{"id":"https://openalex.org/W7214754875","doi":"https://doi.org/10.48550/arxiv.2609.32448","title":"ForkLeft: Entropy-First Rollouts for Prefix-Aligned Autoregressive-to-Diffusion Distillation","display_name":"ForkLeft: Entropy-First Rollouts for Prefix-Aligned Autoregressive-to-Diffusion Distillation","publication_year":2026,"publication_date":"2026-09-26","ids":{"openalex":"https://openalex.org/W7214754875","doi":"https://doi.org/10.48550/arxiv.2609.32448"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2609.32448","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.32448","pdf_url":"https://arxiv.org/pdf/2609.32448","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/2609.32448","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5153505709","display_name":"Junming Liu","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I165779595","display_name":"The University of Melbourne","ror":"https://ror.org/01ej9dk98","country_code":"AU","type":"education","lineage":["https://openalex.org/I165779595"]}],"countries":["AU"],"is_corresponding":false,"raw_author_name":"Liu, Junming","raw_affiliation_strings":["University of Melbourne"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Melbourne","institution_ids":["https://openalex.org/I165779595"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5153487734","display_name":"Jicheng Wang","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I84218800","display_name":"University of California, Davis","ror":"https://ror.org/05rrcem69","country_code":"US","type":"education","lineage":["https://openalex.org/I84218800"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Wang, Jicheng","raw_affiliation_strings":["University of California , Davis"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of California , Davis","institution_ids":["https://openalex.org/I84218800"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5153509505","display_name":"Yifeng He","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I84218800","display_name":"University of California, Davis","ror":"https://ror.org/05rrcem69","country_code":"US","type":"education","lineage":["https://openalex.org/I84218800"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"He, Yifeng","raw_affiliation_strings":["University of California , Davis"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of California , Davis","institution_ids":["https://openalex.org/I84218800"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5153503869","display_name":"Hao Mark Chen","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I889458895","display_name":"University of Hong Kong","ror":"https://ror.org/02zhqgq86","country_code":"HK","type":"education","lineage":["https://openalex.org/I889458895"]}],"countries":["HK"],"is_corresponding":false,"raw_author_name":"Chen, Hao","raw_affiliation_strings":["University of Hong Kong"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Hong Kong","institution_ids":["https://openalex.org/I889458895"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5135586197","display_name":"Jianzhong Qi","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I165779595","display_name":"The University of Melbourne","ror":"https://ror.org/01ej9dk98","country_code":"AU","type":"education","lineage":["https://openalex.org/I165779595"]}],"countries":["AU"],"is_corresponding":false,"raw_author_name":"Qi, Jianzhong","raw_affiliation_strings":["University of Melbourne"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Melbourne","institution_ids":["https://openalex.org/I165779595"]}]}],"institutions":[],"countries_distinct_count":3,"institutions_distinct_count":3,"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/T10028","display_name":"Topic Modeling","score":0.7836999893188477,"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":0.7836999893188477,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.18649999797344208,"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":0.004900000058114529,"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/diffusion-language-models","display_name":"diffusion language models","score":0.8379999995231628},{"id":"https://openalex.org/keywords/knowledge-distillation","display_name":"knowledge distillation","score":0.8299999833106995},{"id":"https://openalex.org/keywords/next-token-prediction","display_name":"next-token prediction","score":0.8489999771118164},{"id":"https://openalex.org/keywords/parallel-decoding","display_name":"parallel decoding","score":0.7229999899864197},{"id":"https://openalex.org/keywords/reasoning","display_name":"reasoning","score":0.3370000123977661}],"concepts":[{"id":"https://openalex.org/C153180980","wikidata":"https://www.wikidata.org/wiki/Q19776675","display_name":"Commit","level":2,"score":0.7544000148773193},{"id":"https://openalex.org/C204030448","wikidata":"https://www.wikidata.org/wiki/Q101017","display_name":"Distillation","level":2,"score":0.7213000059127808},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.66839998960495},{"id":"https://openalex.org/C55439883","wikidata":"https://www.wikidata.org/wiki/Q360812","display_name":"Correctness","level":2,"score":0.6437000036239624},{"id":"https://openalex.org/C2778755073","wikidata":"https://www.wikidata.org/wiki/Q10858537","display_name":"Scale (ratio)","level":2,"score":0.5403000116348267},{"id":"https://openalex.org/C169590947","wikidata":"https://www.wikidata.org/wiki/Q47506","display_name":"Compiler","level":2,"score":0.526199996471405},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.45239999890327454},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.4472000002861023},{"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/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.38940000534057617},{"id":"https://openalex.org/C141603448","wikidata":"https://www.wikidata.org/wiki/Q134830","display_name":"Prefix","level":2,"score":0.383899986743927},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.30300000309944153},{"id":"https://openalex.org/C2778514511","wikidata":"https://www.wikidata.org/wiki/Q1374194","display_name":"Programmer","level":2,"score":0.30149999260902405},{"id":"https://openalex.org/C122783720","wikidata":"https://www.wikidata.org/wiki/Q183065","display_name":"Interpreter","level":2,"score":0.28940001130104065},{"id":"https://openalex.org/C2780009758","wikidata":"https://www.wikidata.org/wiki/Q6804172","display_name":"Measure (data warehouse)","level":2,"score":0.2870999872684479},{"id":"https://openalex.org/C140779682","wikidata":"https://www.wikidata.org/wiki/Q210868","display_name":"Sampling (signal processing)","level":3,"score":0.2720000147819519},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.2614000141620636},{"id":"https://openalex.org/C57098296","wikidata":"https://www.wikidata.org/wiki/Q12021746","display_name":"Interrogative","level":2,"score":0.2531000077724457},{"id":"https://openalex.org/C199360897","wikidata":"https://www.wikidata.org/wiki/Q9143","display_name":"Programming language","level":1,"score":0.25029999017715454}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2609.32448","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.32448","pdf_url":"https://arxiv.org/pdf/2609.32448","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.2609.32448","doi":"https://doi.org/10.48550/arxiv.2609.32448","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2609.32448","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:2609.32448","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.32448","pdf_url":"https://arxiv.org/pdf/2609.32448","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":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7214754875.pdf","grobid_xml":"https://content.openalex.org/works/W7214754875.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7221780958","https://openalex.org/W7163900200","https://openalex.org/W7208761962","https://openalex.org/W7166870473","https://openalex.org/W7222006676","https://openalex.org/W7131895179","https://openalex.org/W7169878852","https://openalex.org/W7124468437","https://openalex.org/W7214782337","https://openalex.org/W7159019538"],"abstract_inverted_index":{"Autoregressive":[0],"Next-Token":[1],"Prediction":[2],"(NTP)":[3],"has":[4],"enabled":[5],"strong":[6],"reasoning":[7,31,194],"capabilities":[8],"in":[9],"language":[10],"models,":[11],"while":[12],"Diffusion":[13],"Language":[14],"Models":[15],"(DLMs)":[16],"offer":[17],"flexible":[18],"token":[19],"orders":[20],"and":[21,96,107,149,161,175,182,201],"parallel":[22,132,198],"generation.":[23,199],"We":[24,66,100],"ask":[25],"whether":[26],"DLMs":[27,190],"can":[28,59,191],"acquire":[29],"NTP-style":[30,193],"through":[32],"distillation":[33,70],"without":[34,195],"giving":[35],"up":[36],"their":[37],"native":[38,130,197],"generation":[39],"process.":[40],"Direct":[41],"distillation,":[42],"however,":[43],"faces":[44],"a":[45,53,57,69],"fundamental":[46],"mismatch:":[47],"an":[48,109],"autoregressive":[49],"teacher":[50,82,111,159],"predicts":[51],"from":[52,81,145],"left":[54],"prefix,":[55],"whereas":[56],"DLM":[58],"condition":[60],"on":[61,63,139,185],"tokens":[62],"both":[64],"sides.":[65],"introduce":[67],"ForkLeft,":[68],"framework":[71],"that":[72,92,189],"resolves":[73],"this":[74],"mismatch":[75],"by":[76],"separating":[77],"the":[78,86,103,113,120,125,172,179],"student's":[79],"rollout":[80],"supervision.":[83],"During":[84],"training,":[85],"student":[87,105,126],"first":[88],"performs":[89],"entropy-first":[90],"rollouts":[91],"commit":[93],"uncertain":[94],"positions":[95],"expose":[97],"potential":[98],"forks.":[99],"then":[101],"fix":[102],"resulting":[104],"prefix":[106],"distill":[108],"NTP":[110],"under":[112],"same":[114],"context,":[115],"with":[116,158,165],"answer":[117],"correctness":[118],"determining":[119],"supervision":[121],"source.":[122],"At":[123,169],"inference,":[124],"returns":[127],"to":[128,147,163],"its":[129],"confidence-first":[131],"decoding.":[133],"With":[134],"Qwen3-30B-A3B-Base,":[135],"ForkLeft":[136],"improves":[137],"Efficient-DLM-4B":[138],"all":[140],"ten":[141],"benchmarks,":[142,187],"raising":[143],"MATH500":[144],"72.60%":[146],"79.60%":[148],"consistently":[150],"outperforming":[151],"three":[152],"alternative":[153],"designs.":[154],"The":[155],"gains":[156],"scale":[157],"strength":[160],"generalize":[162],"SDAR-4B":[164],"only":[166],"$500$":[167],"updates.":[168],"matched":[170],"scale,":[171],"distilled":[173],"4B":[174],"8B":[176],"students":[177],"exceed":[178],"published":[180],"SDAR-Chat":[181],"OPDLM":[183],"models":[184],"seven":[186],"showing":[188],"learn":[192],"sacrificing":[196],"Code":[200],"datasets":[202],"will":[203],"be":[204],"released":[205],"upon":[206],"acceptance.":[207]},"counts_by_year":[],"updated_date":"2026-10-28T10:33:24.094632","created_date":"2026-09-30T00:00:00"}