{"id":"https://openalex.org/W7208750677","doi":"https://doi.org/10.48550/arxiv.2609.02998","title":"Verify Before You Distill: Prompt-Level Teacher Gating for On-Policy Distillation","display_name":"Verify Before You Distill: Prompt-Level Teacher Gating for On-Policy Distillation","publication_year":2026,"publication_date":"2026-09-02","ids":{"openalex":"https://openalex.org/W7208750677","doi":"https://doi.org/10.48550/arxiv.2609.02998"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2609.02998","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.02998","pdf_url":"https://arxiv.org/pdf/2609.02998","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":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2609.02998","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5102935455","display_name":"Zhiwei Zhang","orcid":"https://orcid.org/0009-0009-2894-7179","observed_orcids":["https://orcid.org/0009-0009-2894-7179"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Zhiwei","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5150023529","display_name":"Zechen Sun","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Sun, Zechen","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5149986699","display_name":"Fei Zhao","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhao, Fei","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5147615397","display_name":"Kang Peng","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Peng, Kang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5126971590","display_name":"Bin Liang","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liang, Bin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5153593230","display_name":"Huayu Deng","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Deng, Huayu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5149991522","display_name":"Yao Hu","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Hu, Yao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5008208316","display_name":"Kam\u2010Fai Wong","orcid":"https://orcid.org/0000-0002-9427-5659","observed_orcids":["https://orcid.org/0000-0002-9427-5659"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wong, Kam-Fai","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5148448611","display_name":"Mu Chuan","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chuan, Mu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":0,"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/T10462","display_name":"Reinforcement Learning in Robotics","score":0.35030001401901245,"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/T10462","display_name":"Reinforcement Learning in Robotics","score":0.35030001401901245,"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/T10028","display_name":"Topic Modeling","score":0.21240000426769257,"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":0.1720000058412552,"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-distillation","display_name":"on-policy distillation","score":0.9980000257492065},{"id":"https://openalex.org/keywords/group-relative-policy-optimization","display_name":"Group Relative Policy Optimization","score":0.4790000021457672},{"id":"https://openalex.org/keywords/post-training","display_name":"post-training","score":0.33899998664855957}],"concepts":[{"id":"https://openalex.org/C43214815","wikidata":"https://www.wikidata.org/wiki/Q7310987","display_name":"Reliability (semiconductor)","level":3,"score":0.7361000180244446},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5967000126838684},{"id":"https://openalex.org/C204030448","wikidata":"https://www.wikidata.org/wiki/Q101017","display_name":"Distillation","level":2,"score":0.5658000111579895},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.5415999889373779},{"id":"https://openalex.org/C2780009758","wikidata":"https://www.wikidata.org/wiki/Q6804172","display_name":"Measure (data warehouse)","level":2,"score":0.45570001006126404},{"id":"https://openalex.org/C200601418","wikidata":"https://www.wikidata.org/wiki/Q2193887","display_name":"Reliability engineering","level":1,"score":0.44929999113082886},{"id":"https://openalex.org/C106301342","wikidata":"https://www.wikidata.org/wiki/Q4117933","display_name":"Entropy (arrow of time)","level":2,"score":0.4456999897956848},{"id":"https://openalex.org/C151319957","wikidata":"https://www.wikidata.org/wiki/Q752739","display_name":"Asynchronous communication","level":2,"score":0.42590001225471497},{"id":"https://openalex.org/C165838908","wikidata":"https://www.wikidata.org/wiki/Q736777","display_name":"Calibration","level":2,"score":0.3264000117778778},{"id":"https://openalex.org/C21880701","wikidata":"https://www.wikidata.org/wiki/Q2144042","display_name":"Process engineering","level":1,"score":0.31360000371932983},{"id":"https://openalex.org/C137209882","wikidata":"https://www.wikidata.org/wiki/Q1403517","display_name":"Measurement uncertainty","level":2,"score":0.30550000071525574},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.2904999852180481},{"id":"https://openalex.org/C161664118","wikidata":"https://www.wikidata.org/wiki/Q1089933","display_name":"Churning","level":2,"score":0.28200000524520874},{"id":"https://openalex.org/C165064840","wikidata":"https://www.wikidata.org/wiki/Q1321061","display_name":"Matching (statistics)","level":2,"score":0.2757999897003174},{"id":"https://openalex.org/C9679016","wikidata":"https://www.wikidata.org/wiki/Q1417473","display_name":"Principle of maximum entropy","level":2,"score":0.2676999866962433}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2609.02998","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.02998","pdf_url":"https://arxiv.org/pdf/2609.02998","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":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2609.02998","doi":"https://doi.org/10.48550/arxiv.2609.02998","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2609.02998","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2609.02998","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.02998","pdf_url":"https://arxiv.org/pdf/2609.02998","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":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[{"score":0.9235000014305115,"id":"https://metadata.un.org/sdg/4","display_name":"Quality education"}],"study_designs":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7208750677.pdf","grobid_xml":"https://content.openalex.org/works/W7208750677.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7167615039","https://openalex.org/W7164018557","https://openalex.org/W7213995943","https://openalex.org/W7220388840","https://openalex.org/W7202257096","https://openalex.org/W7222213547","https://openalex.org/W7222327894","https://openalex.org/W7162992044","https://openalex.org/W7213997239","https://openalex.org/W7204908525"],"abstract_inverted_index":{"On-policy":[0],"distillation":[1],"(OPD)":[2],"accelerates":[3],"post-training":[4],"by":[5],"providing":[6],"dense":[7,94,115],"token-level":[8],"supervision":[9,23,95],"from":[10,101,179],"a":[11,42,48,102],"frozen":[12],"teacher":[13,31,45,84,107,161],"on":[14,80],"the":[15,30,81,90,118,184],"student's":[16],"own":[17],"rollouts.":[18],"Vanilla":[19,140],"OPD":[20,116,141],"applies":[21],"this":[22],"uniformly":[24],"across":[25],"prompts,":[26],"without":[27],"checking":[28],"whether":[29],"is":[32,40,96],"reliable":[33],"for":[34,163],"each":[35,111],"prompt.":[36],"Because":[37],"reverse":[38],"KL":[39],"mode-seeking,":[41],"confidently":[43],"wrong":[44],"can":[46],"induce":[47],"strong":[49],"yet":[50],"misleading":[51],"update.":[52],"Distributional":[53],"proxies,":[54],"such":[55],"as":[56],"entropy":[57],"or":[58,64,122],"teacher-student":[59],"likelihood":[60],"agreement,":[61],"measure":[62],"uncertainty":[63],"agreement":[65],"but":[66],"do":[67],"not":[68],"directly":[69],"verify":[70],"outcome":[71],"correctness.":[72],"We":[73],"introduce":[74],"Teacher-Gated":[75],"On-Policy":[76],"Distillation":[77],"(TGOPD),":[78],"built":[79],"principle":[82],"that":[83],"reliability":[85,100,119,164],"should":[86],"be":[87],"verified":[88],"at":[89,152],"prompt":[91,112],"level":[92],"before":[93],"admitted.":[97],"TGOPD":[98,138,166],"estimates":[99],"small":[103],"set":[104],"of":[105],"verifier-scored":[106],"probes":[108],"and":[109,129,135,147],"routes":[110],"exclusively":[113],"to":[114,123,181],"when":[117],"check":[120],"passes":[121],"verifier-grounded":[124],"GRPO":[125],"otherwise.":[126],"Across":[127],"4B":[128,186],"35B":[130],"students":[131],"in":[132,142,172,183],"mathematics,":[133],"code,":[134],"instruction":[136],"following,":[137],"outperforms":[139],"all":[143],"six":[144],"single-domain":[145,187],"settings":[146],"achieves":[148],"higher":[149],"seven-benchmark":[150],"averages":[151],"both":[153],"scales":[154],"under":[155],"multi-domain":[156],"training.":[157],"By":[158],"using":[159],"otherwise-idle":[160],"capacity":[162],"estimation,":[165],"also":[167],"reduces":[168],"teacher-side":[169],"compute":[170],"waste":[171],"asynchronous":[173],"OPD,":[174],"increasing":[175],"teacher-node":[176],"GPU":[177],"utilization":[178],"9.8%":[180],"78.9%":[182],"measured":[185],"run.":[188]},"counts_by_year":[],"updated_date":"2026-10-28T10:33:24.094632","created_date":"2026-09-05T00:00:00"}