# Budgeted KV-Cache Repair for Evolving Contexts

arXiv: https://arxiv.org/abs/2610.17338 · HTML: https://mlsanity.com/paper/2610.17338/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in KV cache optimization, one of 12 directions in Efficient LLM inference and serving.
- Builds on 22 of its 28 references that are in the map.
- First posted on arXiv 12 Oct 2026.

## Placed in

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

## Builds on (most cited first, 22 of 22)

- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [Qwen3 Technical Report](https://mlsanity.com/paper/2505.09388/) (2025)
- [ReAct: Synergizing Reasoning and Acting in Language Models](https://mlsanity.com/paper/2210.03629/) (2022)
- [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://mlsanity.com/paper/2104.09864/) (2021)
- [HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answering](https://mlsanity.com/paper/1809.09600/) (2018)
- [FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness](https://mlsanity.com/paper/2205.14135/) (2022)
- [MemGPT: Towards LLMs as Operating Systems](https://mlsanity.com/paper/2310.08560/) (2023)
- [Constructing A Multi-hop QA Dataset for Comprehensive Evaluation of Reasoning Steps](https://mlsanity.com/paper/2011.01060/) (2020)
- [MuSiQue: Multihop Questions via Single-hop Question Composition](https://mlsanity.com/paper/2108.00573/) (2021)
- [SnapKV: LLM Knows What You are Looking for Before Generation](https://mlsanity.com/paper/2404.14469/) (2024)
- [GLM-4.5: Agentic, Reasoning, and Coding (ARC) Foundation Models](https://mlsanity.com/paper/2508.06471/) (2025)
- [Prompt Cache: Modular Attention Reuse for Low-Latency Inference](https://mlsanity.com/paper/2311.04934/) (2023)
- [CacheBlend: Fast Large Language Model Serving for RAG with Cached Knowledge Fusion](https://mlsanity.com/paper/2405.16444/) (2024)
- [IIRC: A Dataset of Incomplete Information Reading Comprehension Questions](https://mlsanity.com/paper/2011.07127/) (2020)
- [MEMENTO: Teaching LLMs to Manage Their Own Context](https://mlsanity.com/paper/2604.09852/) (2026)
- [Let the Code LLM Edit Itself When You Edit the Code](https://mlsanity.com/paper/2407.03157/) (2024)
- [A3: Attention-Aware Accurate KV Cache Fusion for Fast Large Language Model Serving](https://mlsanity.com/paper/2511.17560/) (2025)
- [KVEraser: Learning to Steer KV Cache for Efficient Localized Context Erasing](https://mlsanity.com/paper/2606.17034/) (2026)
- [Models Take Notes at Prefill: KV Cache Can Be Editable and Composable](https://mlsanity.com/paper/2606.17107/) (2026)
- [InfoFlow KV: Information-Flow-Aware KV Recomputation for Long Context](https://mlsanity.com/paper/2603.05353/) (2026)
- [Adaptive KV Cache Reuse for Fast Long-Context LLM Serving](https://mlsanity.com/paper/2605.24022/) (2026)
- [QCFuse: Query-Aware Cache Fusion via Compressed View for Efficient RAG Serving](https://mlsanity.com/paper/2606.05875/) (2026)

## Abstract (arXiv)

Language-model agents revise their working context as information becomes obsolete, but deleting text does not erase its influence on cached computation. Later key-value (KV) states remain conditioned on the deleted content, while recomputing the entire retained suffix can be costly. We introduce budgeted KV-cache repair: recovering task performance by recomputing a user-chosen fraction of the retained suffix. Comparing repair allocations under identical relevance scores and token budgets, we find that refreshing coherent regions can substantially outperform refreshing isolated tokens. This finding motivates Semantic-Unit Refresh (SUR), which allocates the budget to query-relevant text regions using inexpensive embedding scores, without training or an auxiliary transformer forward. To execute these repairs efficiently, a layer-packed executor processes selected states together at each layer while preserving full causal cache access. Across five models, SUR achieves strong recovery in controlled deletion tasks while recomputing only 2-5% of the suffix. On four natural multi-hop QA datasets, it achieves near-full-recompute exact match at a 5% budget with lower post-edit latency. The executor accelerates fragmented refresh by up to 16.1x over interval-wise execution.