{"id":"https://openalex.org/W4411120344","doi":"https://doi.org/10.18653/v1/2025.findings-naacl.462","title":"Hierarchical Speculative Decoding with Dynamic Window","display_name":"Hierarchical Speculative Decoding with Dynamic Window","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4411120344","doi":"https://doi.org/10.18653/v1/2025.findings-naacl.462"},"language":"en","primary_location":{"id":"doi:10.18653/v1/2025.findings-naacl.462","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-naacl.462","pdf_url":"https://aclanthology.org/2025.findings-naacl.462.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: NAACL 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-naacl.462.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5054647497","display_name":"Shen-Sian Syu","orcid":"https://orcid.org/0009-0002-5307-3572"},"institutions":[{"id":"https://openalex.org/I16733864","display_name":"National Taiwan University","ror":"https://ror.org/05bqach95","country_code":"TW","type":"education","lineage":["https://openalex.org/I16733864"]}],"countries":["TW"],"is_corresponding":false,"raw_author_name":"Shensian Syu","raw_affiliation_strings":["National Taiwan University , Taiwan"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"National Taiwan University , Taiwan","institution_ids":["https://openalex.org/I16733864"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5040508737","display_name":"Hung-yi Lee","orcid":"https://orcid.org/0000-0002-9654-5747"},"institutions":[{"id":"https://openalex.org/I16733864","display_name":"National Taiwan University","ror":"https://ror.org/05bqach95","country_code":"TW","type":"education","lineage":["https://openalex.org/I16733864"]}],"countries":["TW"],"is_corresponding":false,"raw_author_name":"Hung-yi Lee","raw_affiliation_strings":["National Taiwan University , Taiwan"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"National Taiwan University , Taiwan","institution_ids":["https://openalex.org/I16733864"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I16733864"],"apc_list":null,"apc_paid":null,"fwci":1.4204,"has_fulltext":true,"cited_by_count":1,"citation_normalized_percentile":{"value":0.79028359,"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":"8260","last_page":"8273"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10320","display_name":"Neural Networks and Applications","score":0.6173999905586243,"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/T10320","display_name":"Neural Networks and Applications","score":0.6173999905586243,"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":0.5861999988555908,"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/T12162","display_name":"Cellular Automata and Applications","score":0.5248000025749207,"subfield":{"id":"https://openalex.org/subfields/1703","display_name":"Computational Theory and Mathematics"},"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/decoding-methods","display_name":"Decoding methods","score":0.7278677821159363},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.7142106294631958},{"id":"https://openalex.org/keywords/window","display_name":"Window (computing)","score":0.678045928478241},{"id":"https://openalex.org/keywords/parallel-computing","display_name":"Parallel computing","score":0.3657422661781311},{"id":"https://openalex.org/keywords/algorithm","display_name":"Algorithm","score":0.15933552384376526},{"id":"https://openalex.org/keywords/operating-system","display_name":"Operating system","score":0.11133220791816711}],"concepts":[{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.7278677821159363},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7142106294631958},{"id":"https://openalex.org/C2778751112","wikidata":"https://www.wikidata.org/wiki/Q835016","display_name":"Window (computing)","level":2,"score":0.678045928478241},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.3657422661781311},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.15933552384376526},{"id":"https://openalex.org/C111919701","wikidata":"https://www.wikidata.org/wiki/Q9135","display_name":"Operating system","level":1,"score":0.11133220791816711}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2025.findings-naacl.462","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-naacl.462","pdf_url":"https://aclanthology.org/2025.findings-naacl.462.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: NAACL 2025","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2025.findings-naacl.462","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-naacl.462","pdf_url":"https://aclanthology.org/2025.findings-naacl.462.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: NAACL 2025","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4411120344.pdf","grobid_xml":"https://content.openalex.org/works/W4411120344.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4391375266","https://openalex.org/W2899084033","https://openalex.org/W2748952813","https://openalex.org/W2390901981","https://openalex.org/W2390279801","https://openalex.org/W2109115373","https://openalex.org/W4391913857","https://openalex.org/W2358668433","https://openalex.org/W4396701345","https://openalex.org/W2376932109"],"abstract_inverted_index":{"Speculative":[0,79],"Decoding":[1,80],"(SD)":[2],"utilizes":[3],"an":[4],"efficient":[5],"draft":[6,103],"model":[7,104],"to":[8,52,70,105,109],"generate":[9],"multiple":[10],"tokens,":[11],"which":[12,61],"are":[13],"subsequently":[14],"verified":[15],"in":[16,30],"parallel":[17],"by":[18,113],"a":[19,85,97,115],"target":[20],"model.This":[21],"approach":[22],"has":[23],"shown":[24],"significant":[25],"potential":[26],"for":[27,92],"accelerating":[28],"inference":[29],"large":[31],"language":[32],"models":[33,123],"(LLMs),":[34],"with":[35,81],"performance":[36,138],"heavily":[37],"reliant":[38],"on":[39,49,147,151],"the":[40,58,90,102,120,130,134],"hyperparameter":[41],"K-the":[42],"window":[43,59],"size.However,":[44],"previous":[45],"methods":[46],"often":[47],"depend":[48],"simple":[50],"heuristics":[51],"select":[53],"K":[54],"or":[55,66],"dynamically":[56],"adjust":[57],"size,":[60],"may":[62],"necessitate":[63],"additional":[64,93],"training":[65],"careful":[67],"resource":[68],"management":[69],"avoid":[71],"competition.To":[72],"address":[73],"these":[74],"challenges,":[75],"we":[76,95,127],"propose":[77],"Hierarchical":[78],"Dynamic":[82],"Window":[83],"(HSDDW),":[84],"straightforward":[86],"framework":[87],"that":[88,100,118],"eliminates":[89],"need":[91],"training.Specifically,":[94],"introduce":[96],"self-verify":[98],"mechanism":[99],"enables":[101],"autonomously":[106],"decide":[107],"when":[108],"stop":[110],"generating":[111],"tokens.Additionally,":[112],"integrating":[114],"hierarchical":[116],"structure":[117],"leverages":[119],"capabilities":[121],"of":[122,124,133,145],"different":[125],"sizes,":[126],"significantly":[128],"enhance":[129],"overall":[131],"speed":[132],"system.HSDDW":[135],"demonstrates":[136],"competitive":[137],"across":[139],"four":[140],"datasets,":[141],"achieving":[142],"notable":[143],"speedups":[144],"2.91\u00d7":[146],"MT-Bench":[148],"and":[149],"2.99\u00d7":[150],"Alpaca,":[152],"outperforming":[153],"existing":[154],"state-of-the-art":[155],"methods.1.":[156],"SD":[157],"13":[158],":":[159],"Apache":[160],"License":[161],"2.0":[162]},"counts_by_year":[{"year":2026,"cited_by_count":1}],"updated_date":"2026-08-24T07:32:12.397491","created_date":"2025-10-10T00:00:00"}