{"id":"https://openalex.org/W4412887924","doi":"https://doi.org/10.18653/v1/2025.findings-acl.1043","title":"TailorKV: A Hybrid Framework for Long-Context Inference via Tailored KV Cache Optimization","display_name":"TailorKV: A Hybrid Framework for Long-Context Inference via Tailored KV Cache Optimization","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4412887924","doi":"https://doi.org/10.18653/v1/2025.findings-acl.1043"},"language":"en","primary_location":{"id":"doi:10.18653/v1/2025.findings-acl.1043","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.1043","pdf_url":"https://aclanthology.org/2025.findings-acl.1043.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: ACL 2025","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2025.findings-acl.1043.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":null,"display_name":"Dingyu Yao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Dingyu Yao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101952163","display_name":"Bowen Shen","orcid":"https://orcid.org/0000-0002-7913-4786"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Bowen Shen","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5004434493","display_name":"Zheng Lin","orcid":"https://orcid.org/0009-0006-4977-1385"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zheng Lin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5114949151","display_name":"Wei Liu","orcid":"https://orcid.org/0000-0001-5756-2264"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wei Liu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5051287176","display_name":"Jian Luan","orcid":"https://orcid.org/0000-0002-3851-0829"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jian Luan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5073236866","display_name":"Bin Wang","orcid":"https://orcid.org/0000-0001-8861-571X"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Bin Wang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5115591448","display_name":"Weiping Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Weiping Wang","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":true,"cited_by_count":1,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"20340","last_page":"20359"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11478","display_name":"Caching and Content Delivery","score":0.9726999998092651,"subfield":{"id":"https://openalex.org/subfields/1705","display_name":"Computer Networks and Communications"},"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/T11478","display_name":"Caching and Content Delivery","score":0.9726999998092651,"subfield":{"id":"https://openalex.org/subfields/1705","display_name":"Computer Networks and Communications"},"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/T10901","display_name":"Advanced Data Compression Techniques","score":0.9546999931335449,"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/T11269","display_name":"Algorithms and Data Compression","score":0.9373000264167786,"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":0.7690144777297974},{"id":"https://openalex.org/keywords/cache","display_name":"Cache","score":0.7270379066467285},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.6660548448562622},{"id":"https://openalex.org/keywords/context","display_name":"Context (archaeology)","score":0.5358445048332214},{"id":"https://openalex.org/keywords/computer-architecture","display_name":"Computer architecture","score":0.43164175748825073},{"id":"https://openalex.org/keywords/cache-algorithms","display_name":"Cache algorithms","score":0.41615211963653564},{"id":"https://openalex.org/keywords/parallel-computing","display_name":"Parallel computing","score":0.323682963848114},{"id":"https://openalex.org/keywords/cpu-cache","display_name":"CPU cache","score":0.3168046474456787},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.19350627064704895}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7690144777297974},{"id":"https://openalex.org/C115537543","wikidata":"https://www.wikidata.org/wiki/Q165596","display_name":"Cache","level":2,"score":0.7270379066467285},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.6660548448562622},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.5358445048332214},{"id":"https://openalex.org/C118524514","wikidata":"https://www.wikidata.org/wiki/Q173212","display_name":"Computer architecture","level":1,"score":0.43164175748825073},{"id":"https://openalex.org/C38556500","wikidata":"https://www.wikidata.org/wiki/Q13404475","display_name":"Cache algorithms","level":4,"score":0.41615211963653564},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.323682963848114},{"id":"https://openalex.org/C189783530","wikidata":"https://www.wikidata.org/wiki/Q352090","display_name":"CPU cache","level":3,"score":0.3168046474456787},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.19350627064704895},{"id":"https://openalex.org/C86803240","wikidata":"https://www.wikidata.org/wiki/Q420","display_name":"Biology","level":0,"score":0.0},{"id":"https://openalex.org/C151730666","wikidata":"https://www.wikidata.org/wiki/Q7205","display_name":"Paleontology","level":1,"score":0.0}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2025.findings-acl.1043","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.1043","pdf_url":"https://aclanthology.org/2025.findings-acl.1043.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: ACL 2025","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2025.findings-acl.1043","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.1043","pdf_url":"https://aclanthology.org/2025.findings-acl.1043.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: ACL 2025","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[],"funders":[{"id":"https://openalex.org/F4320321001","display_name":"National Natural Science Foundation of China","ror":"https://ror.org/01h0zpd94"}],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4412887924.pdf","grobid_xml":"https://content.openalex.org/works/W4412887924.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"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":{"The":[0],"Key-Value":[1],"(KV)":[2],"cache":[3,28],"in":[4,37,51],"generative":[5],"large":[6],"language":[7],"models":[8],"(LLMs)":[9],"introduces":[10],"substantial":[11,80],"memory":[12],"overhead.Existing":[13],"works":[14],"mitigate":[15],"this":[16,102],"burden":[17],"by":[18],"offloading":[19],"or":[20],"compressing":[21],"the":[22,26,52,145,147],"KV":[23],"cache.However,":[24],"loading":[25,90],"entire":[27],"incurs":[29],"significant":[30],"latency":[31],"due":[32],"to":[33,55,85],"PCIe":[34],"bandwidth":[35],"bottlenecks":[36],"CPU-GPU":[38],"communication,":[39],"while":[40],"aggressive":[41,141],"compression":[42,108,142],"causes":[43],"notable":[44],"performance":[45,139],"degradation.We":[46],"identify":[47],"that":[48,77,89,126,134],"certain":[49],"layers":[50,67],"LLM":[53],"need":[54],"maintain":[56],"global":[57],"information":[58],"and":[59,93,115],"are":[60],"unsuitable":[61],"for":[62],"selective":[63],"loading.In":[64],"contrast,":[65],"other":[66],"primarily":[68],"focus":[69],"on":[70,101],"a":[71,86,106,123,156],"few":[72],"tokens":[73,92,96],"with":[74,122,149],"dominant":[75,91],"activations":[76],"potentially":[78],"incur":[79],"quantization":[81,114],"error.This":[82],"observation":[83],"leads":[84],"key":[87],"insight":[88],"quantizing":[94],"all":[95],"can":[97,152],"complement":[98],"each":[99],"other.Building":[100],"insight,":[103],"we":[104],"propose":[105],"hybrid":[107],"method,":[109],"TailorKV,":[110],"which":[111],"seamlessly":[112],"integrates":[113],"offloading.TailorKV":[116],"develops":[117],"an":[118],"inference":[119],"framework":[120],"along":[121],"hardware-friendly":[124],"implementation":[125],"leverages":[127],"these":[128],"complementary":[129],"characteristics.Extensive":[130],"long-context":[131],"evaluations":[132],"exhibit":[133],"TailorKV":[135],"achieves":[136],"nearly":[137],"lossless":[138],"under":[140],"settings,":[143],"outperforming":[144],"state-of-the-art.Particularly,":[146],"Llama-3.1-8B":[148],"128k":[150],"context":[151],"be":[153],"served":[154],"within":[155],"single":[157],"RTX":[158],"3090":[159],"GPU,":[160],"reaching":[161],"82":[162],"ms":[163],"per":[164],"token":[165],"during":[166],"decoding":[167],"1":[168],".":[169]},"counts_by_year":[{"year":2026,"cited_by_count":1}],"updated_date":"2026-07-14T23:27:15.235271","created_date":"2025-10-10T00:00:00"}