{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:V72IZK3SNEYI4H22VBUF6765NP","short_pith_number":"pith:V72IZK3S","schema_version":"1.0","canonical_sha256":"aff48cab7269308e1f5aa8685f7fdd6be59fdf00192d03016aa8920937d047d8","source":{"kind":"arxiv","id":"2410.14442","version":2},"attestation_state":"computed","paper":{"title":"A Systematic Study of Cross-Layer KV Sharing for Efficient LLM Inference","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Haoyi Wu, Kewei Tu, You Wu","submitted_at":"2024-10-18T13:01:14Z","abstract_excerpt":"Recently, sharing key-value (KV) cache across layers has been found effective in efficient inference of large language models (LLMs). To systematically investigate different techniques of cross-layer KV sharing, we propose a unified framework that covers several recent methods and their novel variants. We conduct comprehensive experiments on all the configurations of the framework, evaluating their generation throughput and performance in language modeling and downstream tasks. We find that when reducing the size of the KV cache by 2$\\times$, most configurations can achieve higher throughput t"},"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":"2410.14442","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-10-18T13:01:14Z","cross_cats_sorted":[],"title_canon_sha256":"a58403e1dc85b4ad6d230cb5a1e64a250f45c3f18b2a155cc82bfb2eff4b266f","abstract_canon_sha256":"cbb45036bc7e1c747b1bb4dd11ad6a526875fadbb5fccc08aea9fdfbf3f005b2"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:09:49.150609Z","signature_b64":"cBEvIrStvPbd3Qzl2aMhPABR1x45xI6OAFPeItPwzGtu+i7vv2QJToUe51xpy9yK56+WGl9tDyjnyiyKhVZpCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"aff48cab7269308e1f5aa8685f7fdd6be59fdf00192d03016aa8920937d047d8","last_reissued_at":"2026-07-05T10:09:49.150122Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:09:49.150122Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"A Systematic Study of Cross-Layer KV Sharing for Efficient LLM Inference","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Haoyi Wu, Kewei Tu, You Wu","submitted_at":"2024-10-18T13:01:14Z","abstract_excerpt":"Recently, sharing key-value (KV) cache across layers has been found effective in efficient inference of large language models (LLMs). To systematically investigate different techniques of cross-layer KV sharing, we propose a unified framework that covers several recent methods and their novel variants. We conduct comprehensive experiments on all the configurations of the framework, evaluating their generation throughput and performance in language modeling and downstream tasks. We find that when reducing the size of the KV cache by 2$\\times$, most configurations can achieve higher throughput t"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.14442","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/2410.14442/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":"2410.14442","created_at":"2026-07-05T10:09:49.150179+00:00"},{"alias_kind":"arxiv_version","alias_value":"2410.14442v2","created_at":"2026-07-05T10:09:49.150179+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.14442","created_at":"2026-07-05T10:09:49.150179+00:00"},{"alias_kind":"pith_short_12","alias_value":"V72IZK3SNEYI","created_at":"2026-07-05T10:09:49.150179+00:00"},{"alias_kind":"pith_short_16","alias_value":"V72IZK3SNEYI4H22","created_at":"2026-07-05T10:09:49.150179+00:00"},{"alias_kind":"pith_short_8","alias_value":"V72IZK3S","created_at":"2026-07-05T10:09:49.150179+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2508.15881","citing_title":"TPLA: Tensor Parallel Latent Attention for Efficient Disaggregated Prefill and Decode Inference","ref_index":33,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/V72IZK3SNEYI4H22VBUF6765NP","json":"https://pith.science/pith/V72IZK3SNEYI4H22VBUF6765NP.json","graph_json":"https://pith.science/api/pith-number/V72IZK3SNEYI4H22VBUF6765NP/graph.json","events_json":"https://pith.science/api/pith-number/V72IZK3SNEYI4H22VBUF6765NP/events.json","paper":"https://pith.science/paper/V72IZK3S"},"agent_actions":{"view_html":"https://pith.science/pith/V72IZK3SNEYI4H22VBUF6765NP","download_json":"https://pith.science/pith/V72IZK3SNEYI4H22VBUF6765NP.json","view_paper":"https://pith.science/paper/V72IZK3S","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2410.14442&json=true","fetch_graph":"https://pith.science/api/pith-number/V72IZK3SNEYI4H22VBUF6765NP/graph.json","fetch_events":"https://pith.science/api/pith-number/V72IZK3SNEYI4H22VBUF6765NP/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/V72IZK3SNEYI4H22VBUF6765NP/action/timestamp_anchor","attest_storage":"https://pith.science/pith/V72IZK3SNEYI4H22VBUF6765NP/action/storage_attestation","attest_author":"https://pith.science/pith/V72IZK3SNEYI4H22VBUF6765NP/action/author_attestation","sign_citation":"https://pith.science/pith/V72IZK3SNEYI4H22VBUF6765NP/action/citation_signature","submit_replication":"https://pith.science/pith/V72IZK3SNEYI4H22VBUF6765NP/action/replication_record"}},"created_at":"2026-07-05T10:09:49.150179+00:00","updated_at":"2026-07-05T10:09:49.150179+00:00"}