{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:QQQUQSKZWJ7IXWYBW6HBGALYHD","short_pith_number":"pith:QQQUQSKZ","schema_version":"1.0","canonical_sha256":"8421484959b27e8bdb01b78e13017838cf2aebdd9369953810557d4a6e9e3140","source":{"kind":"arxiv","id":"2406.09297","version":3},"attestation_state":"computed","paper":{"title":"MLKV: Multi-Layer Key-Value Heads for Memory Efficient Transformer Decoding","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Alham Fikri Aji, Ayu Purwarianti, Muhammad Farid Adilazuarda, Zayd Muhammad Kawakibi Zuhri","submitted_at":"2024-06-13T16:33:44Z","abstract_excerpt":"Auto-regressive inference of transformers benefit greatly from Key-Value (KV) caching, but can lead to major memory bottlenecks as model size, batch size, and sequence length grow at scale. We introduce Multi-Layer Key-Value (MLKV) sharing, a novel approach extending KV sharing across transformer layers to reduce memory usage beyond what was possible with Multi-Query Attention (MQA) and Grouped-Query Attention (GQA). Evaluations on various NLP benchmarks and inference metrics using uptrained Pythia-160M variants demonstrate that MLKV significantly reduces memory usage with minimal performance "},"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":"2406.09297","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2024-06-13T16:33:44Z","cross_cats_sorted":[],"title_canon_sha256":"aaea7180f1940b3bc68d8b310d6345b181cfdddd2a535492fc111b49744e300b","abstract_canon_sha256":"0ea8acb4b3c4e8f8fab949af65e23c586b3fefbcc8043c80f081d31b16b2be22"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:20:44.239213Z","signature_b64":"E0ufWdiF6yoZcsN63HcLdEzzthVPnmau5tJMNzOz+ZSRaQfyy1Fie2cfZxIvGoHgNMs27lmAmLPVEiaqbXF5BQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"8421484959b27e8bdb01b78e13017838cf2aebdd9369953810557d4a6e9e3140","last_reissued_at":"2026-07-05T09:20:44.238764Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:20:44.238764Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"MLKV: Multi-Layer Key-Value Heads for Memory Efficient Transformer Decoding","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Alham Fikri Aji, Ayu Purwarianti, Muhammad Farid Adilazuarda, Zayd Muhammad Kawakibi Zuhri","submitted_at":"2024-06-13T16:33:44Z","abstract_excerpt":"Auto-regressive inference of transformers benefit greatly from Key-Value (KV) caching, but can lead to major memory bottlenecks as model size, batch size, and sequence length grow at scale. We introduce Multi-Layer Key-Value (MLKV) sharing, a novel approach extending KV sharing across transformer layers to reduce memory usage beyond what was possible with Multi-Query Attention (MQA) and Grouped-Query Attention (GQA). Evaluations on various NLP benchmarks and inference metrics using uptrained Pythia-160M variants demonstrate that MLKV significantly reduces memory usage with minimal performance "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2406.09297","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/2406.09297/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":"2406.09297","created_at":"2026-07-05T09:20:44.238822+00:00"},{"alias_kind":"arxiv_version","alias_value":"2406.09297v3","created_at":"2026-07-05T09:20:44.238822+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2406.09297","created_at":"2026-07-05T09:20:44.238822+00:00"},{"alias_kind":"pith_short_12","alias_value":"QQQUQSKZWJ7I","created_at":"2026-07-05T09:20:44.238822+00:00"},{"alias_kind":"pith_short_16","alias_value":"QQQUQSKZWJ7IXWYB","created_at":"2026-07-05T09:20:44.238822+00:00"},{"alias_kind":"pith_short_8","alias_value":"QQQUQSKZ","created_at":"2026-07-05T09:20:44.238822+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2508.10824","citing_title":"Memory-Augmented Transformers: A Systematic Review from Neuroscience Principles to Enhanced Model Architectures","ref_index":123,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/QQQUQSKZWJ7IXWYBW6HBGALYHD","json":"https://pith.science/pith/QQQUQSKZWJ7IXWYBW6HBGALYHD.json","graph_json":"https://pith.science/api/pith-number/QQQUQSKZWJ7IXWYBW6HBGALYHD/graph.json","events_json":"https://pith.science/api/pith-number/QQQUQSKZWJ7IXWYBW6HBGALYHD/events.json","paper":"https://pith.science/paper/QQQUQSKZ"},"agent_actions":{"view_html":"https://pith.science/pith/QQQUQSKZWJ7IXWYBW6HBGALYHD","download_json":"https://pith.science/pith/QQQUQSKZWJ7IXWYBW6HBGALYHD.json","view_paper":"https://pith.science/paper/QQQUQSKZ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2406.09297&json=true","fetch_graph":"https://pith.science/api/pith-number/QQQUQSKZWJ7IXWYBW6HBGALYHD/graph.json","fetch_events":"https://pith.science/api/pith-number/QQQUQSKZWJ7IXWYBW6HBGALYHD/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/QQQUQSKZWJ7IXWYBW6HBGALYHD/action/timestamp_anchor","attest_storage":"https://pith.science/pith/QQQUQSKZWJ7IXWYBW6HBGALYHD/action/storage_attestation","attest_author":"https://pith.science/pith/QQQUQSKZWJ7IXWYBW6HBGALYHD/action/author_attestation","sign_citation":"https://pith.science/pith/QQQUQSKZWJ7IXWYBW6HBGALYHD/action/citation_signature","submit_replication":"https://pith.science/pith/QQQUQSKZWJ7IXWYBW6HBGALYHD/action/replication_record"}},"created_at":"2026-07-05T09:20:44.238822+00:00","updated_at":"2026-07-05T09:20:44.238822+00:00"}