Research map / Architectures and efficiency / Attention and transformers
Retrospective Sparse Attention for Efficient Long-Context Generation
Seonghwan Choi, Beomseok Kang, Dongwon Jo, Jae-Joon Kim
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 retrospective attention updates | Quest on Llama-3.1-8B · LongGenBench GSM8K mean | accuracy: 52.6 -> 56.5 | Improved |
| add retrospective attention updates | Quest on Llama-3.1-8B · LongGenBench MMLU mean | accuracy: 54.8 -> 55.3 | Improved |
| add retrospective attention updates | Quest on Llama-3.1-8B · LongGenBench CSQA mean | accuracy: 56.8 -> 60.3 | Improved |
| add retrospective attention updates | Quest on Llama-3.1-8B · LongGenBench CSQA n=15 | accuracy: 70.9 -> 68.8 | Hurt |
| add retrospective attention updates | Quest on Llama-3.1-8B · LongGenBench CSQA n=45 | accuracy: 46.1 -> 53.0 | Improved |
| add retrospective attention updates | Quest on Qwen2.5-14B-Instruct · LongGenBench CSQA mean | accuracy: 58.1 -> 65.5 | Improved |
| add retrospective attention updates | Quest on Qwen2.5-32B-Instruct · LongGenBench GSM8K mean | accuracy: 83.8 -> 87.8 | Improved |
| add retrospective attention updates | Quest on Qwen2.5-32B-Instruct · LongGenBench MMLU mean | accuracy: 80.0 -> 80.4 | No clear effect |
| replace top-k reuse (f=2) with retrospective update (ablation) | Quest on Llama-3.1-8B · LongGenBench GSM8K mean | accuracy: 25.2 -> 56.5 | Improved |
| add top-k selection reuse with refresh every 2 steps (ablation) | Quest on Llama-3.1-8B · LongGenBench GSM8K mean | accuracy: 52.6 -> 25.2 | Hurt |
| use permanent eviction (StreamingLLM) instead of Quest | Llama-3.1-8B · LongGenBench GSM8K mean | accuracy: 52.6 -> 0.0 | Hurt |
| increase retrospective window w from 0 to 8 (ablation) | RetroAttention · Throughput 32k batch 1 | batch/sec: 48.7 -> 46.2 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Quest — QUEST: Query-Aware Sparsity for Efficient Long-Context LLM Inference (ICML 2024) · better on 9 of 9 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
- DuoAttention: Efficient Long-Context LLM Inference with Retrieval and Streaming Heads (ICLR 2025 · similarity 0.91)
- ArkVale: Efficient Generative LLM Inference with Recallable Key-Value Eviction (NeurIPS 2024 · similarity 0.91)
- Model Tells You What to Discard: Adaptive KV Cache Compression for LLMs (ICLR 2024 · similarity 0.91)
- CateKV: On Sequential Consistency for Long-Context LLM Inference Acceleration (ICML 2025 · similarity 0.90)
- RazorAttention: Efficient KV Cache Compression Through Retrieval Heads (ICLR 2025 · similarity 0.90)
- SCBench: A KV Cache-Centric Analysis of Long-Context Methods (ICLR 2025 · similarity 0.90)
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.
