{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:Z6426YSSMXWF4YXGUJGX5XJ5G6","short_pith_number":"pith:Z6426YSS","schema_version":"1.0","canonical_sha256":"cfb9af625265ec5e62e6a24d7edd3d37acf4cddd97b68dd3d774932cb63d3731","source":{"kind":"arxiv","id":"2505.16990","version":2},"attestation_state":"computed","paper":{"title":"Dimple: Discrete Diffusion Multimodal Large Language Model with Parallel Decoding","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Runpeng Yu, Xinchao Wang, Xinyin Ma","submitted_at":"2025-05-22T17:55:04Z","abstract_excerpt":"In this work, we propose Dimple, the first Discrete Diffusion Multimodal Large Language Model (DMLLM). We observe that training with a purely discrete diffusion approach leads to significant training instability, suboptimal performance, and severe length bias issues. To address these challenges, we design a novel training paradigm that combines an initial autoregressive phase with a subsequent diffusion phase. This approach yields the Dimple-7B model, trained on the same dataset and using a similar training pipeline as LLaVA-NEXT. Dimple-7B ultimately surpasses LLaVA-NEXT in performance by 3.9"},"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":"2505.16990","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2025-05-22T17:55:04Z","cross_cats_sorted":[],"title_canon_sha256":"d5d0f869602bb04c1a3c0c2b718530ac74e37adf687f5814d1a03105e0f40341","abstract_canon_sha256":"f51b22e25357f2bd1c1b5a8ef278fb1cf1be555bf4aa60016742bd3763918ff7"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:09:15.507288Z","signature_b64":"EP7V8rt/wMasObaqPTio5YrmESR9gI2fkUtjQnC94DeQqIdE2hfqBqytp865wdS520f5SQe2RzwmlBXN+PAuCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"cfb9af625265ec5e62e6a24d7edd3d37acf4cddd97b68dd3d774932cb63d3731","last_reissued_at":"2026-07-05T11:09:15.506801Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:09:15.506801Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Dimple: Discrete Diffusion Multimodal Large Language Model with Parallel Decoding","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Runpeng Yu, Xinchao Wang, Xinyin Ma","submitted_at":"2025-05-22T17:55:04Z","abstract_excerpt":"In this work, we propose Dimple, the first Discrete Diffusion Multimodal Large Language Model (DMLLM). We observe that training with a purely discrete diffusion approach leads to significant training instability, suboptimal performance, and severe length bias issues. To address these challenges, we design a novel training paradigm that combines an initial autoregressive phase with a subsequent diffusion phase. This approach yields the Dimple-7B model, trained on the same dataset and using a similar training pipeline as LLaVA-NEXT. Dimple-7B ultimately surpasses LLaVA-NEXT in performance by 3.9"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.16990","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/2505.16990/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":"2505.16990","created_at":"2026-07-05T11:09:15.506868+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.16990v2","created_at":"2026-07-05T11:09:15.506868+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.16990","created_at":"2026-07-05T11:09:15.506868+00:00"},{"alias_kind":"pith_short_12","alias_value":"Z6426YSSMXWF","created_at":"2026-07-05T11:09:15.506868+00:00"},{"alias_kind":"pith_short_16","alias_value":"Z6426YSSMXWF4YXG","created_at":"2026-07-05T11:09:15.506868+00:00"},{"alias_kind":"pith_short_8","alias_value":"Z6426YSS","created_at":"2026-07-05T11:09:15.506868+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":20,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.05722","citing_title":"Nemotron-Labs-Diffusion: A Tri-Mode Language Model Unifying Autoregressive, Diffusion, and Self-Speculation Decoding","ref_index":37,"is_internal_anchor":true},{"citing_arxiv_id":"2607.00208","citing_title":"SLIM-RL: Risk-Budgeted Random-Masking RL for Diffusion LLMs Without Trajectory Slicing","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2607.00773","citing_title":"Accelerating Discrete Diffusion Models with Parallel-In-Time Sampling","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2606.02955","citing_title":"Fast-dLLM++: Fr\\'{e}chet Profile Decoding for Faster Diffusion LLM Inference","ref_index":31,"is_internal_anchor":false},{"citing_arxiv_id":"2605.23163","citing_title":"Fast-dDrive: Efficient Block-Diffusion VLM for Autonomous Driving","ref_index":24,"is_internal_anchor":false},{"citing_arxiv_id":"2605.25820","citing_title":"Visual-Redundancy-Controlled Parallel Decoding for Diffusion-Based Multimodal Large Language Models","ref_index":28,"is_internal_anchor":false},{"citing_arxiv_id":"2605.29613","citing_title":"Decoding Strategies for Diffusion-Based ASR: A Systematic Evaluation of Confidence-Based Thresholding","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2605.30876","citing_title":"dMoE: dLLMs with Learnable Block Experts","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2602.18176","citing_title":"Improving Sampling for Masked Diffusion Models via Information Gain","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2605.23163","citing_title":"Fast-dDrive: Efficient Block-Diffusion VLM for Autonomous Driving","ref_index":24,"is_internal_anchor":false},{"citing_arxiv_id":"2604.09450","citing_title":"ECHO: Efficient Chest X-ray Report Generation with One-step Block Diffusion","ref_index":58,"is_internal_anchor":false},{"citing_arxiv_id":"2605.20316","citing_title":"FullFlow: Upgrading Text-to-Image Flow Matching Models for Bidirectional Vision--Language Generation","ref_index":59,"is_internal_anchor":false},{"citing_arxiv_id":"2604.08302","citing_title":"DMax: Aggressive Parallel Decoding for dLLMs","ref_index":96,"is_internal_anchor":false},{"citing_arxiv_id":"2510.06133","citing_title":"CreditDecoding: Accelerating Parallel Decoding in Diffusion Large Language Models with Trace Credit","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06548","citing_title":"Continuous Latent Diffusion Language Model","ref_index":106,"is_internal_anchor":false},{"citing_arxiv_id":"2604.09450","citing_title":"ECHO: Efficient Chest X-ray Report Generation with One-step Block Diffusion","ref_index":58,"is_internal_anchor":false},{"citing_arxiv_id":"2604.08302","citing_title":"DMax: Aggressive Parallel Decoding for dLLMs","ref_index":96,"is_internal_anchor":false},{"citing_arxiv_id":"2604.05497","citing_title":"Thinking Diffusion: Penalize and Guide Visual-Grounded Reasoning in Diffusion Multimodal Language Models","ref_index":43,"is_internal_anchor":false},{"citing_arxiv_id":"2604.16514","citing_title":"BARD: Bridging AutoRegressive and Diffusion Vision-Language Models Via Highly Efficient Progressive Block Merging and Stage-Wise Distillation","ref_index":27,"is_internal_anchor":false},{"citing_arxiv_id":"2604.13413","citing_title":"Dataset-Level Metrics Attenuate Non-Determinism: A Fine-Grained Non-Determinism Evaluation in Diffusion Language Models","ref_index":26,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/Z6426YSSMXWF4YXGUJGX5XJ5G6","json":"https://pith.science/pith/Z6426YSSMXWF4YXGUJGX5XJ5G6.json","graph_json":"https://pith.science/api/pith-number/Z6426YSSMXWF4YXGUJGX5XJ5G6/graph.json","events_json":"https://pith.science/api/pith-number/Z6426YSSMXWF4YXGUJGX5XJ5G6/events.json","paper":"https://pith.science/paper/Z6426YSS"},"agent_actions":{"view_html":"https://pith.science/pith/Z6426YSSMXWF4YXGUJGX5XJ5G6","download_json":"https://pith.science/pith/Z6426YSSMXWF4YXGUJGX5XJ5G6.json","view_paper":"https://pith.science/paper/Z6426YSS","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.16990&json=true","fetch_graph":"https://pith.science/api/pith-number/Z6426YSSMXWF4YXGUJGX5XJ5G6/graph.json","fetch_events":"https://pith.science/api/pith-number/Z6426YSSMXWF4YXGUJGX5XJ5G6/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/Z6426YSSMXWF4YXGUJGX5XJ5G6/action/timestamp_anchor","attest_storage":"https://pith.science/pith/Z6426YSSMXWF4YXGUJGX5XJ5G6/action/storage_attestation","attest_author":"https://pith.science/pith/Z6426YSSMXWF4YXGUJGX5XJ5G6/action/author_attestation","sign_citation":"https://pith.science/pith/Z6426YSSMXWF4YXGUJGX5XJ5G6/action/citation_signature","submit_replication":"https://pith.science/pith/Z6426YSSMXWF4YXGUJGX5XJ5G6/action/replication_record"}},"created_at":"2026-07-05T11:09:15.506868+00:00","updated_at":"2026-07-05T11:09:15.506868+00:00"}