{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:PHIZODRQIPHHCQIWNPMVOVBUAJ","short_pith_number":"pith:PHIZODRQ","schema_version":"1.0","canonical_sha256":"79d1970e3043ce7141166bd957543402417ea2008984e5e982deb862830a3d52","source":{"kind":"arxiv","id":"2506.11886","version":1},"attestation_state":"computed","paper":{"title":"Beyond Homogeneous Attention: Memory-Efficient LLMs via Fourier-Approximated KV Cache","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Linlin Li, Qipeng Guo, Qiqi Wang, Qun Liu, Ruixiao Li, Siyang He, Xiaoran Liu, Xipeng Qiu, Yuerong Song, Zengfeng Huang, Zhigeng Liu, Ziwei He","submitted_at":"2025-06-13T15:35:54Z","abstract_excerpt":"Large Language Models struggle with memory demands from the growing Key-Value (KV) cache as context lengths increase. Existing compression methods homogenize head dimensions or rely on attention-guided token pruning, often sacrificing accuracy or introducing computational overhead. We propose FourierAttention, a training-free framework that exploits the heterogeneous roles of transformer head dimensions: lower dimensions prioritize local context, while upper ones capture long-range dependencies. By projecting the long-context-insensitive dimensions onto orthogonal Fourier bases, FourierAttenti"},"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":"2506.11886","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-06-13T15:35:54Z","cross_cats_sorted":[],"title_canon_sha256":"571eae3ceaca8ff057df6a6e4ddf0bbcab1a3082481fdcd70b911465e7644259","abstract_canon_sha256":"5a57269cb11e3ac5b1a3ecde057d11fc57b332ec9489f07e57db777fd925f2e9"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:21:12.280899Z","signature_b64":"vI8k78Jh/1xr9MgfsJuEiuRVv6ldjE7WuFQUKF5iYdf38QEkh8Vz18M91A7plW+QBna/XZZtj2RKrF8qTS4lCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"79d1970e3043ce7141166bd957543402417ea2008984e5e982deb862830a3d52","last_reissued_at":"2026-07-05T11:21:12.280470Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:21:12.280470Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Beyond Homogeneous Attention: Memory-Efficient LLMs via Fourier-Approximated KV Cache","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Linlin Li, Qipeng Guo, Qiqi Wang, Qun Liu, Ruixiao Li, Siyang He, Xiaoran Liu, Xipeng Qiu, Yuerong Song, Zengfeng Huang, Zhigeng Liu, Ziwei He","submitted_at":"2025-06-13T15:35:54Z","abstract_excerpt":"Large Language Models struggle with memory demands from the growing Key-Value (KV) cache as context lengths increase. Existing compression methods homogenize head dimensions or rely on attention-guided token pruning, often sacrificing accuracy or introducing computational overhead. We propose FourierAttention, a training-free framework that exploits the heterogeneous roles of transformer head dimensions: lower dimensions prioritize local context, while upper ones capture long-range dependencies. By projecting the long-context-insensitive dimensions onto orthogonal Fourier bases, FourierAttenti"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.11886","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/2506.11886/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":"2506.11886","created_at":"2026-07-05T11:21:12.280526+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.11886v1","created_at":"2026-07-05T11:21:12.280526+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.11886","created_at":"2026-07-05T11:21:12.280526+00:00"},{"alias_kind":"pith_short_12","alias_value":"PHIZODRQIPHH","created_at":"2026-07-05T11:21:12.280526+00:00"},{"alias_kind":"pith_short_16","alias_value":"PHIZODRQIPHHCQIW","created_at":"2026-07-05T11:21:12.280526+00:00"},{"alias_kind":"pith_short_8","alias_value":"PHIZODRQ","created_at":"2026-07-05T11:21:12.280526+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.05699","citing_title":"When Quantization Is Free: An int4 KV Cache That Outruns fp16 on Apple Silicon","ref_index":20,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/PHIZODRQIPHHCQIWNPMVOVBUAJ","json":"https://pith.science/pith/PHIZODRQIPHHCQIWNPMVOVBUAJ.json","graph_json":"https://pith.science/api/pith-number/PHIZODRQIPHHCQIWNPMVOVBUAJ/graph.json","events_json":"https://pith.science/api/pith-number/PHIZODRQIPHHCQIWNPMVOVBUAJ/events.json","paper":"https://pith.science/paper/PHIZODRQ"},"agent_actions":{"view_html":"https://pith.science/pith/PHIZODRQIPHHCQIWNPMVOVBUAJ","download_json":"https://pith.science/pith/PHIZODRQIPHHCQIWNPMVOVBUAJ.json","view_paper":"https://pith.science/paper/PHIZODRQ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.11886&json=true","fetch_graph":"https://pith.science/api/pith-number/PHIZODRQIPHHCQIWNPMVOVBUAJ/graph.json","fetch_events":"https://pith.science/api/pith-number/PHIZODRQIPHHCQIWNPMVOVBUAJ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/PHIZODRQIPHHCQIWNPMVOVBUAJ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/PHIZODRQIPHHCQIWNPMVOVBUAJ/action/storage_attestation","attest_author":"https://pith.science/pith/PHIZODRQIPHHCQIWNPMVOVBUAJ/action/author_attestation","sign_citation":"https://pith.science/pith/PHIZODRQIPHHCQIWNPMVOVBUAJ/action/citation_signature","submit_replication":"https://pith.science/pith/PHIZODRQIPHHCQIWNPMVOVBUAJ/action/replication_record"}},"created_at":"2026-07-05T11:21:12.280526+00:00","updated_at":"2026-07-05T11:21:12.280526+00:00"}