{"id":"https://openalex.org/W4413640445","doi":"https://doi.org/10.23919/wiopt66569.2025.11123401","title":"Large Language Model Partitioning for Low-Latency Inference at the Edge","display_name":"Large Language Model Partitioning for Low-Latency Inference at the Edge","publication_year":2025,"publication_date":"2025-05-26","ids":{"openalex":"https://openalex.org/W4413640445","doi":"https://doi.org/10.23919/wiopt66569.2025.11123401"},"language":"en","primary_location":{"id":"doi:10.23919/wiopt66569.2025.11123401","is_oa":false,"landing_page_url":"https://doi.org/10.23919/wiopt66569.2025.11123401","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2025 23rd International Symposium on Modeling and Optimization in Mobile, Ad Hoc, and Wireless Networks (WiOpt)","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":false,"oa_status":"closed","oa_url":null,"any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5030663399","display_name":"Dimitrios Kafetzis","orcid":"https://orcid.org/0009-0003-1090-6017"},"institutions":[{"id":"https://openalex.org/I73142707","display_name":"Athens University of Economics and Business","ror":"https://ror.org/03s262162","country_code":"GR","type":"education","lineage":["https://openalex.org/I73142707"]}],"countries":["GR"],"is_corresponding":false,"raw_author_name":"Dimitrios Kafetzis","raw_affiliation_strings":["Athens University of Economics and Business,Department of Informatics,Athens,Greece"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Athens University of Economics and Business,Department of Informatics,Athens,Greece","institution_ids":["https://openalex.org/I73142707"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5020778576","display_name":"Ramin Khalili","orcid":"https://orcid.org/0000-0003-2463-7033"},"institutions":[{"id":"https://openalex.org/I4210166625","display_name":"Huawei German Research Center","ror":"https://ror.org/00z59w514","country_code":"DE","type":"facility","lineage":["https://openalex.org/I2250955327","https://openalex.org/I4210129353","https://openalex.org/I4210166625"]}],"countries":["DE"],"is_corresponding":false,"raw_author_name":"Ramin Khalili","raw_affiliation_strings":["Huawei European Research Center,Munich,Germany"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Huawei European Research Center,Munich,Germany","institution_ids":["https://openalex.org/I4210166625"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5104579646","display_name":"Iordanis Koutsopoulos","orcid":null},"institutions":[{"id":"https://openalex.org/I73142707","display_name":"Athens University of Economics and Business","ror":"https://ror.org/03s262162","country_code":"GR","type":"education","lineage":["https://openalex.org/I73142707"]}],"countries":["GR"],"is_corresponding":false,"raw_author_name":"Iordanis Koutsopoulos","raw_affiliation_strings":["Athens University of Economics and Business,Department of Informatics,Athens,Greece"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Athens University of Economics and Business,Department of Informatics,Athens,Greece","institution_ids":["https://openalex.org/I73142707"]}]}],"institutions":[],"countries_distinct_count":2,"institutions_distinct_count":2,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":7.4602,"has_fulltext":false,"cited_by_count":5,"citation_normalized_percentile":{"value":0.97533314,"is_in_top_1_percent":false,"is_in_top_10_percent":true},"cited_by_percentile_year":{"min":95,"max":99},"biblio":{"volume":null,"issue":null,"first_page":"1","last_page":"8"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.6848999857902527,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.6848999857902527,"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/computer-science","display_name":"Computer science","score":0.7703685760498047},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.6191499829292297},{"id":"https://openalex.org/keywords/enhanced-data-rates-for-gsm-evolution","display_name":"Enhanced Data Rates for GSM Evolution","score":0.4720500409603119},{"id":"https://openalex.org/keywords/latency","display_name":"Latency (audio)","score":0.41622740030288696},{"id":"https://openalex.org/keywords/parallel-computing","display_name":"Parallel computing","score":0.3435111939907074},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.2873500883579254},{"id":"https://openalex.org/keywords/telecommunications","display_name":"Telecommunications","score":0.07440617680549622}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7703685760498047},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.6191499829292297},{"id":"https://openalex.org/C162307627","wikidata":"https://www.wikidata.org/wiki/Q204833","display_name":"Enhanced Data Rates for GSM Evolution","level":2,"score":0.4720500409603119},{"id":"https://openalex.org/C82876162","wikidata":"https://www.wikidata.org/wiki/Q17096504","display_name":"Latency (audio)","level":2,"score":0.41622740030288696},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.3435111939907074},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.2873500883579254},{"id":"https://openalex.org/C76155785","wikidata":"https://www.wikidata.org/wiki/Q418","display_name":"Telecommunications","level":1,"score":0.07440617680549622}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.23919/wiopt66569.2025.11123401","is_oa":false,"landing_page_url":"https://doi.org/10.23919/wiopt66569.2025.11123401","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2025 23rd International Symposium on Modeling and Optimization in Mobile, Ad Hoc, and Wireless Networks (WiOpt)","raw_type":"proceedings-article"}],"best_oa_location":null,"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":9,"referenced_works":["https://openalex.org/W2969388332","https://openalex.org/W4293195477","https://openalex.org/W4312681979","https://openalex.org/W4327521519","https://openalex.org/W4361029763","https://openalex.org/W4392910546","https://openalex.org/W4400810992","https://openalex.org/W4401508667","https://openalex.org/W4405934565"],"related_works":["https://openalex.org/W4391375266","https://openalex.org/W2899084033","https://openalex.org/W2748952813","https://openalex.org/W2390279801","https://openalex.org/W4391913857","https://openalex.org/W2358668433","https://openalex.org/W4396701345","https://openalex.org/W2376932109","https://openalex.org/W3128807919","https://openalex.org/W3176411177"],"abstract_inverted_index":{"Large":[0],"Language":[1],"Models":[2],"(LLMs)":[3],"based":[4,133],"on":[5,134,158],"autoregressive,":[6],"decoder-only":[7],"Transformers":[8],"generate":[9],"text":[10],"one":[11],"token":[12,18,29,119],"at":[13,115,191],"a":[14,17,20,103,124,154],"time,":[15],"where":[16,109],"represents":[19],"discrete":[21],"unit":[22],"of":[23,58,156,178,223,250],"text.":[24],"As":[25,68],"each":[26,162,196],"newly":[27],"produced":[28],"is":[30,113,123,132,149,166],"appended":[31],"to":[32,49,97,216,271],"the":[33,37,43,50,63,110,128,147,176,189,192,244],"partial":[34],"output":[35],"sequence,":[36],"length":[38],"grows":[39],"and":[40,45,75,142,160,181,203,226,267],"so":[41,174],"does":[42],"memory":[44,74,87,268],"compute":[46,76],"load,":[47],"due":[48],"expanding":[51],"key-value":[52],"(K/V)":[53],"caches-which":[54],"store":[55],"intermediate":[56],"representations":[57],"all":[59],"previously":[60],"generated":[61],"tokens-in":[62],"multi-head":[64],"attention":[65,193,197,214,224],"(MHA)":[66],"layer.":[67],"this":[69],"iterative":[70],"process":[71],"steadily":[72],"increases":[73],"demands,":[77],"layer-based":[78,273],"partitioning":[79,107,111,274],"in":[80,86,127,161,232,239,257,264],"resource-constrained":[81],"edge":[82],"environments":[83],"often":[84],"results":[85],"overload":[88],"or":[89],"high":[90],"inference":[91,99,182,233,265],"latency.":[92],"To":[93],"address":[94],"this,":[95],"aiming":[96],"reduce":[98],"latency,":[100,255],"we":[101,219],"propose":[102],"resource-aware":[104],"Transformer":[105],"architecture":[106],"algorithm,":[108],"decision":[112],"updated":[114],"regular":[116],"intervals":[117],"during":[118],"generation.":[120],"The":[121],"approach":[122,187],"myopic":[125],"algorithm":[126,148],"sense":[129],"that":[130,175,238],"it":[131,152,165,168,260],"instantaneously":[135],"available":[136],"information":[137],"about":[138],"device":[139],"resources":[140,208],"availability":[141],"network":[143],"link":[144],"bandwidths.":[145],"When":[146],"first":[150],"executed,":[151,167],"generates":[153],"placement":[155],"blocks":[157,171],"devices,":[159,218],"consecutive":[163],"time":[164],"migrates":[169],"these":[170],"among":[172],"devices":[173],"sum":[177],"migration":[179],"delay":[180,183],"remains":[184],"low.":[185],"Our":[186,235],"partitions":[188],"decoder":[190],"head-level,":[194],"co-locating":[195],"head":[198],"with":[199],"its":[200],"K/V":[201],"cache":[202],"allowing":[204],"dynamic":[205],"migrations":[206],"whenever":[207],"become":[209],"tight.":[210],"By":[211],"allocating":[212],"different":[213,217],"heads":[215,225],"exploit":[220],"parallel":[221],"execution":[222],"thus":[227],"allow":[228],"for":[229],"substantial":[230],"reductions":[231],"delays.":[234],"experiments":[236],"show":[237],"small-scale":[240],"settings":[241],"(3-5":[242],"devices),":[243],"proposed":[245],"method":[246],"achieves":[247,261],"within":[248],"15-20%":[249],"an":[251],"exact":[252],"optimal":[253],"solver's":[254],"while":[256],"larger-scale":[258],"tests":[259],"notable":[262],"improvements":[263],"speed":[266],"usage":[269],"compared":[270],"state-of-the-art":[272],"approaches.":[275]},"counts_by_year":[{"year":2026,"cited_by_count":3},{"year":2025,"cited_by_count":2}],"updated_date":"2026-08-11T07:18:39.950985","created_date":"2025-10-10T00:00:00"}