{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:ZEE7CCL5KRH3UMTVUB5YYDMKUF","short_pith_number":"pith:ZEE7CCL5","schema_version":"1.0","canonical_sha256":"c909f1097d544fba3275a07b8c0d8aa14b168f10e7a92c6ff44dfa6a3a80b49a","source":{"kind":"arxiv","id":"2606.30406","version":1},"attestation_state":"computed","paper":{"title":"MOPD: Multi-Teacher On-Policy Distillation for Capability Integration in LLM Post-Training","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Bangjun Xiao, Bofei Gao, Fuli Luo, Hailin Zhang, Jianyu Wei, Jinhao Dong, Lei Li, Liang Zhao, Qibin Yang, Rang Li, Wenhan Ma, Yudong Wang, Zhifang Sui","submitted_at":"2026-06-29T14:51:28Z","abstract_excerpt":"Modern large language models (LLMs) rely on reinforcement learning during post-training to push specific capabilities, yet integrating multiple capabilities into one model remains hard. Existing methods, such as Off-Policy Finetune and Mix-RL, are either inefficient or lose performance. In this work, we propose Multi-teacher On-Policy Distillation (MOPD), a post-training paradigm for combining the capabilities of multiple domain RL teachers: we first run per-domain specialised RL to obtain a set of domain teachers, then distill these teachers into the student on its own rollouts. This eliminat"},"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":"2606.30406","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-06-29T14:51:28Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"a4c8e6ba6f75f4625b309e5303123f2bb8487dd612204dea1ae468bb6b1f9682","abstract_canon_sha256":"57e66cdfe3c4b1a0d870324b125612c48fc1023189e96a6f6ce24de77d7e919b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-30T02:18:14.066347Z","signature_b64":"/cIaB0TQ7WDVL0uULYFyjyXtdPGiqjwBRJfgPI93b0VB+Po3BfxNU41QxbE2YeV/e4RD77O83x2A4eIHvC6KCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c909f1097d544fba3275a07b8c0d8aa14b168f10e7a92c6ff44dfa6a3a80b49a","last_reissued_at":"2026-06-30T02:18:14.065835Z","signature_status":"signed_v1","first_computed_at":"2026-06-30T02:18:14.065835Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"MOPD: Multi-Teacher On-Policy Distillation for Capability Integration in LLM Post-Training","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Bangjun Xiao, Bofei Gao, Fuli Luo, Hailin Zhang, Jianyu Wei, Jinhao Dong, Lei Li, Liang Zhao, Qibin Yang, Rang Li, Wenhan Ma, Yudong Wang, Zhifang Sui","submitted_at":"2026-06-29T14:51:28Z","abstract_excerpt":"Modern large language models (LLMs) rely on reinforcement learning during post-training to push specific capabilities, yet integrating multiple capabilities into one model remains hard. Existing methods, such as Off-Policy Finetune and Mix-RL, are either inefficient or lose performance. In this work, we propose Multi-teacher On-Policy Distillation (MOPD), a post-training paradigm for combining the capabilities of multiple domain RL teachers: we first run per-domain specialised RL to obtain a set of domain teachers, then distill these teachers into the student on its own rollouts. This eliminat"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.30406","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/2606.30406/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":"2606.30406","created_at":"2026-06-30T02:18:14.065914+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.30406v1","created_at":"2026-06-30T02:18:14.065914+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.30406","created_at":"2026-06-30T02:18:14.065914+00:00"},{"alias_kind":"pith_short_12","alias_value":"ZEE7CCL5KRH3","created_at":"2026-06-30T02:18:14.065914+00:00"},{"alias_kind":"pith_short_16","alias_value":"ZEE7CCL5KRH3UMTV","created_at":"2026-06-30T02:18:14.065914+00:00"},{"alias_kind":"pith_short_8","alias_value":"ZEE7CCL5","created_at":"2026-06-30T02:18:14.065914+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.07050","citing_title":"When Top-K Misses the Decision: Tool-Call Drift in Multi-Teacher On-Policy Distillation","ref_index":12,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ZEE7CCL5KRH3UMTVUB5YYDMKUF","json":"https://pith.science/pith/ZEE7CCL5KRH3UMTVUB5YYDMKUF.json","graph_json":"https://pith.science/api/pith-number/ZEE7CCL5KRH3UMTVUB5YYDMKUF/graph.json","events_json":"https://pith.science/api/pith-number/ZEE7CCL5KRH3UMTVUB5YYDMKUF/events.json","paper":"https://pith.science/paper/ZEE7CCL5"},"agent_actions":{"view_html":"https://pith.science/pith/ZEE7CCL5KRH3UMTVUB5YYDMKUF","download_json":"https://pith.science/pith/ZEE7CCL5KRH3UMTVUB5YYDMKUF.json","view_paper":"https://pith.science/paper/ZEE7CCL5","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.30406&json=true","fetch_graph":"https://pith.science/api/pith-number/ZEE7CCL5KRH3UMTVUB5YYDMKUF/graph.json","fetch_events":"https://pith.science/api/pith-number/ZEE7CCL5KRH3UMTVUB5YYDMKUF/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ZEE7CCL5KRH3UMTVUB5YYDMKUF/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ZEE7CCL5KRH3UMTVUB5YYDMKUF/action/storage_attestation","attest_author":"https://pith.science/pith/ZEE7CCL5KRH3UMTVUB5YYDMKUF/action/author_attestation","sign_citation":"https://pith.science/pith/ZEE7CCL5KRH3UMTVUB5YYDMKUF/action/citation_signature","submit_replication":"https://pith.science/pith/ZEE7CCL5KRH3UMTVUB5YYDMKUF/action/replication_record"}},"created_at":"2026-06-30T02:18:14.065914+00:00","updated_at":"2026-06-30T02:18:14.065914+00:00"}