{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:KKTEJASMAJHVKPVKTDABXTVNKA","short_pith_number":"pith:KKTEJASM","schema_version":"1.0","canonical_sha256":"52a644824c024f553eaa98c01bcead50025d9f9d5bc7591dc51e97284e06f1e1","source":{"kind":"arxiv","id":"2507.03865","version":2},"attestation_state":"computed","paper":{"title":"OrthoRank: Token Selection via Sink Token Orthogonality for Efficient LLM inference","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Dokwan Oh, Jaehoon Oh, Seungjun Shin","submitted_at":"2025-07-05T02:29:23Z","abstract_excerpt":"Attention mechanisms are central to the success of large language models (LLMs), enabling them to capture intricate token dependencies and implicitly assign importance to each token. Recent studies have revealed the sink token, which receives disproportionately high attention despite their limited semantic role. In this paper, we first expand the relationship between the sink token and other tokens, moving beyond attention to explore their similarity in hidden states, considering the layer depth. We observe that as the layers get deeper, the cosine similarity between the normalized hidden stat"},"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":"2507.03865","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-07-05T02:29:23Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"f9506ba53dbaa0c69e04824f45b25e0f6fe7a819a077433b2cacb28847136551","abstract_canon_sha256":"101afc50eca19c21f0d341f4a9188bf7c5ce12abf79b3a947207d6ec82a7978d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:54:48.977473Z","signature_b64":"ou32ptOdKXhQftlF/QyikfsXpb4XY00+Hl/LNtAaf3kpmtyn5ea4fp2it3bgvyQhSAzEG/Eqm1KkdAoE5iNhBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"52a644824c024f553eaa98c01bcead50025d9f9d5bc7591dc51e97284e06f1e1","last_reissued_at":"2026-07-05T11:54:48.977037Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:54:48.977037Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"OrthoRank: Token Selection via Sink Token Orthogonality for Efficient LLM inference","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Dokwan Oh, Jaehoon Oh, Seungjun Shin","submitted_at":"2025-07-05T02:29:23Z","abstract_excerpt":"Attention mechanisms are central to the success of large language models (LLMs), enabling them to capture intricate token dependencies and implicitly assign importance to each token. Recent studies have revealed the sink token, which receives disproportionately high attention despite their limited semantic role. In this paper, we first expand the relationship between the sink token and other tokens, moving beyond attention to explore their similarity in hidden states, considering the layer depth. We observe that as the layers get deeper, the cosine similarity between the normalized hidden stat"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2507.03865","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/2507.03865/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":"2507.03865","created_at":"2026-07-05T11:54:48.977092+00:00"},{"alias_kind":"arxiv_version","alias_value":"2507.03865v2","created_at":"2026-07-05T11:54:48.977092+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.03865","created_at":"2026-07-05T11:54:48.977092+00:00"},{"alias_kind":"pith_short_12","alias_value":"KKTEJASMAJHV","created_at":"2026-07-05T11:54:48.977092+00:00"},{"alias_kind":"pith_short_16","alias_value":"KKTEJASMAJHVKPVK","created_at":"2026-07-05T11:54:48.977092+00:00"},{"alias_kind":"pith_short_8","alias_value":"KKTEJASM","created_at":"2026-07-05T11:54:48.977092+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.07604","citing_title":"Contribution Weights: A Geometrical Analysis of Self-Attention Transformers","ref_index":171,"is_internal_anchor":false},{"citing_arxiv_id":"2606.07604","citing_title":"Contribution Weights: A Geometrical Analysis of Self-Attention Transformers","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2602.01203","citing_title":"Attention Sink Forges Native MoE in Attention Layers: Sink-Aware Training to Address Head Collapse","ref_index":29,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/KKTEJASMAJHVKPVKTDABXTVNKA","json":"https://pith.science/pith/KKTEJASMAJHVKPVKTDABXTVNKA.json","graph_json":"https://pith.science/api/pith-number/KKTEJASMAJHVKPVKTDABXTVNKA/graph.json","events_json":"https://pith.science/api/pith-number/KKTEJASMAJHVKPVKTDABXTVNKA/events.json","paper":"https://pith.science/paper/KKTEJASM"},"agent_actions":{"view_html":"https://pith.science/pith/KKTEJASMAJHVKPVKTDABXTVNKA","download_json":"https://pith.science/pith/KKTEJASMAJHVKPVKTDABXTVNKA.json","view_paper":"https://pith.science/paper/KKTEJASM","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2507.03865&json=true","fetch_graph":"https://pith.science/api/pith-number/KKTEJASMAJHVKPVKTDABXTVNKA/graph.json","fetch_events":"https://pith.science/api/pith-number/KKTEJASMAJHVKPVKTDABXTVNKA/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/KKTEJASMAJHVKPVKTDABXTVNKA/action/timestamp_anchor","attest_storage":"https://pith.science/pith/KKTEJASMAJHVKPVKTDABXTVNKA/action/storage_attestation","attest_author":"https://pith.science/pith/KKTEJASMAJHVKPVKTDABXTVNKA/action/author_attestation","sign_citation":"https://pith.science/pith/KKTEJASMAJHVKPVKTDABXTVNKA/action/citation_signature","submit_replication":"https://pith.science/pith/KKTEJASMAJHVKPVKTDABXTVNKA/action/replication_record"}},"created_at":"2026-07-05T11:54:48.977092+00:00","updated_at":"2026-07-05T11:54:48.977092+00:00"}