KVPress Benchmark (Qwen3-8B (LLM), RULER-4K, Compression 88%: 94.6 Acc.)</p>\n","updatedAt":"2026-08-05T13:58:48.355Z","author":{"_id":"66c2f277a2213740f48ff641","avatarUrl":"/avatars/118cd48dc1692db21f281b9a9f13d539.svg","fullname":"changwoo baek","name":"higokri","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.33810800313949585},"editors":["higokri"],"editorAvatarUrls":["/avatars/118cd48dc1692db21f281b9a9f13d539.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2608.01247","authors":[{"_id":"6a71790eec5082b9f872cf0c","user":{"_id":"66c2f277a2213740f48ff641","avatarUrl":"/avatars/118cd48dc1692db21f281b9a9f13d539.svg","isPro":false,"fullname":"changwoo baek","user":"higokri","type":"user","name":"higokri"},"name":"Changwoo Baek","status":"claimed_verified","statusLastChangedAt":"2026-08-04T08:45:04.725Z","hidden":false},{"_id":"6a71790eec5082b9f872cf0d","name":"Seungjun Shin","hidden":false},{"_id":"6a71790eec5082b9f872cf0e","name":"Kyeongbo Kong","hidden":false}],"publishedAt":"2026-08-02T00:00:00.000Z","submittedOnDailyAt":"2026-08-05T00:00:00.000Z","title":"RestoreKV: Recovering Full-Cache Behavior Under Aggressive Query-Agnostic KV Cache Eviction","submittedOnDailyBy":{"_id":"66c2f277a2213740f48ff641","avatarUrl":"/avatars/118cd48dc1692db21f281b9a9f13d539.svg","isPro":false,"fullname":"changwoo baek","user":"higokri","type":"user","name":"higokri"},"summary":"Query-agnostic KV cache eviction compresses a context once and reuses the resulting cache for arbitrary future queries, but performance can collapse under tight budgets. Existing methods primarily improve which original KV pairs are retained. We introduce RestoreKV, which complements this selection-based formulation with learned restoration under the same total KV budget. Our key insight is that, although the information lost through eviction is context-specific, the mechanism for generating its compact complement can be shared across contexts. After context prefill, a few restore tokens attend to the full KV cache in a single LoRA-adapted pass, generating a compact, context-conditioned restore cache. The base importance scorer and eviction rule remain unchanged, and the adapters are disabled for all subsequent queries and decoding. RestoreKV is trained through parameter-efficient self-distillation from the frozen full-cache model, optimizing only 0.4% of the parameters and requiring no task-specific tuning. Across four backbones and four long-context benchmarks, RestoreKV substantially reduces compression-induced degradation. On Qwen3-4B, it improves 59 of 60 paired, budget-matched settings across five base eviction methods; at a 5% budget, it raises KVzip from 38.2 to 73.2 on RULER-4K. Applied to KVzip+, RestoreKV reaches 86.4 RULER accuracy at 16times compression on the KVPress Benchmark, while adding less than 0.5% one-time cache-construction overhead in a 32K-context evaluation. Our project page is available at https://paper.pnu-cvsp.com/RestoreKV/","upvotes":5,"discussionId":"6a71790fec5082b9f872cf0f","projectPage":"https://paper.pnu-cvsp.com/RestoreKV/"},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"696a2e2cb4cb5e359740af1d","avatarUrl":"/avatars/d40e6882a95ab92e99cfe29b98e07baf.svg","isPro":false,"fullname":"Beomkwon Cho","user":"chobk5194","type":"user"},{"_id":"66c2f277a2213740f48ff641","avatarUrl":"/avatars/118cd48dc1692db21f281b9a9f13d539.svg","isPro":false,"fullname":"changwoo baek","user":"higokri","type":"user"},{"_id":"6974eca019d789b7f50400a4","avatarUrl":"/avatars/c98f7f2e3077f9b46b31613754126eaf.svg","isPro":false,"fullname":"Gatsby Abessolo","user":"gatsby13579","type":"user"},{"_id":"63ac5701c21e60a3e9b58aa7","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/63ac5701c21e60a3e9b58aa7/g6EX7diOpuA94R2ab-rZC.png","isPro":true,"fullname":"Dipankar Sarkar","user":"dipankarsarkar","type":"user"},{"_id":"661ab1f1fa3b144a381fa454","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/661ab1f1fa3b144a381fa454/IlpZBb9NCjo7ntFwMIH53.png","isPro":false,"fullname":"Urro","user":"urroxyz","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2608/2608.01247.md","query":{}}">
RestoreKV: Recovering Full-Cache Behavior Under Aggressive Query-Agnostic KV Cache Eviction
Abstract
Query-agnostic KV cache eviction compresses a context once and reuses the resulting cache for arbitrary future queries, but performance can collapse under tight budgets. Existing methods primarily improve which original KV pairs are retained. We introduce RestoreKV, which complements this selection-based formulation with learned restoration under the same total KV budget. Our key insight is that, although the information lost through eviction is context-specific, the mechanism for generating its compact complement can be shared across contexts. After context prefill, a few restore tokens attend to the full KV cache in a single LoRA-adapted pass, generating a compact, context-conditioned restore cache. The base importance scorer and eviction rule remain unchanged, and the adapters are disabled for all subsequent queries and decoding. RestoreKV is trained through parameter-efficient self-distillation from the frozen full-cache model, optimizing only 0.4% of the parameters and requiring no task-specific tuning. Across four backbones and four long-context benchmarks, RestoreKV substantially reduces compression-induced degradation. On Qwen3-4B, it improves 59 of 60 paired, budget-matched settings across five base eviction methods; at a 5% budget, it raises KVzip from 38.2 to 73.2 on RULER-4K. Applied to KVzip+, RestoreKV reaches 86.4 RULER accuracy at 16times compression on the KVPress Benchmark, while adding less than 0.5% one-time cache-construction overhead in a 32K-context evaluation. Our project page is available at https://paper.pnu-cvsp.com/RestoreKV/
Community
KVPress Benchmark (Qwen3-8B (LLM), RULER-4K, Compression 88%: 94.6 Acc.)
Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.
Tap or paste here to upload images
Cite arxiv.org/abs/2608.01247 in a dataset README.md to link it from this page.
Cite arxiv.org/abs/2608.01247 in a Space README.md to link it from this page.
Discussion (0)
Sign in to join the discussion. Free account, 30 seconds — email code or GitHub.
Sign in →No comments yet. Sign in and be the first to say something.