{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:EKUNAA7XSMP6DPLBATEPOC6YUO","short_pith_number":"pith:EKUNAA7X","schema_version":"1.0","canonical_sha256":"22a8d003f7931fe1bd6104c8f70bd8a38cf9f5bf39604a5d0c845c88e466b01b","source":{"kind":"arxiv","id":"2607.16872","version":1},"attestation_state":"computed","paper":{"title":"Trace-Based On-Policy Distillation for Masked Diffusion Language Models","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Chenhao Yuan, Haolin Ren, Jun Zhao, Kang Liu, Ziyang Huang","submitted_at":"2026-07-18T16:25:17Z","abstract_excerpt":"Diffusion large language models (dLLMs) are a promising alternative to autoregressive generation. However, reasoning-oriented post-training for dLLMs remains challenging. Supervised fine-tuning (SFT) for dLLMs requires dense but often off-policy masked states, while reinforcement learning (RL) relies on sparse rewards or value modeling. This paper proposes \\textbf{trace-based on-policy distillation (TOPD)}, a teacher-supervised framework that transfers reasoning ability to a target dLLM without reward estimation. The key idea is to supervise a dLLM on its own denoising trajectory, focusing on "},"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":"2607.16872","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.CL","submitted_at":"2026-07-18T16:25:17Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"96fb1b47eb3bd29b225a2e014114a8d3016bd7797c16c55175e71c738b90d788","abstract_canon_sha256":"01d051f9d1613e568b239d88ececbc543af80e5c7b8e166cee42be87487d0335"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-21T01:21:03.282727Z","signature_b64":"qEiH9/WoDcC1+5MEmdjRJAk6naEi9LMS6MgYVktQ0blmBF2wrTuXzZQBYzpFsRVtHGstW0+/yQzVGZvMKx12BA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"22a8d003f7931fe1bd6104c8f70bd8a38cf9f5bf39604a5d0c845c88e466b01b","last_reissued_at":"2026-07-21T01:21:03.281832Z","signature_status":"signed_v1","first_computed_at":"2026-07-21T01:21:03.281832Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Trace-Based On-Policy Distillation for Masked Diffusion Language Models","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Chenhao Yuan, Haolin Ren, Jun Zhao, Kang Liu, Ziyang Huang","submitted_at":"2026-07-18T16:25:17Z","abstract_excerpt":"Diffusion large language models (dLLMs) are a promising alternative to autoregressive generation. However, reasoning-oriented post-training for dLLMs remains challenging. Supervised fine-tuning (SFT) for dLLMs requires dense but often off-policy masked states, while reinforcement learning (RL) relies on sparse rewards or value modeling. This paper proposes \\textbf{trace-based on-policy distillation (TOPD)}, a teacher-supervised framework that transfers reasoning ability to a target dLLM without reward estimation. The key idea is to supervise a dLLM on its own denoising trajectory, focusing on "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2607.16872","kind":"arxiv","version":1},"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/2607.16872/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":"2607.16872","created_at":"2026-07-21T01:21:03.282297+00:00"},{"alias_kind":"arxiv_version","alias_value":"2607.16872v1","created_at":"2026-07-21T01:21:03.282297+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2607.16872","created_at":"2026-07-21T01:21:03.282297+00:00"},{"alias_kind":"pith_short_12","alias_value":"EKUNAA7XSMP6","created_at":"2026-07-21T01:21:03.282297+00:00"},{"alias_kind":"pith_short_16","alias_value":"EKUNAA7XSMP6DPLB","created_at":"2026-07-21T01:21:03.282297+00:00"},{"alias_kind":"pith_short_8","alias_value":"EKUNAA7X","created_at":"2026-07-21T01:21:03.282297+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/EKUNAA7XSMP6DPLBATEPOC6YUO","json":"https://pith.science/pith/EKUNAA7XSMP6DPLBATEPOC6YUO.json","graph_json":"https://pith.science/api/pith-number/EKUNAA7XSMP6DPLBATEPOC6YUO/graph.json","events_json":"https://pith.science/api/pith-number/EKUNAA7XSMP6DPLBATEPOC6YUO/events.json","paper":"https://pith.science/paper/EKUNAA7X"},"agent_actions":{"view_html":"https://pith.science/pith/EKUNAA7XSMP6DPLBATEPOC6YUO","download_json":"https://pith.science/pith/EKUNAA7XSMP6DPLBATEPOC6YUO.json","view_paper":"https://pith.science/paper/EKUNAA7X","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2607.16872&json=true","fetch_graph":"https://pith.science/api/pith-number/EKUNAA7XSMP6DPLBATEPOC6YUO/graph.json","fetch_events":"https://pith.science/api/pith-number/EKUNAA7XSMP6DPLBATEPOC6YUO/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/EKUNAA7XSMP6DPLBATEPOC6YUO/action/timestamp_anchor","attest_storage":"https://pith.science/pith/EKUNAA7XSMP6DPLBATEPOC6YUO/action/storage_attestation","attest_author":"https://pith.science/pith/EKUNAA7XSMP6DPLBATEPOC6YUO/action/author_attestation","sign_citation":"https://pith.science/pith/EKUNAA7XSMP6DPLBATEPOC6YUO/action/citation_signature","submit_replication":"https://pith.science/pith/EKUNAA7XSMP6DPLBATEPOC6YUO/action/replication_record"}},"created_at":"2026-07-21T01:21:03.282297+00:00","updated_at":"2026-07-21T01:21:03.282297+00:00"}