{"id":"https://openalex.org/W7164972282","doi":"https://doi.org/10.48550/arxiv.2606.16067","title":"Stepwise Token Selection for Efficient Multimodal Large Language Models","display_name":"Stepwise Token Selection for Efficient Multimodal Large Language Models","publication_year":2026,"publication_date":"2026-06-14","ids":{"openalex":"https://openalex.org/W7164972282","doi":"https://doi.org/10.48550/arxiv.2606.16067"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2606.16067","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.16067","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.2606.16067","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5128646065","display_name":"Landi He","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"He, Landi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5127926984","display_name":"Shawn Young","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Young, Shawn","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5015286750","display_name":"Lei Xu","orcid":"https://orcid.org/0009-0003-1055-9318"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xu, Lijian","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":0.6764000058174133,"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":0.6764000058174133,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.09939999878406525,"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/T10028","display_name":"Topic Modeling","score":0.05380000174045563,"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/security-token","display_name":"Security token","score":0.7110999822616577},{"id":"https://openalex.org/keywords/pruning","display_name":"Pruning","score":0.5419999957084656},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.5252000093460083},{"id":"https://openalex.org/keywords/selection","display_name":"Selection (genetic algorithm)","score":0.5235999822616577},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.46790000796318054},{"id":"https://openalex.org/keywords/interpolation","display_name":"Interpolation (computer graphics)","score":0.39570000767707825},{"id":"https://openalex.org/keywords/abstraction","display_name":"Abstraction","score":0.3840000033378601},{"id":"https://openalex.org/keywords/reduction","display_name":"Reduction (mathematics)","score":0.3659000098705292}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7789000272750854},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.7110999822616577},{"id":"https://openalex.org/C108010975","wikidata":"https://www.wikidata.org/wiki/Q500094","display_name":"Pruning","level":2,"score":0.5419999957084656},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5412999987602234},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.5252000093460083},{"id":"https://openalex.org/C81917197","wikidata":"https://www.wikidata.org/wiki/Q628760","display_name":"Selection (genetic algorithm)","level":2,"score":0.5235999822616577},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.46790000796318054},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.43320000171661377},{"id":"https://openalex.org/C137800194","wikidata":"https://www.wikidata.org/wiki/Q11713455","display_name":"Interpolation (computer graphics)","level":3,"score":0.39570000767707825},{"id":"https://openalex.org/C124304363","wikidata":"https://www.wikidata.org/wiki/Q673661","display_name":"Abstraction","level":2,"score":0.3840000033378601},{"id":"https://openalex.org/C111335779","wikidata":"https://www.wikidata.org/wiki/Q3454686","display_name":"Reduction (mathematics)","level":2,"score":0.3659000098705292},{"id":"https://openalex.org/C35651441","wikidata":"https://www.wikidata.org/wiki/Q625303","display_name":"Independence (probability theory)","level":2,"score":0.35040000081062317},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.3424000144004822},{"id":"https://openalex.org/C2776502983","wikidata":"https://www.wikidata.org/wiki/Q690182","display_name":"Contrast (vision)","level":2,"score":0.3273000121116638},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.3158999979496002},{"id":"https://openalex.org/C28855332","wikidata":"https://www.wikidata.org/wiki/Q198099","display_name":"Quantization (signal processing)","level":2,"score":0.3127000033855438},{"id":"https://openalex.org/C171836373","wikidata":"https://www.wikidata.org/wiki/Q2266329","display_name":"Linear interpolation","level":3,"score":0.303600013256073},{"id":"https://openalex.org/C118505674","wikidata":"https://www.wikidata.org/wiki/Q42586063","display_name":"Encoder","level":2,"score":0.29660001397132874},{"id":"https://openalex.org/C141603448","wikidata":"https://www.wikidata.org/wiki/Q134830","display_name":"Prefix","level":2,"score":0.29089999198913574},{"id":"https://openalex.org/C2781039887","wikidata":"https://www.wikidata.org/wiki/Q1391724","display_name":"Factor (programming language)","level":2,"score":0.2816999852657318},{"id":"https://openalex.org/C2780385302","wikidata":"https://www.wikidata.org/wiki/Q367158","display_name":"Protocol (science)","level":3,"score":0.27459999918937683},{"id":"https://openalex.org/C99498987","wikidata":"https://www.wikidata.org/wiki/Q2210247","display_name":"Noise (video)","level":3,"score":0.26179999113082886},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.2531999945640564}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2606.16067","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.16067","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.2606.16067","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.16067","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":[{"id":"https://metadata.un.org/sdg/16","score":0.5262434482574463,"display_name":"Peace, Justice and strong institutions"}],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"In":[0,59],"multimodal":[1],"large":[2],"language":[3,19,120],"models":[4],"(MLLMs),":[5],"inference":[6],"cost":[7],"is":[8],"largely":[9],"dominated":[10],"by":[11],"the":[12,18,108,140,176],"visual":[13,38,64,169],"token":[14,22,50,65],"prefix":[15],"rather":[16],"than":[17],"backbone,":[20],"making":[21],"reduction":[23],"a":[24,42,53,68,75,98,125,130,181],"key":[25],"factor":[26],"for":[27],"improving":[28],"efficiency.":[29],"Existing":[30],"approaches":[31],"typically":[32],"assign":[33],"independent":[34],"importance":[35],"scores":[36],"to":[37,95,137],"tokens":[39],"and":[40,52,91,111,148],"retain":[41],"fixed":[43],"number":[44],"of":[45,106,168,175],"top-ranked":[46],"tokens,":[47,83,170],"implicitly":[48],"assuming":[49],"independence":[51],"uniform":[54],"compression":[55,160],"ratio":[56],"across":[57,158],"inputs.":[58],"this":[60],"work,":[61],"we":[62,73,123],"reformulate":[63],"pruning":[66,164],"as":[67],"sequential":[69],"decision-making":[70],"process.":[71,143],"Specifically,":[72],"introduce":[74],"pointer-style":[76],"selection":[77,142],"mechanism":[78],"that":[79,151,165],"iteratively":[80],"chooses":[81],"informative":[82],"conditioning":[84],"each":[85],"decision":[86],"on":[87,129,146],"previously":[88],"selected":[89,109],"ones,":[90],"dynamically":[92],"determines":[93],"when":[94],"stop":[96],"via":[97],"learned":[99],"termination":[100],"action.":[101],"This":[102],"enables":[103],"joint":[104],"optimization":[105],"both":[107],"subset":[110],"its":[112],"size.":[113],"To":[114],"enable":[115],"end-to-end":[116],"training":[117],"under":[118],"standard":[119],"modeling":[121],"objectives,":[122],"design":[124],"differentiable":[126],"relaxation":[127],"based":[128],"variance-preserving":[131],"noise":[132],"interpolation":[133],"scheme,":[134],"allowing":[135],"gradients":[136],"propagate":[138],"through":[139],"discrete":[141],"Extensive":[144],"experiments":[145],"LLaVA-v1.5-7B":[147],"Qwen2.5-VL-7B":[149],"demonstrate":[150],"our":[152,171],"approach":[153],"consistently":[154],"outperforms":[155],"fixed-ratio":[156],"baselines":[157],"different":[159],"levels.":[161],"Under":[162],"aggressive":[163],"removes":[166],"88.9%":[167],"method":[172],"preserves":[173],"94.6%":[174],"original":[177],"accuracy":[178],"while":[179],"achieving":[180],"1.88x":[182],"speed-up":[183],"in":[184],"prefill":[185],"latency.":[186]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-06-17T00:00:00"}