{"id":"https://openalex.org/W7165613460","doi":"https://doi.org/10.48550/arxiv.2606.23607","title":"Scaling Linear Mode Connectivity and Merging to Billion Parameter Pretrained Transformers","display_name":"Scaling Linear Mode Connectivity and Merging to Billion Parameter Pretrained Transformers","publication_year":2026,"publication_date":"2026-06-22","ids":{"openalex":"https://openalex.org/W7165613460","doi":"https://doi.org/10.48550/arxiv.2606.23607"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2606.23607","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.23607","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.2606.23607","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5139163273","display_name":"Tianyi Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Tianyi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5139176898","display_name":"Zhiqiang Shen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Shen, Zhiqiang","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/T10036","display_name":"Advanced Neural Network Applications","score":0.21789999306201935,"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/T10036","display_name":"Advanced Neural Network Applications","score":0.21789999306201935,"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":0.21359999477863312,"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":0.08449999988079071,"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/scalability","display_name":"Scalability","score":0.670799970626831},{"id":"https://openalex.org/keywords/interpolation","display_name":"Interpolation (computer graphics)","score":0.5813999772071838},{"id":"https://openalex.org/keywords/linear-interpolation","display_name":"Linear interpolation","score":0.5468999743461609},{"id":"https://openalex.org/keywords/parameterized-complexity","display_name":"Parameterized complexity","score":0.5353000164031982},{"id":"https://openalex.org/keywords/scaling","display_name":"Scaling","score":0.4997999966144562},{"id":"https://openalex.org/keywords/transformer","display_name":"Transformer","score":0.48570001125335693},{"id":"https://openalex.org/keywords/linear-model","display_name":"Linear model","score":0.39660000801086426},{"id":"https://openalex.org/keywords/linear-programming","display_name":"Linear programming","score":0.3944999873638153}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7092999815940857},{"id":"https://openalex.org/C48044578","wikidata":"https://www.wikidata.org/wiki/Q727490","display_name":"Scalability","level":2,"score":0.670799970626831},{"id":"https://openalex.org/C137800194","wikidata":"https://www.wikidata.org/wiki/Q11713455","display_name":"Interpolation (computer graphics)","level":3,"score":0.5813999772071838},{"id":"https://openalex.org/C171836373","wikidata":"https://www.wikidata.org/wiki/Q2266329","display_name":"Linear interpolation","level":3,"score":0.5468999743461609},{"id":"https://openalex.org/C165464430","wikidata":"https://www.wikidata.org/wiki/Q1570441","display_name":"Parameterized complexity","level":2,"score":0.5353000164031982},{"id":"https://openalex.org/C99844830","wikidata":"https://www.wikidata.org/wiki/Q102441924","display_name":"Scaling","level":2,"score":0.4997999966144562},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.48570001125335693},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.41749998927116394},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4032000005245209},{"id":"https://openalex.org/C163175372","wikidata":"https://www.wikidata.org/wiki/Q3339222","display_name":"Linear model","level":2,"score":0.39660000801086426},{"id":"https://openalex.org/C41045048","wikidata":"https://www.wikidata.org/wiki/Q202843","display_name":"Linear programming","level":2,"score":0.3944999873638153},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3720000088214874},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.310699999332428},{"id":"https://openalex.org/C2777735758","wikidata":"https://www.wikidata.org/wiki/Q817765","display_name":"Path (computing)","level":2,"score":0.30630001425743103},{"id":"https://openalex.org/C188198153","wikidata":"https://www.wikidata.org/wiki/Q1613840","display_name":"Limiting","level":2,"score":0.29499998688697815},{"id":"https://openalex.org/C108583219","wikidata":"https://www.wikidata.org/wiki/Q197536","display_name":"Deep learning","level":2,"score":0.2833999991416931},{"id":"https://openalex.org/C206729178","wikidata":"https://www.wikidata.org/wiki/Q2271896","display_name":"Scheduling (production processes)","level":2,"score":0.28119999170303345},{"id":"https://openalex.org/C170122806","wikidata":"https://www.wikidata.org/wiki/Q1914828","display_name":"Linear scale","level":2,"score":0.2720000147819519},{"id":"https://openalex.org/C203332170","wikidata":"https://www.wikidata.org/wiki/Q6334079","display_name":"Multivariate interpolation","level":3,"score":0.26649999618530273},{"id":"https://openalex.org/C113775141","wikidata":"https://www.wikidata.org/wiki/Q428691","display_name":"Computer engineering","level":1,"score":0.2563000023365021},{"id":"https://openalex.org/C185798385","wikidata":"https://www.wikidata.org/wiki/Q1161707","display_name":"Benchmark (surveying)","level":2,"score":0.25200000405311584},{"id":"https://openalex.org/C49766605","wikidata":"https://www.wikidata.org/wiki/Q207643","display_name":"Linear map","level":2,"score":0.25189998745918274}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2606.23607","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.23607","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.2606.23607","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.23607","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":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Linear":[0],"mode":[1],"connectivity":[2,132],"(LMC)":[3],"provides":[4],"a":[5,40,70,84],"promising":[6],"foundation":[7],"for":[8,34,45,116],"understanding":[9],"and":[10,32,42,68,96,175],"merging":[11,49,100],"independently":[12],"trained":[13],"neural":[14],"networks,":[15],"but":[16],"existing":[17],"methods":[18],"typically":[19],"optimize":[20],"the":[21,126,137,148],"interpolation":[22,87,94,149,184],"path":[23],"from":[24],"only":[25,156],"one":[26],"model":[27,48],"endpoint,":[28],"limiting":[29],"their":[30,80],"scalability":[31],"effectiveness":[33],"large":[35,169],"pretrained":[36,53,170],"transformers.":[37],"We":[38],"propose":[39],"novel":[41],"scalable":[43],"framework":[44],"enabling":[46],"LMC-based":[47],"to":[50,63,123,172],"{\\em":[51],"billion-parameter":[52,153],"transformers}.":[54],"Our":[55],"method":[56],"applies":[57],"properly":[58,164],"parameterized":[59],"functionality-preserving":[60],"weight":[61],"transformations":[62,82],"align":[64],"functionally":[65],"equivalent":[66],"solutions,":[67],"introduces":[69],"dual":[71],"learning":[72],"procedure":[73],"in":[74],"which":[75],"both":[76],"models":[77,118],"jointly":[78],"learn":[79],"corresponding":[81],"toward":[83],"shared":[85],"linear":[86,131,179],"path.":[88],"This":[89],"bidirectional":[90],"optimization":[91],"substantially":[92,182],"reduces":[93],"barriers":[95,113],"enables":[97,168],"more":[98],"reliable":[99],"across":[101],"large-scale":[102],"architectures.":[103],"Empirically,":[104],"we":[105],"show":[106],"that":[107,163],"our":[108,124],"approach":[109],"achieves":[110],"near-zero":[111],"loss":[112,158],"on":[114],"WikiText":[115],"language":[117],"with":[119,181],"medium-sized":[120],"parameters,":[121],"representing,":[122],"knowledge,":[125],"first":[127],"demonstration":[128],"of":[129],"near-barrier-free":[130],"at":[133],"this":[134],"scale.":[135],"In":[136],"vision":[138],"domain,":[139],"ViT-L":[140],"maintains":[141],"above":[142],"69\\%":[143],"ImageNet":[144],"top-1":[145],"accuracy":[146],"throughout":[147],"path,":[150],"while":[151],"modern":[152],"LLMs":[154],"exhibit":[155],"small":[157],"barriers.":[159],"These":[160],"results":[161],"suggest":[162],"resolving":[165],"parameter":[166],"symmetries":[167],"Transformers":[171],"be":[173],"connected":[174],"merged":[176],"through":[177],"simple":[178],"paths":[180],"improved":[183],"performance.":[185],"Code:":[186],"https://github.com/VILA-Lab/Dual-Learned-Matching":[187],".":[188]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-06-24T00:00:00"}