{"id":"https://openalex.org/W4408355721","doi":"https://doi.org/10.1109/icassp49660.2025.10890367","title":"DynamicAttention: Dynamic KV Cache for Disaggregate LLM Inference","display_name":"DynamicAttention: Dynamic KV Cache for Disaggregate LLM Inference","publication_year":2025,"publication_date":"2025-03-12","ids":{"openalex":"https://openalex.org/W4408355721","doi":"https://doi.org/10.1109/icassp49660.2025.10890367"},"language":"en","primary_location":{"id":"doi:10.1109/icassp49660.2025.10890367","is_oa":false,"landing_page_url":"https://doi.org/10.1109/icassp49660.2025.10890367","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":"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":null,"display_name":"Zhiqiang Ding","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhiqiang Ding","raw_affiliation_strings":["Ant Group,Platform Technology Business Unit,HangZhou,China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Ant Group,Platform Technology Business Unit,HangZhou,China","institution_ids":[]}]},{"author_position":"last","author":{"id":null,"display_name":"Tongkai Yang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tongkai Yang","raw_affiliation_strings":["Ant Group,Platform Technology Business Unit,HangZhou,China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Ant Group,Platform Technology Business Unit,HangZhou,China","institution_ids":[]}]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":1.673,"has_fulltext":false,"cited_by_count":1,"citation_normalized_percentile":{"value":0.79200861,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":{"min":94,"max":97},"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/T10181","display_name":"Natural Language Processing Techniques","score":0.9724000096321106,"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":0.9724000096321106,"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/T10201","display_name":"Speech Recognition and Synthesis","score":0.9362999796867371,"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/T11269","display_name":"Algorithms and Data Compression","score":0.9010000228881836,"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/cache","display_name":"Cache","score":0.84963059425354},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.7748245596885681},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.6096446514129639},{"id":"https://openalex.org/keywords/cache-algorithms","display_name":"Cache algorithms","score":0.44416260719299316},{"id":"https://openalex.org/keywords/parallel-computing","display_name":"Parallel computing","score":0.419432669878006},{"id":"https://openalex.org/keywords/cpu-cache","display_name":"CPU cache","score":0.3305632174015045},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.19376406073570251}],"concepts":[{"id":"https://openalex.org/C115537543","wikidata":"https://www.wikidata.org/wiki/Q165596","display_name":"Cache","level":2,"score":0.84963059425354},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7748245596885681},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.6096446514129639},{"id":"https://openalex.org/C38556500","wikidata":"https://www.wikidata.org/wiki/Q13404475","display_name":"Cache algorithms","level":4,"score":0.44416260719299316},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.419432669878006},{"id":"https://openalex.org/C189783530","wikidata":"https://www.wikidata.org/wiki/Q352090","display_name":"CPU cache","level":3,"score":0.3305632174015045},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.19376406073570251}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1109/icassp49660.2025.10890367","is_oa":false,"landing_page_url":"https://doi.org/10.1109/icassp49660.2025.10890367","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":[{"id":"https://openalex.org/F4320318398","display_name":"Ant Group","ror":null}],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":15,"referenced_works":["https://openalex.org/W2170901118","https://openalex.org/W4281778208","https://openalex.org/W4285504009","https://openalex.org/W4372263604","https://openalex.org/W4382395135","https://openalex.org/W4387321091","https://openalex.org/W4394952072","https://openalex.org/W4394998892","https://openalex.org/W4395112660","https://openalex.org/W4399757647","https://openalex.org/W4400036060","https://openalex.org/W4401211704","https://openalex.org/W4402670433","https://openalex.org/W4409963643","https://openalex.org/W6850927664"],"related_works":["https://openalex.org/W2147122795","https://openalex.org/W2061075966","https://openalex.org/W3147501184","https://openalex.org/W2167303720","https://openalex.org/W2268996566","https://openalex.org/W2109715593","https://openalex.org/W3161817247","https://openalex.org/W2012518269","https://openalex.org/W2068309140","https://openalex.org/W4239937081"],"abstract_inverted_index":{"In":[0],"large":[1],"language":[2],"model":[3],"inference,":[4],"efficient":[5],"utilization":[6,123],"of":[7,11],"GPU":[8,19,38,71,82,121],"memory":[9,20,23,30,39,47,72,88,96,122],"is":[10,24],"utmost":[12],"importance.":[13],"Current":[14],"systems":[15],"suffer":[16],"from":[17],"unreasonable":[18],"allocation:":[21],"excessive":[22],"idle":[25],"under":[26,33,127,132],"low":[27,128],"load,":[28],"while":[29],"shortages":[31],"occur":[32],"high":[34,133],"load.":[35,134],"Moreover,":[36],"existing":[37,116],"management":[40],"methods":[41],"do":[42],"not":[43,78],"consider":[44],"the":[45,55,91,104,115],"different":[46],"requirements":[48],"for":[49,103],"prefill":[50],"and":[51,106,130],"decode":[52],"stages":[53],"in":[54],"inference":[56],"process.":[57],"This":[58],"paper":[59],"proposes":[60],"a":[61,68],"new":[62],"solution":[63],"-":[64],"DynamicAttention.":[65],"It":[66],"allocates":[67,87],"continuous":[69],"virtual":[70],"space":[73],"at":[74],"startup,":[75],"but":[76],"does":[77],"actually":[79],"allocate":[80],"physical":[81],"memory.":[83],"Instead,":[84],"it":[85,136],"dynamically":[86],"according":[89],"to":[90],"demand":[92],"during":[93],"runtime.":[94],"Different":[95],"consumption":[97],"prediction":[98],"algorithms":[99],"are":[100],"designed":[101],"specifically":[102],"Prefill":[105],"Decode":[107],"stages.":[108],"Experimental":[109],"results":[110],"show":[111],"that":[112],"compared":[113],"with":[114],"SOTA":[117],"solutions,":[118],"DynamicAttention":[119],"improves":[120],"by":[124,139],"4":[125],"times":[126],"load":[129],"15%":[131],"Furthermore,":[135],"increases":[137],"throughput":[138],"1.6":[140],"times.":[141]},"counts_by_year":[{"year":2026,"cited_by_count":1}],"updated_date":"2026-07-29T14:22:42.915294","created_date":"2025-10-10T00:00:00"}