{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:44ZSHURFER6AD7MACUTBWZUCXO","short_pith_number":"pith:44ZSHURF","schema_version":"1.0","canonical_sha256":"e73323d225247c01fd8015261b6682bb97ab28e5f4048eaeae510b13d7a199c0","source":{"kind":"arxiv","id":"2606.06021","version":1},"attestation_state":"computed","paper":{"title":"OPRD: On-Policy Representation Distillation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Bowen Song, Gang Chen, Guangcheng Zhu, Haobo Wang, Mingxuan Xia, Shenzhi Yang, Weiqiang Wang, Xing Zheng, Yingfan Ma, Zhongqi Chen","submitted_at":"2026-06-04T11:13:01Z","abstract_excerpt":"On-policy distillation (OPD) supervises the student only in output space by matching next-token probabilities. This output-only paradigm has two limits: (1) sampling variance from Monte Carlo KL estimates over large vocabularies (e.g., Qwen's ~150k tokens) persists throughout training, and (2) it treats the teacher as a black-box, discarding all intermediate hidden states after the LM head. We propose On-Policy Representation Distillation (OPRD), which lifts distillation into hidden-state space by aligning student and teacher representations across selected layers on the same rollouts, bypassi"},"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":"2606.06021","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-06-04T11:13:01Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"0462c6c49ee61628613d511e84cba358fe55cf01e141e39f02fe77a27d6493ba","abstract_canon_sha256":"162fcbf9f051fa1a93475bab43d100f68f3d1c6dbc2da7fee431a086dc24900c"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-05T01:15:30.596375Z","signature_b64":"XuslWn3dn8Uq8xcrXXjSFeLt05OVdCe8GQ+AWLGxyqMOS4xrcHymT4NRPXyvMkRZOHj8eajC2oldFC5DutZxCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"e73323d225247c01fd8015261b6682bb97ab28e5f4048eaeae510b13d7a199c0","last_reissued_at":"2026-06-05T01:15:30.595962Z","signature_status":"signed_v1","first_computed_at":"2026-06-05T01:15:30.595962Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"OPRD: On-Policy Representation Distillation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Bowen Song, Gang Chen, Guangcheng Zhu, Haobo Wang, Mingxuan Xia, Shenzhi Yang, Weiqiang Wang, Xing Zheng, Yingfan Ma, Zhongqi Chen","submitted_at":"2026-06-04T11:13:01Z","abstract_excerpt":"On-policy distillation (OPD) supervises the student only in output space by matching next-token probabilities. This output-only paradigm has two limits: (1) sampling variance from Monte Carlo KL estimates over large vocabularies (e.g., Qwen's ~150k tokens) persists throughout training, and (2) it treats the teacher as a black-box, discarding all intermediate hidden states after the LM head. We propose On-Policy Representation Distillation (OPRD), which lifts distillation into hidden-state space by aligning student and teacher representations across selected layers on the same rollouts, bypassi"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.06021","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/2606.06021/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":"2606.06021","created_at":"2026-06-05T01:15:30.596022+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.06021v1","created_at":"2026-06-05T01:15:30.596022+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.06021","created_at":"2026-06-05T01:15:30.596022+00:00"},{"alias_kind":"pith_short_12","alias_value":"44ZSHURFER6A","created_at":"2026-06-05T01:15:30.596022+00:00"},{"alias_kind":"pith_short_16","alias_value":"44ZSHURFER6AD7MA","created_at":"2026-06-05T01:15:30.596022+00:00"},{"alias_kind":"pith_short_8","alias_value":"44ZSHURF","created_at":"2026-06-05T01:15:30.596022+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/44ZSHURFER6AD7MACUTBWZUCXO","json":"https://pith.science/pith/44ZSHURFER6AD7MACUTBWZUCXO.json","graph_json":"https://pith.science/api/pith-number/44ZSHURFER6AD7MACUTBWZUCXO/graph.json","events_json":"https://pith.science/api/pith-number/44ZSHURFER6AD7MACUTBWZUCXO/events.json","paper":"https://pith.science/paper/44ZSHURF"},"agent_actions":{"view_html":"https://pith.science/pith/44ZSHURFER6AD7MACUTBWZUCXO","download_json":"https://pith.science/pith/44ZSHURFER6AD7MACUTBWZUCXO.json","view_paper":"https://pith.science/paper/44ZSHURF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.06021&json=true","fetch_graph":"https://pith.science/api/pith-number/44ZSHURFER6AD7MACUTBWZUCXO/graph.json","fetch_events":"https://pith.science/api/pith-number/44ZSHURFER6AD7MACUTBWZUCXO/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/44ZSHURFER6AD7MACUTBWZUCXO/action/timestamp_anchor","attest_storage":"https://pith.science/pith/44ZSHURFER6AD7MACUTBWZUCXO/action/storage_attestation","attest_author":"https://pith.science/pith/44ZSHURFER6AD7MACUTBWZUCXO/action/author_attestation","sign_citation":"https://pith.science/pith/44ZSHURFER6AD7MACUTBWZUCXO/action/citation_signature","submit_replication":"https://pith.science/pith/44ZSHURFER6AD7MACUTBWZUCXO/action/replication_record"}},"created_at":"2026-06-05T01:15:30.596022+00:00","updated_at":"2026-06-05T01:15:30.596022+00:00"}