{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:Y3EJWHYLPZOMKGUI3KGYWM2UXT","short_pith_number":"pith:Y3EJWHYL","schema_version":"1.0","canonical_sha256":"c6c89b1f0b7e5cc51a88da8d8b3354bce13f01edc4b070782a73b31617264f05","source":{"kind":"arxiv","id":"2412.09036","version":1},"attestation_state":"computed","paper":{"title":"ZigZagkv: Dynamic KV Cache Compression for Long-context Modeling based on Layer Uncertainty","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Chen Zhang, Kehai Chen, Meizhi Zhong, Min Zhang, Xikai Liu, Yan Gao, Yao Hu, Yikun Lei","submitted_at":"2024-12-12T07:52:56Z","abstract_excerpt":"Large Language models (LLMs) have become a research hotspot. To accelerate the inference of LLMs, storing computed caches in memory has become the standard technique. However, as the inference length increases, growing KV caches might lead to out-of-memory issues. Many existing methods address this issue through KV cache compression, primarily by preserving key tokens throughout all layers to reduce information loss. Most of them allocate a uniform budget size for each layer to retain. However, we observe that the minimum budget sizes needed to retain essential information vary across layers a"},"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":"2412.09036","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-12-12T07:52:56Z","cross_cats_sorted":[],"title_canon_sha256":"a077095d61fa278a70b838f54cca0bcb9e931fa2614f34b3824050bf8ec22cf7","abstract_canon_sha256":"8d985ce7a1913a5998548a7cf56e2d49107d647628f34a4c9f68070bc614e750"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:48:18.720933Z","signature_b64":"a8rzR99n8tG07+qZ5wwy4V8pw0L63mW5iIUZ+myHy2anm9UZ/qGE03YiNiY2YngZbYnBWpftHGDt6KyEVUaTDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c6c89b1f0b7e5cc51a88da8d8b3354bce13f01edc4b070782a73b31617264f05","last_reissued_at":"2026-07-05T09:48:18.720422Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:48:18.720422Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"ZigZagkv: Dynamic KV Cache Compression for Long-context Modeling based on Layer Uncertainty","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Chen Zhang, Kehai Chen, Meizhi Zhong, Min Zhang, Xikai Liu, Yan Gao, Yao Hu, Yikun Lei","submitted_at":"2024-12-12T07:52:56Z","abstract_excerpt":"Large Language models (LLMs) have become a research hotspot. To accelerate the inference of LLMs, storing computed caches in memory has become the standard technique. However, as the inference length increases, growing KV caches might lead to out-of-memory issues. Many existing methods address this issue through KV cache compression, primarily by preserving key tokens throughout all layers to reduce information loss. Most of them allocate a uniform budget size for each layer to retain. However, we observe that the minimum budget sizes needed to retain essential information vary across layers a"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2412.09036","kind":"arxiv","version":1},"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/2412.09036/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":"2412.09036","created_at":"2026-07-05T09:48:18.720488+00:00"},{"alias_kind":"arxiv_version","alias_value":"2412.09036v1","created_at":"2026-07-05T09:48:18.720488+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2412.09036","created_at":"2026-07-05T09:48:18.720488+00:00"},{"alias_kind":"pith_short_12","alias_value":"Y3EJWHYLPZOM","created_at":"2026-07-05T09:48:18.720488+00:00"},{"alias_kind":"pith_short_16","alias_value":"Y3EJWHYLPZOMKGUI","created_at":"2026-07-05T09:48:18.720488+00:00"},{"alias_kind":"pith_short_8","alias_value":"Y3EJWHYL","created_at":"2026-07-05T09:48:18.720488+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.24786","citing_title":"CONF-KV: Confidence-Aware KV Cache Eviction with Mixed-Precision Storage for Long-Horizon LLM","ref_index":27,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/Y3EJWHYLPZOMKGUI3KGYWM2UXT","json":"https://pith.science/pith/Y3EJWHYLPZOMKGUI3KGYWM2UXT.json","graph_json":"https://pith.science/api/pith-number/Y3EJWHYLPZOMKGUI3KGYWM2UXT/graph.json","events_json":"https://pith.science/api/pith-number/Y3EJWHYLPZOMKGUI3KGYWM2UXT/events.json","paper":"https://pith.science/paper/Y3EJWHYL"},"agent_actions":{"view_html":"https://pith.science/pith/Y3EJWHYLPZOMKGUI3KGYWM2UXT","download_json":"https://pith.science/pith/Y3EJWHYLPZOMKGUI3KGYWM2UXT.json","view_paper":"https://pith.science/paper/Y3EJWHYL","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2412.09036&json=true","fetch_graph":"https://pith.science/api/pith-number/Y3EJWHYLPZOMKGUI3KGYWM2UXT/graph.json","fetch_events":"https://pith.science/api/pith-number/Y3EJWHYLPZOMKGUI3KGYWM2UXT/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/Y3EJWHYLPZOMKGUI3KGYWM2UXT/action/timestamp_anchor","attest_storage":"https://pith.science/pith/Y3EJWHYLPZOMKGUI3KGYWM2UXT/action/storage_attestation","attest_author":"https://pith.science/pith/Y3EJWHYLPZOMKGUI3KGYWM2UXT/action/author_attestation","sign_citation":"https://pith.science/pith/Y3EJWHYLPZOMKGUI3KGYWM2UXT/action/citation_signature","submit_replication":"https://pith.science/pith/Y3EJWHYLPZOMKGUI3KGYWM2UXT/action/replication_record"}},"created_at":"2026-07-05T09:48:18.720488+00:00","updated_at":"2026-07-05T09:48:18.720488+00:00"}