{"id":"https://openalex.org/W7155965484","doi":"https://doi.org/10.1016/j.neucom.2026.133760","title":"Adaptive hybrid speculative decoding for accelerating large language model inference","display_name":"Adaptive hybrid speculative decoding for accelerating large language model inference","publication_year":2026,"publication_date":"2026-04-27","ids":{"openalex":"https://openalex.org/W7155965484","doi":"https://doi.org/10.1016/j.neucom.2026.133760"},"language":"en","primary_location":{"id":"doi:10.1016/j.neucom.2026.133760","is_oa":true,"landing_page_url":"https://doi.org/10.1016/j.neucom.2026.133760","pdf_url":null,"source":{"id":"https://openalex.org/S45693802","display_name":"Neurocomputing","issn_l":"0925-2312","issn":["0925-2312","1872-8286"],"is_oa":false,"is_in_doaj":false,"is_core":true,"host_organization":"https://openalex.org/P4310320990","host_organization_name":"Elsevier BV","host_organization_lineage":["https://openalex.org/P4310320990"],"host_organization_lineage_names":["Elsevier BV"],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Neurocomputing","raw_type":"journal-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"hybrid","oa_url":"https://doi.org/10.1016/j.neucom.2026.133760","any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5101709269","display_name":"Yang Yong","orcid":"https://orcid.org/0000-0002-3309-4643"},"institutions":[{"id":"https://openalex.org/I4210136793","display_name":"Peng Cheng Laboratory","ror":"https://ror.org/03qdqbt06","country_code":"CN","type":"facility","lineage":["https://openalex.org/I4210136793"]},{"id":"https://openalex.org/I4210165038","display_name":"University of Chinese Academy of Sciences","ror":"https://ror.org/05qbk4x57","country_code":"CN","type":"education","lineage":["https://openalex.org/I19820366","https://openalex.org/I4210165038"]}],"countries":["CN"],"is_corresponding":true,"raw_author_name":"Yang Yong","raw_affiliation_strings":["Pengcheng Laboratory, Shenzhen, China","School of Electronic, Electrical and Communication Engineering, University of Chinese Academy of Sciences, Beijing, China"],"raw_orcid":"https://orcid.org/0000-0002-3309-4643","affiliations":[{"raw_affiliation_string":"Pengcheng Laboratory, Shenzhen, China","institution_ids":["https://openalex.org/I4210136793"]},{"raw_affiliation_string":"School of Electronic, Electrical and Communication Engineering, University of Chinese Academy of Sciences, Beijing, China","institution_ids":["https://openalex.org/I4210165038"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101702056","display_name":"Ting Yang","orcid":"https://orcid.org/0000-0001-8586-1675"},"institutions":[{"id":"https://openalex.org/I4210136793","display_name":"Peng Cheng Laboratory","ror":"https://ror.org/03qdqbt06","country_code":"CN","type":"facility","lineage":["https://openalex.org/I4210136793"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Ting Ting Yang","raw_affiliation_strings":["Pengcheng Laboratory, Shenzhen, China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Pengcheng Laboratory, Shenzhen, China","institution_ids":["https://openalex.org/I4210136793"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5126730561","display_name":"Shao Shuai Gao","orcid":null},"institutions":[{"id":"https://openalex.org/I4210165038","display_name":"University of Chinese Academy of Sciences","ror":"https://ror.org/05qbk4x57","country_code":"CN","type":"education","lineage":["https://openalex.org/I19820366","https://openalex.org/I4210165038"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Shao Shuai Gao","raw_affiliation_strings":["School of Electronic, Electrical and Communication Engineering, University of Chinese Academy of Sciences, Beijing, China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"School of Electronic, Electrical and Communication Engineering, University of Chinese Academy of Sciences, Beijing, China","institution_ids":["https://openalex.org/I4210165038"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5134706087","display_name":"Ye Hua Yin","orcid":null},"institutions":[{"id":"https://openalex.org/I4210136793","display_name":"Peng Cheng Laboratory","ror":"https://ror.org/03qdqbt06","country_code":"CN","type":"facility","lineage":["https://openalex.org/I4210136793"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Ye Hua Yin","raw_affiliation_strings":["Pengcheng Laboratory, Shenzhen, China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Pengcheng Laboratory, Shenzhen, China","institution_ids":["https://openalex.org/I4210136793"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"corresponding_author_ids":["https://openalex.org/A5101709269"],"corresponding_institution_ids":["https://openalex.org/I4210136793","https://openalex.org/I4210165038"],"apc_list":{"value":2470,"currency":"USD","value_usd":2470},"apc_paid":{"value":2470,"currency":"USD","value_usd":2470},"fwci":0.0,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":{"value":0.4909922,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":"689","issue":null,"first_page":"133760","last_page":"133760"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10201","display_name":"Speech Recognition and Synthesis","score":0.4481000006198883,"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/T10201","display_name":"Speech Recognition and Synthesis","score":0.4481000006198883,"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/T10775","display_name":"Generative Adversarial Networks and Image Synthesis","score":0.09109999984502792,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.060100000351667404,"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/decoding-methods","display_name":"Decoding methods","score":0.6182000041007996},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.5760999917984009},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.541700005531311},{"id":"https://openalex.org/keywords/encoding","display_name":"Encoding (memory)","score":0.29170000553131104},{"id":"https://openalex.org/keywords/natural-language","display_name":"Natural language","score":0.27239999175071716},{"id":"https://openalex.org/keywords/approximate-inference","display_name":"Approximate inference","score":0.26750001311302185}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7882999777793884},{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.6182000041007996},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.5760999917984009},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.541700005531311},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5062000155448914},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.4002000093460083},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.3488999903202057},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.31779998540878296},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.31130000948905945},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.29170000553131104},{"id":"https://openalex.org/C28490314","wikidata":"https://www.wikidata.org/wiki/Q189436","display_name":"Speech recognition","level":1,"score":0.2784999907016754},{"id":"https://openalex.org/C195324797","wikidata":"https://www.wikidata.org/wiki/Q33742","display_name":"Natural language","level":2,"score":0.27239999175071716},{"id":"https://openalex.org/C2777472644","wikidata":"https://www.wikidata.org/wiki/Q16968992","display_name":"Approximate inference","level":3,"score":0.26750001311302185},{"id":"https://openalex.org/C40743351","wikidata":"https://www.wikidata.org/wiki/Q7002049","display_name":"Neural decoding","level":3,"score":0.25279998779296875},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.25099998712539673},{"id":"https://openalex.org/C60048249","wikidata":"https://www.wikidata.org/wiki/Q37437","display_name":"Syntax","level":2,"score":0.25040000677108765},{"id":"https://openalex.org/C184337299","wikidata":"https://www.wikidata.org/wiki/Q1437428","display_name":"Semantics (computer science)","level":2,"score":0.2500999867916107}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1016/j.neucom.2026.133760","is_oa":true,"landing_page_url":"https://doi.org/10.1016/j.neucom.2026.133760","pdf_url":null,"source":{"id":"https://openalex.org/S45693802","display_name":"Neurocomputing","issn_l":"0925-2312","issn":["0925-2312","1872-8286"],"is_oa":false,"is_in_doaj":false,"is_core":true,"host_organization":"https://openalex.org/P4310320990","host_organization_name":"Elsevier BV","host_organization_lineage":["https://openalex.org/P4310320990"],"host_organization_lineage_names":["Elsevier BV"],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Neurocomputing","raw_type":"journal-article"}],"best_oa_location":{"id":"doi:10.1016/j.neucom.2026.133760","is_oa":true,"landing_page_url":"https://doi.org/10.1016/j.neucom.2026.133760","pdf_url":null,"source":{"id":"https://openalex.org/S45693802","display_name":"Neurocomputing","issn_l":"0925-2312","issn":["0925-2312","1872-8286"],"is_oa":false,"is_in_doaj":false,"is_core":true,"host_organization":"https://openalex.org/P4310320990","host_organization_name":"Elsevier BV","host_organization_lineage":["https://openalex.org/P4310320990"],"host_organization_lineage_names":["Elsevier BV"],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Neurocomputing","raw_type":"journal-article"},"sustainable_development_goals":[],"awards":[],"funders":[{"id":"https://openalex.org/F4320318558","display_name":"Peng Cheng Laboratory","ror":"https://ror.org/03qdqbt06"}],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":3,"referenced_works":["https://openalex.org/W4285173593","https://openalex.org/W4402683901","https://openalex.org/W4404782770"],"related_works":[],"abstract_inverted_index":{"Processing":[0],"long":[1],"sequences":[2,175,198],"efficiently":[3,93,163],"is":[4,124,130],"crucial":[5],"for":[6,237,245],"modern":[7,165],"large":[8],"language":[9],"models":[10,15],"(LLM),":[11],"yet":[12],"serving":[13],"these":[14,159],"incurs":[16],"high":[17],"inference":[18,40],"costs":[19],"due":[20],"to":[21,153,186,226],"computational":[22],"bottlenecks":[23],"in":[24,115,199],"speculative":[25,28],"decoding.":[26],"Existing":[27],"decoding":[29,49],"methods":[30],"leverage":[31],"fixed-granularity":[32],"draft\u2013verification":[33],"strategies,":[34],"which":[35],"introduce":[36],"a":[37,44,55,75,95,127,169,191],"trade-off":[38],"between":[39],"speed":[41],"and":[42,87,107,195,205,216,228],"accuracy:":[43],"larger":[45],"draft":[46,112,146],"step":[47,57],"accelerates":[48],"but":[50,60],"increases":[51],"verification":[52,123,155],"overhead,":[53],"while":[54,135],"smaller":[56],"ensures":[58],"correctness":[59],"reduces":[61,213],"efficiency.":[62],"To":[63,157],"overcome":[64],"this":[65],"trade-off,":[66],"we":[67,167],"propose":[68],"Adaptive":[69,84],"Hybrid":[70],"Speculative":[71],"Decoding":[72],"(AHSD)":[73],",":[74,181],"novel":[76],"framework":[77],"built":[78],"upon":[79],"two":[80],"key":[81],"algorithmic":[82,160],"innovations:":[83],"Feedback":[85],"Thresholds":[86],"Confidence-Driven":[88],"Two-Stage":[89],"Token":[90],"Processing,":[91],"implemented":[92],"via":[94],"bucketed":[96,170],"batching":[97,171],"optimization.":[98],"First,":[99],"the":[100,111,121,140],"adaptive":[101],"feedback":[102],"threshold":[103],"dynamically":[104,177],"adjusts":[105],"rollback":[106],"validation":[108],"by":[109,176,209],"monitoring":[110],"model\u2019s":[113],"confidence":[114,119],"real":[116],"time,":[117],"if":[118],"exceeds":[120],"threshold,":[122],"skipped;":[125],"otherwise,":[126],"refining":[128],"check":[129],"triggered,":[131],"maximizing":[132],"first-pass":[133],"success":[134],"minimizing":[136],"false":[137],"positives.":[138],"Second,":[139],"confidence-driven":[141],"two-stage":[142],"token":[143],"processing":[144],"decouples":[145],"generation":[147],"from":[148],"verification,":[149],"allowing":[150],"high-confidence":[151],"tokens":[152],"bypass":[154],"entirely.":[156],"ensure":[158],"innovations":[161],"execute":[162],"on":[164],"GPUs,":[166],"employ":[168],"strategy":[172],"that":[173,222,239],"groups":[174],"chosen":[178],"block":[179],"sizes":[180],"pads":[182],"each":[183],"homogeneous":[184],"batch":[185],"its":[187],"maximum":[188],"length":[189,210],"with":[190],"unified":[192],"attention":[193],"mask,":[194],"executes":[196],"all":[197],"parallel,":[200],"thereby":[201],"eliminating":[202],"scheduling":[203],"overhead":[204],"branch":[206],"divergence":[207],"caused":[208],"variability.This":[211],"sharply":[212],"re-model":[214],"frequency":[215],"enhances":[217],"parallelism.":[218],"Comprehensive":[219],"simulations":[220],"show":[221],"AHSD":[223],"achieves":[224],"up":[225],"speedup":[227],"throughput":[229],"gain":[230],"over":[231],"existing":[232],"schemes,":[233],"making":[234],"it":[235],"suitable":[236],"applications":[238],"can":[240],"tolerate":[241],"minor":[242],"quality":[243],"loss":[244],"substantially":[246],"faster":[247],"inference.":[248]},"counts_by_year":[],"updated_date":"2026-06-18T08:10:14.011955","created_date":"2026-04-28T00:00:00"}