{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:SPHBHSTURLT5IDHRIVR3XQFSRQ","short_pith_number":"pith:SPHBHSTU","schema_version":"1.0","canonical_sha256":"93ce13ca748ae7d40cf14563bbc0b28c1772854475bfceba8b38b47cfed1ad46","source":{"kind":"arxiv","id":"2605.23751","version":1},"attestation_state":"computed","paper":{"title":"Approaching I/O-optimality for Approximate Attention","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Aleksandros Sobczyk, Anastasios Zouzias, P\\'al Andr\\'as Papp","submitted_at":"2026-05-22T15:23:26Z","abstract_excerpt":"We revisit the I/O complexity of attention in large language models. Given query-key-value matrices $Q,K,V\\in\\mathbb{R}^{n\\times d}$, and a machine with fast memory size $M$, the goal is to compute the \"attention matrix\" $A=\\text{softmax}(Q K ^{\\top}/\\sqrt{d}) V$ with the minimal number of data transfers between fast and slow memory. Existing methods in the literature, most notably FlashAttention and its variants, incur an I/O cost that depends quadratically on $n$, while a trivial lower bound only requires $\\Omega(nd)$ I/O's to read the inputs and write the output. In this work, we present a "},"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":"2605.23751","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-05-22T15:23:26Z","cross_cats_sorted":[],"title_canon_sha256":"5edf548b6a21fae405762f6d20fd12b74600e1c8b1e0f19655b4ae4abd8cd8d3","abstract_canon_sha256":"d7aae02bed5f6b245bba887f0ca7bc9af34b42087ca3d9c2c16f192c3335aa41"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-25T02:02:30.482377Z","signature_b64":"9/fQh0Wf33jjZa2b6F6hEDdkcDD8h309Psnk7n8LELkKfNKL3nl3AW4e/9vA9vA+aHDTrpWsRqb1mlzKB+2gBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"93ce13ca748ae7d40cf14563bbc0b28c1772854475bfceba8b38b47cfed1ad46","last_reissued_at":"2026-05-25T02:02:30.481563Z","signature_status":"signed_v1","first_computed_at":"2026-05-25T02:02:30.481563Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Approaching I/O-optimality for Approximate Attention","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Aleksandros Sobczyk, Anastasios Zouzias, P\\'al Andr\\'as Papp","submitted_at":"2026-05-22T15:23:26Z","abstract_excerpt":"We revisit the I/O complexity of attention in large language models. Given query-key-value matrices $Q,K,V\\in\\mathbb{R}^{n\\times d}$, and a machine with fast memory size $M$, the goal is to compute the \"attention matrix\" $A=\\text{softmax}(Q K ^{\\top}/\\sqrt{d}) V$ with the minimal number of data transfers between fast and slow memory. Existing methods in the literature, most notably FlashAttention and its variants, incur an I/O cost that depends quadratically on $n$, while a trivial lower bound only requires $\\Omega(nd)$ I/O's to read the inputs and write the output. In this work, we present a "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.23751","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/2605.23751/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":"2605.23751","created_at":"2026-05-25T02:02:30.481704+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.23751v1","created_at":"2026-05-25T02:02:30.481704+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.23751","created_at":"2026-05-25T02:02:30.481704+00:00"},{"alias_kind":"pith_short_12","alias_value":"SPHBHSTURLT5","created_at":"2026-05-25T02:02:30.481704+00:00"},{"alias_kind":"pith_short_16","alias_value":"SPHBHSTURLT5IDHR","created_at":"2026-05-25T02:02:30.481704+00:00"},{"alias_kind":"pith_short_8","alias_value":"SPHBHSTU","created_at":"2026-05-25T02:02:30.481704+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/SPHBHSTURLT5IDHRIVR3XQFSRQ","json":"https://pith.science/pith/SPHBHSTURLT5IDHRIVR3XQFSRQ.json","graph_json":"https://pith.science/api/pith-number/SPHBHSTURLT5IDHRIVR3XQFSRQ/graph.json","events_json":"https://pith.science/api/pith-number/SPHBHSTURLT5IDHRIVR3XQFSRQ/events.json","paper":"https://pith.science/paper/SPHBHSTU"},"agent_actions":{"view_html":"https://pith.science/pith/SPHBHSTURLT5IDHRIVR3XQFSRQ","download_json":"https://pith.science/pith/SPHBHSTURLT5IDHRIVR3XQFSRQ.json","view_paper":"https://pith.science/paper/SPHBHSTU","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.23751&json=true","fetch_graph":"https://pith.science/api/pith-number/SPHBHSTURLT5IDHRIVR3XQFSRQ/graph.json","fetch_events":"https://pith.science/api/pith-number/SPHBHSTURLT5IDHRIVR3XQFSRQ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/SPHBHSTURLT5IDHRIVR3XQFSRQ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/SPHBHSTURLT5IDHRIVR3XQFSRQ/action/storage_attestation","attest_author":"https://pith.science/pith/SPHBHSTURLT5IDHRIVR3XQFSRQ/action/author_attestation","sign_citation":"https://pith.science/pith/SPHBHSTURLT5IDHRIVR3XQFSRQ/action/citation_signature","submit_replication":"https://pith.science/pith/SPHBHSTURLT5IDHRIVR3XQFSRQ/action/replication_record"}},"created_at":"2026-05-25T02:02:30.481704+00:00","updated_at":"2026-05-25T02:02:30.481704+00:00"}