{"id":"https://openalex.org/W4414359395","doi":"https://doi.org/10.24963/ijcai.2025/690","title":"RotateKV: Accurate and Robust 2-Bit KV Cache Quantization for LLMs via Outlier-Aware Adaptive Rotations","display_name":"RotateKV: Accurate and Robust 2-Bit KV Cache Quantization for LLMs via Outlier-Aware Adaptive Rotations","publication_year":2025,"publication_date":"2025-09-01","ids":{"openalex":"https://openalex.org/W4414359395","doi":"https://doi.org/10.24963/ijcai.2025/690"},"language":"en","primary_location":{"id":"doi:10.24963/ijcai.2025/690","is_oa":false,"landing_page_url":"https://doi.org/10.24963/ijcai.2025/690","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence","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/A5086331195","display_name":"Zhiyong Su","orcid":"https://orcid.org/0000-0001-9483-5268"},"institutions":[{"id":"https://openalex.org/I3131625388","display_name":"University Town of Shenzhen","ror":"https://ror.org/05f5j6225","country_code":"CN","type":"education","lineage":["https://openalex.org/I3131625388"]},{"id":"https://openalex.org/I99065089","display_name":"Tsinghua University","ror":"https://ror.org/03cve4549","country_code":"CN","type":"education","lineage":["https://openalex.org/I99065089"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Zunhai Su","raw_affiliation_strings":["Shenzhen International Graduate School, Tsinghua University, Shenzhen, China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Shenzhen International Graduate School, Tsinghua University, Shenzhen, China","institution_ids":["https://openalex.org/I3131625388","https://openalex.org/I99065089"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5116053899","display_name":"Hanyu Wei","orcid":null},"institutions":[{"id":"https://openalex.org/I3131625388","display_name":"University Town of Shenzhen","ror":"https://ror.org/05f5j6225","country_code":"CN","type":"education","lineage":["https://openalex.org/I3131625388"]},{"id":"https://openalex.org/I99065089","display_name":"Tsinghua University","ror":"https://ror.org/03cve4549","country_code":"CN","type":"education","lineage":["https://openalex.org/I99065089"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Hanyu Wei","raw_affiliation_strings":["Shenzhen International Graduate School, Tsinghua University, Shenzhen, China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Shenzhen International Graduate School, Tsinghua University, Shenzhen, China","institution_ids":["https://openalex.org/I3131625388","https://openalex.org/I99065089"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100656819","display_name":"Zhe Chen","orcid":"https://orcid.org/0000-0002-3215-2696"},"institutions":[{"id":"https://openalex.org/I2250955327","display_name":"Huawei Technologies (China)","ror":"https://ror.org/00cmhce21","country_code":"CN","type":"company","lineage":["https://openalex.org/I2250955327"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Zhe Chen","raw_affiliation_strings":["Huawei Technologies Co., Ltd., Shenzhen, China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Huawei Technologies Co., Ltd., Shenzhen, China","institution_ids":["https://openalex.org/I2250955327"]}]},{"author_position":"middle","author":{"id":null,"display_name":"Wang Shen","orcid":null},"institutions":[{"id":"https://openalex.org/I2250955327","display_name":"Huawei Technologies (China)","ror":"https://ror.org/00cmhce21","country_code":"CN","type":"company","lineage":["https://openalex.org/I2250955327"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Wang Shen","raw_affiliation_strings":["Huawei Technologies Co., Ltd., Shenzhen, China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Huawei Technologies Co., Ltd., Shenzhen, China","institution_ids":["https://openalex.org/I2250955327"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5116053898","display_name":"Linge Li","orcid":null},"institutions":[{"id":"https://openalex.org/I2250955327","display_name":"Huawei Technologies (China)","ror":"https://ror.org/00cmhce21","country_code":"CN","type":"company","lineage":["https://openalex.org/I2250955327"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Linge Li","raw_affiliation_strings":["Huawei Technologies Co., Ltd., Shenzhen, China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Huawei Technologies Co., Ltd., Shenzhen, China","institution_ids":["https://openalex.org/I2250955327"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5056191914","display_name":"Huangqi Yu","orcid":null},"institutions":[{"id":"https://openalex.org/I2250955327","display_name":"Huawei Technologies (China)","ror":"https://ror.org/00cmhce21","country_code":"CN","type":"company","lineage":["https://openalex.org/I2250955327"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Huangqi Yu","raw_affiliation_strings":["Huawei Technologies Co., Ltd., Shenzhen, China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Huawei Technologies Co., Ltd., Shenzhen, China","institution_ids":["https://openalex.org/I2250955327"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5116053900","display_name":"Kehong Yuan","orcid":null},"institutions":[{"id":"https://openalex.org/I3131625388","display_name":"University Town of Shenzhen","ror":"https://ror.org/05f5j6225","country_code":"CN","type":"education","lineage":["https://openalex.org/I3131625388"]},{"id":"https://openalex.org/I99065089","display_name":"Tsinghua University","ror":"https://ror.org/03cve4549","country_code":"CN","type":"education","lineage":["https://openalex.org/I99065089"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Kehong Yuan","raw_affiliation_strings":["Shenzhen International Graduate School, Tsinghua University, Shenzhen, China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Shenzhen International Graduate School, Tsinghua University, Shenzhen, China","institution_ids":["https://openalex.org/I3131625388","https://openalex.org/I99065089"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":3,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":1.8777,"has_fulltext":false,"cited_by_count":2,"citation_normalized_percentile":{"value":0.88382187,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":{"min":91,"max":97},"biblio":{"volume":null,"issue":null,"first_page":"6200","last_page":"6208"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11181","display_name":"Advanced Data Storage Technologies","score":0.9939000010490417,"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/T11181","display_name":"Advanced Data Storage Technologies","score":0.9939000010490417,"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/T11269","display_name":"Algorithms and Data Compression","score":0.9927999973297119,"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/T11478","display_name":"Caching and Content Delivery","score":0.9922000169754028,"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/quantization","display_name":"Quantization (signal processing)","score":0.7081999778747559},{"id":"https://openalex.org/keywords/robustness","display_name":"Robustness (evolution)","score":0.5753999948501587},{"id":"https://openalex.org/keywords/speedup","display_name":"Speedup","score":0.564300000667572},{"id":"https://openalex.org/keywords/cache","display_name":"Cache","score":0.4894999861717224},{"id":"https://openalex.org/keywords/outlier","display_name":"Outlier","score":0.47360000014305115},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.4571000039577484},{"id":"https://openalex.org/keywords/decoding-methods","display_name":"Decoding methods","score":0.43650001287460327},{"id":"https://openalex.org/keywords/embedding","display_name":"Embedding","score":0.41130000352859497},{"id":"https://openalex.org/keywords/offset","display_name":"Offset (computer science)","score":0.3684999942779541}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7466999888420105},{"id":"https://openalex.org/C28855332","wikidata":"https://www.wikidata.org/wiki/Q198099","display_name":"Quantization (signal processing)","level":2,"score":0.7081999778747559},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.6377999782562256},{"id":"https://openalex.org/C63479239","wikidata":"https://www.wikidata.org/wiki/Q7353546","display_name":"Robustness (evolution)","level":3,"score":0.5753999948501587},{"id":"https://openalex.org/C68339613","wikidata":"https://www.wikidata.org/wiki/Q1549489","display_name":"Speedup","level":2,"score":0.564300000667572},{"id":"https://openalex.org/C115537543","wikidata":"https://www.wikidata.org/wiki/Q165596","display_name":"Cache","level":2,"score":0.4894999861717224},{"id":"https://openalex.org/C79337645","wikidata":"https://www.wikidata.org/wiki/Q779824","display_name":"Outlier","level":2,"score":0.47360000014305115},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.4571000039577484},{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.43650001287460327},{"id":"https://openalex.org/C41608201","wikidata":"https://www.wikidata.org/wiki/Q980509","display_name":"Embedding","level":2,"score":0.41130000352859497},{"id":"https://openalex.org/C175291020","wikidata":"https://www.wikidata.org/wiki/Q1156822","display_name":"Offset (computer science)","level":2,"score":0.3684999942779541},{"id":"https://openalex.org/C46686674","wikidata":"https://www.wikidata.org/wiki/Q466303","display_name":"Boosting (machine learning)","level":2,"score":0.3684000074863434},{"id":"https://openalex.org/C7366592","wikidata":"https://www.wikidata.org/wiki/Q1255620","display_name":"Dram","level":2,"score":0.353300005197525},{"id":"https://openalex.org/C179799912","wikidata":"https://www.wikidata.org/wiki/Q205084","display_name":"Computational complexity theory","level":2,"score":0.3343999981880188},{"id":"https://openalex.org/C189783530","wikidata":"https://www.wikidata.org/wiki/Q352090","display_name":"CPU cache","level":3,"score":0.3093000054359436},{"id":"https://openalex.org/C113775141","wikidata":"https://www.wikidata.org/wiki/Q428691","display_name":"Computer engineering","level":1,"score":0.303600013256073},{"id":"https://openalex.org/C188198153","wikidata":"https://www.wikidata.org/wiki/Q1613840","display_name":"Limiting","level":2,"score":0.30309998989105225},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.30079999566078186},{"id":"https://openalex.org/C111335779","wikidata":"https://www.wikidata.org/wiki/Q3454686","display_name":"Reduction (mathematics)","level":2,"score":0.3003999888896942},{"id":"https://openalex.org/C179518139","wikidata":"https://www.wikidata.org/wiki/Q5140297","display_name":"Coding (social sciences)","level":2,"score":0.29820001125335693},{"id":"https://openalex.org/C49937458","wikidata":"https://www.wikidata.org/wiki/Q2599292","display_name":"Probabilistic logic","level":2,"score":0.2962000072002411},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.29510000348091125},{"id":"https://openalex.org/C74050887","wikidata":"https://www.wikidata.org/wiki/Q848368","display_name":"Rotation (mathematics)","level":2,"score":0.2921000123023987},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.29120001196861267},{"id":"https://openalex.org/C152124472","wikidata":"https://www.wikidata.org/wiki/Q1204361","display_name":"Redundancy (engineering)","level":2,"score":0.2757999897003174},{"id":"https://openalex.org/C116709606","wikidata":"https://www.wikidata.org/wiki/Q1296251","display_name":"Algorithmic efficiency","level":3,"score":0.26910001039505005},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.26840001344680786},{"id":"https://openalex.org/C134537474","wikidata":"https://www.wikidata.org/wiki/Q17144832","display_name":"Naturalness","level":2,"score":0.2556999921798706}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.24963/ijcai.2025/690","is_oa":false,"landing_page_url":"https://doi.org/10.24963/ijcai.2025/690","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence","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":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Key-Value":[0],"(KV)":[1],"cache":[2],"facilitates":[3],"efficient":[4,37],"large":[5],"language":[6],"models":[7],"(LLMs)":[8],"inference":[9],"by":[10],"avoiding":[11],"recomputation":[12],"of":[13,49,53,58,81,121,135],"past":[14],"KVs.":[15],"As":[16],"the":[17,24,34,47,79,98,109,118,122,133,155],"batch":[18,215],"size":[19],"and":[20,63,88,94,144,161,184,217],"context":[21],"length":[22],"increase,":[23],"oversized":[25],"KV":[26,40,86],"caches":[27],"become":[28],"a":[29,50,205,219],"significant":[30,51],"memory":[31,210],"bottleneck,":[32],"highlighting":[33],"need":[35],"for":[36,84],"compression.":[38],"Existing":[39],"quantization":[41,45,87,175],"rely":[42],"on":[43,140,176,192],"fine-grained":[44],"or":[46],"retention":[48],"portion":[52],"high":[54],"bit-widths":[55],"caches,":[56],"both":[57],"which":[59,91,104,131,153],"compromise":[60],"compression":[61],"ratio":[62],"often":[64],"fail":[65],"to":[66,107,111,158],"maintain":[67],"robustness":[68],"at":[69,198],"extremely":[70],"low":[71],"average":[72,200],"bit-widths.":[73,201],"In":[74],"this":[75],"work,":[76],"we":[77],"explore":[78],"potential":[80],"rotation":[82,143],"technique":[83],"2-bit":[85,174],"propose":[89],"RotateKV,":[90],"achieves":[92,166,218],"accurate":[93],"robust":[95],"performance":[96],"through":[97],"following":[99],"innovations:":[100],"(i)":[101],"Outlier-Aware":[102],"Rotation,":[103,130],"utilizes":[105],"channel-reordering":[106],"adapt":[108],"rotations":[110],"varying":[112],"channel-wise":[113],"outlier":[114],"distributions":[115],"without":[116],"sacrificing":[117],"computational":[119],"efficiency":[120],"fast":[123],"Walsh-Hadamard":[124],"transform":[125],"(FWHT);":[126],"(ii)":[127],"Pre-RoPE":[128],"Grouped-Head":[129],"mitigates":[132],"impact":[134],"rotary":[136],"position":[137],"embedding":[138],"(RoPE)":[139],"proposed":[141],"outlier-aware":[142],"further":[145],"smooths":[146],"outliers":[147],"across":[148],"heads;":[149],"(iii)":[150],"Attention-Sink-Aware":[151],"Quantization,":[152],"leverages":[154],"massive":[156],"activations":[157],"precisely":[159],"identify":[160],"protect":[162],"attention":[163],"sinks.":[164],"RotateKV":[165,202],"less":[167,188],"than":[168,189],"0.3":[169],"perplexity":[170],"(PPL)":[171],"degradation":[172,191],"with":[173,187],"WikiText-2":[177],"using":[178],"LLaMA-2-13B,":[179],"maintains":[180],"strong":[181],"CoT":[182],"reasoning":[183],"long-context":[185],"capabilities,":[186],"1.7%":[190],"GSM8K,":[193],"outperforming":[194],"existing":[195],"methods":[196],"even":[197],"lower":[199],"also":[203],"showcases":[204],"3.97\u00d7":[206],"reduction":[207],"in":[208,222],"peak":[209],"usage,":[211],"supports":[212],"5.75\u00d7":[213],"larger":[214],"sizes,":[216],"2.32\u00d7":[220],"speedup":[221],"decoding":[223],"stage.":[224]},"counts_by_year":[{"year":2026,"cited_by_count":1},{"year":2025,"cited_by_count":1}],"updated_date":"2026-05-21T06:26:12.895304","created_date":"2025-10-10T00:00:00"}