{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2025:YCLR3EWRQC3Z5OXPQNXZ3G3AKT","short_pith_number":"pith:YCLR3EWR","canonical_record":{"source":{"id":"2505.16933","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-05-22T17:23:26Z","cross_cats_sorted":["cs.CL","cs.CV"],"title_canon_sha256":"e8c43c6001fae5acb93f23d47e5167dbd812ec4193d9a2e1a3eecee7ca5e9ba6","abstract_canon_sha256":"66a41245fd5d30b449e053aa30fd84a10706e1a55f20b6a50d301cd10e4cedd0"},"schema_version":"1.0"},"canonical_sha256":"c0971d92d180b79ebaef836f9d9b6054dbbe9013c89c383f5479306b6d1cfd0a","source":{"kind":"arxiv","id":"2505.16933","version":2},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2505.16933","created_at":"2026-05-17T23:38:15Z"},{"alias_kind":"arxiv_version","alias_value":"2505.16933v2","created_at":"2026-05-17T23:38:15Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.16933","created_at":"2026-05-17T23:38:15Z"},{"alias_kind":"pith_short_12","alias_value":"YCLR3EWRQC3Z","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"YCLR3EWRQC3Z5OXP","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"YCLR3EWR","created_at":"2026-05-18T12:33:37Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2025:YCLR3EWRQC3Z5OXPQNXZ3G3AKT","target":"record","payload":{"canonical_record":{"source":{"id":"2505.16933","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-05-22T17:23:26Z","cross_cats_sorted":["cs.CL","cs.CV"],"title_canon_sha256":"e8c43c6001fae5acb93f23d47e5167dbd812ec4193d9a2e1a3eecee7ca5e9ba6","abstract_canon_sha256":"66a41245fd5d30b449e053aa30fd84a10706e1a55f20b6a50d301cd10e4cedd0"},"schema_version":"1.0"},"canonical_sha256":"c0971d92d180b79ebaef836f9d9b6054dbbe9013c89c383f5479306b6d1cfd0a","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-17T23:38:15.225440Z","signature_b64":"B8iFByNSK6Bh+ep9dDVLEj9IIaLhI8udxAyYqHdcsdI3MzNOGbuTKhVgVF5wdC+8zWFsluvvOabpD5h0ZvU1Cg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c0971d92d180b79ebaef836f9d9b6054dbbe9013c89c383f5479306b6d1cfd0a","last_reissued_at":"2026-05-17T23:38:15.224793Z","signature_status":"signed_v1","first_computed_at":"2026-05-17T23:38:15.224793Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2505.16933","source_version":2,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T23:38:15Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"Zq0oZU2lwNV8cX0HCLqsXFeED783uX6AQfsNVBwGAY8aoBIzOzxkte5SC9NMrCFtSl90tBYBiHd+wp9KvYXpDQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-31T04:30:48.069591Z"},"content_sha256":"d8053471fcfda73428e256081c04cadecbf6ff03a29589755880295a00c01286","schema_version":"1.0","event_id":"sha256:d8053471fcfda73428e256081c04cadecbf6ff03a29589755880295a00c01286"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2025:YCLR3EWRQC3Z5OXPQNXZ3G3AKT","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"LLaDA-V: Large Language Diffusion Models with Visual Instruction Tuning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"A purely diffusion-based multimodal model matches autoregressive leaders on visual instruction tasks by adding a vision encoder to a language diffusion backbone.","cross_cats":["cs.CL","cs.CV"],"primary_cat":"cs.LG","authors_text":"Chongxuan Li, Ji-Rong Wen, Jun Hu, Jun Zhou, Shen Nie, Xiaolu Zhang, Zebin You, Zhiwu Lu","submitted_at":"2025-05-22T17:23:26Z","abstract_excerpt":"In this work, we introduce LLaDA-V, a purely diffusion-based Multimodal Large Language Model (MLLM) that integrates visual instruction tuning with masked diffusion models, representing a departure from the autoregressive paradigms dominant in current multimodal approaches. Built upon LLaDA, a representative large language diffusion model, LLaDA-V incorporates a vision encoder and MLP connector that projects visual features into the language embedding space, enabling effective multimodal alignment. Our empirical investigation reveals several intriguing results: First, LLaDA-V demonstrates promi"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"LLaDA-V achieves state-of-the-art performance in multimodal understanding compared to existing hybrid autoregressive-diffusion and purely diffusion-based MLLMs.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the empirical gains observed on the chosen instruction-tuning datasets and evaluation benchmarks will generalize beyond the specific training mixture and that the diffusion process can maintain coherent multimodal alignment without the sequential constraints of autoregressive decoding.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"LLaDA-V is a diffusion-based multimodal large language model that reaches competitive or state-of-the-art results on visual instruction tasks while using a non-autoregressive architecture.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A purely diffusion-based multimodal model matches autoregressive leaders on visual instruction tasks by adding a vision encoder to a language diffusion backbone.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"178de71c06e035a4fa34cdeaba118e17706af59e826ca0788899b4b35f6ad09e"},"source":{"id":"2505.16933","kind":"arxiv","version":2},"verdict":{"id":"c7fa6077-f88a-47bc-b507-3c322af454f1","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-17T03:40:53.041337Z","strongest_claim":"LLaDA-V achieves state-of-the-art performance in multimodal understanding compared to existing hybrid autoregressive-diffusion and purely diffusion-based MLLMs.","one_line_summary":"LLaDA-V is a diffusion-based multimodal large language model that reaches competitive or state-of-the-art results on visual instruction tasks while using a non-autoregressive architecture.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the empirical gains observed on the chosen instruction-tuning datasets and evaluation benchmarks will generalize beyond the specific training mixture and that the diffusion process can maintain coherent multimodal alignment without the sequential constraints of autoregressive decoding.","pith_extraction_headline":"A purely diffusion-based multimodal model matches autoregressive leaders on visual instruction tasks by adding a vision encoder to a language diffusion backbone."},"references":{"count":125,"sample":[{"doi":"","year":2023,"title":"Visual instruction tuning","work_id":"4622b96e-3148-4c4d-812a-c4afdea5f469","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Improved baselines with visual instruction tuning,","work_id":"d7798e63-66b9-4475-9452-7f723907e04e","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"LLaVA-OneVision: Easy Visual Task Transfer","work_id":"f5f2452b-f2a9-49ac-b38d-c76e18cdfe49","ref_index":3,"cited_arxiv_id":"2408.03326","is_internal_anchor":true},{"doi":"","year":2024,"title":"Internvl: Scaling up vision foundation models and aligning for generic visual-linguistic tasks,","work_id":"459e59dc-9bb6-4e5b-b2da-ec1327ff8249","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolution","work_id":"8abcfe4f-e0fb-44b7-9123-448fac95f90a","ref_index":5,"cited_arxiv_id":"2409.12191","is_internal_anchor":true}],"resolved_work":125,"snapshot_sha256":"2c03941b0ba0921dd44230e5355b20e99aac880759222e3224f81e32c8551e2b","internal_anchors":39},"formal_canon":{"evidence_count":2,"snapshot_sha256":"e5da34cd29f822726fc7b5cdb630077db695dbdc19388f4f96257f1b82e62863"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"c7fa6077-f88a-47bc-b507-3c322af454f1"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T23:38:15Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"RP96tZ5URrf6u7mJjXX7eNHL6FrYIrfSWb1PZG4yqYM/EEFp3aTndwVsxZ7sBg1BZMP2SsnnbDZO1+NuXtPjAQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-31T04:30:48.070118Z"},"content_sha256":"af6a0f0165258030cc3eeb51b5e404a20010a88138084fff88b7423789b111d3","schema_version":"1.0","event_id":"sha256:af6a0f0165258030cc3eeb51b5e404a20010a88138084fff88b7423789b111d3"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/YCLR3EWRQC3Z5OXPQNXZ3G3AKT/bundle.json","state_url":"https://pith.science/pith/YCLR3EWRQC3Z5OXPQNXZ3G3AKT/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/YCLR3EWRQC3Z5OXPQNXZ3G3AKT/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-05-31T04:30:48Z","links":{"resolver":"https://pith.science/pith/YCLR3EWRQC3Z5OXPQNXZ3G3AKT","bundle":"https://pith.science/pith/YCLR3EWRQC3Z5OXPQNXZ3G3AKT/bundle.json","state":"https://pith.science/pith/YCLR3EWRQC3Z5OXPQNXZ3G3AKT/state.json","well_known_bundle":"https://pith.science/.well-known/pith/YCLR3EWRQC3Z5OXPQNXZ3G3AKT/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2025:YCLR3EWRQC3Z5OXPQNXZ3G3AKT","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"66a41245fd5d30b449e053aa30fd84a10706e1a55f20b6a50d301cd10e4cedd0","cross_cats_sorted":["cs.CL","cs.CV"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-05-22T17:23:26Z","title_canon_sha256":"e8c43c6001fae5acb93f23d47e5167dbd812ec4193d9a2e1a3eecee7ca5e9ba6"},"schema_version":"1.0","source":{"id":"2505.16933","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2505.16933","created_at":"2026-05-17T23:38:15Z"},{"alias_kind":"arxiv_version","alias_value":"2505.16933v2","created_at":"2026-05-17T23:38:15Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.16933","created_at":"2026-05-17T23:38:15Z"},{"alias_kind":"pith_short_12","alias_value":"YCLR3EWRQC3Z","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"YCLR3EWRQC3Z5OXP","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"YCLR3EWR","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:af6a0f0165258030cc3eeb51b5e404a20010a88138084fff88b7423789b111d3","target":"graph","created_at":"2026-05-17T23:38:15Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"LLaDA-V achieves state-of-the-art performance in multimodal understanding compared to existing hybrid autoregressive-diffusion and purely diffusion-based MLLMs."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the empirical gains observed on the chosen instruction-tuning datasets and evaluation benchmarks will generalize beyond the specific training mixture and that the diffusion process can maintain coherent multimodal alignment without the sequential constraints of autoregressive decoding."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"LLaDA-V is a diffusion-based multimodal large language model that reaches competitive or state-of-the-art results on visual instruction tasks while using a non-autoregressive architecture."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"A purely diffusion-based multimodal model matches autoregressive leaders on visual instruction tasks by adding a vision encoder to a language diffusion backbone."}],"snapshot_sha256":"178de71c06e035a4fa34cdeaba118e17706af59e826ca0788899b4b35f6ad09e"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"e5da34cd29f822726fc7b5cdb630077db695dbdc19388f4f96257f1b82e62863"},"paper":{"abstract_excerpt":"In this work, we introduce LLaDA-V, a purely diffusion-based Multimodal Large Language Model (MLLM) that integrates visual instruction tuning with masked diffusion models, representing a departure from the autoregressive paradigms dominant in current multimodal approaches. Built upon LLaDA, a representative large language diffusion model, LLaDA-V incorporates a vision encoder and MLP connector that projects visual features into the language embedding space, enabling effective multimodal alignment. Our empirical investigation reveals several intriguing results: First, LLaDA-V demonstrates promi","authors_text":"Chongxuan Li, Ji-Rong Wen, Jun Hu, Jun Zhou, Shen Nie, Xiaolu Zhang, Zebin You, Zhiwu Lu","cross_cats":["cs.CL","cs.CV"],"headline":"A purely diffusion-based multimodal model matches autoregressive leaders on visual instruction tasks by adding a vision encoder to a language diffusion backbone.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-05-22T17:23:26Z","title":"LLaDA-V: Large Language Diffusion Models with Visual Instruction Tuning"},"references":{"count":125,"internal_anchors":39,"resolved_work":125,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":1,"title":"Visual instruction tuning","work_id":"4622b96e-3148-4c4d-812a-c4afdea5f469","year":2023},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"Improved baselines with visual instruction tuning,","work_id":"d7798e63-66b9-4475-9452-7f723907e04e","year":2024},{"cited_arxiv_id":"2408.03326","doi":"","is_internal_anchor":true,"ref_index":3,"title":"LLaVA-OneVision: Easy Visual Task Transfer","work_id":"f5f2452b-f2a9-49ac-b38d-c76e18cdfe49","year":2024},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"Internvl: Scaling up vision foundation models and aligning for generic visual-linguistic tasks,","work_id":"459e59dc-9bb6-4e5b-b2da-ec1327ff8249","year":2024},{"cited_arxiv_id":"2409.12191","doi":"","is_internal_anchor":true,"ref_index":5,"title":"Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolution","work_id":"8abcfe4f-e0fb-44b7-9123-448fac95f90a","year":2024}],"snapshot_sha256":"2c03941b0ba0921dd44230e5355b20e99aac880759222e3224f81e32c8551e2b"},"source":{"id":"2505.16933","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-17T03:40:53.041337Z","id":"c7fa6077-f88a-47bc-b507-3c322af454f1","model_set":{"reader":"grok-4.3"},"one_line_summary":"LLaDA-V is a diffusion-based multimodal large language model that reaches competitive or state-of-the-art results on visual instruction tasks while using a non-autoregressive architecture.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"A purely diffusion-based multimodal model matches autoregressive leaders on visual instruction tasks by adding a vision encoder to a language diffusion backbone.","strongest_claim":"LLaDA-V achieves state-of-the-art performance in multimodal understanding compared to existing hybrid autoregressive-diffusion and purely diffusion-based MLLMs.","weakest_assumption":"That the empirical gains observed on the chosen instruction-tuning datasets and evaluation benchmarks will generalize beyond the specific training mixture and that the diffusion process can maintain coherent multimodal alignment without the sequential constraints of autoregressive decoding."}},"verdict_id":"c7fa6077-f88a-47bc-b507-3c322af454f1"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:d8053471fcfda73428e256081c04cadecbf6ff03a29589755880295a00c01286","target":"record","created_at":"2026-05-17T23:38:15Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"66a41245fd5d30b449e053aa30fd84a10706e1a55f20b6a50d301cd10e4cedd0","cross_cats_sorted":["cs.CL","cs.CV"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-05-22T17:23:26Z","title_canon_sha256":"e8c43c6001fae5acb93f23d47e5167dbd812ec4193d9a2e1a3eecee7ca5e9ba6"},"schema_version":"1.0","source":{"id":"2505.16933","kind":"arxiv","version":2}},"canonical_sha256":"c0971d92d180b79ebaef836f9d9b6054dbbe9013c89c383f5479306b6d1cfd0a","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"c0971d92d180b79ebaef836f9d9b6054dbbe9013c89c383f5479306b6d1cfd0a","first_computed_at":"2026-05-17T23:38:15.224793Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-17T23:38:15.224793Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"B8iFByNSK6Bh+ep9dDVLEj9IIaLhI8udxAyYqHdcsdI3MzNOGbuTKhVgVF5wdC+8zWFsluvvOabpD5h0ZvU1Cg==","signature_status":"signed_v1","signed_at":"2026-05-17T23:38:15.225440Z","signed_message":"canonical_sha256_bytes"},"source_id":"2505.16933","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:d8053471fcfda73428e256081c04cadecbf6ff03a29589755880295a00c01286","sha256:af6a0f0165258030cc3eeb51b5e404a20010a88138084fff88b7423789b111d3"],"state_sha256":"9490a08f65a0f39bea2013d54a1750a3aecfd62b2920a4daef76e3d285d2e4fc"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"Ud4nm9SDR6//QIWsdIIXi7bPun0MyHt1k/yKQP+ZwJhZs8flq73lUistSl0DDng5PYYT6YJPqGjYYIPC2SDwCg==","signed_message":"bundle_sha256_bytes","signed_at":"2026-05-31T04:30:48.072544Z","bundle_sha256":"b685e3daaf0cf5adfe9087c679d5073c2e1fcd33a3d238672e7bb782d8509d81"}}