{"id":"https://openalex.org/W4408355184","doi":"https://doi.org/10.1109/icassp49660.2025.10887916","title":"CritiPrefill: A Segment-wise Criticality-based Approach for Prefilling Acceleration in LLMs","display_name":"CritiPrefill: A Segment-wise Criticality-based Approach for Prefilling Acceleration in LLMs","publication_year":2025,"publication_date":"2025-03-12","ids":{"openalex":"https://openalex.org/W4408355184","doi":"https://doi.org/10.1109/icassp49660.2025.10887916"},"language":"en","primary_location":{"id":"doi:10.1109/icassp49660.2025.10887916","is_oa":false,"landing_page_url":"https://doi.org/10.1109/icassp49660.2025.10887916","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"ICASSP 2025 - 2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)","raw_type":"proceedings-article"},"type":"article","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/A5025265931","display_name":"Jianfeng Lv","orcid":"https://orcid.org/0000-0003-1773-1638"},"institutions":[{"id":"https://openalex.org/I180662265","display_name":"China Mobile (China)","ror":"https://ror.org/05gftfe97","country_code":"CN","type":"company","lineage":["https://openalex.org/I180662265"]}],"countries":["CN"],"is_corresponding":true,"raw_author_name":"Junlin Lv","raw_affiliation_strings":["Guangdong OPPO Mobile Telecommunications Corp.,Ltd,AI Center,China"],"affiliations":[{"raw_affiliation_string":"Guangdong OPPO Mobile Telecommunications Corp.,Ltd,AI Center,China","institution_ids":["https://openalex.org/I180662265"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5021148181","display_name":"Yuan Feng","orcid":"https://orcid.org/0000-0003-1902-3408"},"institutions":[{"id":"https://openalex.org/I180662265","display_name":"China Mobile (China)","ror":"https://ror.org/05gftfe97","country_code":"CN","type":"company","lineage":["https://openalex.org/I180662265"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Yuan Feng","raw_affiliation_strings":["Guangdong OPPO Mobile Telecommunications Corp.,Ltd,AI Center,China"],"affiliations":[{"raw_affiliation_string":"Guangdong OPPO Mobile Telecommunications Corp.,Ltd,AI Center,China","institution_ids":["https://openalex.org/I180662265"]}]},{"author_position":"middle","author":{"id":null,"display_name":"Xike Xie","orcid":null},"institutions":[{"id":"https://openalex.org/I180662265","display_name":"China Mobile (China)","ror":"https://ror.org/05gftfe97","country_code":"CN","type":"company","lineage":["https://openalex.org/I180662265"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Xike Xie","raw_affiliation_strings":["Guangdong OPPO Mobile Telecommunications Corp.,Ltd,AI Center,China"],"affiliations":[{"raw_affiliation_string":"Guangdong OPPO Mobile Telecommunications Corp.,Ltd,AI Center,China","institution_ids":["https://openalex.org/I180662265"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5107005006","display_name":"Xin Jia","orcid":"https://orcid.org/0000-0002-5144-0765"},"institutions":[{"id":"https://openalex.org/I180662265","display_name":"China Mobile (China)","ror":"https://ror.org/05gftfe97","country_code":"CN","type":"company","lineage":["https://openalex.org/I180662265"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Xin Jia","raw_affiliation_strings":["Guangdong OPPO Mobile Telecommunications Corp.,Ltd,AI Center,China"],"affiliations":[{"raw_affiliation_string":"Guangdong OPPO Mobile Telecommunications Corp.,Ltd,AI Center,China","institution_ids":["https://openalex.org/I180662265"]}]},{"author_position":"middle","author":{"id":null,"display_name":"Qirong Peng","orcid":null},"institutions":[{"id":"https://openalex.org/I180662265","display_name":"China Mobile (China)","ror":"https://ror.org/05gftfe97","country_code":"CN","type":"company","lineage":["https://openalex.org/I180662265"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Qirong Peng","raw_affiliation_strings":["Guangdong OPPO Mobile Telecommunications Corp.,Ltd,AI Center,China"],"affiliations":[{"raw_affiliation_string":"Guangdong OPPO Mobile Telecommunications Corp.,Ltd,AI Center,China","institution_ids":["https://openalex.org/I180662265"]}]},{"author_position":"last","author":{"id":null,"display_name":"Guiming Xie","orcid":null},"institutions":[{"id":"https://openalex.org/I180662265","display_name":"China Mobile (China)","ror":"https://ror.org/05gftfe97","country_code":"CN","type":"company","lineage":["https://openalex.org/I180662265"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Guiming Xie","raw_affiliation_strings":["Guangdong OPPO Mobile Telecommunications Corp.,Ltd,AI Center,China"],"affiliations":[{"raw_affiliation_string":"Guangdong OPPO Mobile Telecommunications Corp.,Ltd,AI Center,China","institution_ids":["https://openalex.org/I180662265"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":6,"corresponding_author_ids":["https://openalex.org/A5025265931"],"corresponding_institution_ids":["https://openalex.org/I180662265"],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":{"value":0.01829346,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"1","last_page":"5"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10215","display_name":"Semantic Web and Ontologies","score":0.8526999950408936,"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/T10215","display_name":"Semantic Web and Ontologies","score":0.8526999950408936,"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/T10703","display_name":"Business Process Modeling and Analysis","score":0.8447999954223633,"subfield":{"id":"https://openalex.org/subfields/1404","display_name":"Management Information Systems"},"field":{"id":"https://openalex.org/fields/14","display_name":"Business, Management and Accounting"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},{"id":"https://openalex.org/T10927","display_name":"Access Control and Trust","score":0.823199987411499,"subfield":{"id":"https://openalex.org/subfields/3312","display_name":"Sociology and Political Science"},"field":{"id":"https://openalex.org/fields/33","display_name":"Social Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/acceleration","display_name":"Acceleration","score":0.7086538076400757},{"id":"https://openalex.org/keywords/criticality","display_name":"Criticality","score":0.6805434226989746},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.5181668996810913},{"id":"https://openalex.org/keywords/physics","display_name":"Physics","score":0.11256510019302368},{"id":"https://openalex.org/keywords/nuclear-physics","display_name":"Nuclear physics","score":0.09861093759536743}],"concepts":[{"id":"https://openalex.org/C117896860","wikidata":"https://www.wikidata.org/wiki/Q11376","display_name":"Acceleration","level":2,"score":0.7086538076400757},{"id":"https://openalex.org/C125611927","wikidata":"https://www.wikidata.org/wiki/Q17008131","display_name":"Criticality","level":2,"score":0.6805434226989746},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5181668996810913},{"id":"https://openalex.org/C121332964","wikidata":"https://www.wikidata.org/wiki/Q413","display_name":"Physics","level":0,"score":0.11256510019302368},{"id":"https://openalex.org/C185544564","wikidata":"https://www.wikidata.org/wiki/Q81197","display_name":"Nuclear physics","level":1,"score":0.09861093759536743},{"id":"https://openalex.org/C74650414","wikidata":"https://www.wikidata.org/wiki/Q11397","display_name":"Classical mechanics","level":1,"score":0.0}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1109/icassp49660.2025.10887916","is_oa":false,"landing_page_url":"https://doi.org/10.1109/icassp49660.2025.10887916","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"ICASSP 2025 - 2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)","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":37,"referenced_works":["https://openalex.org/W2889787757","https://openalex.org/W2963963993","https://openalex.org/W2964110616","https://openalex.org/W3081168214","https://openalex.org/W3120633509","https://openalex.org/W3161820423","https://openalex.org/W4387321091","https://openalex.org/W4389520033","https://openalex.org/W4389520387","https://openalex.org/W4391876619","https://openalex.org/W4393065402","https://openalex.org/W4401024755","https://openalex.org/W4401042382","https://openalex.org/W4402671766","https://openalex.org/W4402671987","https://openalex.org/W4402672007","https://openalex.org/W4404782964","https://openalex.org/W6767997687","https://openalex.org/W6776048684","https://openalex.org/W6779163297","https://openalex.org/W6781533629","https://openalex.org/W6838322825","https://openalex.org/W6850810733","https://openalex.org/W6853192989","https://openalex.org/W6854186793","https://openalex.org/W6854308872","https://openalex.org/W6854866820","https://openalex.org/W6856619137","https://openalex.org/W6857551316","https://openalex.org/W6859925366","https://openalex.org/W6859939998","https://openalex.org/W6861863668","https://openalex.org/W6864878861","https://openalex.org/W6869510815","https://openalex.org/W6870270694","https://openalex.org/W6872656104","https://openalex.org/W6875329788"],"related_works":["https://openalex.org/W4391375266","https://openalex.org/W2899084033","https://openalex.org/W2748952813","https://openalex.org/W2376759283","https://openalex.org/W3008607579","https://openalex.org/W126231364","https://openalex.org/W2279399181","https://openalex.org/W2333336726","https://openalex.org/W2963698000","https://openalex.org/W1668846860"],"abstract_inverted_index":{"Large":[0],"language":[1],"models":[2],"have":[3,36],"achieved":[4],"notable":[5],"success":[6],"across":[7],"various":[8],"domains,":[9],"yet":[10],"efficient":[11],"inference":[12,26],"is":[13],"still":[14],"limited":[15],"by":[16],"the":[17,22,42,45,66,82,102,119,133,136],"quadratic":[18],"computation":[19],"complexity":[20],"of":[21,28,44,69,81],"attention":[23],"mechanism.":[24],"The":[25],"consists":[27],"prefilling":[29,46,67,97,137],"and":[30,106,111,129,156],"decoding":[31],"phases.":[32],"Although":[33],"several":[34],"attempts":[35],"been":[37],"made":[38],"to":[39,76,117,151],"accelerate":[40],"decoding,":[41],"inefficiency":[43],"phase,":[47],"especially":[48],"for":[49,161],"long-context":[50,70,147],"tasks,":[51],"remains":[52],"a":[53,60,94,114,166],"challenge.":[54],"In":[55],"this":[56,89],"paper,":[57],"we":[58,91],"observe":[59],"locality":[61],"in":[62,132],"query":[63,73,120,127],"criticality":[64],"during":[65],"phase":[68],"processing:":[71],"adjacent":[72],"tokens":[74],"tend":[75],"focus":[77],"on":[78,88,145,154,159,165],"similar":[79],"subsets":[80],"past":[83],"Key-Value":[84],"(KV)":[85],"cache.":[86],"Based":[87],"observation,":[90],"propose":[92],"CritiPrefill,":[93],"criticality-based":[95],"segment-wise":[96,115],"method.":[98],"This":[99],"method":[100],"partitions":[101],"input":[103],"sequence\u2019s":[104],"queries":[105],"KV":[107],"cache":[108,130],"into":[109],"segments":[110,128],"blocks,":[112],"utilizing":[113],"algorithm":[116],"estimate":[118],"criticality.":[121],"By":[122],"pruning":[123],"noncritical":[124],"computations":[125],"between":[126],"blocks":[131],"self-attention":[134],"mechanism,":[135],"process":[138],"can":[139],"be":[140],"significantly":[141],"accelerated.":[142],"Extensive":[143],"evaluations":[144],"multiple":[146],"datasets":[148],"show":[149],"up":[150],"2.7x":[152],"speedup":[153,158],"Llama3-8B":[155],"3.0x":[157],"Yi-9B":[160],"128K":[162],"context":[163],"length":[164],"single":[167],"A100":[168],"GPU,":[169],"with":[170],"minimal":[171],"quality":[172],"degradation.":[173]},"counts_by_year":[],"updated_date":"2026-04-21T08:09:41.155169","created_date":"2025-10-10T00:00:00"}