{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:5EY6ZA44WB7URXVE5DHKSIOYGN","short_pith_number":"pith:5EY6ZA44","schema_version":"1.0","canonical_sha256":"e931ec839cb07f48dea4e8cea921d83354b28bde27bc9b22e122dbdbe91ac464","source":{"kind":"arxiv","id":"2507.16018","version":1},"attestation_state":"computed","paper":{"title":"Artifacts and Attention Sinks: Structured Approximations for Efficient Vision Transformers","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Andrew Lu, Huzheng Yang, Jianbo Shi, Liuhui Wang, Wentinn Liao","submitted_at":"2025-07-21T19:29:03Z","abstract_excerpt":"Vision transformers have emerged as a powerful tool across a wide range of applications, yet their inner workings remain only partially understood. In this work, we examine the phenomenon of massive tokens - tokens with exceptionally high activation norms that act as attention sinks - and artifact tokens that emerge as a byproduct during inference. Our analysis reveals that these tokens mutually suppress one another through the attention mechanism, playing a critical role in regulating information flow within the network. Leveraging these insights, we introduce Fast Nystr\\\"om Attention (FNA), "},"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":"2507.16018","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2025-07-21T19:29:03Z","cross_cats_sorted":[],"title_canon_sha256":"5faf6f098f6aef7f3598b3b3fc3d8bc81c99088188854ab529ec0f490042b6c3","abstract_canon_sha256":"ee51a217e6dad585a65ca4efcb2bf93233c563be9c6aa3c5d65f966b5fe577f9"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:41:01.351727Z","signature_b64":"5o5UsopDAG3x0FkU06h7rbCgB1Xh710Qzt5r99nqznlK9R5Ay0sguJuJnpd7lqjcKU/7ZE9DoBParM43JRPPCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"e931ec839cb07f48dea4e8cea921d83354b28bde27bc9b22e122dbdbe91ac464","last_reissued_at":"2026-07-05T11:41:01.350563Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:41:01.350563Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Artifacts and Attention Sinks: Structured Approximations for Efficient Vision Transformers","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Andrew Lu, Huzheng Yang, Jianbo Shi, Liuhui Wang, Wentinn Liao","submitted_at":"2025-07-21T19:29:03Z","abstract_excerpt":"Vision transformers have emerged as a powerful tool across a wide range of applications, yet their inner workings remain only partially understood. In this work, we examine the phenomenon of massive tokens - tokens with exceptionally high activation norms that act as attention sinks - and artifact tokens that emerge as a byproduct during inference. Our analysis reveals that these tokens mutually suppress one another through the attention mechanism, playing a critical role in regulating information flow within the network. Leveraging these insights, we introduce Fast Nystr\\\"om Attention (FNA), "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2507.16018","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/2507.16018/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":"2507.16018","created_at":"2026-07-05T11:41:01.350646+00:00"},{"alias_kind":"arxiv_version","alias_value":"2507.16018v1","created_at":"2026-07-05T11:41:01.350646+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.16018","created_at":"2026-07-05T11:41:01.350646+00:00"},{"alias_kind":"pith_short_12","alias_value":"5EY6ZA44WB7U","created_at":"2026-07-05T11:41:01.350646+00:00"},{"alias_kind":"pith_short_16","alias_value":"5EY6ZA44WB7URXVE","created_at":"2026-07-05T11:41:01.350646+00:00"},{"alias_kind":"pith_short_8","alias_value":"5EY6ZA44","created_at":"2026-07-05T11:41:01.350646+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2604.03316","citing_title":"When Sinks Help or Hurt: Unified Framework for Attention Sink in Large Vision-Language Models","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2604.10098","citing_title":"Attention Sink in Transformers: A Survey on Utilization, Interpretation, and Mitigation","ref_index":119,"is_internal_anchor":false},{"citing_arxiv_id":"2604.20937","citing_title":"Sink-Token-Aware Pruning for Fine-Grained Video Understanding in Efficient Video LLMs","ref_index":27,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/5EY6ZA44WB7URXVE5DHKSIOYGN","json":"https://pith.science/pith/5EY6ZA44WB7URXVE5DHKSIOYGN.json","graph_json":"https://pith.science/api/pith-number/5EY6ZA44WB7URXVE5DHKSIOYGN/graph.json","events_json":"https://pith.science/api/pith-number/5EY6ZA44WB7URXVE5DHKSIOYGN/events.json","paper":"https://pith.science/paper/5EY6ZA44"},"agent_actions":{"view_html":"https://pith.science/pith/5EY6ZA44WB7URXVE5DHKSIOYGN","download_json":"https://pith.science/pith/5EY6ZA44WB7URXVE5DHKSIOYGN.json","view_paper":"https://pith.science/paper/5EY6ZA44","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2507.16018&json=true","fetch_graph":"https://pith.science/api/pith-number/5EY6ZA44WB7URXVE5DHKSIOYGN/graph.json","fetch_events":"https://pith.science/api/pith-number/5EY6ZA44WB7URXVE5DHKSIOYGN/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/5EY6ZA44WB7URXVE5DHKSIOYGN/action/timestamp_anchor","attest_storage":"https://pith.science/pith/5EY6ZA44WB7URXVE5DHKSIOYGN/action/storage_attestation","attest_author":"https://pith.science/pith/5EY6ZA44WB7URXVE5DHKSIOYGN/action/author_attestation","sign_citation":"https://pith.science/pith/5EY6ZA44WB7URXVE5DHKSIOYGN/action/citation_signature","submit_replication":"https://pith.science/pith/5EY6ZA44WB7URXVE5DHKSIOYGN/action/replication_record"}},"created_at":"2026-07-05T11:41:01.350646+00:00","updated_at":"2026-07-05T11:41:01.350646+00:00"}