{"id":"https://openalex.org/W7127589246","doi":"https://doi.org/10.48550/arxiv.2602.03815","title":"Fast-Slow Efficient Training for Multimodal Large Language Models via Visual Token Pruning","display_name":"Fast-Slow Efficient Training for Multimodal Large Language Models via Visual Token Pruning","publication_year":2026,"publication_date":"2026-02-03","ids":{"openalex":"https://openalex.org/W7127589246","doi":"https://doi.org/10.48550/arxiv.2602.03815"},"language":null,"primary_location":{"id":"pmh:doi:10.48550/arxiv.2602.03815","is_oa":true,"landing_page_url":null,"pdf_url":null,"source":{"id":"https://openalex.org/S4406922384","display_name":"Open MIND","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Article"},"type":"article","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":null,"any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5125018191","display_name":"Dingkun Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Dingkun","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5125025091","display_name":"Shuhan Qi","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Qi, Shuhan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5124994486","display_name":"Yulin Wu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wu, Yulin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5125010482","display_name":"Xinyu Xiao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xiao, Xinyu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5125056636","display_name":"Xuan Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Xuan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5125051091","display_name":"Long \u9648\u9f99 Chen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, Long","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":0.0,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":{"value":0.12773666,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"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":0.6362000107765198,"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":0.6362000107765198,"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":0.20479999482631683,"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":0.027699999511241913,"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/security-token","display_name":"Security token","score":0.6344000101089478},{"id":"https://openalex.org/keywords/pruning","display_name":"Pruning","score":0.5364000201225281},{"id":"https://openalex.org/keywords/training","display_name":"Training (meteorology)","score":0.5228999853134155},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.487199991941452},{"id":"https://openalex.org/keywords/focus","display_name":"Focus (optics)","score":0.47850000858306885},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.46880000829696655},{"id":"https://openalex.org/keywords/inefficiency","display_name":"Inefficiency","score":0.42289999127388}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8246999979019165},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.6344000101089478},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.567300021648407},{"id":"https://openalex.org/C108010975","wikidata":"https://www.wikidata.org/wiki/Q500094","display_name":"Pruning","level":2,"score":0.5364000201225281},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.5228999853134155},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.487199991941452},{"id":"https://openalex.org/C192209626","wikidata":"https://www.wikidata.org/wiki/Q190909","display_name":"Focus (optics)","level":2,"score":0.47850000858306885},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.46880000829696655},{"id":"https://openalex.org/C2778869765","wikidata":"https://www.wikidata.org/wiki/Q6028363","display_name":"Inefficiency","level":2,"score":0.42289999127388},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.4129999876022339},{"id":"https://openalex.org/C36464697","wikidata":"https://www.wikidata.org/wiki/Q451553","display_name":"Visualization","level":2,"score":0.3472000062465668},{"id":"https://openalex.org/C2776502983","wikidata":"https://www.wikidata.org/wiki/Q690182","display_name":"Contrast (vision)","level":2,"score":0.34290000796318054},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.3269999921321869},{"id":"https://openalex.org/C4924752","wikidata":"https://www.wikidata.org/wiki/Q184148","display_name":"Plug-in","level":2,"score":0.30809998512268066},{"id":"https://openalex.org/C124304363","wikidata":"https://www.wikidata.org/wiki/Q673661","display_name":"Abstraction","level":2,"score":0.2996000051498413},{"id":"https://openalex.org/C48677424","wikidata":"https://www.wikidata.org/wiki/Q6888088","display_name":"Mode (computer interface)","level":2,"score":0.2989000082015991},{"id":"https://openalex.org/C28490314","wikidata":"https://www.wikidata.org/wiki/Q189436","display_name":"Speech recognition","level":1,"score":0.2791999876499176},{"id":"https://openalex.org/C2778112365","wikidata":"https://www.wikidata.org/wiki/Q3511065","display_name":"Sequence (biology)","level":2,"score":0.2538999915122986}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:doi:10.48550/arxiv.2602.03815","is_oa":true,"landing_page_url":null,"pdf_url":null,"source":{"id":"https://openalex.org/S4406922384","display_name":"Open MIND","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Article"},{"id":"doi:10.48550/arxiv.2602.03815","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2602.03815","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,"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":"pmh:doi:10.48550/arxiv.2602.03815","is_oa":true,"landing_page_url":null,"pdf_url":null,"source":{"id":"https://openalex.org/S4406922384","display_name":"Open MIND","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Article"},"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":{"Multimodal":[0],"Large":[1],"Language":[2],"Models":[3],"(MLLMs)":[4],"suffer":[5],"from":[6,158],"severe":[7],"training":[8,27,58,68,99,168,178],"inefficiency":[9],"issue,":[10],"which":[11,108],"is":[12,104,131,138],"associated":[13],"with":[14,120],"their":[15],"massive":[16],"model":[17,31,137],"sizes":[18,32],"and":[19,170,183],"visual":[20,61,84,117,142],"token":[21,85],"numbers.":[22],"Existing":[23],"efforts":[24],"in":[25,45,71],"efficient":[26,57,98],"focus":[28],"on":[29,81,140],"reducing":[30,60],"or":[33],"trainable":[34],"parameters.":[35],"Inspired":[36],"by":[37,59,181,185],"the":[38,67,105,126,132,136,159,177],"success":[39],"of":[40,100,179],"Visual":[41],"Token":[42],"Pruning":[43],"(VTP)":[44],"improving":[46],"inference":[47],"efficiency,":[48],"we":[49,91],"are":[50],"exploring":[51],"another":[52],"substantial":[53],"research":[54],"direction":[55],"for":[56,97],"tokens.":[62],"However,":[63],"applying":[64],"VTP":[65,111],"at":[66],"stage":[69],"results":[70],"a":[72,94,121],"training-inference":[73,146],"mismatch:":[74],"pruning-trained":[75],"models":[76],"perform":[77],"poorly":[78],"when":[79],"inferring":[80],"non-pruned":[82],"full":[83,141],"sequences.":[86],"To":[87,148],"close":[88],"this":[89],"gap,":[90],"propose":[92],"DualSpeed,":[93],"fast-slow":[95],"framework":[96],"MLLMs.":[101],"The":[102,129],"fast-mode":[103],"primary":[106],"mode,":[107,134],"incorporates":[109],"existing":[110],"methods":[112],"as":[113],"plugins":[114],"to":[115,124,144,156],"reduce":[116],"tokens,":[118],"along":[119],"mode":[122],"isolator":[123],"isolate":[125],"model's":[127],"behaviors.":[128],"slow-mode":[130],"auxiliary":[133],"where":[135],"trained":[139,161],"sequences":[143],"retain":[145],"consistency.":[147],"boost":[149],"its":[150],"training,":[151],"it":[152],"further":[153],"leverages":[154],"self-distillation":[155],"learn":[157],"sufficiently":[160],"fast-mode.":[162],"Together,":[163],"DualSpeed":[164,175],"can":[165],"achieve":[166],"both":[167],"efficiency":[169],"non-degraded":[171],"performance.":[172,190],"Experiments":[173],"show":[174],"accelerates":[176],"LLaVA-1.5":[180],"2.1$\\times$":[182],"LLaVA-NeXT":[184],"4.0$\\times$,":[186],"retaining":[187],"over":[188],"99%":[189],"Code:":[191],"https://github.com/dingkun-zhang/DualSpeed":[192]},"counts_by_year":[],"updated_date":"2026-07-28T07:46:37.118299","created_date":"2026-02-06T00:00:00"}