{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:43IXNE4QYHUGSWBKVHYAH4NBNZ","short_pith_number":"pith:43IXNE4Q","schema_version":"1.0","canonical_sha256":"e6d1769390c1e869582aa9f003f1a16e400c30bc88143780dd876347b02ed20f","source":{"kind":"arxiv","id":"2410.12876","version":3},"attestation_state":"computed","paper":{"title":"In-context KV-Cache Eviction for LLMs via Attention-Gate","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Bokai Lin, Hao Zhang, TianQi Hou, Zhijie Deng, Zihao Zeng","submitted_at":"2024-10-15T05:01:19Z","abstract_excerpt":"The KV-Cache technique has become the standard for the inference of large language models (LLMs). Yet, it is widely criticized that KV-Cache can become a bottleneck of the LLM inference system. This paper enables a novel dynamic KV-Cache eviction policy by injecting a lightweight module called Attention-Gate to the model. It accepts the global context as input and yields eviction flags for each token. The self-attention modules in the model proceed according to the flags and cache only a subset of the KV states for next token prediction. The Attention-Gates can yield various flags for differen"},"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":"2410.12876","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-10-15T05:01:19Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"13bd4ed1800761ac6c04b21806a63e25f0224ef2c60b40bd9697e185a2b2e23d","abstract_canon_sha256":"0e563b5be9b8343433ae5ba04527dd746c34bd3718a4c7782f0ff0ff498c7dec"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:50:09.746890Z","signature_b64":"Uu2pPX00Al8TIYTakn84z4pF9fwyMYaDqW1uyVnSz7KMPqo1wdU5LzypcW/HL9kbT6Vg4VhoCl59fOi9AT2dAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"e6d1769390c1e869582aa9f003f1a16e400c30bc88143780dd876347b02ed20f","last_reissued_at":"2026-07-05T10:50:09.746484Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:50:09.746484Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"In-context KV-Cache Eviction for LLMs via Attention-Gate","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Bokai Lin, Hao Zhang, TianQi Hou, Zhijie Deng, Zihao Zeng","submitted_at":"2024-10-15T05:01:19Z","abstract_excerpt":"The KV-Cache technique has become the standard for the inference of large language models (LLMs). Yet, it is widely criticized that KV-Cache can become a bottleneck of the LLM inference system. This paper enables a novel dynamic KV-Cache eviction policy by injecting a lightweight module called Attention-Gate to the model. It accepts the global context as input and yields eviction flags for each token. The self-attention modules in the model proceed according to the flags and cache only a subset of the KV states for next token prediction. The Attention-Gates can yield various flags for differen"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.12876","kind":"arxiv","version":3},"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/2410.12876/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":"2410.12876","created_at":"2026-07-05T10:50:09.746539+00:00"},{"alias_kind":"arxiv_version","alias_value":"2410.12876v3","created_at":"2026-07-05T10:50:09.746539+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.12876","created_at":"2026-07-05T10:50:09.746539+00:00"},{"alias_kind":"pith_short_12","alias_value":"43IXNE4QYHUG","created_at":"2026-07-05T10:50:09.746539+00:00"},{"alias_kind":"pith_short_16","alias_value":"43IXNE4QYHUGSWBK","created_at":"2026-07-05T10:50:09.746539+00:00"},{"alias_kind":"pith_short_8","alias_value":"43IXNE4Q","created_at":"2026-07-05T10:50:09.746539+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.28831","citing_title":"HARD-KV: Head-Adaptive Regularization for Decoding-time KV Compression","ref_index":33,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/43IXNE4QYHUGSWBKVHYAH4NBNZ","json":"https://pith.science/pith/43IXNE4QYHUGSWBKVHYAH4NBNZ.json","graph_json":"https://pith.science/api/pith-number/43IXNE4QYHUGSWBKVHYAH4NBNZ/graph.json","events_json":"https://pith.science/api/pith-number/43IXNE4QYHUGSWBKVHYAH4NBNZ/events.json","paper":"https://pith.science/paper/43IXNE4Q"},"agent_actions":{"view_html":"https://pith.science/pith/43IXNE4QYHUGSWBKVHYAH4NBNZ","download_json":"https://pith.science/pith/43IXNE4QYHUGSWBKVHYAH4NBNZ.json","view_paper":"https://pith.science/paper/43IXNE4Q","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2410.12876&json=true","fetch_graph":"https://pith.science/api/pith-number/43IXNE4QYHUGSWBKVHYAH4NBNZ/graph.json","fetch_events":"https://pith.science/api/pith-number/43IXNE4QYHUGSWBKVHYAH4NBNZ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/43IXNE4QYHUGSWBKVHYAH4NBNZ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/43IXNE4QYHUGSWBKVHYAH4NBNZ/action/storage_attestation","attest_author":"https://pith.science/pith/43IXNE4QYHUGSWBKVHYAH4NBNZ/action/author_attestation","sign_citation":"https://pith.science/pith/43IXNE4QYHUGSWBKVHYAH4NBNZ/action/citation_signature","submit_replication":"https://pith.science/pith/43IXNE4QYHUGSWBKVHYAH4NBNZ/action/replication_record"}},"created_at":"2026-07-05T10:50:09.746539+00:00","updated_at":"2026-07-05T10:50:09.746539+00:00"}