{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:BCC2U5KAPOIXA3U5MXUUBK35XY","short_pith_number":"pith:BCC2U5KA","schema_version":"1.0","canonical_sha256":"0885aa75407b91706e9d65e940ab7dbe3a259334ed215f812f1fa3c08b8663ff","source":{"kind":"arxiv","id":"2605.16359","version":1},"attestation_state":"computed","paper":{"title":"How Many Visual Tokens Do Multimodal Language Models Need? Scaling Visual Token Pruning with F^3A","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CV","authors_text":"Daling Wang, Junzhao Huang, Shi Feng, Xiaocui Yang, Yifei Zhang, Yijie Huang, Yiqun Zhang, Yongkang Liu, Zhuoyue Jia, Zihan Wang","submitted_at":"2026-05-09T13:13:04Z","abstract_excerpt":"Vision-language models improve perception by feeding increasingly long visual token sequences into language backbones, but the resulting inference cost raises a basic scaling question: as multimodal models grow, how many visual tokens are actually needed, and how should they be allocated under a fixed visual token budget? Existing training-free pruning methods typically answer this with one-shot proxies such as decoder attention, visual similarity, or conditional diversity. We argue that visual token pruning is better viewed as task-conditioned evidence search, especially under aggressive comp"},"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":"2605.16359","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-05-09T13:13:04Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"64ba929f0ec968f7f5859866f9603a088e8c0c261c74c9ced60b8f22f6a422d2","abstract_canon_sha256":"3386ecce2da3abed05cf7aaf390f1651e6a25519798a2da764d461d278687570"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:02:18.289904Z","signature_b64":"MIy/AzPDTS74yTOiWCTyW62BosFg50HBtf36pe0+0Xbxl9KZTcZqguBf9yw32BXyvZa6qA6msif3x6U1xqvBAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"0885aa75407b91706e9d65e940ab7dbe3a259334ed215f812f1fa3c08b8663ff","last_reissued_at":"2026-05-20T00:02:18.289362Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:02:18.289362Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"How Many Visual Tokens Do Multimodal Language Models Need? Scaling Visual Token Pruning with F^3A","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CV","authors_text":"Daling Wang, Junzhao Huang, Shi Feng, Xiaocui Yang, Yifei Zhang, Yijie Huang, Yiqun Zhang, Yongkang Liu, Zhuoyue Jia, Zihan Wang","submitted_at":"2026-05-09T13:13:04Z","abstract_excerpt":"Vision-language models improve perception by feeding increasingly long visual token sequences into language backbones, but the resulting inference cost raises a basic scaling question: as multimodal models grow, how many visual tokens are actually needed, and how should they be allocated under a fixed visual token budget? Existing training-free pruning methods typically answer this with one-shot proxies such as decoder attention, visual similarity, or conditional diversity. We argue that visual token pruning is better viewed as task-conditioned evidence search, especially under aggressive comp"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.16359","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/2605.16359/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"ai_meta_artifact","ran_at":"2026-05-19T20:40:55.520349Z","status":"skipped","version":"1.0.0","findings_count":0}],"snapshot_sha256":"5af3bb9a32bf23724a76d334d56a4f4329175dd90c9d67a64bddfa49b1c616d1"},"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":"2605.16359","created_at":"2026-05-20T00:02:18.289443+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.16359v1","created_at":"2026-05-20T00:02:18.289443+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.16359","created_at":"2026-05-20T00:02:18.289443+00:00"},{"alias_kind":"pith_short_12","alias_value":"BCC2U5KAPOIX","created_at":"2026-05-20T00:02:18.289443+00:00"},{"alias_kind":"pith_short_16","alias_value":"BCC2U5KAPOIXA3U5","created_at":"2026-05-20T00:02:18.289443+00:00"},{"alias_kind":"pith_short_8","alias_value":"BCC2U5KA","created_at":"2026-05-20T00:02:18.289443+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/BCC2U5KAPOIXA3U5MXUUBK35XY","json":"https://pith.science/pith/BCC2U5KAPOIXA3U5MXUUBK35XY.json","graph_json":"https://pith.science/api/pith-number/BCC2U5KAPOIXA3U5MXUUBK35XY/graph.json","events_json":"https://pith.science/api/pith-number/BCC2U5KAPOIXA3U5MXUUBK35XY/events.json","paper":"https://pith.science/paper/BCC2U5KA"},"agent_actions":{"view_html":"https://pith.science/pith/BCC2U5KAPOIXA3U5MXUUBK35XY","download_json":"https://pith.science/pith/BCC2U5KAPOIXA3U5MXUUBK35XY.json","view_paper":"https://pith.science/paper/BCC2U5KA","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.16359&json=true","fetch_graph":"https://pith.science/api/pith-number/BCC2U5KAPOIXA3U5MXUUBK35XY/graph.json","fetch_events":"https://pith.science/api/pith-number/BCC2U5KAPOIXA3U5MXUUBK35XY/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/BCC2U5KAPOIXA3U5MXUUBK35XY/action/timestamp_anchor","attest_storage":"https://pith.science/pith/BCC2U5KAPOIXA3U5MXUUBK35XY/action/storage_attestation","attest_author":"https://pith.science/pith/BCC2U5KAPOIXA3U5MXUUBK35XY/action/author_attestation","sign_citation":"https://pith.science/pith/BCC2U5KAPOIXA3U5MXUUBK35XY/action/citation_signature","submit_replication":"https://pith.science/pith/BCC2U5KAPOIXA3U5MXUUBK35XY/action/replication_record"}},"created_at":"2026-05-20T00:02:18.289443+00:00","updated_at":"2026-05-20T00:02:18.289443+00:00"}