{"id":"https://openalex.org/W7214907809","doi":"https://doi.org/10.48550/arxiv.2609.37915","title":"Overcoming Scaling Limits in On-Policy Self-Distillation for LLM Reasoning","display_name":"Overcoming Scaling Limits in On-Policy Self-Distillation for LLM Reasoning","publication_year":2026,"publication_date":"2026-09-29","ids":{"openalex":"https://openalex.org/W7214907809","doi":"https://doi.org/10.48550/arxiv.2609.37915"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2609.37915","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.37915","pdf_url":"https://arxiv.org/pdf/2609.37915","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"parent_works":[],"lineage":["https://openalex.org/W7214907809"],"child_works_count":0,"type":"preprint","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2609.37915","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5087605642","display_name":"Md. Ismail Hossain","orcid":"https://orcid.org/0000-0002-4665-0530","observed_orcids":["https://orcid.org/0000-0002-4665-0530"]},"institutions":[{"id":"https://openalex.org/I157386601","display_name":"North South University","ror":"https://ror.org/05wdbfp45","country_code":"BD","type":"education","lineage":["https://openalex.org/I157386601"]}],"countries":["BD"],"is_corresponding":false,"raw_author_name":"Hossain, Md. Ismail","raw_affiliation_strings":["North South University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"North South University","institution_ids":["https://openalex.org/I157386601"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5055808404","display_name":"Humaira Kousar","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I157485424","display_name":"Korea Advanced Institute of Science and Technology","ror":"https://ror.org/05apxxy63","country_code":"KR","type":"education","lineage":["https://openalex.org/I157485424"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Kousar, Humaira","raw_affiliation_strings":["KAIST"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"KAIST","institution_ids":["https://openalex.org/I157485424"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5047809949","display_name":"Isidora Chara Tourni","orcid":"https://orcid.org/0000-0003-0636-7519","observed_orcids":["https://orcid.org/0000-0003-0636-7519"]},"institutions":[{"id":"https://openalex.org/I157485424","display_name":"Korea Advanced Institute of Science and Technology","ror":"https://ror.org/05apxxy63","country_code":"KR","type":"education","lineage":["https://openalex.org/I157485424"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Tourni, Isidora Chara","raw_affiliation_strings":["KAIST"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"KAIST","institution_ids":["https://openalex.org/I157485424"]}]}],"institutions":[],"countries_distinct_count":2,"institutions_distinct_count":2,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":1,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.5156000256538391,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.5156000256538391,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.11500000208616257,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T12072","display_name":"Machine Learning and Algorithms","score":0.0934000015258789,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/on-policy-self-distillation","display_name":"on-policy self-distillation","score":0.9929999709129333},{"id":"https://openalex.org/keywords/large-language-model-reasoning","display_name":"large language model reasoning","score":0.7839999794960022},{"id":"https://openalex.org/keywords/oasis","display_name":"OASIS","score":0.8059999942779541},{"id":"https://openalex.org/keywords/aime","display_name":"AIME","score":0.4180000126361847}],"concepts":[{"id":"https://openalex.org/C55439883","wikidata":"https://www.wikidata.org/wiki/Q360812","display_name":"Correctness","level":2,"score":0.7024999856948853},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.6725999712944031},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6262000203132629},{"id":"https://openalex.org/C126388530","wikidata":"https://www.wikidata.org/wiki/Q1131737","display_name":"Imitation","level":2,"score":0.5415999889373779},{"id":"https://openalex.org/C42058472","wikidata":"https://www.wikidata.org/wiki/Q810214","display_name":"Base (topology)","level":2,"score":0.510200023651123},{"id":"https://openalex.org/C99844830","wikidata":"https://www.wikidata.org/wiki/Q102441924","display_name":"Scaling","level":2,"score":0.4489000141620636},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.42089998722076416},{"id":"https://openalex.org/C12713177","wikidata":"https://www.wikidata.org/wiki/Q1900281","display_name":"Perspective (graphical)","level":2,"score":0.28360000252723694},{"id":"https://openalex.org/C110121322","wikidata":"https://www.wikidata.org/wiki/Q865811","display_name":"Distribution (mathematics)","level":2,"score":0.27950000762939453},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.2745000123977661},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.27390000224113464},{"id":"https://openalex.org/C112972136","wikidata":"https://www.wikidata.org/wiki/Q7595718","display_name":"Stability (learning theory)","level":2,"score":0.2644999921321869},{"id":"https://openalex.org/C127705205","wikidata":"https://www.wikidata.org/wiki/Q5748245","display_name":"Heuristics","level":2,"score":0.2637999951839447}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2609.37915","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.37915","pdf_url":"https://arxiv.org/pdf/2609.37915","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2609.37915","doi":"https://doi.org/10.48550/arxiv.2609.37915","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2609.37915","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2609.37915","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.37915","pdf_url":"https://arxiv.org/pdf/2609.37915","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[{"score":0.4156000018119812,"display_name":"Quality education","id":"https://metadata.un.org/sdg/4"}],"study_designs":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7214907809.pdf","grobid_xml":"https://content.openalex.org/works/W7214907809.grobid-xml"},"referenced_works_count":13,"referenced_works":["https://openalex.org/W1690739335","https://openalex.org/W1821462560","https://openalex.org/W2736601468","https://openalex.org/W4310290453","https://openalex.org/W4385970164","https://openalex.org/W4391631327","https://openalex.org/W4414939355","https://openalex.org/W7134063327","https://openalex.org/W7161147084","https://openalex.org/W7161991833","https://openalex.org/W7167214743","https://openalex.org/W7201885121","https://openalex.org/W7202181542"],"related_works":["https://openalex.org/W7201842969","https://openalex.org/W7214590818","https://openalex.org/W7204470223","https://openalex.org/W7220699741","https://openalex.org/W7202213136","https://openalex.org/W7214887446","https://openalex.org/W7223070783","https://openalex.org/W7215074916","https://openalex.org/W7125861034","https://openalex.org/W7161054869"],"abstract_inverted_index":{"On-policy":[0],"self-distillation":[1,201],"(OPSD)":[2],"trains":[3],"a":[4,8,34,42,51],"student":[5,24],"to":[6,22,73,85,179],"match":[7],"privileged":[9,31],"teacher":[10,29,68,99,138],"distribution":[11],"along":[12],"its":[13],"own":[14,105],"sampled":[15],"trajectory.":[16],"Standard":[17],"OPSD":[18,83,118,188],"applies":[19],"this":[20,110],"supervision":[21],"unverified":[23,88,133],"rollouts":[25],"while":[26,170],"conditioning":[27],"the":[28,67,74,98,103,117,137,162,198],"on":[30,54,102,109,151,168],"context,":[32],"typically":[33],"reference":[35],"solution.":[36],"We":[37],"separate":[38],"these":[39],"roles":[40],"in":[41],"factorial":[43],"analysis":[44],"and":[45,128,149,156],"find":[46],"that":[47,193],"scaffold":[48],"correctness":[49],"has":[50],"stronger":[52],"effect":[53],"downstream":[55],"accuracy":[56],"than":[57],"context":[58],"correctness.":[59],"Unverified":[60],"scaffolds":[61,93,196],"create":[62],"an":[63],"imitation":[64],"gap":[65,77],"because":[66],"can":[69],"use":[70],"information":[71],"unavailable":[72],"student.":[75],"This":[76],"shrinks":[78],"with":[79,132],"model":[80,164],"scale,":[81],"yet":[82],"continues":[84],"supervise":[86],"mostly":[87,122],"trajectories.":[89],"In":[90],"contrast,":[91],"verified":[92,123,194],"remain":[94],"effective":[95],"even":[96],"when":[97],"is":[100],"conditioned":[101],"student's":[104],"unsuccessful":[106],"rollout.":[107],"Based":[108],"finding,":[111],"we":[112],"introduce":[113],"OASIS,":[114],"which":[115],"retains":[116],"objective":[119],"but":[120],"supervises":[121],"by":[124,165,189],"label":[125],"on-policy":[126,195],"trajectories":[127],"replaces":[129],"written":[130],"solutions":[131],"model-generated":[134],"attempts":[135],"as":[136,202],"context.":[139],"OASIS":[140,159,185],"therefore":[141],"requires":[142],"only":[143],"final-answer":[144],"labels.":[145],"Across":[146],"Qwen3-1.7B,":[147],"4B,":[148],"8B":[150],"AIME":[152,154],"2024,":[153],"2025,":[155,158],"HMMT":[157],"improves":[160,186],"over":[161,187],"base":[163],"3.2--3.8":[166],"points":[167,176],"average,":[169],"OPSD's":[171],"gain":[172],"falls":[173],"from":[174],"3.05":[175,190],"at":[177,181],"1.7B":[178],"0.14":[180],"8B.":[182],"At":[183],"8B,":[184],"points,":[191],"showing":[192],"preserve":[197],"effectiveness":[199],"of":[200],"models":[203],"scale.":[204]},"counts_by_year":[{"year":2026,"cited_by_count":1}],"updated_date":"2026-11-10T11:39:43.021295","created_date":"2026-10-01T00:00:00"}