{"id":"https://openalex.org/W7152384572","doi":"https://doi.org/10.48550/arxiv.2604.06832","title":"Fast-dVLM: Efficient Block-Diffusion VLM via Direct Conversion from Autoregressive VLM","display_name":"Fast-dVLM: Efficient Block-Diffusion VLM via Direct Conversion from Autoregressive VLM","publication_year":2026,"publication_date":"2026-04-08","ids":{"openalex":"https://openalex.org/W7152384572","doi":"https://doi.org/10.48550/arxiv.2604.06832"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2604.06832","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.06832","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.2604.06832","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5133307286","display_name":"Chengyue Wu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wu, Chengyue","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5017304880","display_name":"Shiyi Lan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lan, Shiyi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5061572132","display_name":"Yonggan Fu","orcid":"https://orcid.org/0000-0002-7483-2921"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Fu, Yonggan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5114236320","display_name":"Sensen Gao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Gao, Sensen","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5133276140","display_name":"Jin Wang","orcid":"https://orcid.org/0000-0002-8298-4378"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Jin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5133282507","display_name":"Jincheng Yu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yu, Jincheng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5082207468","display_name":"Jos\u00e9 Alvarez","orcid":"https://orcid.org/0000-0003-3558-8302"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Alvarez, Jose M.","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5133249364","display_name":"Pavlo Molchanov","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Molchanov, Pavlo","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5133243235","display_name":"Ping Luo","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Luo, Ping","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5133307728","display_name":"Song Han","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Han, Song","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5133295069","display_name":"Ligeng Zhu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhu, Ligeng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5133302653","display_name":"Enze Xie","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xie, Enze","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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.7506999969482422,"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"}},"topics":[{"id":"https://openalex.org/T11714","display_name":"Multimodal Machine Learning Applications","score":0.7506999969482422,"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/T10775","display_name":"Generative Adversarial Networks and Image Synthesis","score":0.10920000076293945,"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.037300001829862595,"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/autoregressive-model","display_name":"Autoregressive model","score":0.6463000178337097},{"id":"https://openalex.org/keywords/context","display_name":"Context (archaeology)","score":0.63919997215271},{"id":"https://openalex.org/keywords/block","display_name":"Block (permutation group theory)","score":0.619700014591217},{"id":"https://openalex.org/keywords/decoding-methods","display_name":"Decoding methods","score":0.5931000113487244},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.5931000113487244},{"id":"https://openalex.org/keywords/speedup","display_name":"Speedup","score":0.48910000920295715},{"id":"https://openalex.org/keywords/humanoid-robot","display_name":"Humanoid robot","score":0.40369999408721924}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7400000095367432},{"id":"https://openalex.org/C159877910","wikidata":"https://www.wikidata.org/wiki/Q2202883","display_name":"Autoregressive model","level":2,"score":0.6463000178337097},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.63919997215271},{"id":"https://openalex.org/C2777210771","wikidata":"https://www.wikidata.org/wiki/Q4927124","display_name":"Block (permutation group theory)","level":2,"score":0.619700014591217},{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.5931000113487244},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.5931000113487244},{"id":"https://openalex.org/C68339613","wikidata":"https://www.wikidata.org/wiki/Q1549489","display_name":"Speedup","level":2,"score":0.48910000920295715},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4657000005245209},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.45210000872612},{"id":"https://openalex.org/C60692881","wikidata":"https://www.wikidata.org/wiki/Q584529","display_name":"Humanoid robot","level":3,"score":0.40369999408721924},{"id":"https://openalex.org/C79581498","wikidata":"https://www.wikidata.org/wiki/Q1367530","display_name":"Suite","level":2,"score":0.32350000739097595},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.3133000135421753},{"id":"https://openalex.org/C165696696","wikidata":"https://www.wikidata.org/wiki/Q11287","display_name":"Exploit","level":2,"score":0.3093000054359436},{"id":"https://openalex.org/C162307627","wikidata":"https://www.wikidata.org/wiki/Q204833","display_name":"Enhanced Data Rates for GSM Evolution","level":2,"score":0.3089999854564667},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.29679998755455017},{"id":"https://openalex.org/C5465570","wikidata":"https://www.wikidata.org/wiki/Q5326898","display_name":"Early stopping","level":3,"score":0.29490000009536743},{"id":"https://openalex.org/C2777472644","wikidata":"https://www.wikidata.org/wiki/Q16968992","display_name":"Approximate inference","level":3,"score":0.2721000015735626},{"id":"https://openalex.org/C34413123","wikidata":"https://www.wikidata.org/wiki/Q170978","display_name":"Robotics","level":3,"score":0.2700999975204468},{"id":"https://openalex.org/C160145156","wikidata":"https://www.wikidata.org/wiki/Q778586","display_name":"Executable","level":2,"score":0.2700999975204468},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.2632000148296356},{"id":"https://openalex.org/C31972630","wikidata":"https://www.wikidata.org/wiki/Q844240","display_name":"Computer vision","level":1,"score":0.26269999146461487},{"id":"https://openalex.org/C157764524","wikidata":"https://www.wikidata.org/wiki/Q1383412","display_name":"Throughput","level":3,"score":0.26179999113082886}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2604.06832","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.06832","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.2604.06832","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.06832","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":{"Vision-language":[0],"models":[1],"(VLMs)":[2],"predominantly":[3],"rely":[4],"on":[5,39],"autoregressive":[6,210],"decoding,":[7],"which":[8],"generates":[9],"tokens":[10,85],"one":[11,143],"at":[12,42],"a":[13,94,116,133,172],"time":[14],"and":[15,32,50,82,102,132,186,218],"fundamentally":[16],"limits":[17],"inference":[18,107,226],"throughput.":[19],"This":[20],"limitation":[21],"is":[22],"especially":[23],"acute":[24],"in":[25,142,196,212],"physical":[26],"AI":[27],"scenarios":[28],"such":[29],"as":[30,166],"robotics":[31],"autonomous":[33],"driving,":[34],"where":[35],"VLMs":[36,69],"are":[37],"deployed":[38],"edge":[40],"devices":[41],"batch":[43],"size":[44,179],"one,":[45],"making":[46],"AR":[47,140,230],"decoding":[48,101,105],"memory-bandwidth-bound":[49],"leaving":[51],"hardware":[52],"parallelism":[53],"underutilized.":[54],"While":[55],"block-wise":[56],"discrete":[57,83],"diffusion":[58,127,176,195],"has":[59],"shown":[60],"promise":[61],"for":[62,106],"parallel":[63,100],"text":[64,84],"generation,":[65],"extending":[66],"it":[67,165],"to":[68,73,76],"remains":[70],"challenging":[71],"due":[72],"the":[74,122,138,157,197,229],"need":[75],"jointly":[77],"handle":[78],"continuous":[79],"visual":[80],"representations":[81],"while":[86],"preserving":[87],"pretrained":[88],"multimodal":[89,130,175,204],"capabilities.":[90],"We":[91,109,170],"present":[92],"Fast-dVLM,":[93],"block-diffusion-based":[95],"VLM":[96,141,198],"that":[97,119,136,190],"enables":[98],"KV-cache-compatible":[99],"speculative":[103],"block":[104,178,194],"acceleration.":[108],"systematically":[110],"compare":[111],"two":[112],"AR-to-diffusion":[113],"conversion":[114,150],"strategies:":[115],"two-stage":[117],"approach":[118,135],"first":[120],"adapts":[121],"LLM":[123],"backbone":[124],"with":[125],"text-only":[126],"fine-tuning":[128],"before":[129],"training,":[131],"direct":[134,149],"converts":[137],"full":[139],"stage.":[144],"Under":[145],"comparable":[146],"training":[147],"budgets,":[148],"proves":[151],"substantially":[152],"more":[153],"efficient":[154,188],"by":[155],"leveraging":[156],"already":[158],"multimodally":[159],"aligned":[160],"VLM;":[161],"we":[162],"therefore":[163],"adopt":[164],"our":[167],"recommended":[168],"recipe.":[169],"introduce":[171],"suite":[173],"of":[174],"adaptations,":[177],"annealing,":[180],"causal":[181],"context":[182],"attention,":[183],"auto-truncation":[184],"masking,":[185],"vision":[187],"concatenation,":[189],"collectively":[191],"enable":[192],"effective":[193],"setting.":[199],"Extensive":[200],"experiments":[201],"across":[202],"11":[203],"benchmarks":[205],"show":[206],"Fast-dVLM":[207,221],"matches":[208],"its":[209],"counterpart":[211],"generation":[213],"quality.":[214],"With":[215],"SGLang":[216],"integration":[217],"FP8":[219],"quantization,":[220],"achieves":[222],"over":[223,228],"6x":[224],"end-to-end":[225],"speedup":[227],"baseline.":[231]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-04-10T00:00:00"}