention instead of full attention | LLaMA3.1-8B-Instruct · short-context GSM8K | accuracy: 83.85 -> 84.23 | No clear effect |
| use TCA-Attention instead of full attention | Qwen2.5-7B-Instruct · MT-Bench-101 | average score: 8.90 -> 8.97 | Improved |
| use TCA-Attention instead of full attention | LLaMA3.1-8B-Instruct · OlympiadBench | average score: 11.84 -> 12.15 | No clear effect |
| use TCA-Attention instead of full attention | LLaMA3.1-8B-Instruct · decoding KV cache | KV cache memory: 8.00GB -> 3.12GB | Improved |
| vary balance parameter alpha from 0.5 to 0.1 (ablation) | TCA-Attention · LongBench-E | average score: 51.51 -> 51.14 | Hurt |
| vary balance parameter alpha from 0.5 to 0.9 (ablation) | TCA-Attention · LongBench-E | average score: 51.51 -> 51.13 | Hurt |
| change calibration dataset to GovReport (ablation) | TCA-Attention · LongBench-E | average score: 51.54 -> 51.56 | No clear effect |
| increase local window size to 4096 (ablation) | TCA-Attention · LongBench-E | average score: 38.8ms -> 41.5ms | Improved |
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.
- XAttention — XAttention: Block Sparse Attention with Antidiagonal Scoring (ICML 2025) · better on 6 of 6 shared results
- FlexPrefill — FlexPrefill: A Context-Aware Sparse Attention Mechanism for Efficient Long-Sequence Inference (ICLR 2025) · better on 5 of 5 shared results
- DuoAttention — DuoAttention: Efficient Long-Context LLM Inference with Retrieval and Streaming Heads (ICLR 2025) · better on 1 of 1 shared result
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
- Full Attention Strikes Back: Transferring Full Attention into Sparse within Hundred Training Steps (NeurIPS 2026 · similarity 0.93)
- CateKV: On Sequential Consistency for Long-Context LLM Inference Acceleration (ICML 2025 · similarity 0.92)
- A Little Goes a Long Way: Efficient Long Context Training and Inference with Partial Contexts (ICLR 2025 · similarity 0.92)
- FlexPrefill: A Context-Aware Sparse Attention Mechanism for Efficient Long-Sequence Inference (ICLR 2025 · similarity 0.92)
- Model Tells You What to Discard: Adaptive KV Cache Compression for LLMs (ICLR 2024 · similarity 0.92)
- QUEST: Query-Aware Sparsity for Efficient Long-Context LLM Inference (ICML 2024 · similarity 0.92)
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.
