{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:YXNXTX5XH2BT3RTTQGFJRSHSG7","short_pith_number":"pith:YXNXTX5X","schema_version":"1.0","canonical_sha256":"c5db79dfb73e833dc673818a98c8f237f7fa0ecc417f1305b7550433744acb80","source":{"kind":"arxiv","id":"2505.18952","version":1},"attestation_state":"computed","paper":{"title":"Online Knowledge Distillation with Reward Guidance","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Chen Jia","submitted_at":"2025-05-25T02:56:18Z","abstract_excerpt":"This work studies knowledge distillation (KD) for large language models (LLMs) through preference optimization. We propose a reward-guided imitation learning framework for sequential KD, formulating a min-max optimization problem between the policy and reward model (RM) to minimize the performance gap between the student and teacher policies. Specifically, the reward optimization is constrained to achieve near-optimality within a confidence set for preference alignment. For preference data construction, we explore both offline and online preference-based KD. Additionally, we reformulate the RM"},"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":"2505.18952","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-05-25T02:56:18Z","cross_cats_sorted":[],"title_canon_sha256":"0c808edd1852b0e3619d2d23171e244b1035138c373f21687f5979db14acecae","abstract_canon_sha256":"82d627e51838b9a3df0be355bb659f5c2eab6253b9e5292844da1dc9f05b40f8"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:09:22.523086Z","signature_b64":"a1YpjXqCUC66Ji7I9bGN2oN3wchEUKVoTgm3gFwgyShcU9tgmT4Q6vrpu4CsHWbBKLGYFJSBqfSHqijvYRLCDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c5db79dfb73e833dc673818a98c8f237f7fa0ecc417f1305b7550433744acb80","last_reissued_at":"2026-07-05T11:09:22.522597Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:09:22.522597Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Online Knowledge Distillation with Reward Guidance","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Chen Jia","submitted_at":"2025-05-25T02:56:18Z","abstract_excerpt":"This work studies knowledge distillation (KD) for large language models (LLMs) through preference optimization. We propose a reward-guided imitation learning framework for sequential KD, formulating a min-max optimization problem between the policy and reward model (RM) to minimize the performance gap between the student and teacher policies. Specifically, the reward optimization is constrained to achieve near-optimality within a confidence set for preference alignment. For preference data construction, we explore both offline and online preference-based KD. Additionally, we reformulate the RM"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.18952","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/2505.18952/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":"2505.18952","created_at":"2026-07-05T11:09:22.522657+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.18952v1","created_at":"2026-07-05T11:09:22.522657+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.18952","created_at":"2026-07-05T11:09:22.522657+00:00"},{"alias_kind":"pith_short_12","alias_value":"YXNXTX5XH2BT","created_at":"2026-07-05T11:09:22.522657+00:00"},{"alias_kind":"pith_short_16","alias_value":"YXNXTX5XH2BT3RTT","created_at":"2026-07-05T11:09:22.522657+00:00"},{"alias_kind":"pith_short_8","alias_value":"YXNXTX5X","created_at":"2026-07-05T11:09:22.522657+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/YXNXTX5XH2BT3RTTQGFJRSHSG7","json":"https://pith.science/pith/YXNXTX5XH2BT3RTTQGFJRSHSG7.json","graph_json":"https://pith.science/api/pith-number/YXNXTX5XH2BT3RTTQGFJRSHSG7/graph.json","events_json":"https://pith.science/api/pith-number/YXNXTX5XH2BT3RTTQGFJRSHSG7/events.json","paper":"https://pith.science/paper/YXNXTX5X"},"agent_actions":{"view_html":"https://pith.science/pith/YXNXTX5XH2BT3RTTQGFJRSHSG7","download_json":"https://pith.science/pith/YXNXTX5XH2BT3RTTQGFJRSHSG7.json","view_paper":"https://pith.science/paper/YXNXTX5X","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.18952&json=true","fetch_graph":"https://pith.science/api/pith-number/YXNXTX5XH2BT3RTTQGFJRSHSG7/graph.json","fetch_events":"https://pith.science/api/pith-number/YXNXTX5XH2BT3RTTQGFJRSHSG7/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/YXNXTX5XH2BT3RTTQGFJRSHSG7/action/timestamp_anchor","attest_storage":"https://pith.science/pith/YXNXTX5XH2BT3RTTQGFJRSHSG7/action/storage_attestation","attest_author":"https://pith.science/pith/YXNXTX5XH2BT3RTTQGFJRSHSG7/action/author_attestation","sign_citation":"https://pith.science/pith/YXNXTX5XH2BT3RTTQGFJRSHSG7/action/citation_signature","submit_replication":"https://pith.science/pith/YXNXTX5XH2BT3RTTQGFJRSHSG7/action/replication_record"}},"created_at":"2026-07-05T11:09:22.522657+00:00","updated_at":"2026-07-05T11:09:22.522657+00:00"}