{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:MN2EQJXQG54WQ4L7IWMEGMWGWK","short_pith_number":"pith:MN2EQJXQ","schema_version":"1.0","canonical_sha256":"63744826f0377968717f45984332c6b298819ba5d74ad8597e990b25ff4f2a92","source":{"kind":"arxiv","id":"2502.03805","version":2},"attestation_state":"computed","paper":{"title":"CriticalKV: Optimizing KV Cache Eviction from an Output Perturbation Perspective","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Haoyu Guo, Junlin Lv, S Kevin Zhou, Xike Xie, Yuan Feng, Yukun Cao","submitted_at":"2025-02-06T06:31:47Z","abstract_excerpt":"Large language models have revolutionized natural language processing but face significant challenges of high storage and runtime costs, due to the transformer architecture's reliance on self-attention, particularly the large KV cache for long-sequence inference. Recent efforts to reduce KV cache size by pruning less critical entries based on attention weights remain empirical and lack formal grounding. This paper presents a formal study on identifying critical KV cache entries by analyzing attention output perturbation. Our analysis reveals that, beyond attention weights, the value states wit"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2502.03805","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-02-06T06:31:47Z","cross_cats_sorted":[],"title_canon_sha256":"fe576bcc2b56dad45d962889287c4e0eede3801a3d1abb489fc3b709145c35c2","abstract_canon_sha256":"7ca7d3bb9be2858a7ec7e49d0a648e7ce60ce8377b0dc377188c8a3396c2b64f"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-29T01:04:51.892844Z","signature_b64":"dcXGTLnWEBhMZdes4FU24bz6f1DYq8wX3hsvnG3XHLo9LKuNllfJ9OLQ9aX2l/eEb9VN/cImku0clbqXXeB3AQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"63744826f0377968717f45984332c6b298819ba5d74ad8597e990b25ff4f2a92","last_reissued_at":"2026-05-29T01:04:51.892373Z","signature_status":"signed_v1","first_computed_at":"2026-05-29T01:04:51.892373Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"CriticalKV: Optimizing KV Cache Eviction from an Output Perturbation Perspective","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Haoyu Guo, Junlin Lv, S Kevin Zhou, Xike Xie, Yuan Feng, Yukun Cao","submitted_at":"2025-02-06T06:31:47Z","abstract_excerpt":"Large language models have revolutionized natural language processing but face significant challenges of high storage and runtime costs, due to the transformer architecture's reliance on self-attention, particularly the large KV cache for long-sequence inference. Recent efforts to reduce KV cache size by pruning less critical entries based on attention weights remain empirical and lack formal grounding. This paper presents a formal study on identifying critical KV cache entries by analyzing attention output perturbation. Our analysis reveals that, beyond attention weights, the value states wit"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.03805","kind":"arxiv","version":2},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2502.03805/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2502.03805","created_at":"2026-05-29T01:04:51.892433+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.03805v2","created_at":"2026-05-29T01:04:51.892433+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.03805","created_at":"2026-05-29T01:04:51.892433+00:00"},{"alias_kind":"pith_short_12","alias_value":"MN2EQJXQG54W","created_at":"2026-05-29T01:04:51.892433+00:00"},{"alias_kind":"pith_short_16","alias_value":"MN2EQJXQG54WQ4L7","created_at":"2026-05-29T01:04:51.892433+00:00"},{"alias_kind":"pith_short_8","alias_value":"MN2EQJXQ","created_at":"2026-05-29T01:04:51.892433+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":6,"internal_anchor_count":6,"sample":[{"citing_arxiv_id":"2605.20600","citing_title":"Head-Aware Key-Value Compression for Efficient Autoregressive Image Generation","ref_index":9,"is_internal_anchor":true},{"citing_arxiv_id":"2407.11550","citing_title":"Ada-KV: Optimizing KV Cache Eviction by Adaptive Budget Allocation for Efficient LLM Inference","ref_index":39,"is_internal_anchor":true},{"citing_arxiv_id":"2605.08317","citing_title":"RDKV: Rate-Distortion Bit Allocation for Joint Eviction and Quantization of the KV Cache","ref_index":30,"is_internal_anchor":true},{"citing_arxiv_id":"2605.08234","citing_title":"When Does Value-Aware KV Eviction Help? A Fixed-Contract Diagnostic for Non-Monotone Cache Compression","ref_index":18,"is_internal_anchor":true},{"citing_arxiv_id":"2604.25975","citing_title":"Rethinking KV Cache Eviction via a Unified Information-Theoretic Objective","ref_index":8,"is_internal_anchor":true},{"citing_arxiv_id":"2605.07234","citing_title":"Reformulating KV Cache Eviction Problem for Long-Context LLM Inference","ref_index":12,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/MN2EQJXQG54WQ4L7IWMEGMWGWK","json":"https://pith.science/pith/MN2EQJXQG54WQ4L7IWMEGMWGWK.json","graph_json":"https://pith.science/api/pith-number/MN2EQJXQG54WQ4L7IWMEGMWGWK/graph.json","events_json":"https://pith.science/api/pith-number/MN2EQJXQG54WQ4L7IWMEGMWGWK/events.json","paper":"https://pith.science/paper/MN2EQJXQ"},"agent_actions":{"view_html":"https://pith.science/pith/MN2EQJXQG54WQ4L7IWMEGMWGWK","download_json":"https://pith.science/pith/MN2EQJXQG54WQ4L7IWMEGMWGWK.json","view_paper":"https://pith.science/paper/MN2EQJXQ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.03805&json=true","fetch_graph":"https://pith.science/api/pith-number/MN2EQJXQG54WQ4L7IWMEGMWGWK/graph.json","fetch_events":"https://pith.science/api/pith-number/MN2EQJXQG54WQ4L7IWMEGMWGWK/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/MN2EQJXQG54WQ4L7IWMEGMWGWK/action/timestamp_anchor","attest_storage":"https://pith.science/pith/MN2EQJXQG54WQ4L7IWMEGMWGWK/action/storage_attestation","attest_author":"https://pith.science/pith/MN2EQJXQG54WQ4L7IWMEGMWGWK/action/author_attestation","sign_citation":"https://pith.science/pith/MN2EQJXQG54WQ4L7IWMEGMWGWK/action/citation_signature","submit_replication":"https://pith.science/pith/MN2EQJXQG54WQ4L7IWMEGMWGWK/action/replication_record"}},"created_at":"2026-05-29T01:04:51.892433+00:00","updated_at":"2026-05-29T01:04:51.892433+00:00"}