Research map / Architectures and efficiency / Attention and transformers
LouisKV: Efficient KV Cache Retrieval for Long Input-Output Sequences
Wenbo Wu, Qingyi Si, Xiurui Pan, Ye Wang, Jie Zhang
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 |
|---|---|---|---|
| retrieve at adaptive semantic boundaries instead of every decoding step (system optimization SR) (ablation) | LouisKV · Qwen3-8B, 32K+16K | speedup: nearly 2.6x speedup | Improved |
| add group-consistent selection (ablation) | LouisKV · Qwen3-8B, 32K+16K | speedup: +13.1% | Improved |
| add custom retrieval CUDA kernel (ablation) | LouisKV · Qwen3-8B, 32K+16K | speedup: +15.7% | Improved |
| use adaptive-stride retrieval instead of fixed-stride retrieval (ablation) | LouisKV · LongBench and long-output reasoning tasks | accuracy | Improved |
| use fine-grained clusters/segments instead of page-level management (ablation) | LouisKV · LongBench and long-output reasoning tasks | accuracy | Improved |
| raise similarity threshold tau (more frequent retrieval) (ablation) | LouisKV · long-input and long-output tasks | accuracy and latency | Mixed |
| use LouisKV instead of Arkvale at budget 512 on long-input tasks | Llama-3.1-8B-Instruct · LongBench | average accuracy: +1.1% over Arkvale, +3.3% over Quest | Improved |
| use LouisKV instead of Arkvale at budget 1024 on long-output tasks | Qwen3-8B · long-output reasoning | average accuracy: +2.0% over Arkvale, +4.8% over Quest | Improved |
| use KV dropping (H2O, RaaS) instead of KV retrieval | Llama-3.1-8B-Instruct · LongBench budget 512 | accuracy: Qasper 44.38 -> 24.50 (H2O) | Hurt |
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.
- Arkvale — ArkVale: Efficient Generative LLM Inference with Recallable Key-Value Eviction (NeurIPS 2024) · better on 5 of 5 shared results
- Quest — QUEST: Query-Aware Sparsity for Efficient Long-Context LLM Inference (ICML 2024) · better on 4 of 4 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
- EchoKV: Efficient KV Cache Compression via Similarity-Based Reconstruction (NeurIPS 2026 · similarity 0.91)
- RazorAttention: Efficient KV Cache Compression Through Retrieval Heads (ICLR 2025 · similarity 0.90)
- SpeCache: Speculative Key-Value Caching for Efficient Generation of LLMs (ICML 2025 · similarity 0.90)
- SCBench: A KV Cache-Centric Analysis of Long-Context Methods (ICLR 2025 · similarity 0.89)
- RocketKV: Accelerating Long-Context LLM Inference via Two-Stage KV Cache Compression (ICML 2025 · similarity 0.89)
- SqueezeAttention: 2D Management of KV-Cache in LLM Inference via Layer-wise Optimal Budget (ICLR 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.
