{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:VFLFREZPN6POFUG3NPVHGIOBY4","short_pith_number":"pith:VFLFREZP","schema_version":"1.0","canonical_sha256":"a95658932f6f9ee2d0db6bea7321c1c71826e3bb411c55a1db19227701c45158","source":{"kind":"arxiv","id":"2605.11182","version":2},"attestation_state":"computed","paper":{"title":"The Many Faces of On-Policy Distillation: Pitfalls, Mechanisms, and Fixes","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"On-policy distillation fails in LLMs due to distribution mismatch, biased gradients, and privileged information aggregation but targeted fixes restore effectiveness.","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Ge Liu, Hongyu Lu, Siqi Zhu, Weiye Shi, Xuyan Ye","submitted_at":"2026-05-11T19:44:59Z","abstract_excerpt":"On-policy distillation (OPD) and on-policy self-distillation (OPSD) have emerged as promising post-training methods for large language models, offering dense token-level supervision on trajectories sampled from the model's own policy. However, existing results on their effectiveness remain mixed: while OP(S)D has shown promise in system prompt and knowledge internalization, recent studies also report instability and degradation. In this work, we present a comprehensive empirical study of when OPD and OPSD work, when they fail, and why. We find that OPD on mathematical reasoning is highly sensi"},"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":"2605.11182","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-11T19:44:59Z","cross_cats_sorted":[],"title_canon_sha256":"b248cb51e667023b57534f1c43b260f0bb7b7d2661aab3993870fca78c420d80","abstract_canon_sha256":"de0bcebca2a87255dcfef7f5533ad79685999585cf31a57ff359212454eb5959"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-26T01:03:33.409959Z","signature_b64":"7trjWTxlCnNDCdEdnrPq1XMKZyi9NBY/HqfKqfQxdFIPIu/zbAEYKbjkqJtSf44W119BDXpUGBSB3BSLG5XqCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a95658932f6f9ee2d0db6bea7321c1c71826e3bb411c55a1db19227701c45158","last_reissued_at":"2026-05-26T01:03:33.409214Z","signature_status":"signed_v1","first_computed_at":"2026-05-26T01:03:33.409214Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"The Many Faces of On-Policy Distillation: Pitfalls, Mechanisms, and Fixes","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"On-policy distillation fails in LLMs due to distribution mismatch, biased gradients, and privileged information aggregation but targeted fixes restore effectiveness.","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Ge Liu, Hongyu Lu, Siqi Zhu, Weiye Shi, Xuyan Ye","submitted_at":"2026-05-11T19:44:59Z","abstract_excerpt":"On-policy distillation (OPD) and on-policy self-distillation (OPSD) have emerged as promising post-training methods for large language models, offering dense token-level supervision on trajectories sampled from the model's own policy. However, existing results on their effectiveness remain mixed: while OP(S)D has shown promise in system prompt and knowledge internalization, recent studies also report instability and degradation. In this work, we present a comprehensive empirical study of when OPD and OPSD work, when they fail, and why. We find that OPD on mathematical reasoning is highly sensi"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"We identify three failure mechanisms: (1) distribution mismatch between teacher and student caused by conditioning on student-generated prefixes, (2) optimization instability from biased TopK reverse-KL gradients, and (3) an OPSD-specific limitation where the student learns a PI-free policy that aggregates PI-conditioned teachers, which is insufficient when PI is instance-specific. We further show that stop-gradient TopK objectives, RLVR-adapted teachers, and SFT-stabilized students mitigate these failures.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The tested settings (mathematical reasoning trajectories and system-prompt/alignment PI) are representative enough that the three failure mechanisms and proposed fixes will apply to other LLM tasks, model scales, and data distributions without additional confounding factors.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"On-policy distillation for LLMs is sensitive to teacher choice and loss design, while self-distillation fails on instance-specific information but succeeds on shared rules, with stop-gradient TopK, adapted teachers, and SFT stabilization as mitigations.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"On-policy distillation fails in LLMs due to distribution mismatch, biased gradients, and privileged information aggregation but targeted fixes restore effectiveness.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"832de19b3cd4f2b63119cd12226e252ddd28bde55f16c215f062eb805aecbf04"},"source":{"id":"2605.11182","kind":"arxiv","version":2},"verdict":{"id":"25146744-6cee-46de-be8c-ce7581fd0189","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-13T02:13:55.930380Z","strongest_claim":"We identify three failure mechanisms: (1) distribution mismatch between teacher and student caused by conditioning on student-generated prefixes, (2) optimization instability from biased TopK reverse-KL gradients, and (3) an OPSD-specific limitation where the student learns a PI-free policy that aggregates PI-conditioned teachers, which is insufficient when PI is instance-specific. We further show that stop-gradient TopK objectives, RLVR-adapted teachers, and SFT-stabilized students mitigate these failures.","one_line_summary":"On-policy distillation for LLMs is sensitive to teacher choice and loss design, while self-distillation fails on instance-specific information but succeeds on shared rules, with stop-gradient TopK, adapted teachers, and SFT stabilization as mitigations.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The tested settings (mathematical reasoning trajectories and system-prompt/alignment PI) are representative enough that the three failure mechanisms and proposed fixes will apply to other LLM tasks, model scales, and data distributions without additional confounding factors.","pith_extraction_headline":"On-policy distillation fails in LLMs due to distribution mismatch, biased gradients, and privileged information aggregation but targeted fixes restore effectiveness."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.11182/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"claim_evidence","ran_at":"2026-05-20T04:42:00.971795Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T13:33:52.618077Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T10:01:17.612652Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T08:42:08.036066Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"e3f86565937dc8f02f8385eeaa5b3f6abf8d1dad995e7e7b0150e5577541d36e"},"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":"2605.11182","created_at":"2026-05-26T01:03:33.409339+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.11182v2","created_at":"2026-05-26T01:03:33.409339+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.11182","created_at":"2026-05-26T01:03:33.409339+00:00"},{"alias_kind":"pith_short_12","alias_value":"VFLFREZPN6PO","created_at":"2026-05-26T01:03:33.409339+00:00"},{"alias_kind":"pith_short_16","alias_value":"VFLFREZPN6POFUG3","created_at":"2026-05-26T01:03:33.409339+00:00"},{"alias_kind":"pith_short_8","alias_value":"VFLFREZP","created_at":"2026-05-26T01:03:33.409339+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/VFLFREZPN6POFUG3NPVHGIOBY4","json":"https://pith.science/pith/VFLFREZPN6POFUG3NPVHGIOBY4.json","graph_json":"https://pith.science/api/pith-number/VFLFREZPN6POFUG3NPVHGIOBY4/graph.json","events_json":"https://pith.science/api/pith-number/VFLFREZPN6POFUG3NPVHGIOBY4/events.json","paper":"https://pith.science/paper/VFLFREZP"},"agent_actions":{"view_html":"https://pith.science/pith/VFLFREZPN6POFUG3NPVHGIOBY4","download_json":"https://pith.science/pith/VFLFREZPN6POFUG3NPVHGIOBY4.json","view_paper":"https://pith.science/paper/VFLFREZP","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.11182&json=true","fetch_graph":"https://pith.science/api/pith-number/VFLFREZPN6POFUG3NPVHGIOBY4/graph.json","fetch_events":"https://pith.science/api/pith-number/VFLFREZPN6POFUG3NPVHGIOBY4/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/VFLFREZPN6POFUG3NPVHGIOBY4/action/timestamp_anchor","attest_storage":"https://pith.science/pith/VFLFREZPN6POFUG3NPVHGIOBY4/action/storage_attestation","attest_author":"https://pith.science/pith/VFLFREZPN6POFUG3NPVHGIOBY4/action/author_attestation","sign_citation":"https://pith.science/pith/VFLFREZPN6POFUG3NPVHGIOBY4/action/citation_signature","submit_replication":"https://pith.science/pith/VFLFREZPN6POFUG3NPVHGIOBY4/action/replication_record"}},"created_at":"2026-05-26T01:03:33.409339+00:00","updated_at":"2026-05-26T01:03:33.409339+00:00"}