Research map / Architectures and efficiency / Attention and transformers
RESA: Bringing Back What Sparse Attention Ignores with Residual Estimation
Weihao Yang, Hao Huang, Ningke Li, Shihao Wang, Darong Yang, Yanqi Pan, Wen Xia, Shiyi Li, Xiangyu Zou
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 |
|---|---|---|---|
| add RESA residual estimation | Quest · RULER MK3 (Llama-3.1-8B, 8K) | accuracy: 24 -> 40 | Improved |
| add RESA residual estimation | Quest · RULER MK3 (Llama-3.2-3B, 8K) | accuracy: 3 -> 6 | Improved |
| add RESA residual estimation | Quest · RULER MK3 (Mistral-7B, 8K) | accuracy: 9 -> 27 | Improved |
| add RESA residual estimation | Ideal-TopK · RULER MK3 (LWM-Text-7B) | accuracy: 50 -> 70 | Improved |
| add RESA residual estimation | Ideal-TopK · RULER MK3 (Mistral-7B) | accuracy: 99 -> 99 | No clear effect |
| add RESA residual estimation | ArkVale · LongBench PR (Llama-3.1-8B) | score: 89.19 -> 94.84 | Improved |
| add RESA residual estimation | Quest · LongBench 2WMQA (Mistral-7B) | score: 22.41 -> 30.63 | Improved |
| use mean of queries rather than last query as typical query (ablation) | ArkVale · RULER MK3 (Llama3.2-3B) | accuracy: 2 -> 6 | Improved |
| use mean of queries rather than last query as typical query (ablation) | Quest · RULER MK3 (Llama3.2-3B) | accuracy: 3 -> 5 | Improved |
| add RESA at 5% budget | Quest · RULER MK3 (Llama3.1-8B) | accuracy: 50 -> 69 | Improved |
| increase prior weight lambda (ablation) | RESA · RULER summary/comprehension tasks | accuracy | Mixed |
| add RESA to PSA/Ada-KV budgeting | PSA · KV budget at <=2% accuracy loss | KV budget: up to 33.2% budget reduction | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- Quest — QUEST: Query-Aware Sparsity for Efficient Long-Context LLM Inference (ICML 2024) · better on 6 of 6 shared results
- ArkVale — ArkVale: Efficient Generative LLM Inference with Recallable Key-Value Eviction (NeurIPS 2024) · better on 5 of 5 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
- Efficient Low Rank Attention for Long-Context Inference in Large Language Models (NeurIPS 2025 · similarity 0.90)
- More Value per Key: Asymmetric Sparse Attention for Faster LLM Decoding (NeurIPS 2026 · similarity 0.89)
- Unveiling and Harnessing Hidden Attention Sinks: Enhancing Large Language Models without Training through Attention Calibration (ICML 2024 · similarity 0.89)
- FlexPrefill: A Context-Aware Sparse Attention Mechanism for Efficient Long-Sequence Inference (ICLR 2025 · similarity 0.89)
- Virtual Head Attention (NeurIPS 2026 · similarity 0.89)
- MVA: Linear Attention with High-order Query-Keys Integration and Multi-level Vocabulary Decomposition (ICML 2025 · similarity 0.89)
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.
