{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:ZVDHXWBK6V5ZTWDHVF5BQWTO3O","short_pith_number":"pith:ZVDHXWBK","schema_version":"1.0","canonical_sha256":"cd467bd82af57b99d867a97a185a6edb887387e0c09efc3d60da8c29872409b0","source":{"kind":"arxiv","id":"2608.05802","version":1},"attestation_state":"computed","paper":{"title":"On-Policy Delta Distillation for Multilingual Math Reasoning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Byeongho Heo, Dongyoon Han, Jaehui Hwang, Sangdoo Yun","submitted_at":"2026-08-06T09:37:49Z","abstract_excerpt":"On-Policy Distillation (OPD) is emerging as a promising alternative to reinforcement learning for LLM post-training, yet its effectiveness in multilingual settings remains underexplored. We study OPD and its advanced variant, On-Policy Delta Distillation (OPD$^2$), for mathematical reasoning in English, Korean, and Japanese. OPD$^2$ improves OPD by using the probability gap between a post-trained teacher and its base model as the learning signal. Experiments with Qwen3 show that OPD$^2$ consistently outperforms the original OPD, with particularly strong improvements in Korean and Japanese, and"},"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":"2608.05802","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-08-06T09:37:49Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"1acf91eaeefa6eeae72446db8f67a2768446c15feab5ca51eda0588b6ac38218","abstract_canon_sha256":"cedef610dbb3a86f3ea9f68f5dd8aa0ca7dab2615e6c107f24ad2fa5acaf1b4d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-08-07T00:52:46.863217Z","signature_b64":"SbxFPTf8vOJGUrphlM+ZYVwJRODi7fq7XtSVZlOQTm2CjhFfUjq68guZPo7AcnX36KHxBu6F2cEN3PykejuRCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"cd467bd82af57b99d867a97a185a6edb887387e0c09efc3d60da8c29872409b0","last_reissued_at":"2026-08-07T00:52:46.861800Z","signature_status":"signed_v1","first_computed_at":"2026-08-07T00:52:46.861800Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"On-Policy Delta Distillation for Multilingual Math Reasoning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Byeongho Heo, Dongyoon Han, Jaehui Hwang, Sangdoo Yun","submitted_at":"2026-08-06T09:37:49Z","abstract_excerpt":"On-Policy Distillation (OPD) is emerging as a promising alternative to reinforcement learning for LLM post-training, yet its effectiveness in multilingual settings remains underexplored. We study OPD and its advanced variant, On-Policy Delta Distillation (OPD$^2$), for mathematical reasoning in English, Korean, and Japanese. OPD$^2$ improves OPD by using the probability gap between a post-trained teacher and its base model as the learning signal. Experiments with Qwen3 show that OPD$^2$ consistently outperforms the original OPD, with particularly strong improvements in Korean and Japanese, and"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2608.05802","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/2608.05802/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":"2608.05802","created_at":"2026-08-07T00:52:46.863322+00:00"},{"alias_kind":"arxiv_version","alias_value":"2608.05802v1","created_at":"2026-08-07T00:52:46.863322+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2608.05802","created_at":"2026-08-07T00:52:46.863322+00:00"},{"alias_kind":"pith_short_12","alias_value":"ZVDHXWBK6V5Z","created_at":"2026-08-07T00:52:46.863322+00:00"},{"alias_kind":"pith_short_16","alias_value":"ZVDHXWBK6V5ZTWDH","created_at":"2026-08-07T00:52:46.863322+00:00"},{"alias_kind":"pith_short_8","alias_value":"ZVDHXWBK","created_at":"2026-08-07T00:52:46.863322+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/ZVDHXWBK6V5ZTWDHVF5BQWTO3O","json":"https://pith.science/pith/ZVDHXWBK6V5ZTWDHVF5BQWTO3O.json","graph_json":"https://pith.science/api/pith-number/ZVDHXWBK6V5ZTWDHVF5BQWTO3O/graph.json","events_json":"https://pith.science/api/pith-number/ZVDHXWBK6V5ZTWDHVF5BQWTO3O/events.json","paper":"https://pith.science/paper/ZVDHXWBK"},"agent_actions":{"view_html":"https://pith.science/pith/ZVDHXWBK6V5ZTWDHVF5BQWTO3O","download_json":"https://pith.science/pith/ZVDHXWBK6V5ZTWDHVF5BQWTO3O.json","view_paper":"https://pith.science/paper/ZVDHXWBK","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2608.05802&json=true","fetch_graph":"https://pith.science/api/pith-number/ZVDHXWBK6V5ZTWDHVF5BQWTO3O/graph.json","fetch_events":"https://pith.science/api/pith-number/ZVDHXWBK6V5ZTWDHVF5BQWTO3O/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ZVDHXWBK6V5ZTWDHVF5BQWTO3O/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ZVDHXWBK6V5ZTWDHVF5BQWTO3O/action/storage_attestation","attest_author":"https://pith.science/pith/ZVDHXWBK6V5ZTWDHVF5BQWTO3O/action/author_attestation","sign_citation":"https://pith.science/pith/ZVDHXWBK6V5ZTWDHVF5BQWTO3O/action/citation_signature","submit_replication":"https://pith.science/pith/ZVDHXWBK6V5ZTWDHVF5BQWTO3O/action/replication_record"}},"created_at":"2026-08-07T00:52:46.863322+00:00","updated_at":"2026-08-07T00:52:46.863322+00:00"}