{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:YBXQPUUMWQBMWFUTUJMTAHHGX6","short_pith_number":"pith:YBXQPUUM","schema_version":"1.0","canonical_sha256":"c06f07d28cb402cb1693a259301ce6bf98839b3ac10393407e980d5de4c4cd0b","source":{"kind":"arxiv","id":"2601.07568","version":3},"attestation_state":"computed","paper":{"title":"d3LLM: Ultra-Fast Diffusion LLM using Pseudo-Trajectory Distillation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Hao Zhang, Junda Su, Lanxiang Hu, Peiyuan Zhang, Peng Zhao, Yu-Yang Qian, Zhijie Deng","submitted_at":"2026-01-12T14:25:36Z","abstract_excerpt":"Diffusion large language models (dLLMs) offer capabilities beyond those of autoregressive (AR) LLMs, such as parallel decoding and random-order generation. However, realizing these benefits in practice is non-trivial, as dLLMs inherently face an accuracy-parallelism trade-off. Despite increasing interest, existing methods typically focus on only one-side of the coin, targeting either efficiency or accuracy. To address this limitation, we propose d3LLM (Pseudo-Distilled Diffusion Large Language Model), striking a balance between accuracy and parallelism: (i) during training, we introduce pseudo"},"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":"2601.07568","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-01-12T14:25:36Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"9e02ed4ccff2d125c56cd905a0672f3e6064eda1eacfed51114324351631cf57","abstract_canon_sha256":"b251b269fc104376d37d050dad9028dcdd69558dc9e1179ef51cecd232dcd224"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-08-07T01:36:12.288641Z","signature_b64":"tx25d7b7tOx9RcB7Dbmi1/lD+VpDAScIrXOg4YAKeW8oPgHpu1357+xpikER+PUUfxwG3/WpDiSj5Klj1WUPAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c06f07d28cb402cb1693a259301ce6bf98839b3ac10393407e980d5de4c4cd0b","last_reissued_at":"2026-08-07T01:36:12.286638Z","signature_status":"signed_v1","first_computed_at":"2026-08-07T01:36:12.286638Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"d3LLM: Ultra-Fast Diffusion LLM using Pseudo-Trajectory Distillation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Hao Zhang, Junda Su, Lanxiang Hu, Peiyuan Zhang, Peng Zhao, Yu-Yang Qian, Zhijie Deng","submitted_at":"2026-01-12T14:25:36Z","abstract_excerpt":"Diffusion large language models (dLLMs) offer capabilities beyond those of autoregressive (AR) LLMs, such as parallel decoding and random-order generation. However, realizing these benefits in practice is non-trivial, as dLLMs inherently face an accuracy-parallelism trade-off. Despite increasing interest, existing methods typically focus on only one-side of the coin, targeting either efficiency or accuracy. To address this limitation, we propose d3LLM (Pseudo-Distilled Diffusion Large Language Model), striking a balance between accuracy and parallelism: (i) during training, we introduce pseudo"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2601.07568","kind":"arxiv","version":3},"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/2601.07568/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":"2601.07568","created_at":"2026-08-07T01:36:12.289281+00:00"},{"alias_kind":"arxiv_version","alias_value":"2601.07568v3","created_at":"2026-08-07T01:36:12.289281+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2601.07568","created_at":"2026-08-07T01:36:12.289281+00:00"},{"alias_kind":"pith_short_12","alias_value":"YBXQPUUMWQBM","created_at":"2026-08-07T01:36:12.289281+00:00"},{"alias_kind":"pith_short_16","alias_value":"YBXQPUUMWQBMWFUT","created_at":"2026-08-07T01:36:12.289281+00:00"},{"alias_kind":"pith_short_8","alias_value":"YBXQPUUM","created_at":"2026-08-07T01:36:12.289281+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":15,"internal_anchor_count":15,"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":67,"is_internal_anchor":true},{"citing_arxiv_id":"2606.25331","citing_title":"Improved Large Language Diffusion Models","ref_index":31,"is_internal_anchor":true},{"citing_arxiv_id":"2606.19534","citing_title":"PerceptionDLM: Parallel Region Perception with Multimodal Diffusion Language Models","ref_index":35,"is_internal_anchor":true},{"citing_arxiv_id":"2606.18394","citing_title":"JetSpec: Breaking the Scaling Ceiling of Speculative Decoding with Parallel Tree Drafting","ref_index":35,"is_internal_anchor":true},{"citing_arxiv_id":"2606.18195","citing_title":"Learning from the Self-future: On-policy Self-distillation for dLLMs","ref_index":20,"is_internal_anchor":true},{"citing_arxiv_id":"2606.04236","citing_title":"Supportive Token Revealing for Fast Diffusion Language Model Decoding","ref_index":5,"is_internal_anchor":true},{"citing_arxiv_id":"2606.29215","citing_title":"Multi-Block Diffusion Language Models","ref_index":41,"is_internal_anchor":true},{"citing_arxiv_id":"2606.29215","citing_title":"Multi-Block Diffusion Language Models","ref_index":41,"is_internal_anchor":true},{"citing_arxiv_id":"2606.08411","citing_title":"AsyncLane: Decoupling Refinement from Advancement in Diffusion Language Model Decoding","ref_index":15,"is_internal_anchor":true},{"citing_arxiv_id":"2602.08404","citing_title":"TEAM: Temporal-Spatial Consistency Guided Expert Activation for MoE Diffusion Language Model Acceleration","ref_index":18,"is_internal_anchor":true},{"citing_arxiv_id":"2604.09450","citing_title":"ECHO: Efficient Chest X-ray Report Generation with One-step Block Diffusion","ref_index":36,"is_internal_anchor":true},{"citing_arxiv_id":"2604.08302","citing_title":"DMax: Aggressive Parallel Decoding for dLLMs","ref_index":62,"is_internal_anchor":true},{"citing_arxiv_id":"2605.09536","citing_title":"TAD: Temporal-Aware Trajectory Self-Distillation for Fast and Accurate Diffusion LLM","ref_index":9,"is_internal_anchor":true},{"citing_arxiv_id":"2604.09450","citing_title":"ECHO: Efficient Chest X-ray Report Generation with One-step Block Diffusion","ref_index":36,"is_internal_anchor":true},{"citing_arxiv_id":"2604.08302","citing_title":"DMax: Aggressive Parallel Decoding for dLLMs","ref_index":62,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/YBXQPUUMWQBMWFUTUJMTAHHGX6","json":"https://pith.science/pith/YBXQPUUMWQBMWFUTUJMTAHHGX6.json","graph_json":"https://pith.science/api/pith-number/YBXQPUUMWQBMWFUTUJMTAHHGX6/graph.json","events_json":"https://pith.science/api/pith-number/YBXQPUUMWQBMWFUTUJMTAHHGX6/events.json","paper":"https://pith.science/paper/YBXQPUUM"},"agent_actions":{"view_html":"https://pith.science/pith/YBXQPUUMWQBMWFUTUJMTAHHGX6","download_json":"https://pith.science/pith/YBXQPUUMWQBMWFUTUJMTAHHGX6.json","view_paper":"https://pith.science/paper/YBXQPUUM","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2601.07568&json=true","fetch_graph":"https://pith.science/api/pith-number/YBXQPUUMWQBMWFUTUJMTAHHGX6/graph.json","fetch_events":"https://pith.science/api/pith-number/YBXQPUUMWQBMWFUTUJMTAHHGX6/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/YBXQPUUMWQBMWFUTUJMTAHHGX6/action/timestamp_anchor","attest_storage":"https://pith.science/pith/YBXQPUUMWQBMWFUTUJMTAHHGX6/action/storage_attestation","attest_author":"https://pith.science/pith/YBXQPUUMWQBMWFUTUJMTAHHGX6/action/author_attestation","sign_citation":"https://pith.science/pith/YBXQPUUMWQBMWFUTUJMTAHHGX6/action/citation_signature","submit_replication":"https://pith.science/pith/YBXQPUUMWQBMWFUTUJMTAHHGX6/action/replication_record"}},"created_at":"2026-08-07T01:36:12.289281+00:00","updated_at":"2026-08-07T01:36:12.289281+00:00"}