{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:ZOZN4RW4KFQWLHZUGIBLX747NF","short_pith_number":"pith:ZOZN4RW4","schema_version":"1.0","canonical_sha256":"cbb2de46dc5161659f343202bbff9f695c120373f248854c3373cf9d66795322","source":{"kind":"arxiv","id":"2405.10637","version":2},"attestation_state":"computed","paper":{"title":"Layer-Condensed KV Cache for Efficient Inference of Large Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Haoyi Wu, Kewei Tu","submitted_at":"2024-05-17T08:59:46Z","abstract_excerpt":"Huge memory consumption has been a major bottleneck for deploying high-throughput large language models in real-world applications. In addition to the large number of parameters, the key-value (KV) cache for the attention mechanism in the transformer architecture consumes a significant amount of memory, especially when the number of layers is large for deep language models. In this paper, we propose a novel method that only computes and caches the KVs of a small number of layers, thus significantly saving memory consumption and improving inference throughput. Our experiments on large language "},"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":"2405.10637","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-05-17T08:59:46Z","cross_cats_sorted":[],"title_canon_sha256":"e8acc1c3805dde83c83423834019c2bce6865718e62f3bd5d6eb160b1bc7936f","abstract_canon_sha256":"e6b24c461e59acfa7afdc9bfc5b2953a921524d725338a2995517ffd71767215"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:26:44.814774Z","signature_b64":"8xfudO31r6Ag5oW33yym5nuMygi8bAlptQR+GAlnYQEc/fXXk+JPF+y+2Ye9CKoe8YJq+qlicCa3htrrvIiqCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"cbb2de46dc5161659f343202bbff9f695c120373f248854c3373cf9d66795322","last_reissued_at":"2026-07-05T08:26:44.814327Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:26:44.814327Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Layer-Condensed KV Cache for Efficient Inference of Large Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Haoyi Wu, Kewei Tu","submitted_at":"2024-05-17T08:59:46Z","abstract_excerpt":"Huge memory consumption has been a major bottleneck for deploying high-throughput large language models in real-world applications. In addition to the large number of parameters, the key-value (KV) cache for the attention mechanism in the transformer architecture consumes a significant amount of memory, especially when the number of layers is large for deep language models. In this paper, we propose a novel method that only computes and caches the KVs of a small number of layers, thus significantly saving memory consumption and improving inference throughput. Our experiments on large language "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2405.10637","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/2405.10637/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":"2405.10637","created_at":"2026-07-05T08:26:44.814393+00:00"},{"alias_kind":"arxiv_version","alias_value":"2405.10637v2","created_at":"2026-07-05T08:26:44.814393+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2405.10637","created_at":"2026-07-05T08:26:44.814393+00:00"},{"alias_kind":"pith_short_12","alias_value":"ZOZN4RW4KFQW","created_at":"2026-07-05T08:26:44.814393+00:00"},{"alias_kind":"pith_short_16","alias_value":"ZOZN4RW4KFQWLHZU","created_at":"2026-07-05T08:26:44.814393+00:00"},{"alias_kind":"pith_short_8","alias_value":"ZOZN4RW4","created_at":"2026-07-05T08:26:44.814393+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.02780","citing_title":"Do Value Vectors in Deep Layers Need Context from the Residual Stream?","ref_index":106,"is_internal_anchor":false},{"citing_arxiv_id":"2502.01941","citing_title":"Semantic Integrity Matters: Benchmarking and Preserving High-Density Reasoning in KV Cache Compression","ref_index":57,"is_internal_anchor":false},{"citing_arxiv_id":"2511.22681","citing_title":"CacheTrap: Unveiling a Stealthier Gray-Box Trojan against LLMs","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2410.10781","citing_title":"When Attention Sink Emerges in Language Models: An Empirical View","ref_index":53,"is_internal_anchor":false},{"citing_arxiv_id":"2604.23553","citing_title":"ClusterFusion++: Expanding Cluster-Level Fusion to Full Transformer-Block Decoding","ref_index":11,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ZOZN4RW4KFQWLHZUGIBLX747NF","json":"https://pith.science/pith/ZOZN4RW4KFQWLHZUGIBLX747NF.json","graph_json":"https://pith.science/api/pith-number/ZOZN4RW4KFQWLHZUGIBLX747NF/graph.json","events_json":"https://pith.science/api/pith-number/ZOZN4RW4KFQWLHZUGIBLX747NF/events.json","paper":"https://pith.science/paper/ZOZN4RW4"},"agent_actions":{"view_html":"https://pith.science/pith/ZOZN4RW4KFQWLHZUGIBLX747NF","download_json":"https://pith.science/pith/ZOZN4RW4KFQWLHZUGIBLX747NF.json","view_paper":"https://pith.science/paper/ZOZN4RW4","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2405.10637&json=true","fetch_graph":"https://pith.science/api/pith-number/ZOZN4RW4KFQWLHZUGIBLX747NF/graph.json","fetch_events":"https://pith.science/api/pith-number/ZOZN4RW4KFQWLHZUGIBLX747NF/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ZOZN4RW4KFQWLHZUGIBLX747NF/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ZOZN4RW4KFQWLHZUGIBLX747NF/action/storage_attestation","attest_author":"https://pith.science/pith/ZOZN4RW4KFQWLHZUGIBLX747NF/action/author_attestation","sign_citation":"https://pith.science/pith/ZOZN4RW4KFQWLHZUGIBLX747NF/action/citation_signature","submit_replication":"https://pith.science/pith/ZOZN4RW4KFQWLHZUGIBLX747NF/action/replication_record"}},"created_at":"2026-07-05T08:26:44.814393+00:00","updated_at":"2026-07-05T08:26:44.814393+00:00"}