{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:Z7J6R26O2WMKTIDAPIVAZTPWWO","short_pith_number":"pith:Z7J6R26O","schema_version":"1.0","canonical_sha256":"cfd3e8ebced598a9a0607a2a0ccdf6b3880c85160ddfe080754e7dfd7ec6e7ba","source":{"kind":"arxiv","id":"2605.00809","version":2},"attestation_state":"computed","paper":{"title":"Let ViT Speak: Generative Language-Image Pre-training","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"A ViT can learn to generate language tokens from visual tokens using only a language modeling objective, aligning it with autoregressive LLMs without contrastive batches or a separate text decoder.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Mengcheng Lan, Qi She, Weixian Lei, Yan Fang, Yao Zhao, Yingchen Yu, Yujie Zhong, Yunchao Wei, Yunqing Zhao, Zilong Huang","submitted_at":"2026-05-01T17:51:38Z","abstract_excerpt":"In this paper, we present \\textbf{Gen}erative \\textbf{L}anguage-\\textbf{I}mage \\textbf{P}re-training (GenLIP), a minimalist generative pretraining framework for Vision Transformers (ViTs) designed for multimodal large language models (MLLMs). To better align vision encoders with the autoregressive nature of LLMs, GenLIP trains a ViT to predict language tokens directly from visual tokens using a standard language modeling objective, without contrastive batch construction or an additional text decoder. This design offers three key advantages: (1) \\textbf{Simplicity}: a single transformer jointly"},"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":"2605.00809","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.CV","submitted_at":"2026-05-01T17:51:38Z","cross_cats_sorted":[],"title_canon_sha256":"928e2a93b338332a1a6ae9c2c6c6ad21678f39ee6c7230cefc2936ba17eeb327","abstract_canon_sha256":"fab2a47d18f8a0fe7e8c6b249310094018ee16cca82076d63b06ce86bd610cfd"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-10T01:10:02.636520Z","signature_b64":"EqUon7kM3cU7hofmnMZO6EnHWCwmEpCrG3E4migsgGhOfYYt9jmJJLo7yr7rkIt5zHYtxp8wNMBl/gkDfB9cDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"cfd3e8ebced598a9a0607a2a0ccdf6b3880c85160ddfe080754e7dfd7ec6e7ba","last_reissued_at":"2026-06-10T01:10:02.635414Z","signature_status":"signed_v1","first_computed_at":"2026-06-10T01:10:02.635414Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Let ViT Speak: Generative Language-Image Pre-training","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"A ViT can learn to generate language tokens from visual tokens using only a language modeling objective, aligning it with autoregressive LLMs without contrastive batches or a separate text decoder.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Mengcheng Lan, Qi She, Weixian Lei, Yan Fang, Yao Zhao, Yingchen Yu, Yujie Zhong, Yunchao Wei, Yunqing Zhao, Zilong Huang","submitted_at":"2026-05-01T17:51:38Z","abstract_excerpt":"In this paper, we present \\textbf{Gen}erative \\textbf{L}anguage-\\textbf{I}mage \\textbf{P}re-training (GenLIP), a minimalist generative pretraining framework for Vision Transformers (ViTs) designed for multimodal large language models (MLLMs). To better align vision encoders with the autoregressive nature of LLMs, GenLIP trains a ViT to predict language tokens directly from visual tokens using a standard language modeling objective, without contrastive batch construction or an additional text decoder. This design offers three key advantages: (1) \\textbf{Simplicity}: a single transformer jointly"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Trained on 8B samples from Recap-DataComp-1B, GenLIP matches or surpasses strong baselines despite using substantially less pretraining data. After continued pretraining on multi-resolution images at native aspect ratios, GenLIP further improves on detail-sensitive tasks such as OCR and chart understanding.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That training a ViT to predict language tokens directly from visual tokens using only a language modeling objective will produce a vision encoder that aligns effectively with autoregressive LLMs without needing contrastive batch construction or an additional text decoder.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"GenLIP pretrains ViTs to generate language tokens from visual tokens via autoregressive language modeling, matching strong baselines on multimodal tasks with less data.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A ViT can learn to generate language tokens from visual tokens using only a language modeling objective, aligning it with autoregressive LLMs without contrastive batches or a separate text decoder.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"37db47e83e491e57c877e035697f97c7aaa32e8fa769b261b6652a58510634ce"},"source":{"id":"2605.00809","kind":"arxiv","version":2},"verdict":{"id":"7e04d2db-b33e-497a-a045-a60eb4c954ca","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-09T18:53:36.754036Z","strongest_claim":"Trained on 8B samples from Recap-DataComp-1B, GenLIP matches or surpasses strong baselines despite using substantially less pretraining data. After continued pretraining on multi-resolution images at native aspect ratios, GenLIP further improves on detail-sensitive tasks such as OCR and chart understanding.","one_line_summary":"GenLIP pretrains ViTs to generate language tokens from visual tokens via autoregressive language modeling, matching strong baselines on multimodal tasks with less data.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That training a ViT to predict language tokens directly from visual tokens using only a language modeling objective will produce a vision encoder that aligns effectively with autoregressive LLMs without needing contrastive batch construction or an additional text decoder.","pith_extraction_headline":"A ViT can learn to generate language tokens from visual tokens using only a language modeling objective, aligning it with autoregressive LLMs without contrastive batches or a separate text decoder."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.00809/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"ai_meta_artifact","ran_at":"2026-05-20T19:33:29.536387Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T17:48:59.422737Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"87f41b401d52c132de956a21acbf104e1d30a05eeeaee098f27c8e0b6531bd5e"},"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":"2605.00809","created_at":"2026-06-10T01:10:02.635587+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.00809v2","created_at":"2026-06-10T01:10:02.635587+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.00809","created_at":"2026-06-10T01:10:02.635587+00:00"},{"alias_kind":"pith_short_12","alias_value":"Z7J6R26O2WMK","created_at":"2026-06-10T01:10:02.635587+00:00"},{"alias_kind":"pith_short_16","alias_value":"Z7J6R26O2WMKTIDA","created_at":"2026-06-10T01:10:02.635587+00:00"},{"alias_kind":"pith_short_8","alias_value":"Z7J6R26O","created_at":"2026-06-10T01:10:02.635587+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/Z7J6R26O2WMKTIDAPIVAZTPWWO","json":"https://pith.science/pith/Z7J6R26O2WMKTIDAPIVAZTPWWO.json","graph_json":"https://pith.science/api/pith-number/Z7J6R26O2WMKTIDAPIVAZTPWWO/graph.json","events_json":"https://pith.science/api/pith-number/Z7J6R26O2WMKTIDAPIVAZTPWWO/events.json","paper":"https://pith.science/paper/Z7J6R26O"},"agent_actions":{"view_html":"https://pith.science/pith/Z7J6R26O2WMKTIDAPIVAZTPWWO","download_json":"https://pith.science/pith/Z7J6R26O2WMKTIDAPIVAZTPWWO.json","view_paper":"https://pith.science/paper/Z7J6R26O","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.00809&json=true","fetch_graph":"https://pith.science/api/pith-number/Z7J6R26O2WMKTIDAPIVAZTPWWO/graph.json","fetch_events":"https://pith.science/api/pith-number/Z7J6R26O2WMKTIDAPIVAZTPWWO/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/Z7J6R26O2WMKTIDAPIVAZTPWWO/action/timestamp_anchor","attest_storage":"https://pith.science/pith/Z7J6R26O2WMKTIDAPIVAZTPWWO/action/storage_attestation","attest_author":"https://pith.science/pith/Z7J6R26O2WMKTIDAPIVAZTPWWO/action/author_attestation","sign_citation":"https://pith.science/pith/Z7J6R26O2WMKTIDAPIVAZTPWWO/action/citation_signature","submit_replication":"https://pith.science/pith/Z7J6R26O2WMKTIDAPIVAZTPWWO/action/replication_record"}},"created_at":"2026-06-10T01:10:02.635587+00:00","updated_at":"2026-06-10T01:10:02.635587+00:00"}