Research map / Architectures and efficiency / Attention and transformers
ParisKV: Fast and Drift-Robust KV-Cache Retrieval for Long-Context LLMs
Yanlin Qi, Xinhang Chen, Huiqiang Jiang, Qitong Wang, Botao Peng, Themis Palpanas
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 normalization, rotation and theoretical spherical centroids (ablation) | ParisKV coarse stage · AIME long decoding | coarse Recall@100 (%): 6 -> 16.1 | Improved |
| add normalization, rotation and theoretical spherical centroids (ablation) | ParisKV · AIME long decoding | final Recall@100 after reranking (%): 36.5 -> 64.3 | Improved |
| use ParisKV instead of PQCache | Qwen3-4B · GPQA-Diamond | accuracy: 38.38 -> 72.22 | Improved |
| use ParisKV instead of PQCache | Qwen3-4B · MATH500 | accuracy: 58.80 -> 92.80 | Improved |
| use ParisKV instead of PQCache | Qwen3-4B · AIME25 | pass@8: 3.33 -> 80.00 | Improved |
| use ParisKV instead of full attention | Qwen3-8B · MATH500 | accuracy: 87.40 -> 93.00 | Improved |
| use ParisKV instead of full attention | Qwen3-8B · AIME25 | pass@8: 83.33 -> 73.33 | Hurt |
| use ParisKV instead of full attention | Qwen3-4B · AIME25 | pass@8: 86.67 -> 80.00 | Hurt |
| use ParisKV instead of PQCache and MagicPIG | DeepSeek-R1-Llama-8B · LongBench-V2 overall | score: 19.90 -> 28.43 | Improved |
| use ParisKV instead of ShadowKV | Sparse retrieval · RULER 128K | average score: 80.29 -> 83.49 | Improved |
| use ParisKV instead of full attention for decoding | Llama3.1-8B and Qwen3-8B · 64K-256K decoding throughput | peak throughput: 2.1x to 2.8x | Improved |
| use ParisKV instead of PQCache | Llama3.1-8B · 1024K decode latency | ms/step: 2179 -> 49 | 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.
- MagicPIG — MagicPIG: LSH Sampling for Efficient LLM Generation (ICLR 2025) · better on 10 of 10 shared results
- ShadowKV — ShadowKV: KV Cache in Shadows for High-Throughput Long-Context LLM Inference (ICML 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
- RocketKV: Accelerating Long-Context LLM Inference via Two-Stage KV Cache Compression (ICML 2025 · similarity 0.93)
- ShadowKV: KV Cache in Shadows for High-Throughput Long-Context LLM Inference (ICML 2025 · similarity 0.92)
- SCBench: A KV Cache-Centric Analysis of Long-Context Methods (ICLR 2025 · similarity 0.92)
- SmallKV: Small Model Assisted Compensation of KV Cache Compression for Efficient LLM Inference (NeurIPS 2025 · similarity 0.91)
- ChunkKV: Semantic-Preserving KV Cache Compression for Efficient Long-Context LLM Inference (NeurIPS 2025 · similarity 0.91)
- QUEST: Query-Aware Sparsity for Efficient Long-Context LLM Inference (ICML 2024 · similarity 0.91)
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.
