{"id":"https://openalex.org/W7213205698","doi":"https://doi.org/10.1016/j.compeleceng.2026.111505","title":"Component-aware self-speculative decoding for hybrid language models: An architectural viability study","display_name":"Component-aware self-speculative decoding for hybrid language models: An architectural viability study","publication_year":2026,"publication_date":"2026-09-14","ids":{"openalex":"https://openalex.org/W7213205698","doi":"https://doi.org/10.1016/j.compeleceng.2026.111505"},"language":"en","primary_location":{"id":"doi:10.1016/j.compeleceng.2026.111505","doi":"https://doi.org/10.1016/j.compeleceng.2026.111505","is_oa":true,"landing_page_url":"https://doi.org/10.1016/j.compeleceng.2026.111505","pdf_url":"https://www.sciencedirect.com/science/article/pii/S0045790626005756/pdf","source":{"id":"https://openalex.org/S121340289","display_name":"Computers & Electrical Engineering","issn_l":"0045-7906","issn":["0045-7906","1879-0755"],"is_oa":false,"is_in_doaj":false,"is_core":true,"listed_in":["cwts-core","dongbi-d","jufo-1","norway-1","poland-70","russia-white-list-1"],"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":"Computers and Electrical Engineering","raw_type":"journal-article"},"volume":null,"type":"article","is_traditional":true,"indexed_in":["arxiv","crossref","datacite"],"open_access":{"is_oa":true,"oa_status":"hybrid","oa_url":"https://www.sciencedirect.com/science/article/pii/S0045790626005756/pdf","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5135336248","display_name":"Hector Borobia","orcid":"https://orcid.org/0009-0009-9460-4967","observed_orcids":["https://orcid.org/0009-0009-9460-4967"]},"institutions":[{"id":"https://openalex.org/I4210131846","display_name":"Artificial Intelligence Research Institute","ror":"https://ror.org/03c0ach84","country_code":"ES","type":"facility","lineage":["https://openalex.org/I134820265","https://openalex.org/I4210131846"]},{"id":"https://openalex.org/I60053951","display_name":"Universitat Polit\u00e8cnica de Val\u00e8ncia","ror":"https://ror.org/01460j859","country_code":"ES","type":"education","lineage":["https://openalex.org/I60053951"]}],"countries":["ES"],"is_corresponding":true,"raw_author_name":"Hector Borobia","raw_affiliation_strings":["VRAIN \u2013 Valencian Research Institute for Artificial Intelligence, Universitat Polit\u00e8cnica de Val\u00e8ncia, Val\u00e8ncia, Spain"],"raw_orcid":"https://orcid.org/0009-0009-9460-4967","affiliations":[{"raw_affiliation_string":"VRAIN \u2013 Valencian Research Institute for Artificial Intelligence, Universitat Polit\u00e8cnica de Val\u00e8ncia, Val\u00e8ncia, Spain","institution_ids":["https://openalex.org/I4210131846","https://openalex.org/I60053951"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5037110970","display_name":"Elies Segu\u00ed\u2010Mas","orcid":"https://orcid.org/0000-0001-6985-4758","observed_orcids":["https://orcid.org/0000-0001-6985-4758"]},"institutions":[{"id":"https://openalex.org/I60053951","display_name":"Universitat Polit\u00e8cnica de Val\u00e8ncia","ror":"https://ror.org/01460j859","country_code":"ES","type":"education","lineage":["https://openalex.org/I60053951"]}],"countries":["ES"],"is_corresponding":false,"raw_author_name":"Elies Segu\u00ed-Mas","raw_affiliation_strings":["Department of Economics and Social Sciences, Universitat Polit\u00e8cnica de Val\u00e8ncia, Val\u00e8ncia, Spain"],"raw_orcid":"https://orcid.org/0000-0001-6985-4758","affiliations":[{"raw_affiliation_string":"Department of Economics and Social Sciences, Universitat Polit\u00e8cnica de Val\u00e8ncia, Val\u00e8ncia, Spain","institution_ids":["https://openalex.org/I60053951"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5002533063","display_name":"Guillermina Tormo\u2010Carb\u00f3","orcid":"https://orcid.org/0000-0001-5865-9631","observed_orcids":["https://orcid.org/0000-0001-5865-9631"]},"institutions":[{"id":"https://openalex.org/I60053951","display_name":"Universitat Polit\u00e8cnica de Val\u00e8ncia","ror":"https://ror.org/01460j859","country_code":"ES","type":"education","lineage":["https://openalex.org/I60053951"]}],"countries":["ES"],"is_corresponding":false,"raw_author_name":"Guillermina Tormo-Carb\u00f3","raw_affiliation_strings":["Department of Business Organisation, Universitat Polit\u00e8cnica de Val\u00e8ncia, Val\u00e8ncia, Spain"],"raw_orcid":"https://orcid.org/0000-0001-5865-9631","affiliations":[{"raw_affiliation_string":"Department of Business Organisation, Universitat Polit\u00e8cnica de Val\u00e8ncia, Val\u00e8ncia, Spain","institution_ids":["https://openalex.org/I60053951"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"corresponding_author_ids":["https://openalex.org/A5135336248"],"corresponding_institution_ids":["https://openalex.org/I4210131846","https://openalex.org/I60053951"],"apc_list":{"value":2880,"currency":"USD","value_usd":2880},"apc_paid":{"value":2880,"currency":"USD","value_usd":2880},"fwci":0.0,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":{"value":0.0,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":"140","issue":null,"first_page":"111505","last_page":"111505"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.8314999938011169,"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":0.8314999938011169,"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":0.12809999287128448,"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/T10036","display_name":"Advanced Neural Network Applications","score":0.009100000374019146,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/hybrid-language-models","display_name":"hybrid language models","score":0.9300000071525574},{"id":"https://openalex.org/keywords/state-space-models","display_name":"state-space model","score":0.6809999942779541},{"id":"https://openalex.org/keywords/linear-attention","display_name":"linear attention","score":0.8059999942779541},{"id":"https://openalex.org/keywords/speculative-decoding","display_name":"speculative decoding","score":0.5239999890327454},{"id":"https://openalex.org/keywords/inference-latency","display_name":"inference latency","score":0.7400000095367432},{"id":"https://openalex.org/keywords/acceptance-rate","display_name":"acceptance rate","score":0.5709999799728394},{"id":"https://openalex.org/keywords/qwen3-5","display_name":"Qwen3.5","score":0.9369999766349792},{"id":"https://openalex.org/keywords/granite","display_name":"granite","score":0.3779999911785126},{"id":"https://openalex.org/keywords/transformer","display_name":"transformer models","score":0.3019999861717224}],"concepts":[{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.7935000061988831},{"id":"https://openalex.org/C68339613","wikidata":"https://www.wikidata.org/wiki/Q1549489","display_name":"Speedup","level":2,"score":0.7720999717712402},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6320000290870667},{"id":"https://openalex.org/C100279451","wikidata":"https://www.wikidata.org/wiki/Q372193","display_name":"Perplexity","level":3,"score":0.599399983882904},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.544700026512146},{"id":"https://openalex.org/C50897621","wikidata":"https://www.wikidata.org/wiki/Q2665508","display_name":"Hybrid system","level":2,"score":0.36570000648498535},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.34130001068115234},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.3395000100135803},{"id":"https://openalex.org/C3261483","wikidata":"https://www.wikidata.org/wiki/Q119565","display_name":"Frame rate","level":2,"score":0.31459999084472656},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.3111000061035156},{"id":"https://openalex.org/C2776036281","wikidata":"https://www.wikidata.org/wiki/Q48769818","display_name":"Constraint (computer-aided design)","level":2,"score":0.3082999885082245},{"id":"https://openalex.org/C66882249","wikidata":"https://www.wikidata.org/wiki/Q169336","display_name":"Homogeneous","level":2,"score":0.29589998722076416},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.2784999907016754},{"id":"https://openalex.org/C126042441","wikidata":"https://www.wikidata.org/wiki/Q1324888","display_name":"Frame (networking)","level":2,"score":0.2727000117301941},{"id":"https://openalex.org/C82876162","wikidata":"https://www.wikidata.org/wiki/Q17096504","display_name":"Latency (audio)","level":2,"score":0.27239999175071716},{"id":"https://openalex.org/C134121241","wikidata":"https://www.wikidata.org/wiki/Q899301","display_name":"Yield (engineering)","level":2,"score":0.26660001277923584},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.2612999975681305},{"id":"https://openalex.org/C187834632","wikidata":"https://www.wikidata.org/wiki/Q188804","display_name":"Factorization","level":2,"score":0.2540000081062317},{"id":"https://openalex.org/C94375191","wikidata":"https://www.wikidata.org/wiki/Q11205","display_name":"Arithmetic","level":1,"score":0.25029999017715454}],"mesh":[],"locations_count":4,"locations":[{"id":"doi:10.1016/j.compeleceng.2026.111505","doi":"https://doi.org/10.1016/j.compeleceng.2026.111505","is_oa":true,"landing_page_url":"https://doi.org/10.1016/j.compeleceng.2026.111505","pdf_url":"https://www.sciencedirect.com/science/article/pii/S0045790626005756/pdf","source":{"id":"https://openalex.org/S121340289","display_name":"Computers & Electrical Engineering","issn_l":"0045-7906","issn":["0045-7906","1879-0755"],"is_oa":false,"is_in_doaj":false,"is_core":true,"listed_in":["cwts-core","dongbi-d","jufo-1","norway-1","poland-70","russia-white-list-1"],"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":"Computers and Electrical Engineering","raw_type":"journal-article"},{"id":"pmh:oai:riunet.upv.es:10251/241967","doi":null,"is_oa":true,"landing_page_url":"https://riunet.upv.es/handle/10251/241967","pdf_url":null,"source":{"id":"https://openalex.org/S4306400639","display_name":"RiuNet (Universitat Polit\u00e8cnica de Val\u00e8ncia)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I60053951","host_organization_name":"Universitat Polit\u00e8cnica de Val\u00e8ncia","host_organization_lineage":["https://openalex.org/I60053951"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":null,"raw_type":"info:eu-repo/semantics/article"},{"id":"pmh:oai:arXiv.org:2605.01106","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2605.01106","pdf_url":"https://arxiv.org/pdf/2605.01106","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2605.01106","doi":"https://doi.org/10.48550/arxiv.2605.01106","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.01106","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.1016/j.compeleceng.2026.111505","doi":"https://doi.org/10.1016/j.compeleceng.2026.111505","is_oa":true,"landing_page_url":"https://doi.org/10.1016/j.compeleceng.2026.111505","pdf_url":"https://www.sciencedirect.com/science/article/pii/S0045790626005756/pdf","source":{"id":"https://openalex.org/S121340289","display_name":"Computers & Electrical Engineering","issn_l":"0045-7906","issn":["0045-7906","1879-0755"],"is_oa":false,"is_in_doaj":false,"is_core":true,"listed_in":["cwts-core","dongbi-d","jufo-1","norway-1","poland-70","russia-white-list-1"],"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":"Computers and Electrical Engineering","raw_type":"journal-article"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[{"id":"https://openalex.org/F4320334905","display_name":"Universitat Polit\u00e8cnica de Val\u00e8ncia","ror":"https://ror.org/01460j859"}],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7213205698.pdf","grobid_xml":"https://content.openalex.org/works/W7213205698.grobid-xml"},"referenced_works_count":17,"referenced_works":["https://openalex.org/W3210904070","https://openalex.org/W4319166707","https://openalex.org/W4387321091","https://openalex.org/W4393399080","https://openalex.org/W4399655853","https://openalex.org/W4402671987","https://openalex.org/W4403345584","https://openalex.org/W4414856349","https://openalex.org/W4414895251","https://openalex.org/W4415327314","https://openalex.org/W4416036444","https://openalex.org/W4416047838","https://openalex.org/W4416102280","https://openalex.org/W6929450882","https://openalex.org/W7131434869","https://openalex.org/W7140274764","https://openalex.org/W7154742567"],"related_works":["https://openalex.org/W7160281113","https://openalex.org/W7163187171","https://openalex.org/W7166886618","https://openalex.org/W7172426877","https://openalex.org/W7222112453","https://openalex.org/W4416058011","https://openalex.org/W4403346564","https://openalex.org/W7138867972","https://openalex.org/W7169850532","https://openalex.org/W7155965484"],"abstract_inverted_index":{"Speculative":[0],"decoding":[1,46,68],"reduces":[2],"language-model":[3],"inference":[4],"latency":[5],"when":[6],"a":[7,63,74,140,195,252],"cheap":[8],"draft":[9,31],"distribution":[10],"matches":[11],"the":[12,69,84,96,106,123,131,149,187,207,210,228,235,242],"target.":[13],"Existing":[14],"self-speculative":[15,45],"methods":[16],"skip":[17],"layers":[18],"in":[19,241],"homogeneous":[20],"Transformer":[21],"stacks;":[22],"we":[23],"ask":[24],"whether":[25],"hybrid":[26,71,86,137],"language":[27],"models":[28],"can":[29],"instead":[30],"from":[32],"their":[33],"internal":[34,236],"state-space":[35,190,237],"or":[36],"linear-attention":[37],"components":[38],"as":[39,105,230],"parameter-free":[40],"drafters.":[41],"We":[42,226],"introduce":[43],"component-aware":[44],"and":[47,58,133,151,183,255],"evaluate":[48],"it":[49,167,247],"on":[50],"nine":[51],"models:":[52],"eight":[53,124],"hybrids":[54],"across":[55],"five":[56],"families":[57],"both":[59],"integration":[60],"paradigms,":[61],"plus":[62],"pure-Transformer":[64],"control.":[65],"Under":[66],"greedy":[67],"parallel":[70,141],"Falcon-H1-0.5B":[72],"reaches":[73,88,194],"mean":[75],"acceptance":[76,103,214],"rate":[77],"A":[78],"2":[79],"=":[80,115,120,127,158,203],"0.680":[81],",":[82,118],"whereas":[83],"sequential":[85,136],"Qwen3.5-0.8B":[87],"only":[89],"0.038.":[90],"The":[91,144],"governing":[92],"factor":[93],"is":[94,146,163,213,239,251],"not":[95,169],"parallel/sequential":[97],"label":[98],"but":[99,166,245],"functional":[100],"attention-dependence":[101],":":[102],"falls":[104],"perplexity":[107],"increase":[108],"under":[109],"attention":[110],"suppression":[111],"grows":[112],"(Spearman":[113],"\u03c1":[114,126],"\u2212":[116,128],"0.79":[117],"p":[119,157],"0.028":[121],"over":[122],"hybrids;":[125],"0.75":[129],"with":[130],"control),":[132],"an":[134,231],"attention-light":[135],"(granite)":[138],"outperforms":[139],"one":[142,199],"(Falcon-H1-1.5B).":[143],"separation":[145],"stable":[147],"at":[148],"0.5B":[150],"3B":[152],"scales":[153],"tested":[154],"(permutation":[155],"test,":[156],"0.962":[159],").":[160,205],"Acceptance":[161],"viability":[162,233],"thus":[164],"demonstrated,":[165],"does":[168],"by":[170],"itself":[171],"yield":[172],"speedup.":[173],"Using":[174],"hardware-measured":[175],"draft/verify":[176],"cost":[177],"ratios":[178],"(":[179],"c":[180],"eff":[181],"),":[182],"even":[184],"after":[185],"activating":[186],"official":[188],"accelerated":[189],"kernels,":[191],"no":[192],"drafter":[193,238],"theoretical":[196],"speedup":[197,250],"above":[198],"(best":[200],"S":[201],"theory":[202],"0.986":[204],"With":[206],"kernels":[208],"active":[209],"binding":[211],"constraint":[212],"rather":[215],"than":[216],"drafting":[217],"cost,":[218],"which":[219],"has":[220],"fallen":[221],"to":[222],"its":[223],"parameter-count":[224],"floor.":[225],"frame":[227],"work":[229],"architectural":[232],"study:":[234],"acceptance-viable":[240],"right":[243],"architectures,":[244],"turning":[246],"into":[248],"wall-clock":[249],"separate":[253],"execution-stack":[254],"modeling":[256],"problem.":[257]},"counts_by_year":[],"updated_date":"2026-11-10T11:39:43.021295","created_date":"2026-09-15T00:00:00"}