{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:KGPGLEZR2WAFQBPVL2KNJSHXYR","short_pith_number":"pith:KGPGLEZR","schema_version":"1.0","canonical_sha256":"519e659331d5805805f55e94d4c8f7c47d540f4437eefd30134ee9e6a16c6e3c","source":{"kind":"arxiv","id":"2504.00927","version":2},"attestation_state":"computed","paper":{"title":"Multi-Token Attention","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Jason Weston, Olga Golovneva, Sainbayar Sukhbaatar, Tianlu Wang","submitted_at":"2025-04-01T15:59:32Z","abstract_excerpt":"Soft attention is a critical mechanism powering LLMs to locate relevant parts within a given context. However, individual attention weights are determined by the similarity of only a single query and key token vector. This \"single token attention\" bottlenecks the amount of information used in distinguishing a relevant part from the rest of the context. To address this issue, we propose a new attention method, Multi-Token Attention (MTA), which allows LLMs to condition their attention weights on multiple query and key vectors simultaneously. This is achieved by applying convolution operations o"},"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":"2504.00927","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-04-01T15:59:32Z","cross_cats_sorted":[],"title_canon_sha256":"157fdd2541710c5b832ed3bdd628d001d41967775c2a96c39d6dfddbd6257a41","abstract_canon_sha256":"c107ba0531c9d551dc31c66bebc44ebcd6950e872ed18be4ad08bb3166e8a16a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:35:34.328587Z","signature_b64":"2+5sZpIX4TsYAFNDgD7vSWs6yIRO9f6LjPw+q1dqPxRme5+q9IlwWrKDxaBG7oesDwgCowTIupBuAmHnNKgHCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"519e659331d5805805f55e94d4c8f7c47d540f4437eefd30134ee9e6a16c6e3c","last_reissued_at":"2026-07-05T11:35:34.328135Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:35:34.328135Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Multi-Token Attention","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Jason Weston, Olga Golovneva, Sainbayar Sukhbaatar, Tianlu Wang","submitted_at":"2025-04-01T15:59:32Z","abstract_excerpt":"Soft attention is a critical mechanism powering LLMs to locate relevant parts within a given context. However, individual attention weights are determined by the similarity of only a single query and key token vector. This \"single token attention\" bottlenecks the amount of information used in distinguishing a relevant part from the rest of the context. To address this issue, we propose a new attention method, Multi-Token Attention (MTA), which allows LLMs to condition their attention weights on multiple query and key vectors simultaneously. This is achieved by applying convolution operations o"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.00927","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/2504.00927/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":"2504.00927","created_at":"2026-07-05T11:35:34.328190+00:00"},{"alias_kind":"arxiv_version","alias_value":"2504.00927v2","created_at":"2026-07-05T11:35:34.328190+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.00927","created_at":"2026-07-05T11:35:34.328190+00:00"},{"alias_kind":"pith_short_12","alias_value":"KGPGLEZR2WAF","created_at":"2026-07-05T11:35:34.328190+00:00"},{"alias_kind":"pith_short_16","alias_value":"KGPGLEZR2WAFQBPV","created_at":"2026-07-05T11:35:34.328190+00:00"},{"alias_kind":"pith_short_8","alias_value":"KGPGLEZR","created_at":"2026-07-05T11:35:34.328190+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2510.23641","citing_title":"Spatially Aware Linear Transformer (SAL-T) for Particle Jet Tagging","ref_index":23,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/KGPGLEZR2WAFQBPVL2KNJSHXYR","json":"https://pith.science/pith/KGPGLEZR2WAFQBPVL2KNJSHXYR.json","graph_json":"https://pith.science/api/pith-number/KGPGLEZR2WAFQBPVL2KNJSHXYR/graph.json","events_json":"https://pith.science/api/pith-number/KGPGLEZR2WAFQBPVL2KNJSHXYR/events.json","paper":"https://pith.science/paper/KGPGLEZR"},"agent_actions":{"view_html":"https://pith.science/pith/KGPGLEZR2WAFQBPVL2KNJSHXYR","download_json":"https://pith.science/pith/KGPGLEZR2WAFQBPVL2KNJSHXYR.json","view_paper":"https://pith.science/paper/KGPGLEZR","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2504.00927&json=true","fetch_graph":"https://pith.science/api/pith-number/KGPGLEZR2WAFQBPVL2KNJSHXYR/graph.json","fetch_events":"https://pith.science/api/pith-number/KGPGLEZR2WAFQBPVL2KNJSHXYR/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/KGPGLEZR2WAFQBPVL2KNJSHXYR/action/timestamp_anchor","attest_storage":"https://pith.science/pith/KGPGLEZR2WAFQBPVL2KNJSHXYR/action/storage_attestation","attest_author":"https://pith.science/pith/KGPGLEZR2WAFQBPVL2KNJSHXYR/action/author_attestation","sign_citation":"https://pith.science/pith/KGPGLEZR2WAFQBPVL2KNJSHXYR/action/citation_signature","submit_replication":"https://pith.science/pith/KGPGLEZR2WAFQBPVL2KNJSHXYR/action/replication_record"}},"created_at":"2026-07-05T11:35:34.328190+00:00","updated_at":"2026-07-05T11:35:34.328190+00:00"}