{"id":"https://openalex.org/W7160946237","doi":"https://doi.org/10.48550/arxiv.2605.08873","title":"CoDistill-GRPO: A Co-Distillation Recipe for Efficient Group Relative Policy Optimization","display_name":"CoDistill-GRPO: A Co-Distillation Recipe for Efficient Group Relative Policy Optimization","publication_year":2026,"publication_date":"2026-05-09","ids":{"openalex":"https://openalex.org/W7160946237","doi":"https://doi.org/10.48550/arxiv.2605.08873"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.08873","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.08873","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2605.08873","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5104002781","display_name":"Soo Min Kwon","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Kwon, Soo Min","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5001128596","display_name":"Ziteng Sun","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Sun, Ziteng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5103890688","display_name":"Ananda Theertha Suresh","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Suresh, Ananda Theertha","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135980565","display_name":"Himanshu Jain","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jain, Himanshu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5135984778","display_name":"Sanjiv Kumar","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Kumar, Sanjiv","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":false,"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":0.3833000063896179,"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":0.3833000063896179,"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":0.06790000200271606,"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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.06729999929666519,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"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/overhead","display_name":"Overhead (engineering)","score":0.5573999881744385},{"id":"https://openalex.org/keywords/base","display_name":"Base (topology)","score":0.5037999749183655},{"id":"https://openalex.org/keywords/train","display_name":"Train","score":0.4864000082015991},{"id":"https://openalex.org/keywords/training","display_name":"Training (meteorology)","score":0.4165000021457672},{"id":"https://openalex.org/keywords/group","display_name":"Group (periodic table)","score":0.3926999866962433},{"id":"https://openalex.org/keywords/recipe","display_name":"Recipe","score":0.351500004529953},{"id":"https://openalex.org/keywords/percentage-point","display_name":"Percentage point","score":0.33880001306533813},{"id":"https://openalex.org/keywords/point","display_name":"Point (geometry)","score":0.334199994802475}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7103000283241272},{"id":"https://openalex.org/C2779960059","wikidata":"https://www.wikidata.org/wiki/Q7113681","display_name":"Overhead (engineering)","level":2,"score":0.5573999881744385},{"id":"https://openalex.org/C42058472","wikidata":"https://www.wikidata.org/wiki/Q810214","display_name":"Base (topology)","level":2,"score":0.5037999749183655},{"id":"https://openalex.org/C190839683","wikidata":"https://www.wikidata.org/wiki/Q2448197","display_name":"Train","level":2,"score":0.4864000082015991},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.4165000021457672},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.40369999408721924},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3944000005722046},{"id":"https://openalex.org/C2781311116","wikidata":"https://www.wikidata.org/wiki/Q83306","display_name":"Group (periodic table)","level":2,"score":0.3926999866962433},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.36340001225471497},{"id":"https://openalex.org/C2778671685","wikidata":"https://www.wikidata.org/wiki/Q219239","display_name":"Recipe","level":2,"score":0.351500004529953},{"id":"https://openalex.org/C44648626","wikidata":"https://www.wikidata.org/wiki/Q1049848","display_name":"Percentage point","level":2,"score":0.33880001306533813},{"id":"https://openalex.org/C28719098","wikidata":"https://www.wikidata.org/wiki/Q44946","display_name":"Point (geometry)","level":2,"score":0.334199994802475},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.3310000002384186},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.32690000534057617},{"id":"https://openalex.org/C101454708","wikidata":"https://www.wikidata.org/wiki/Q17106019","display_name":"Standard Model (mathematical formulation)","level":3,"score":0.31459999084472656},{"id":"https://openalex.org/C77553402","wikidata":"https://www.wikidata.org/wiki/Q13222579","display_name":"Upper and lower bounds","level":2,"score":0.30550000071525574},{"id":"https://openalex.org/C204030448","wikidata":"https://www.wikidata.org/wiki/Q101017","display_name":"Distillation","level":2,"score":0.3027999997138977},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.29030001163482666},{"id":"https://openalex.org/C165064840","wikidata":"https://www.wikidata.org/wiki/Q1321061","display_name":"Matching (statistics)","level":2,"score":0.29019999504089355},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.2856999933719635},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.2768999934196472},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.26100000739097595},{"id":"https://openalex.org/C114289077","wikidata":"https://www.wikidata.org/wiki/Q3284399","display_name":"Statistical model","level":2,"score":0.25270000100135803},{"id":"https://openalex.org/C4554734","wikidata":"https://www.wikidata.org/wiki/Q593744","display_name":"Knowledge base","level":2,"score":0.251800000667572},{"id":"https://openalex.org/C17648541","wikidata":"https://www.wikidata.org/wiki/Q2265984","display_name":"Efficiency","level":3,"score":0.250900000333786}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.08873","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.08873","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2605.08873","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.08873","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Group":[0],"Relative":[1],"Policy":[2],"Optimization":[3],"(GRPO)":[4],"has":[5],"emerged":[6],"as":[7,217],"a":[8,40,83,89,92,218],"powerful":[9],"algorithm":[10,85],"for":[11,47,223],"improving":[12],"the":[13,62,108,119,124,133,140,181,193,197],"reasoning":[14],"capabilities":[15],"of":[16,64,143,175,234],"language":[17],"models,":[18,225],"but":[19],"often":[20],"fails":[21],"to":[22,27,44,50,116,221],"improve":[23],"small":[24,93,109,134,152],"models":[25,103],"due":[26],"sparse":[28],"rewards":[29],"on":[30,158,192,211],"difficult":[31],"tasks.":[32],"Existing":[33],"works":[34],"mitigate":[35],"this":[36,60,78],"issue":[37],"by":[38,95,132],"leveraging":[39],"larger":[41,198,224],"model,":[42],"either":[43],"provide":[45,51],"hints":[46],"rollouts":[48,130],"or":[49],"dense":[52],"reward":[53,115],"signals":[54],"through":[55],"knowledge":[56],"distillation":[57],"(KD).":[58],"However,":[59],"assumes":[61],"existence":[63],"such":[65],"an":[66,112,172,185,227],"oracle,":[67],"and":[68,91,164,184],"training":[69,75,210],"one":[70],"can":[71],"significantly":[72],"increase":[73,174],"total":[74],"time.":[76],"In":[77],"work,":[79],"we":[80,170],"propose":[81],"CoDistill-GRPO,":[82],"co-distillation":[84],"that":[86,148],"simultaneously":[87],"trains":[88],"large":[90,120,125],"model":[94,110,126,135,153,183,199],"maximizing":[96],"carefully":[97],"designed":[98],"GRPO":[99,157,191,207,222],"objectives.":[100],"The":[101],"two":[102],"learn":[104,117],"from":[105,118],"each":[106],"other:":[107],"uses":[111],"on-policy":[113],"KD":[114],"model's":[121],"distribution,":[122],"while":[123],"is":[127],"updated":[128],"using":[129],"generated":[131],"with":[136,168,202],"importance":[137],"reweighting,":[138],"reducing":[139],"computational":[141],"overhead":[142],"rollout":[144],"generation.":[145],"We":[146],"show":[147],"CoDistill-GRPO":[149,203,216],"substantially":[150],"improves":[151],"performance":[154,208],"over":[155,176,180,190],"standard":[156,206],"mathematical":[159],"benchmarks":[160],"across":[161],"both":[162],"Qwen":[163],"Llama":[165],"models.":[166],"Specifically,":[167],"Qwen2.5-Math-1.5B,":[169],"observe":[171],"accuracy":[173],"11.6":[177],"percentage":[178,188],"points":[179,189],"base":[182],"additional":[186],"6.0":[187],"Minerva":[194],"dataset.":[195],"Interestingly,":[196],"(Qwen2.5-Math-7B)":[200],"trained":[201],"nearly":[204],"matches":[205],"despite":[209],"small-model":[212],"rollouts.":[213],"This":[214],"highlights":[215],"cost-effective":[219],"alternative":[220],"yielding":[226],"approximate":[228],"18%":[229],"speedup,":[230],"which":[231],"may":[232],"be":[233],"independent":[235],"interest.":[236]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-05-13T00:00:00"}