{"id":"https://openalex.org/W4415330563","doi":"https://doi.org/10.48550/arxiv.2508.09204","title":"PrismServe: Efficient LLM Serving on Heterogeneous Accelerator Clusters","display_name":"PrismServe: Efficient LLM Serving on Heterogeneous Accelerator Clusters","publication_year":2025,"publication_date":"2025-08-09","ids":{"openalex":"https://openalex.org/W4415330563","doi":"https://doi.org/10.48550/arxiv.2508.09204"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2508.09204","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2508.09204","pdf_url":"https://arxiv.org/pdf/2508.09204","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,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2508.09204","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5022694116","display_name":"Jinhao Zhang","orcid":"https://orcid.org/0009-0001-5722-2726","observed_orcids":["https://orcid.org/0009-0001-5722-2726"]},"institutions":[{"id":"https://openalex.org/I139759216","display_name":"Beijing University of Posts and Telecommunications","ror":"https://ror.org/04w9fbh59","country_code":"CN","type":"education","lineage":["https://openalex.org/I139759216"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Zhang, Jinhao","raw_affiliation_strings":["Beijing University of Posts and Telecommunications , Beijing , China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Beijing University of Posts and Telecommunications , Beijing , China","institution_ids":["https://openalex.org/I139759216"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5123220988","display_name":"Zicheng Yan","orcid":"https://orcid.org/0009-0002-0886-8405","observed_orcids":["https://orcid.org/0009-0002-0886-8405"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yan, Zicheng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100677423","display_name":"Boyang Zhang","orcid":"https://orcid.org/0000-0002-5065-8762","observed_orcids":["https://orcid.org/0000-0002-5065-8762"]},"institutions":[{"id":"https://openalex.org/I4210136793","display_name":"Peng Cheng Laboratory","ror":"https://ror.org/03qdqbt06","country_code":"CN","type":"facility","lineage":["https://openalex.org/I4210136793"]},{"id":"https://openalex.org/I4210165038","display_name":"University of Chinese Academy of Sciences","ror":"https://ror.org/05qbk4x57","country_code":"CN","type":"education","lineage":["https://openalex.org/I19820366","https://openalex.org/I4210165038"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Zhang, Boyang","raw_affiliation_strings":["Peng Cheng Laboratory , Shenzhen , China","University of Chinese Academy of Sciences , Beijing , China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Peng Cheng Laboratory , Shenzhen , China","institution_ids":["https://openalex.org/I4210136793"]},{"raw_affiliation_string":"University of Chinese Academy of Sciences , Beijing , China","institution_ids":["https://openalex.org/I4210165038"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100376637","display_name":"Zeyu Liu","orcid":"https://orcid.org/0000-0001-9475-1582","observed_orcids":["https://orcid.org/0000-0001-9475-1582","https://orcid.org/0009-0004-9283-0032","https://orcid.org/0009-0004-9725-1996"]},"institutions":[{"id":"https://openalex.org/I135714990","display_name":"North University of China","ror":"https://ror.org/047bp1713","country_code":"CN","type":"education","lineage":["https://openalex.org/I135714990"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Liu, Zeyu","raw_affiliation_strings":["North University of China , Taiyuan , China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"North University of China , Taiyuan , China","institution_ids":["https://openalex.org/I135714990"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5010944123","display_name":"Daning Cheng","orcid":"https://orcid.org/0000-0002-9217-6980","observed_orcids":["https://orcid.org/0000-0002-9217-6980"]},"institutions":[{"id":"https://openalex.org/I19820366","display_name":"Chinese Academy of Sciences","ror":"https://ror.org/034t30j35","country_code":"CN","type":"government","lineage":["https://openalex.org/I19820366"]},{"id":"https://openalex.org/I4210090176","display_name":"Institute of Computing Technology","ror":"https://ror.org/0090r4d87","country_code":"CN","type":"facility","lineage":["https://openalex.org/I19820366","https://openalex.org/I4210090176"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Cheng, Daning","raw_affiliation_strings":["Institute of Computing Technology , Chinese Academy of Sciences , Beijing , China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Institute of Computing Technology , Chinese Academy of Sciences , Beijing , China","institution_ids":["https://openalex.org/I19820366","https://openalex.org/I4210090176"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5001666028","display_name":"Yunquan Zhang","orcid":"https://orcid.org/0000-0002-2618-5088","observed_orcids":["https://orcid.org/0000-0002-2618-5088","https://orcid.org/0000-0001-7520-9640"]},"institutions":[{"id":"https://openalex.org/I19820366","display_name":"Chinese Academy of Sciences","ror":"https://ror.org/034t30j35","country_code":"CN","type":"government","lineage":["https://openalex.org/I19820366"]},{"id":"https://openalex.org/I4210090176","display_name":"Institute of Computing Technology","ror":"https://ror.org/0090r4d87","country_code":"CN","type":"facility","lineage":["https://openalex.org/I19820366","https://openalex.org/I4210090176"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Zhang, Yunquan","raw_affiliation_strings":["Institute of Computing Technology , Chinese Academy of Sciences , Beijing , China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Institute of Computing Technology , Chinese Academy of Sciences , Beijing , China","institution_ids":["https://openalex.org/I19820366","https://openalex.org/I4210090176"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5160621116","display_name":"Guangming Tan","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tan, Guangming","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5160596601","display_name":"Xiang Gao","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Gao, Xiang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":6,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":5,"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":0.9980000257492065,"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":0.9980000257492065,"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":0.0006000000284984708,"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":0.00019999999494757503,"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/quantization","display_name":"quantization","score":0.5680000185966492},{"id":"https://openalex.org/keywords/mixture-of-experts","display_name":"mixture of experts","score":0.8799999952316284},{"id":"https://openalex.org/keywords/dynamic-routing","display_name":"dynamic routing","score":0.7710000276565552},{"id":"https://openalex.org/keywords/model-efficiency","display_name":"model efficiency","score":0.503000020980835},{"id":"https://openalex.org/keywords/resource-constrained-devices","display_name":"resource-constrained devices","score":0.7300000190734863},{"id":"https://openalex.org/keywords/inference-latency","display_name":"inference latency","score":0.5329999923706055},{"id":"https://openalex.org/keywords/image-classification","display_name":"image classification","score":0.5120000243186951},{"id":"https://openalex.org/keywords/natural-language-processing","display_name":"natural language processing","score":0.8109999895095825}],"concepts":[{"id":"https://openalex.org/C28855332","wikidata":"https://www.wikidata.org/wiki/Q198099","display_name":"Quantization (signal processing)","level":2,"score":0.8679999709129333},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.6881999969482422},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6643000245094299},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4970000088214874},{"id":"https://openalex.org/C93372532","wikidata":"https://www.wikidata.org/wiki/Q6552455","display_name":"Linde\u2013Buzo\u2013Gray algorithm","level":3,"score":0.42170000076293945},{"id":"https://openalex.org/C185798385","wikidata":"https://www.wikidata.org/wiki/Q1161707","display_name":"Benchmark (surveying)","level":2,"score":0.40470001101493835},{"id":"https://openalex.org/C199833920","wikidata":"https://www.wikidata.org/wiki/Q612536","display_name":"Vector quantization","level":2,"score":0.40459999442100525},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.39800000190734863},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.37540000677108765},{"id":"https://openalex.org/C108583219","wikidata":"https://www.wikidata.org/wiki/Q197536","display_name":"Deep learning","level":2,"score":0.3644999861717224},{"id":"https://openalex.org/C105339364","wikidata":"https://www.wikidata.org/wiki/Q2297740","display_name":"Software deployment","level":2,"score":0.3472000062465668},{"id":"https://openalex.org/C67186912","wikidata":"https://www.wikidata.org/wiki/Q367664","display_name":"Data modeling","level":2,"score":0.29159998893737793},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.2784000039100647},{"id":"https://openalex.org/C2775896111","wikidata":"https://www.wikidata.org/wiki/Q642560","display_name":"Router","level":2,"score":0.25859999656677246},{"id":"https://openalex.org/C98045186","wikidata":"https://www.wikidata.org/wiki/Q205663","display_name":"Process (computing)","level":2,"score":0.25760000944137573}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2508.09204","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2508.09204","pdf_url":"https://arxiv.org/pdf/2508.09204","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,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2508.09204","doi":"https://doi.org/10.48550/arxiv.2508.09204","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2508.09204","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,"listed_in":[],"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"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2508.09204","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2508.09204","pdf_url":"https://arxiv.org/pdf/2508.09204","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,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4415330563.pdf","grobid_xml":"https://content.openalex.org/works/W4415330563.grobid-xml"},"referenced_works_count":6,"referenced_works":["https://openalex.org/W1593114658","https://openalex.org/W2581624817","https://openalex.org/W3035855442","https://openalex.org/W3122499249","https://openalex.org/W4307934016","https://openalex.org/W4392904203"],"related_works":["https://openalex.org/W4416123559","https://openalex.org/W4415343303","https://openalex.org/W7223246138","https://openalex.org/W7152605290","https://openalex.org/W4387389714","https://openalex.org/W4399657447","https://openalex.org/W7213505222","https://openalex.org/W4416010244","https://openalex.org/W4415062079","https://openalex.org/W4412944360"],"abstract_inverted_index":{"Serving":[0],"large":[1],"language":[2],"models":[3],"across":[4,13,41],"heterogeneous":[5,42],"accelerators":[6,64,122],"is":[7,155],"challenging":[8],"because":[9],"partitioning":[10],"a":[11,34,45,60,66,70,84],"replica":[12,85],"device":[14],"types":[15],"introduces":[16],"communication":[17],"overhead":[18],"during":[19],"autoregressive":[20],"decoding,":[21],"while":[22,69,145],"deploying":[23],"complete":[24],"quantized":[25,48],"replicas":[26],"can":[27],"degrade":[28],"model":[29],"quality.":[30],"We":[31],"present":[32],"PrismServe,":[33],"serving":[35,118],"system":[36],"that":[37],"combines":[38],"request-level":[39],"collaboration":[40],"hardware":[43],"with":[44],"mixture":[46],"of":[47,62,101,150],"experts":[49],"derived":[50],"from":[51,123],"the":[52,99,137],"same":[53],"foundation":[54],"model.":[55],"Each":[56],"expert":[57,75],"executes":[58],"within":[59,147],"group":[61],"same-type":[63],"in":[65],"single":[67,139],"node,":[68],"two-level":[71],"router":[72],"selects":[73,83],"an":[74],"based":[76],"on":[77,93,120,131],"predicted":[78],"response":[79],"quality":[80],"and":[81,89,108,129],"then":[82],"according":[86],"to":[87],"load":[88],"prefix-cache":[90],"locality.":[91],"Implemented":[92],"unmodified":[94],"vLLM,":[95],"PrismServe":[96,135],"achieves":[97],"2.0--2.6$\\times$":[98],"throughput":[100],"cross-node":[102],"or":[103],"cross-vendor":[104],"full-precision":[105,151],"pipelined":[106],"deployments":[107],"reduces":[109],"P99":[110],"time":[111],"per":[112],"output":[113],"token":[114],"by":[115,142],"55--65\\%":[116],"when":[117],"Qwen3-14B":[119],"48":[121],"three":[124],"vendors.":[125],"Across":[126],"Qwen3-8B,":[127],"Qwen3-14B,":[128],"Qwen3-32B":[130],"five":[132],"downstream":[133],"tasks,":[134],"outperforms":[136],"best":[138],"quantization":[140],"method":[141],"2.0--4.3":[143],"points":[144,149],"remaining":[146],"1.4":[148],"accuracy.":[152],"Our":[153],"code":[154],"available":[156],"at":[157],"https://anonymous.4open.science/r/LLM-Inference-System-B000/":[158]},"counts_by_year":[{"year":2026,"cited_by_count":4},{"year":2025,"cited_by_count":1}],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2025-10-19T00:00:00"}