{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:3VRARFE3CFBERUCMJGDAWSCMAE","short_pith_number":"pith:3VRARFE3","schema_version":"1.0","canonical_sha256":"dd6208949b114248d04c49860b484c0111cb4c4846d85c7e209cecadf9d5ae6e","source":{"kind":"arxiv","id":"2606.01249","version":1},"attestation_state":"computed","paper":{"title":"Trust Region On-Policy Distillation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Boyan Gao, Haoqing Wang, Xingrun Xing, Yehui Tang, Ziheng Li","submitted_at":"2026-05-31T14:04:51Z","abstract_excerpt":"On-Policy Distillation (OPD) is a fundamental technique for efficient post-training of large language models (LLMs), with broad applications in agent learning, multi-task enhancement, and model compression. However, OPD training becomes unstable when the teacher and student distributions differ substantially, as teacher supervision on student-generated tokens may yield unreliable policy gradients and even cause optimization failure. This work addresses reliable on-policy token-level supervision through credit assignment strategies, and proposes Trust Region On-Policy Distillation, TrOPD. It fe"},"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":"2606.01249","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-05-31T14:04:51Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"7576a155ff7a288ed2816c16e8f23259c1e8c413a4bf15148062033f0e745ce0","abstract_canon_sha256":"e0993e93880fcb67cc0585c716e8e99f3b59f3586757b809c78bd58bfc42ad3a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-02T02:04:27.930453Z","signature_b64":"bd+tL7xddXkYBrkVSODBUpNWwesxnIKHmv37GAI1aQmhBscJFoDptG/TDZQSV6PBgiCJkwvwnv3vRBZ0FBzyCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"dd6208949b114248d04c49860b484c0111cb4c4846d85c7e209cecadf9d5ae6e","last_reissued_at":"2026-06-02T02:04:27.930021Z","signature_status":"signed_v1","first_computed_at":"2026-06-02T02:04:27.930021Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Trust Region On-Policy Distillation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Boyan Gao, Haoqing Wang, Xingrun Xing, Yehui Tang, Ziheng Li","submitted_at":"2026-05-31T14:04:51Z","abstract_excerpt":"On-Policy Distillation (OPD) is a fundamental technique for efficient post-training of large language models (LLMs), with broad applications in agent learning, multi-task enhancement, and model compression. However, OPD training becomes unstable when the teacher and student distributions differ substantially, as teacher supervision on student-generated tokens may yield unreliable policy gradients and even cause optimization failure. This work addresses reliable on-policy token-level supervision through credit assignment strategies, and proposes Trust Region On-Policy Distillation, TrOPD. It fe"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.01249","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/2606.01249/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":"2606.01249","created_at":"2026-06-02T02:04:27.930078+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.01249v1","created_at":"2026-06-02T02:04:27.930078+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.01249","created_at":"2026-06-02T02:04:27.930078+00:00"},{"alias_kind":"pith_short_12","alias_value":"3VRARFE3CFBE","created_at":"2026-06-02T02:04:27.930078+00:00"},{"alias_kind":"pith_short_16","alias_value":"3VRARFE3CFBERUCM","created_at":"2026-06-02T02:04:27.930078+00:00"},{"alias_kind":"pith_short_8","alias_value":"3VRARFE3","created_at":"2026-06-02T02:04:27.930078+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/3VRARFE3CFBERUCMJGDAWSCMAE","json":"https://pith.science/pith/3VRARFE3CFBERUCMJGDAWSCMAE.json","graph_json":"https://pith.science/api/pith-number/3VRARFE3CFBERUCMJGDAWSCMAE/graph.json","events_json":"https://pith.science/api/pith-number/3VRARFE3CFBERUCMJGDAWSCMAE/events.json","paper":"https://pith.science/paper/3VRARFE3"},"agent_actions":{"view_html":"https://pith.science/pith/3VRARFE3CFBERUCMJGDAWSCMAE","download_json":"https://pith.science/pith/3VRARFE3CFBERUCMJGDAWSCMAE.json","view_paper":"https://pith.science/paper/3VRARFE3","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.01249&json=true","fetch_graph":"https://pith.science/api/pith-number/3VRARFE3CFBERUCMJGDAWSCMAE/graph.json","fetch_events":"https://pith.science/api/pith-number/3VRARFE3CFBERUCMJGDAWSCMAE/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/3VRARFE3CFBERUCMJGDAWSCMAE/action/timestamp_anchor","attest_storage":"https://pith.science/pith/3VRARFE3CFBERUCMJGDAWSCMAE/action/storage_attestation","attest_author":"https://pith.science/pith/3VRARFE3CFBERUCMJGDAWSCMAE/action/author_attestation","sign_citation":"https://pith.science/pith/3VRARFE3CFBERUCMJGDAWSCMAE/action/citation_signature","submit_replication":"https://pith.science/pith/3VRARFE3CFBERUCMJGDAWSCMAE/action/replication_record"}},"created_at":"2026-06-02T02:04:27.930078+00:00","updated_at":"2026-06-02T02:04:27.930078+00:00"}