{"id":"https://openalex.org/W7163900200","doi":"https://doi.org/10.48550/arxiv.2606.06712","title":"Data-Efficient Autoregressive-to-Diffusion Language Models via On-Policy Distillation","display_name":"Data-Efficient Autoregressive-to-Diffusion Language Models via On-Policy Distillation","publication_year":2026,"publication_date":"2026-06-04","ids":{"openalex":"https://openalex.org/W7163900200","doi":"https://doi.org/10.48550/arxiv.2606.06712"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2606.06712","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.06712","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":"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"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2606.06712","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5138115394","display_name":"Xingyu Su","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Su, Xingyu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5138130797","display_name":"Jacob Helwig","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Helwig, Jacob","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5024670621","display_name":"Shubham Parashar","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Parashar, Shubham","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5138185940","display_name":"Atharv Chagi","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chagi, Atharv","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5138124397","display_name":"Lakshmi Jotsna","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jotsna, Lakshmi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5012354604","display_name":"Degui Zhi","orcid":"https://orcid.org/0000-0001-7754-1890"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhi, Degui","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5138187206","display_name":"James Caverlee","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Caverlee, James","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5138118305","display_name":"Dileep Kalathil","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Kalathil, Dileep","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5138160477","display_name":"Shuiwang Ji","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ji, Shuiwang","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/T10028","display_name":"Topic Modeling","score":0.22210000455379486,"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":0.22210000455379486,"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/T10775","display_name":"Generative Adversarial Networks and Image Synthesis","score":0.1881999969482422,"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"}},{"id":"https://openalex.org/T13702","display_name":"Machine Learning in Healthcare","score":0.07569999992847443,"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/language-model","display_name":"Language model","score":0.6212999820709229},{"id":"https://openalex.org/keywords/distillation","display_name":"Distillation","score":0.5982999801635742},{"id":"https://openalex.org/keywords/variety","display_name":"Variety (cybernetics)","score":0.5067999958992004},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.49639999866485596},{"id":"https://openalex.org/keywords/transformation","display_name":"Transformation (genetics)","score":0.46230000257492065},{"id":"https://openalex.org/keywords/autoregressive-model","display_name":"Autoregressive model","score":0.4361000061035156},{"id":"https://openalex.org/keywords/train","display_name":"Train","score":0.38119998574256897}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7465999722480774},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.6212999820709229},{"id":"https://openalex.org/C204030448","wikidata":"https://www.wikidata.org/wiki/Q101017","display_name":"Distillation","level":2,"score":0.5982999801635742},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5720000267028809},{"id":"https://openalex.org/C136197465","wikidata":"https://www.wikidata.org/wiki/Q1729295","display_name":"Variety (cybernetics)","level":2,"score":0.5067999958992004},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.49639999866485596},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.48159998655319214},{"id":"https://openalex.org/C204241405","wikidata":"https://www.wikidata.org/wiki/Q461499","display_name":"Transformation (genetics)","level":3,"score":0.46230000257492065},{"id":"https://openalex.org/C159877910","wikidata":"https://www.wikidata.org/wiki/Q2202883","display_name":"Autoregressive model","level":2,"score":0.4361000061035156},{"id":"https://openalex.org/C190839683","wikidata":"https://www.wikidata.org/wiki/Q2448197","display_name":"Train","level":2,"score":0.38119998574256897},{"id":"https://openalex.org/C2983448237","wikidata":"https://www.wikidata.org/wiki/Q1078276","display_name":"Language understanding","level":2,"score":0.3546000123023987},{"id":"https://openalex.org/C77618280","wikidata":"https://www.wikidata.org/wiki/Q1155772","display_name":"Scheme (mathematics)","level":2,"score":0.30790001153945923},{"id":"https://openalex.org/C22367795","wikidata":"https://www.wikidata.org/wiki/Q7625208","display_name":"Structured prediction","level":2,"score":0.29280000925064087},{"id":"https://openalex.org/C28490314","wikidata":"https://www.wikidata.org/wiki/Q189436","display_name":"Speech recognition","level":1,"score":0.2888000011444092},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.2840000092983246},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.2680000066757202},{"id":"https://openalex.org/C114289077","wikidata":"https://www.wikidata.org/wiki/Q3284399","display_name":"Statistical model","level":2,"score":0.2635999917984009},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.26350000500679016},{"id":"https://openalex.org/C18762648","wikidata":"https://www.wikidata.org/wiki/Q42213","display_name":"Work (physics)","level":2,"score":0.2551000118255615},{"id":"https://openalex.org/C67186912","wikidata":"https://www.wikidata.org/wiki/Q367664","display_name":"Data modeling","level":2,"score":0.25}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2606.06712","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.06712","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2606.06712","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.06712","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":"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"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"We":[0],"study":[1],"the":[2,21,32,62,75,86,124,136,138,161,169,176,201],"transformation":[3,210],"of":[4,197,204,214],"autoregressive":[5],"models":[6,11],"(ARLMs)":[7],"into":[8],"diffusion":[9],"language":[10],"(DLMs).":[12],"Rather":[13],"than":[14,85],"pretraining":[15,206],"from":[16,48,70,168,175],"scratch,":[17],"prior":[18],"work":[19],"replaces":[20],"causal":[22],"attention":[23,28],"in":[24,107,155,164],"ARLMs":[25],"with":[26,128,190],"bidirectional":[27,129],"and":[29,135,207],"then":[30],"trains":[31],"resulting":[33],"model":[34,171],"using":[35],"a":[36,49,54,71,194,212],"DLM":[37,55,205,209],"objective.":[38],"However,":[39],"these":[40,150],"approaches":[41],"incur":[42],"two":[43],"distribution":[44],"shifts.":[45],"First,":[46],"transitioning":[47],"next-token":[50],"prediction":[51],"objective":[52,56],"to":[53,185],"can":[57],"discard":[58],"knowledge":[59,144,173],"acquired":[60],"by":[61,92,145],"ARLM":[63,127,215],"during":[64],"training.":[65],"Second,":[66],"standard":[67],"DLMs":[68],"suffer":[69],"train-inference":[72,162],"mismatch,":[73],"as":[74,211],"training":[76,153,188],"loss":[77],"is":[78,112,119],"defined":[79],"on":[80,149],"randomly":[81],"masked":[82],"sequences":[83],"rather":[84],"trajectories":[87],"encountered":[88],"at":[89],"inference":[90],"produced":[91],"confidence-based":[93],"decoding.":[94],"To":[95],"address":[96],"both":[97],"challenges,":[98],"we":[99],"introduce":[100],"an":[101,126,156],"On-Policy":[102,109],"Diffusion":[103],"Language":[104],"Model":[105],"(OPDLM)":[106],"which":[108],"Distillation":[110],"(OPD)":[111],"employed":[113],"for":[114],"ARLM-to-DLM":[115],"transformation.":[116],"Specifically,":[117],"OPDLM":[118,159,182,199],"trained":[120],"via":[121],"self-OPD,":[122],"where":[123],"student,":[125],"attention,":[130],"generates":[131],"its":[132,143],"own":[133],"trajectories,":[134],"teacher,":[137],"original":[139,170],"frozen":[140],"ARLM,":[141],"distills":[142],"providing":[146],"target":[147],"logits":[148],"trajectories.":[151],"By":[152],"directly":[154],"on-policy":[157],"manner,":[158],"eliminates":[160],"mismatch":[163],"DLMs,":[165],"while":[166],"distillation":[167],"enhances":[172],"retention":[174],"ARLM.":[177],"Empirical":[178],"results":[179],"demonstrate":[180],"that":[181],"requires":[183],"15x":[184],"7,000x":[186],"fewer":[187],"tokens":[189],"strong":[191],"performance":[192],"across":[193],"wide":[195],"variety":[196],"tasks.":[198],"avoids":[200],"prohibitive":[202],"cost":[203],"positions":[208],"form":[213],"post-training.":[216]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-06-09T00:00:00"}