{"id":"https://openalex.org/W7166874233","doi":"https://doi.org/10.18653/v1/2026.acl-long.1746","title":"River-LLM: Large Language Model Seamless Exit Based on KV Share","display_name":"River-LLM: Large Language Model Seamless Exit Based on KV Share","publication_year":2026,"publication_date":"2026-01-01","ids":{"openalex":"https://openalex.org/W7166874233","doi":"https://doi.org/10.18653/v1/2026.acl-long.1746"},"language":null,"primary_location":{"id":"doi:10.18653/v1/2026.acl-long.1746","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.acl-long.1746","pdf_url":"https://aclanthology.org/2026.acl-long.1746.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":"Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2026.acl-long.1746.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5079024501","display_name":"Yingtao Shen","orcid":"https://orcid.org/0009-0001-7868-5340"},"institutions":[{"id":"https://openalex.org/I183067930","display_name":"Shanghai Jiao Tong University","ror":"https://ror.org/0220qvk04","country_code":"CN","type":"education","lineage":["https://openalex.org/I183067930"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Yingtao Shen","raw_affiliation_strings":["Shanghai Jiao Tong University Shanghai , China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Shanghai Jiao Tong University Shanghai , China","institution_ids":["https://openalex.org/I183067930"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5121563724","display_name":"An Zou","orcid":null},"institutions":[{"id":"https://openalex.org/I183067930","display_name":"Shanghai Jiao Tong University","ror":"https://ror.org/0220qvk04","country_code":"CN","type":"education","lineage":["https://openalex.org/I183067930"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"An Zou","raw_affiliation_strings":["Shanghai Jiao Tong University Shanghai , China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Shanghai Jiao Tong University Shanghai , China","institution_ids":["https://openalex.org/I183067930"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I183067930"],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":{"value":0.83649784,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"37633","last_page":"37645"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.5210000276565552,"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":0.5210000276565552,"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/T10028","display_name":"Topic Modeling","score":0.27129998803138733,"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/T11273","display_name":"Advanced Graph Neural Networks","score":0.035100001841783524,"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/set","display_name":"Set (abstract data type)","score":0.39320001006126404},{"id":"https://openalex.org/keywords/modeling-language","display_name":"Modeling language","score":0.30820000171661377},{"id":"https://openalex.org/keywords/key","display_name":"Key (lock)","score":0.28279998898506165},{"id":"https://openalex.org/keywords/work","display_name":"Work (physics)","score":0.2824000120162964},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.2815999984741211},{"id":"https://openalex.org/keywords/field","display_name":"Field (mathematics)","score":0.27900001406669617}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6234999895095825},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.39320001006126404},{"id":"https://openalex.org/C179603123","wikidata":"https://www.wikidata.org/wiki/Q1941921","display_name":"Modeling language","level":3,"score":0.30820000171661377},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.28279998898506165},{"id":"https://openalex.org/C18762648","wikidata":"https://www.wikidata.org/wiki/Q42213","display_name":"Work (physics)","level":2,"score":0.2824000120162964},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.2815999984741211},{"id":"https://openalex.org/C9652623","wikidata":"https://www.wikidata.org/wiki/Q190109","display_name":"Field (mathematics)","level":2,"score":0.27900001406669617},{"id":"https://openalex.org/C67186912","wikidata":"https://www.wikidata.org/wiki/Q367664","display_name":"Data modeling","level":2,"score":0.2777999937534332},{"id":"https://openalex.org/C2777904410","wikidata":"https://www.wikidata.org/wiki/Q7397","display_name":"Software","level":2,"score":0.2639000117778778},{"id":"https://openalex.org/C147494362","wikidata":"https://www.wikidata.org/wiki/Q2078905","display_name":"Troubleshooting","level":2,"score":0.25619998574256897},{"id":"https://openalex.org/C168167062","wikidata":"https://www.wikidata.org/wiki/Q1117970","display_name":"Component (thermodynamics)","level":2,"score":0.25290000438690186},{"id":"https://openalex.org/C2779530757","wikidata":"https://www.wikidata.org/wiki/Q1207505","display_name":"Quality (philosophy)","level":2,"score":0.25189998745918274}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2026.acl-long.1746","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.acl-long.1746","pdf_url":"https://aclanthology.org/2026.acl-long.1746.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":"Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2026.acl-long.1746","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.acl-long.1746","pdf_url":"https://aclanthology.org/2026.acl-long.1746.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":"Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","raw_type":"proceedings-article"},"sustainable_development_goals":[{"score":0.4018631875514984,"id":"https://metadata.un.org/sdg/13","display_name":"Climate action"}],"awards":[{"id":"https://openalex.org/G505931191","display_name":null,"funder_award_id":"2025KF1B0010","funder_id":"https://openalex.org/F4320318558","funder_display_name":"Peng Cheng Laboratory"}],"funders":[{"id":"https://openalex.org/F4320318558","display_name":"Peng Cheng Laboratory","ror":"https://ror.org/03qdqbt06"}],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7166874233.pdf","grobid_xml":"https://content.openalex.org/works/W7166874233.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Large":[0],"Language":[1],"Models":[2],"(LLMs)":[3],"have":[4],"demonstrated":[5],"exceptional":[6],"performance":[7],"across":[8],"diverse":[9],"domains":[10],"but":[11],"are":[12],"increasingly":[13],"constrained":[14],"by":[15,29,45],"high":[16,174],"inference":[17,28],"latency.Early":[18],"Exit":[19,41,111],"has":[20],"emerged":[21],"as":[22,66],"a":[23,98,108],"promising":[24],"solution":[25],"to":[26,55,81,120,145,168],"accelerate":[27],"dynamically":[30],"bypassing":[31],"redundant":[32],"layers.However,":[33],"in":[34],"decoder-only":[35],"architectures,":[36],"the":[37,46,57,83,115,127,131],"efficiency":[38],"of":[39],"Early":[40,105],"is":[42],"severely":[43],"bottlenecked":[44],"KV":[47,118,148],"Cache":[48],"Absence":[49],"problem,":[50],"where":[51],"skipped":[52],"layers":[53],"fail":[54],"provide":[56],"necessary":[58],"historical":[59],"states":[60],"for":[61,133],"subsequent":[62],"tokens.Existing":[63],"solutions,":[64],"such":[65],"recomputation":[67],"or":[68,75],"masking,":[69],"either":[70],"introduce":[71],"significant":[72],"latency":[73],"overhead":[74],"incur":[76],"severe":[77],"precision":[78],"loss,":[79],"failing":[80],"bridge":[82],"gap":[84],"between":[85],"theoretical":[86],"layer":[87],"reduction":[88],"and":[89,124,150,159],"practical":[90,170],"wall-clock":[91],"speedup.In":[92],"this":[93],"paper,":[94],"we":[95,137],"propose":[96],"River-LLM,":[97],"training-free":[99],"framework":[100],"that":[101,113,164],"enables":[102],"seamless":[103],"token-level":[104],"Exit.River-LLM":[106],"introduces":[107],"lightweight":[109],"KV-Shared":[110],"River":[112],"allows":[114],"backbone's":[116],"missing":[117],"cache":[119],"be":[121],"naturally":[122],"generated":[123],"preserved":[125],"during":[126],"exit":[128,153],"process,":[129],"eliminating":[130],"need":[132],"costly":[134],"recovery":[135],"operations.Furthermore,":[136],"utilize":[138],"state":[139],"transition":[140],"similarity":[141],"within":[142],"decoder":[143],"blocks":[144],"predict":[146],"cumulative":[147],"errors":[149],"guide":[151],"precise":[152],"decisions.Extensive":[154],"experiments":[155],"on":[156],"mathematical":[157],"reasoning":[158],"code":[160],"generation":[161,175],"tasks":[162],"demonstrate":[163],"River-LLM":[165],"achieves":[166],"1.71\u00d7":[167],"2.16\u00d7":[169],"speedup":[171],"while":[172],"maintaining":[173],"quality.":[176]},"counts_by_year":[],"updated_date":"2026-08-21T09:56:20.448147","created_date":"2026-07-02T00:00:00"}