# LLMs Know What to Drop: Self-Attention Guided KV Cache Eviction for Efficient Long-Context Inference

arXiv: https://arxiv.org/abs/2503.08879 · HTML: https://mlsanity.com/paper/2503.08879/

## Key facts

- Cited by 20 papers in the map (data through 14 Oct 2026).
- Placed in KV cache optimization, one of 12 directions in Efficient LLM inference and serving.
- Builds on 19 of its 20 references that are in the map.
- First posted on arXiv 11 Mar 2025.

## Placed in

- Efficient LLM inference and serving / [KV cache optimization](https://mlsanity.com/inference/kv-cache-optimization/)

## Builds on (most cited first, 19 of 19)

- [GPT-4 Technical Report](https://mlsanity.com/paper/2303.08774/) (2023)
- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [Gemini 1.5: Unlocking multimodal understanding across millions of tokens of context](https://mlsanity.com/paper/2403.05530/) (2024)
- [Efficient Streaming Language Models with Attention Sinks](https://mlsanity.com/paper/2309.17453/) (2023)
- [LongBench: A Bilingual, Multitask Benchmark for Long Context Understanding](https://mlsanity.com/paper/2308.14508/) (2023)
- [Qwen2.5-Coder Technical Report](https://mlsanity.com/paper/2409.12186/) (2024)
- [H2O: Heavy-Hitter Oracle for Efficient Generative Inference of Large Language Models](https://mlsanity.com/paper/2306.14048/) (2023)
- [Quest: Query-Aware Sparsity for Efficient Long-Context LLM Inference](https://mlsanity.com/paper/2406.10774/) (2024)
- [NV-Embed: Improved Techniques for Training LLMs as Generalist Embedding Models](https://mlsanity.com/paper/2405.17428/) (2024)
- [LLM2Vec: Large Language Models Are Secretly Powerful Text Encoders](https://mlsanity.com/paper/2404.05961/) (2024)
- [DuoAttention: Efficient Long-Context LLM Inference with Retrieval and Streaming Heads](https://mlsanity.com/paper/2410.10819/) (2024)
- [InfiniGen: Efficient Generative Inference of Large Language Models with Dynamic KV Cache Management](https://mlsanity.com/paper/2406.19707/) (2024)
- [A Survey on Efficient Inference for Large Language Models](https://mlsanity.com/paper/2404.14294/) (2024)
- [InfLLM: Training-Free Long-Context Extrapolation for LLMs with an Efficient Context Memory](https://mlsanity.com/paper/2402.04617/) (2024)
- [How to Train Long-Context Language Models (Effectively)](https://mlsanity.com/paper/2410.02660/) (2024)
- [ShadowKV: KV Cache in Shadows for High-Throughput Long-Context LLM Inference](https://mlsanity.com/paper/2410.21465/) (2024)
- [A Survey on Large Language Model Acceleration based on KV Cache Management](https://mlsanity.com/paper/2412.19442/) (2024)
- [RetrievalAttention: Accelerating Long-Context LLM Inference via Vector Retrieval](https://mlsanity.com/paper/2409.10516/) (2024)
- [FastDecode: High-Throughput GPU-Efficient LLM Serving using Heterogeneous Pipelines](https://mlsanity.com/paper/2403.11421/) (2024)

## Built on by (most cited first, 20 of 20)

- [Cache What Lasts: Token Retention for Memory-Bounded KV Cache in LLMs](https://mlsanity.com/paper/2512.03324/) (2025)
- [SlimInfer: Accelerating Long-Context LLM Inference via Dynamic Token Pruning](https://mlsanity.com/paper/2508.06447/) (2025)
- [Attention Sink in Transformers: A Survey on Utilization, Interpretation, and Mitigation](https://mlsanity.com/paper/2604.10098/) (2026)
- [Make Each Token Count: Towards Improving Long-Context Performance with KV Cache Eviction](https://mlsanity.com/paper/2605.09649/) (2026)
- [Cache Me If You Can: How Many KVs Do You Need for Effective Long-Context LMs?](https://mlsanity.com/paper/2506.17121/) (2025)
- [LKV: End-to-End Learning of Head-wise Budgets and Token Selection for LLM KV Cache Eviction](https://mlsanity.com/paper/2605.06676/) (2026)
- [SnapStream: Efficient Long Sequence Decoding on Dataflow Accelerators](https://mlsanity.com/paper/2511.03092/) (2025)
- [Cross-Family Speculative Prefill: Training-Free Long-Context Compression with Small Draft Models](https://mlsanity.com/paper/2603.02631/) (2026)
- [SPHERICAL KV: Angle-Domain Attention and Rate-Distortion Retention for Efficient Long-Context Inference](https://mlsanity.com/paper/2605.18856/) (2026)
- [Prefix Sliding for efficient test-time scaling](https://mlsanity.com/paper/2608.26070/) (2026)
- [S3-Attention:Attention-Aligned Endogenous Retrieval for Memory-Bounded Long-Context Inference](https://mlsanity.com/paper/2601.17702/) (2026)
- [More Than a Quick Glance: Overcoming the Greedy Bias in KV-Cache Compression](https://mlsanity.com/paper/2602.02199/) (2026)
- [UT-ACA: Uncertainty-Triggered Adaptive Context Allocation for Long-Context Inference](https://mlsanity.com/paper/2603.18446/) (2026)
- [Adaptive Filtering of the KV Cache: Diagnosing and Correcting Structural-Role Bias in LLM Inference](https://mlsanity.com/paper/2607.13205/) (2026)
- [Monte Carlo Estimation for KV Cache Eviction](https://mlsanity.com/paper/2610.07643/) (2026)
- [PatternKV: Flattening KV Representation Expands Quantization Headroom](https://mlsanity.com/paper/2510.05176/) (2025)
- [APCE: Adaptive Progressive Context Expansion for Long Context Processing](https://mlsanity.com/paper/2510.12051/) (2025)
- [KV Cache Recycling to Expand Usable Context Capacity in Low Parameter LLMs](https://mlsanity.com/paper/2512.11851/) (2025)
- [Benchmarking KV-Cache Optimizations across Task Quality and System Performance for Long-Context Serving](https://mlsanity.com/paper/2607.05399/) (2026)
- [Learning how to Forget: Fine-tuning for Long-Context Sparse Attention](https://mlsanity.com/paper/2608.19920/) (2026)

## Abstract (arXiv)

Efficient long-context inference is critical as large language models (LLMs) adopt context windows of ranging from 128K to 1M tokens. However, the growing key-value (KV) cache and the high computational complexity of attention create significant bottlenecks in memory usage and latency. In this paper, we find that attention in diverse long-context tasks exhibits sparsity, and LLMs implicitly "know" which tokens can be dropped or evicted at the head level after the pre-filling stage. Based on this insight, we propose Self-Attention Guided Eviction~(SAGE-KV), a simple and effective KV eviction cache method for long-context inference. After prefilling, our method performs a one-time top-k selection at both the token and head levels to compress the KV cache, enabling efficient inference with the reduced cache. Evaluations on LongBench and three long-context LLMs (Llama3.1-8B-Instruct-128k, Llama3-8B-Prolong-512k-Instruct, and Qwen2.5-7B-Instruct-128k) show that SAGE-KV maintains accuracy comparable to full attention while significantly improving efficiency. Specifically, SAGE-KV achieves 4x higher memory efficiency with improved accuracy over the static KV cache selection method StreamLLM, and 2x higher memory efficiency with better accuracy than the dynamic KV cache selection method Quest.