{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:NPD6UFJ2AKI3UWWHFCQ6ESE7XK","short_pith_number":"pith:NPD6UFJ2","schema_version":"1.0","canonical_sha256":"6bc7ea153a0291ba5ac728a1e2489fba90b4aa35b1ff1cbb461d0a1b92955dd1","source":{"kind":"arxiv","id":"2503.22655","version":2},"attestation_state":"computed","paper":{"title":"Text-Only Data Synthesis for Vision Language Model Training","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":["cs.CV","cs.MM"],"primary_cat":"cs.AI","authors_text":"Chengwei Qin, Hui Xiong, Wenjie Zhang, Xiaomin Yu, Ziyue Qiao","submitted_at":"2025-03-28T17:43:00Z","abstract_excerpt":"Training vision-language models (VLMs) typically requires large-scale, high-quality image-text pairs, but collecting or synthesizing such data is costly. In contrast, text data is abundant and inexpensive, prompting the question: can high-quality multimodal training data be synthesized purely from text? To tackle this, we propose a cross-integrated three-stage multimodal data synthesis framework, which generates two datasets: Unicorn-1.2M and Unicorn-471K-Instruction. In Stage 1: Diverse Caption Data Synthesis, we construct 1.2M semantically diverse high-quality captions by expanding sparse ca"},"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":"2503.22655","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.AI","submitted_at":"2025-03-28T17:43:00Z","cross_cats_sorted":["cs.CV","cs.MM"],"title_canon_sha256":"ab74feb25373d63b5e078395d5c6abbffe354264deabdd289fa92b2b485ee5c7","abstract_canon_sha256":"c6906c2e6d60c142b904ee8cac3c92dffcec7af526444dcad4e1fa401a4786a9"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-28T01:04:27.516419Z","signature_b64":"1jUin5C5UKTskBuXacH/q4Jq8EDpdiqKeoQ+2snngJnnAxeC50qRhyCyWgn1iln+Rd7DJhJTac76q0swnXPjAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"6bc7ea153a0291ba5ac728a1e2489fba90b4aa35b1ff1cbb461d0a1b92955dd1","last_reissued_at":"2026-05-28T01:04:27.514655Z","signature_status":"signed_v1","first_computed_at":"2026-05-28T01:04:27.514655Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Text-Only Data Synthesis for Vision Language Model Training","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":["cs.CV","cs.MM"],"primary_cat":"cs.AI","authors_text":"Chengwei Qin, Hui Xiong, Wenjie Zhang, Xiaomin Yu, Ziyue Qiao","submitted_at":"2025-03-28T17:43:00Z","abstract_excerpt":"Training vision-language models (VLMs) typically requires large-scale, high-quality image-text pairs, but collecting or synthesizing such data is costly. In contrast, text data is abundant and inexpensive, prompting the question: can high-quality multimodal training data be synthesized purely from text? To tackle this, we propose a cross-integrated three-stage multimodal data synthesis framework, which generates two datasets: Unicorn-1.2M and Unicorn-471K-Instruction. In Stage 1: Diverse Caption Data Synthesis, we construct 1.2M semantically diverse high-quality captions by expanding sparse ca"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.22655","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/2503.22655/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":"2503.22655","created_at":"2026-05-28T01:04:27.514735+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.22655v2","created_at":"2026-05-28T01:04:27.514735+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.22655","created_at":"2026-05-28T01:04:27.514735+00:00"},{"alias_kind":"pith_short_12","alias_value":"NPD6UFJ2AKI3","created_at":"2026-05-28T01:04:27.514735+00:00"},{"alias_kind":"pith_short_16","alias_value":"NPD6UFJ2AKI3UWWH","created_at":"2026-05-28T01:04:27.514735+00:00"},{"alias_kind":"pith_short_8","alias_value":"NPD6UFJ2","created_at":"2026-05-28T01:04:27.514735+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":4,"sample":[{"citing_arxiv_id":"2602.07026","citing_title":"Modality Gap-Driven Subspace Alignment Training Paradigm For Multimodal Large Language Models","ref_index":36,"is_internal_anchor":true},{"citing_arxiv_id":"2605.11405","citing_title":"20/20 Vision Language Models: A Prescription for Better VLMs through Data Curation Alone","ref_index":23,"is_internal_anchor":true},{"citing_arxiv_id":"2605.11405","citing_title":"20/20 Vision Language Models: A Prescription for Better VLMs through Data Curation Alone","ref_index":23,"is_internal_anchor":true},{"citing_arxiv_id":"2605.07825","citing_title":"Anisotropic Modality Align","ref_index":16,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/NPD6UFJ2AKI3UWWHFCQ6ESE7XK","json":"https://pith.science/pith/NPD6UFJ2AKI3UWWHFCQ6ESE7XK.json","graph_json":"https://pith.science/api/pith-number/NPD6UFJ2AKI3UWWHFCQ6ESE7XK/graph.json","events_json":"https://pith.science/api/pith-number/NPD6UFJ2AKI3UWWHFCQ6ESE7XK/events.json","paper":"https://pith.science/paper/NPD6UFJ2"},"agent_actions":{"view_html":"https://pith.science/pith/NPD6UFJ2AKI3UWWHFCQ6ESE7XK","download_json":"https://pith.science/pith/NPD6UFJ2AKI3UWWHFCQ6ESE7XK.json","view_paper":"https://pith.science/paper/NPD6UFJ2","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.22655&json=true","fetch_graph":"https://pith.science/api/pith-number/NPD6UFJ2AKI3UWWHFCQ6ESE7XK/graph.json","fetch_events":"https://pith.science/api/pith-number/NPD6UFJ2AKI3UWWHFCQ6ESE7XK/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/NPD6UFJ2AKI3UWWHFCQ6ESE7XK/action/timestamp_anchor","attest_storage":"https://pith.science/pith/NPD6UFJ2AKI3UWWHFCQ6ESE7XK/action/storage_attestation","attest_author":"https://pith.science/pith/NPD6UFJ2AKI3UWWHFCQ6ESE7XK/action/author_attestation","sign_citation":"https://pith.science/pith/NPD6UFJ2AKI3UWWHFCQ6ESE7XK/action/citation_signature","submit_replication":"https://pith.science/pith/NPD6UFJ2AKI3UWWHFCQ6ESE7XK/action/replication_record"}},"created_at":"2026-05-28T01:04:27.514735+00:00","updated_at":"2026-05-28T01:04:27.514735+00:00"}