{"id":"https://openalex.org/W7154231987","doi":"https://doi.org/10.48550/arxiv.2604.09940","title":"New Hybrid Fine-Tuning Paradigm for LLMs: Algorithm Design and Convergence Analysis Framework","display_name":"New Hybrid Fine-Tuning Paradigm for LLMs: Algorithm Design and Convergence Analysis Framework","publication_year":2026,"publication_date":"2026-04-10","ids":{"openalex":"https://openalex.org/W7154231987","doi":"https://doi.org/10.48550/arxiv.2604.09940"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2604.09940","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.09940","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":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.2604.09940","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5133622647","display_name":"Shaocong Ma","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ma, Shaocong","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5133597822","display_name":"Peiran Yu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yu, Peiran","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5133605699","display_name":"Heng Huang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Huang, Heng","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/T10028","display_name":"Topic Modeling","score":0.1932000070810318,"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":0.1932000070810318,"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":0.09430000185966492,"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/T12535","display_name":"Machine Learning and Data Classification","score":0.08389999717473984,"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/convergence","display_name":"Convergence (economics)","score":0.6919999718666077},{"id":"https://openalex.org/keywords/smoothness","display_name":"Smoothness","score":0.5810999870300293},{"id":"https://openalex.org/keywords/joint","display_name":"Joint (building)","score":0.42320001125335693},{"id":"https://openalex.org/keywords/optimization-algorithm","display_name":"Optimization algorithm","score":0.3343000113964081},{"id":"https://openalex.org/keywords/sequence","display_name":"Sequence (biology)","score":0.3327000141143799},{"id":"https://openalex.org/keywords/empirical-research","display_name":"Empirical research","score":0.2922999858856201}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7168999910354614},{"id":"https://openalex.org/C2777303404","wikidata":"https://www.wikidata.org/wiki/Q759757","display_name":"Convergence (economics)","level":2,"score":0.6919999718666077},{"id":"https://openalex.org/C102634674","wikidata":"https://www.wikidata.org/wiki/Q868473","display_name":"Smoothness","level":2,"score":0.5810999870300293},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.47360000014305115},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.4690999984741211},{"id":"https://openalex.org/C18555067","wikidata":"https://www.wikidata.org/wiki/Q8375051","display_name":"Joint (building)","level":2,"score":0.42320001125335693},{"id":"https://openalex.org/C2987595161","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Optimization algorithm","level":2,"score":0.3343000113964081},{"id":"https://openalex.org/C2778112365","wikidata":"https://www.wikidata.org/wiki/Q3511065","display_name":"Sequence (biology)","level":2,"score":0.3327000141143799},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3089999854564667},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3059999942779541},{"id":"https://openalex.org/C120936955","wikidata":"https://www.wikidata.org/wiki/Q2155640","display_name":"Empirical research","level":2,"score":0.2922999858856201},{"id":"https://openalex.org/C137836250","wikidata":"https://www.wikidata.org/wiki/Q984063","display_name":"Optimization problem","level":2,"score":0.290800005197525},{"id":"https://openalex.org/C62469222","wikidata":"https://www.wikidata.org/wiki/Q17092103","display_name":"Hybrid algorithm (constraint satisfaction)","level":5,"score":0.28929999470710754},{"id":"https://openalex.org/C106516650","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm design","level":2,"score":0.28760001063346863},{"id":"https://openalex.org/C57869625","wikidata":"https://www.wikidata.org/wiki/Q1783502","display_name":"Rate of convergence","level":3,"score":0.2637999951839447},{"id":"https://openalex.org/C22367795","wikidata":"https://www.wikidata.org/wiki/Q7625208","display_name":"Structured prediction","level":2,"score":0.26190000772476196},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.25029999017715454}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2604.09940","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.09940","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":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.2604.09940","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.09940","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[{"score":0.4507986009120941,"display_name":"Quality Education","id":"https://metadata.un.org/sdg/4"}],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Fine-tuning":[0],"Large":[1],"Language":[2],"Models":[3],"(LLMs)":[4],"typically":[5],"involves":[6],"either":[7],"full":[8,32],"fine-tuning,":[9],"which":[10,19,95],"updates":[11,62],"all":[12],"model":[13,141,162],"parameters,":[14],"or":[15],"Parameter-Efficient":[16],"Fine-Tuning":[17],"(PEFT),":[18],"adjusts":[20],"a":[21,55,69,84,113,156],"small":[22],"subset":[23],"of":[24,71,91,101,120],"parameters.":[25],"However,":[26],"both":[27,63],"approaches":[28],"have":[29],"inherent":[30],"limitations:":[31],"fine-tuning":[33,58],"is":[34],"computationally":[35],"expensive,":[36],"while":[37],"PEFT":[38,66,109],"often":[39],"struggles":[40],"to":[41],"learn":[42],"new":[43,80],"knowledge":[44],"and":[45,65,73,108,128,140],"exhibits":[46],"suboptimal":[47],"performance.":[48],"To":[49,77],"overcome":[50],"these":[51],"issues,":[52],"we":[53,82],"propose":[54],"novel":[56],"hybrid":[57,92],"approach":[59,155],"that":[60],"jointly":[61],"LLMs":[64],"modules":[67],"using":[68],"combination":[70],"zeroth-order":[72],"first-order":[74],"optimization":[75,103],"methods.":[76],"analyze":[78],"our":[79,147],"algorithm,":[81],"develop":[83],"theoretical":[85],"framework":[86],"centered":[87],"on":[88],"the":[89,98,102,118,144,154],"concept":[90],"smoothness":[93],"condition,":[94],"accounts":[96],"for":[97,117,159],"heterogeneous":[99],"nature":[100],"landscape":[104],"in":[105],"joint":[106],"LLM":[107],"training.":[110],"We":[111],"derive":[112],"rigorous":[114],"convergence":[115,119],"analysis":[116],"reshuffling-type":[121],"SGD":[122],"algorithm":[123],"under":[124],"multiple":[125],"learning":[126],"rates":[127],"demonstrate":[129,149],"its":[130],"effectiveness":[131],"through":[132],"extensive":[133],"empirical":[134],"studies":[135],"across":[136],"various":[137],"downstream":[138],"tasks":[139],"architectures.":[142],"On":[143],"practical":[145],"side,":[146],"results":[148],"consistent":[150],"performance":[151],"improvement,":[152],"making":[153],"viable":[157],"solution":[158],"large-scale":[160],"language":[161],"fine-tuning.":[163]},"counts_by_year":[],"updated_date":"2026-07-28T07:46:37.118299","created_date":"2026-04-15T00:00:00"}