{"id":"https://openalex.org/W4416034896","doi":"https://doi.org/10.18653/v1/2025.findings-emnlp.429","title":"More Tokens, Lower Precision: Towards the Optimal Token-Precision Trade-off in KV Cache Compression","display_name":"More Tokens, Lower Precision: Towards the Optimal Token-Precision Trade-off in KV Cache Compression","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4416034896","doi":"https://doi.org/10.18653/v1/2025.findings-emnlp.429"},"language":null,"primary_location":{"id":"doi:10.18653/v1/2025.findings-emnlp.429","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-emnlp.429","pdf_url":"https://aclanthology.org/2025.findings-emnlp.429.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: EMNLP 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-emnlp.429.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5070520200","display_name":"Jiebin Zhang","orcid":"https://orcid.org/0000-0002-2650-6822"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jiebin Zhang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100567405","display_name":"Dawei Zhu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Dawei Zhu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5065232236","display_name":"Yifan Song","orcid":"https://orcid.org/0009-0004-6205-2655"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yifan Song","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5023056692","display_name":"Wenhao Wu","orcid":"https://orcid.org/0000-0002-0736-6561"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wenhao Wu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Chuqiao Kuang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chuqiao Kuang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100373860","display_name":"Xiaoguang Li","orcid":"https://orcid.org/0000-0002-1839-1607"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xiaoguang Li","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5046228314","display_name":"Lifeng Shang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lifeng Shang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101973489","display_name":"Qun Liu","orcid":"https://orcid.org/0000-0002-5772-4929"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Qun Liu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5058353424","display_name":"Sujian Li","orcid":"https://orcid.org/0000-0001-7493-0786"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Sujian Li","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":2,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"8092","last_page":"8105"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.304500013589859,"subfield":{"id":"https://openalex.org/subfields/1708","display_name":"Hardware and Architecture"},"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/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.304500013589859,"subfield":{"id":"https://openalex.org/subfields/1708","display_name":"Hardware and Architecture"},"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/T11181","display_name":"Advanced Data Storage Technologies","score":0.1688999980688095,"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/T10772","display_name":"Distributed systems and fault tolerance","score":0.15279999375343323,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/compression","display_name":"Compression (physics)","score":0.427700012922287},{"id":"https://openalex.org/keywords/cache","display_name":"Cache","score":0.3416000008583069},{"id":"https://openalex.org/keywords/data-compression","display_name":"Data compression","score":0.31119999289512634},{"id":"https://openalex.org/keywords/noise","display_name":"Noise (video)","score":0.27160000801086426}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5005000233650208},{"id":"https://openalex.org/C180016635","wikidata":"https://www.wikidata.org/wiki/Q2712821","display_name":"Compression (physics)","level":2,"score":0.427700012922287},{"id":"https://openalex.org/C24326235","wikidata":"https://www.wikidata.org/wiki/Q126095","display_name":"Electronic engineering","level":1,"score":0.35499998927116394},{"id":"https://openalex.org/C115537543","wikidata":"https://www.wikidata.org/wiki/Q165596","display_name":"Cache","level":2,"score":0.3416000008583069},{"id":"https://openalex.org/C78548338","wikidata":"https://www.wikidata.org/wiki/Q2493","display_name":"Data compression","level":2,"score":0.31119999289512634},{"id":"https://openalex.org/C192562407","wikidata":"https://www.wikidata.org/wiki/Q228736","display_name":"Materials science","level":0,"score":0.3066999912261963},{"id":"https://openalex.org/C119599485","wikidata":"https://www.wikidata.org/wiki/Q43035","display_name":"Electrical engineering","level":1,"score":0.27300000190734863},{"id":"https://openalex.org/C99498987","wikidata":"https://www.wikidata.org/wiki/Q2210247","display_name":"Noise (video)","level":3,"score":0.27160000801086426},{"id":"https://openalex.org/C127413603","wikidata":"https://www.wikidata.org/wiki/Q11023","display_name":"Engineering","level":0,"score":0.2596000134944916},{"id":"https://openalex.org/C121332964","wikidata":"https://www.wikidata.org/wiki/Q413","display_name":"Physics","level":0,"score":0.25189998745918274}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2025.findings-emnlp.429","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-emnlp.429","pdf_url":"https://aclanthology.org/2025.findings-emnlp.429.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: EMNLP 2025","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2025.findings-emnlp.429","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-emnlp.429","pdf_url":"https://aclanthology.org/2025.findings-emnlp.429.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: EMNLP 2025","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[{"id":"https://openalex.org/G3336050986","display_name":null,"funder_award_id":"62476010","funder_id":"https://openalex.org/F4320321001","funder_display_name":"National Natural Science Foundation of China"}],"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/W4416034896.pdf","grobid_xml":"https://content.openalex.org/works/W4416034896.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"As":[0],"large":[1],"language":[2],"models":[3],"(LLMs)":[4],"process":[5],"increasing":[6],"context":[7],"windows,":[8],"the":[9,45,58,85,102,115],"memory":[10],"usage":[11],"of":[12,118,122],"KV":[13,23,27,30,64,91,94,103,156,170],"cache":[14,65,104,171],"has":[15],"become":[16],"a":[17,68,74],"critical":[18],"bottleneck":[19],"during":[20],"inference.The":[21],"mainstream":[22],"compression":[24,66,172],"methods,":[25,158],"including":[26],"pruning":[28,92,132,148,157],"and":[29,139,152,161],"quantization,":[31],"primarily":[32],"focus":[33],"on":[34],"either":[35],"token":[36],"or":[37,93],"precision":[38],"dimensions":[39,51],"separately.However,":[40],"these":[41,48],"works":[42],"have":[43],"left":[44],"trade-off":[46,78,125,176],"between":[47],"two":[49],"orthogonal":[50],"largely":[52],"unexplored.In":[53],"this":[54,123],"paper,":[55],"we":[56,109],"leverage":[57],"Information":[59,86],"Bottleneck":[60,87],"principle":[61],"to":[62,89],"formulate":[63],"within":[67,84],"unified":[69],"theoretical":[70],"framework.We":[71],"demonstrate":[72],"that":[73,97,130],"carefully":[75],"managed":[76],"token-precision":[77,124,175],"can":[79],"achieve":[80],"an":[81],"optimal":[82],"point":[83],"compared":[88],"standalone":[90],"quantization.Experiments":[95],"reveal":[96],"storing":[98],"more":[99],"tokens":[100],"in":[101,136],"at":[105],"lower":[106],"precision-a":[107],"strategy":[108],"term":[110],"quantized":[111,131,147],"pruning-can":[112],"significantly":[113],"enhance":[114],"long-context":[116],"performance":[117],"LLMs.An":[119],"in-depth":[120],"analysis":[121],"across":[126,143,154],"key":[127],"aspects":[128],"shows":[129],"achieves":[133],"substantial":[134],"improvements":[135],"retrieval-related":[137],"tasks":[138],"consistently":[140],"performs":[141],"well":[142],"varying":[144],"input":[145],"lengths.Furthermore,":[146],"exhibits":[149],"notable":[150],"stability":[151],"effectiveness":[153],"different":[155],"quantization":[159],"strategies,":[160],"model":[162],"scales.These":[163],"findings":[164],"offer":[165],"valuable":[166],"insights":[167],"into":[168],"optimizing":[169],"through":[173],"balanced":[174],"strategies.":[177]},"counts_by_year":[{"year":2026,"cited_by_count":2}],"updated_date":"2026-07-14T23:27:15.235271","created_date":"2025-11-08T00:00:00"}