{"id":"https://openalex.org/W7162666722","doi":"https://doi.org/10.48550/arxiv.2605.28760","title":"LLM Zeroth-Order Fine-Tuning is an Inference Workload","display_name":"LLM Zeroth-Order Fine-Tuning is an Inference Workload","publication_year":2026,"publication_date":"2026-05-27","ids":{"openalex":"https://openalex.org/W7162666722","doi":"https://doi.org/10.48550/arxiv.2605.28760"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.28760","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.28760","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.2605.28760","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5137281480","display_name":"Zelin Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Zelin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5137194865","display_name":"Caiwen Ding","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ding, Caiwen","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":0.24300000071525574,"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":0.24300000071525574,"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":0.1964000016450882,"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/T11948","display_name":"Machine Learning in Materials Science","score":0.061500001698732376,"subfield":{"id":"https://openalex.org/subfields/2505","display_name":"Materials Chemistry"},"field":{"id":"https://openalex.org/fields/25","display_name":"Materials Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/workload","display_name":"Workload","score":0.7666000127792358},{"id":"https://openalex.org/keywords/path","display_name":"Path (computing)","score":0.5601999759674072},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.46129998564720154},{"id":"https://openalex.org/keywords/training","display_name":"Training (meteorology)","score":0.39899998903274536},{"id":"https://openalex.org/keywords/implementation","display_name":"Implementation","score":0.3869999945163727},{"id":"https://openalex.org/keywords/sequence","display_name":"Sequence (biology)","score":0.385699987411499},{"id":"https://openalex.org/keywords/trajectory","display_name":"Trajectory","score":0.3659999966621399},{"id":"https://openalex.org/keywords/baseline","display_name":"Baseline (sea)","score":0.3578999936580658},{"id":"https://openalex.org/keywords/key","display_name":"Key (lock)","score":0.3513999879360199}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8125},{"id":"https://openalex.org/C2778476105","wikidata":"https://www.wikidata.org/wiki/Q628539","display_name":"Workload","level":2,"score":0.7666000127792358},{"id":"https://openalex.org/C2777735758","wikidata":"https://www.wikidata.org/wiki/Q817765","display_name":"Path (computing)","level":2,"score":0.5601999759674072},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.46129998564720154},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.39899998903274536},{"id":"https://openalex.org/C26713055","wikidata":"https://www.wikidata.org/wiki/Q245962","display_name":"Implementation","level":2,"score":0.3869999945163727},{"id":"https://openalex.org/C2778112365","wikidata":"https://www.wikidata.org/wiki/Q3511065","display_name":"Sequence (biology)","level":2,"score":0.385699987411499},{"id":"https://openalex.org/C79403827","wikidata":"https://www.wikidata.org/wiki/Q3988","display_name":"Real-time computing","level":1,"score":0.3783000111579895},{"id":"https://openalex.org/C13662910","wikidata":"https://www.wikidata.org/wiki/Q193139","display_name":"Trajectory","level":2,"score":0.3659999966621399},{"id":"https://openalex.org/C12725497","wikidata":"https://www.wikidata.org/wiki/Q810247","display_name":"Baseline (sea)","level":2,"score":0.3578999936580658},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.3513999879360199},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.33570000529289246},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.3285999894142151},{"id":"https://openalex.org/C18762648","wikidata":"https://www.wikidata.org/wiki/Q42213","display_name":"Work (physics)","level":2,"score":0.3255000114440918},{"id":"https://openalex.org/C139807058","wikidata":"https://www.wikidata.org/wiki/Q352374","display_name":"Adaptation (eye)","level":2,"score":0.32519999146461487},{"id":"https://openalex.org/C175291020","wikidata":"https://www.wikidata.org/wiki/Q1156822","display_name":"Offset (computer science)","level":2,"score":0.32190001010894775},{"id":"https://openalex.org/C177284502","wikidata":"https://www.wikidata.org/wiki/Q1005390","display_name":"Adapter (computing)","level":2,"score":0.31690001487731934},{"id":"https://openalex.org/C77618280","wikidata":"https://www.wikidata.org/wiki/Q1155772","display_name":"Scheme (mathematics)","level":2,"score":0.31349998712539673},{"id":"https://openalex.org/C187191949","wikidata":"https://www.wikidata.org/wiki/Q1138496","display_name":"Profiling (computer programming)","level":2,"score":0.3100000023841858},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.2808000147342682},{"id":"https://openalex.org/C89992363","wikidata":"https://www.wikidata.org/wiki/Q5961558","display_name":"Track (disk drive)","level":2,"score":0.27469998598098755},{"id":"https://openalex.org/C190839683","wikidata":"https://www.wikidata.org/wiki/Q2448197","display_name":"Train","level":2,"score":0.2718999981880188},{"id":"https://openalex.org/C63479239","wikidata":"https://www.wikidata.org/wiki/Q7353546","display_name":"Robustness (evolution)","level":3,"score":0.2703000009059906},{"id":"https://openalex.org/C99844830","wikidata":"https://www.wikidata.org/wiki/Q102441924","display_name":"Scaling","level":2,"score":0.2646999955177307},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.25920000672340393},{"id":"https://openalex.org/C150202949","wikidata":"https://www.wikidata.org/wiki/Q107602","display_name":"Pointer (user interface)","level":2,"score":0.251800000667572}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.28760","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.28760","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.2605.28760","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.28760","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":{"Zeroth-order":[0],"(ZO)":[1],"fine-tuning":[2,66],"is":[3,32,67],"attractive":[4],"for":[5,47,102],"large":[6],"language":[7],"models":[8],"because":[9],"it":[10],"replaces":[11],"backpropagation":[12],"with":[13],"forward":[14],"objective":[15],"evaluations.":[16],"Existing":[17],"implementations":[18],"nevertheless":[19],"execute":[20,72],"ZO":[21,65,167],"algorithms":[22],"inside":[23],"conventional":[24],"training":[25,97,195],"loops,":[26],"even":[27],"though":[28],"their":[29],"dominant":[30],"work":[31],"repeated":[33,74],"scoring":[34,75],"under":[35,107],"nearby":[36],"parameter":[37],"states.":[38],"This":[39],"creates":[40],"a":[41,55,78,154,174,193],"workload-runtime":[42],"mismatch:":[43],"the":[44,52,84,90,103,108,134,148],"algorithm":[45],"asks":[46],"structured":[48],"inference-style":[49],"scoring,":[50],"while":[51,115,158],"system":[53],"exposes":[54],"sequence":[56],"of":[57],"fragmented":[58],"training-loop":[59],"steps.":[60],"We":[61],"show":[62],"that":[63,147],"LLM":[64],"an":[68,112,187],"inference-dominated":[69],"workload":[70,189],"and":[71,121],"its":[73],"phase":[76],"through":[77],"serving":[79],"runtime.":[80],"On":[81],"OPT-13B":[82],"SST-2,":[83],"resulting":[85],"vLLM":[86],"execution":[87],"path":[88,176],"completes":[89],"20k-step":[91],"LoZO":[92,105],"run":[93],"in":[94],"0.51":[95],"estimated":[96],"hours":[98,101],"versus":[99],"4.15":[100],"official":[104],"baseline":[106],"matched":[109],"LoRA-only":[110],"setting,":[111],"8.13x":[113],"speedup,":[114],"reaching":[116],"0.922":[117],"final":[118,123],"evaluation":[119],"accuracy":[120],"0.931":[122],"full-validation":[124],"accuracy.":[125],"In":[126],"core-step":[127],"scaling":[128],"experiments":[129],"across":[130],"OPT-1.3B":[131],"to":[132,161],"OPT-13B,":[133],"same":[135,149],"runtime":[136,150],"reorganization":[137],"gives":[138],"2.34x--7.72x":[139],"speedups.":[140],"A":[141],"MeZO-style":[142],"high-rank":[143],"factorized":[144],"experiment":[145],"shows":[146],"paradigm":[151],"can":[152,183],"track":[153],"MeZO-like":[155],"loss":[156],"trajectory":[157],"running":[159],"up":[160],"2.55x":[162],"faster.":[163],"More":[164],"broadly,":[165],"representing":[166],"updates":[168],"as":[169,186,192],"dynamic":[170],"adapter":[171],"states":[172],"suggests":[173],"practical":[175],"toward":[177],"inference-time":[178],"training,":[179],"where":[180],"lightweight":[181],"adaptation":[182],"be":[184],"scheduled":[185],"inference-like":[188],"rather":[190],"than":[191],"separate":[194],"job.":[196]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-05-29T00:00:00"}