Research map / Architectures and efficiency / Attention and transformers
CriticalKV: Optimizing KV Cache Eviction from an Output Perturbation Perspective
Yuan Feng, Junlin Lv, Haoyu Guo, Yukun Cao, S Kevin Zhou, Xike Xie
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 CriticalKV selection to SnapKV | SnapKV Qwen2.5-32B · RULER (40% cache) | average score: 63.86 -> 81.09 | Improved |
| add CriticalKV selection to AdaKV | AdaKV Qwen2.5-32B · RULER (40% cache) | average score: 71.09 -> 83.87 | Improved |
| add CriticalKV selection to HeadKV | HeadKV Qwen2.5-32B · RULER (40% cache) | average score: 81.04 -> 90.69 | Improved |
| add CriticalKV selection to AdaKV | AdaKV Mistral-7B · RULER (40% cache) | average score: 34.88 -> 69.17 | Improved |
| add CriticalKV selection to AdaKV | AdaKV Llama-3.1-8B · RULER (20% cache) | average score: 57.48 -> 68.94 | Improved |
| add CriticalKV selection to AdaKV at 40% cache | AdaKV Llama-3.1-8B · LongBench | average loss vs full cache: 6.0% -> 2.4% | Improved |
| add CriticalKV selection to AdaKV at 40% cache | AdaKV Llama-3.1-8B · LongBench Multi-Doc QA | score: 41.36 -> 45.11 | Improved |
| add CriticalKV selection to AdaKV at 40% cache | AdaKV Llama-3.1-8B · SCBench EN.QA | score: 15.71 -> 22.14 | Improved |
| add CriticalKV selection to AdaKV at 20% cache | AdaKV · LongBench code domain | score | Mixed |
| add stage-1 safeguard (alpha=0.5 instead of alpha=0) (ablation) | AdaKV w/ CriticalKV Mistral-7B · LongBench (20% cache) | average score: 31.94 -> 42.85 | Improved |
| add stage-1 safeguard (alpha=0.5 instead of alpha=0) (ablation) | AdaKV w/ CriticalKV Mistral-7B · LongBench (10% cache) | average score: 35.94 -> 37.94 | Improved |
| add stage-1 safeguard alpha (0.5 vs 0) (ablation) | AdaKV w/ CriticalKV Llama-3.1-8B · LongBench (20% cache) | average score: 44.35 -> 43.88 | No clear effect |
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.
- 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
- Ada-KV: Optimizing KV Cache Eviction by Adaptive Budget Allocation for Efficient LLM Inference (NeurIPS 2025 · similarity 0.93)
- CAKE: Cascading and Adaptive KV Cache Eviction with Layer Preferences (ICLR 2025 · similarity 0.92)
- SqueezeAttention: 2D Management of KV-Cache in LLM Inference via Layer-wise Optimal Budget (ICLR 2025 · similarity 0.92)
- Taming the Fragility of KV Cache Eviction in LLM Inference (ICLR 2026 · similarity 0.92)
- DropKV: Decoupling Residual-Output Perturbation for Near-Optimal KV-Cache Eviction (NeurIPS 2026 · similarity 0.92)
- ThinK: Thinner Key Cache by Query-Driven Pruning (ICLR 2025 · 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.
