{"id":"https://openalex.org/W7162430223","doi":"https://doi.org/10.48550/arxiv.2605.26106","title":"Looped Diffusion Language Models","display_name":"Looped Diffusion Language Models","publication_year":2026,"publication_date":"2026-05-25","ids":{"openalex":"https://openalex.org/W7162430223","doi":"https://doi.org/10.48550/arxiv.2605.26106"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.26106","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.26106","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.2605.26106","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5137033600","display_name":"Sanghyun Lee","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lee, Sanghyun","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137004672","display_name":"Chunsan Hong","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Hong, Chunsan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137072761","display_name":"Seungryong Kim","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Kim, Seungryong","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137071791","display_name":"Jonghyun Lee","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lee, Jonghyun","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100783486","display_name":"Jong-Ho Park","orcid":"https://orcid.org/0000-0002-0496-649X"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Park, Jongho","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5137016355","display_name":"Dongmin Park","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Park, Dongmin","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":0.36910000443458557,"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":0.36910000443458557,"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":0.13109999895095825,"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/T13910","display_name":"Computational and Text Analysis Methods","score":0.12359999865293503,"subfield":{"id":"https://openalex.org/subfields/3300","display_name":"General Social Sciences"},"field":{"id":"https://openalex.org/fields/33","display_name":"Social Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.6704999804496765},{"id":"https://openalex.org/keywords/transformer","display_name":"Transformer","score":0.4496999979019165},{"id":"https://openalex.org/keywords/process","display_name":"Process (computing)","score":0.44200000166893005},{"id":"https://openalex.org/keywords/key","display_name":"Key (lock)","score":0.4251999855041504},{"id":"https://openalex.org/keywords/code","display_name":"Code (set theory)","score":0.362199991941452},{"id":"https://openalex.org/keywords/diffusion","display_name":"Diffusion","score":0.3547999858856201}],"concepts":[{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.6704999804496765},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6344000101089478},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.4675999879837036},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.4496999979019165},{"id":"https://openalex.org/C98045186","wikidata":"https://www.wikidata.org/wiki/Q205663","display_name":"Process (computing)","level":2,"score":0.44200000166893005},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.4251999855041504},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.362199991941452},{"id":"https://openalex.org/C69357855","wikidata":"https://www.wikidata.org/wiki/Q163214","display_name":"Diffusion","level":2,"score":0.3547999858856201},{"id":"https://openalex.org/C159877910","wikidata":"https://www.wikidata.org/wiki/Q2202883","display_name":"Autoregressive model","level":2,"score":0.3440000116825104},{"id":"https://openalex.org/C2778755073","wikidata":"https://www.wikidata.org/wiki/Q10858537","display_name":"Scale (ratio)","level":2,"score":0.3165000081062317},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.31290000677108765},{"id":"https://openalex.org/C140779682","wikidata":"https://www.wikidata.org/wiki/Q210868","display_name":"Sampling (signal processing)","level":3,"score":0.3098999857902527},{"id":"https://openalex.org/C46686674","wikidata":"https://www.wikidata.org/wiki/Q466303","display_name":"Boosting (machine learning)","level":2,"score":0.2903999984264374},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.2770000100135803},{"id":"https://openalex.org/C68710425","wikidata":"https://www.wikidata.org/wiki/Q5275442","display_name":"Diffusion process","level":3,"score":0.2678000032901764},{"id":"https://openalex.org/C177774035","wikidata":"https://www.wikidata.org/wiki/Q1246948","display_name":"Granularity","level":2,"score":0.25040000677108765}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.26106","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.26106","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.2605.26106","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.26106","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":[{"score":0.46940699219703674,"display_name":"Quality Education","id":"https://metadata.un.org/sdg/4"}],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Masked":[0,54],"diffusion":[1],"models":[2,12],"(MDMs)":[3],"have":[4],"emerged":[5],"as":[6],"a":[7,67],"promising":[8],"alternative":[9],"to":[10,105,122],"autoregressive":[11],"for":[13,23],"language":[14],"modeling,":[15],"yet":[16],"the":[17,35,75,86,88,98,157,163,168],"effective":[18,144,191],"design":[19],"of":[20,77,100,159,165],"transformer":[21,37],"architectures":[22],"MDMs":[24,102,132,193],"remains":[25],"underexplored.":[26],"In":[27],"this":[28,51],"paper,":[29],"we":[30,185],"show":[31],"that":[32,139,188],"selectively":[33],"looping":[34,62,141,189],"early-middle":[36],"layers":[38,63],"significantly":[39],"improves":[40],"both":[41],"training":[42,108],"efficiency":[43,177],"and":[44,202],"model":[45],"performance":[46,99,113],"in":[47,175,192],"MDMs.":[48],"We":[49],"call":[50],"approach":[52],"LoopMDM(Looped":[53],"Diffusion":[55],"Model),":[56],"which":[57],"brings":[58],"two":[59],"key":[60],"benefits:":[61],"at":[64,79],"training-time":[65],"yields":[66,172],"depth-scaling":[68],"effect":[69],"without":[70],"adding":[71],"parameters,":[72],"while":[73,110,178],"varying":[74],"number":[76,158,164],"loops":[78,166],"inference-time":[80,153],"enables":[81],"flexible":[82],"compute":[83,154,176],"scaling.":[84,148],"Despite":[85],"simplicity,":[87],"results":[89],"are":[90],"striking:":[91],"across":[92],"multiple":[93],"pre-training":[94],"corpora,":[95],"LoopMDM":[96,150],"matches":[97],"same-size":[101],"with":[103,134,182],"up":[104,121],"3.3":[106],"fewer":[107],"FLOPs,":[109],"its":[111],"final":[112],"outperforms":[114],"them":[115],"on":[116,125],"various":[117],"reasoning":[118],"benchmarks,":[119],"including":[120],"8.5":[123],"points":[124],"GSM8K.":[126],"It":[127],"even":[128],"surpasses":[129],"deeper":[130],"non-looped":[131],"trained":[133],"comparable":[135],"per-step":[136],"compute,":[137],"indicating":[138],"selective":[140],"is":[142,190],"more":[143],"than":[145],"naive":[146],"depth":[147],"Furthermore,":[149],"can":[151],"scale":[152],"by":[155,194],"increasing":[156],"loops.":[160],"Adaptively":[161],"adjusting":[162],"throughout":[167],"sampling":[169],"process":[170],"further":[171],"additional":[173],"gains":[174],"maintaining":[179],"performance.":[180],"Lastly,":[181],"attention":[183],"analysis,":[184],"provide":[186],"evidence":[187],"promoting":[195],"interactions":[196],"among":[197],"masked":[198],"positions.":[199],"Our":[200],"code":[201],"weights":[203],"will":[204],"be":[205],"publicly":[206],"released.":[207]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-05-27T00:00:00"}