{"id":"https://openalex.org/W4415328504","doi":"https://doi.org/10.48550/arxiv.2505.15141","title":"BanditSpec: Adaptive Speculative Decoding via Bandit Algorithms","display_name":"BanditSpec: Adaptive Speculative Decoding via Bandit Algorithms","publication_year":2025,"publication_date":"2025-05-21","ids":{"openalex":"https://openalex.org/W4415328504","doi":"https://doi.org/10.48550/arxiv.2505.15141"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2505.15141","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2505.15141","pdf_url":"https://arxiv.org/pdf/2505.15141","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2505.15141","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":null,"display_name":"Hou, Yunlong","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Hou, Yunlong","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5060676898","display_name":"Fengzhuo Zhang","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Fengzhuo","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Du, Cunxiao","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Du, Cunxiao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5043732387","display_name":"Xuan Zhang","orcid":"https://orcid.org/0000-0001-8214-7265","observed_orcids":["https://orcid.org/0000-0001-8214-7265","https://orcid.org/0009-0002-3465-7118"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Xuan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5110004412","display_name":"Jiachun Pan","orcid":"https://orcid.org/0009-0000-2070-8290","observed_orcids":["https://orcid.org/0009-0000-2070-8290"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Pan, Jiachun","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Pang, Tianyu","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Pang, Tianyu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Du, Chao","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Du, Chao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5058345431","display_name":"Vincent Y. F. Tan","orcid":"https://orcid.org/0000-0002-5008-4527","observed_orcids":["https://orcid.org/0000-0002-5008-4527","https://orcid.org/0000-0003-1436-5337"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tan, Vincent Y. F.","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5101727948","display_name":"Zhuoran Yang","orcid":"https://orcid.org/0000-0001-5269-9958","observed_orcids":["https://orcid.org/0000-0001-5269-9958"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yang, Zhuoran","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T12101","display_name":"Advanced Bandit Algorithms Research","score":0.9872999787330627,"subfield":{"id":"https://openalex.org/subfields/1803","display_name":"Management Science and Operations Research"},"field":{"id":"https://openalex.org/fields/18","display_name":"Decision Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},"topics":[{"id":"https://openalex.org/T12101","display_name":"Advanced Bandit Algorithms Research","score":0.9872999787330627,"subfield":{"id":"https://openalex.org/subfields/1803","display_name":"Management Science and Operations Research"},"field":{"id":"https://openalex.org/fields/18","display_name":"Decision Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.009399999864399433,"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":0.001500000013038516,"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/speculative-decoding","display_name":"speculative decoding","score":0.9959999918937683},{"id":"https://openalex.org/keywords/multi-armed-bandits","display_name":"multi-armed bandits","score":0.7940000295639038},{"id":"https://openalex.org/keywords/hyperparameter-optimization","display_name":"hyperparameter optimization","score":0.8500000238418579},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.8700000047683716},{"id":"https://openalex.org/keywords/e-learning","display_name":"online learning","score":0.8370000123977661},{"id":"https://openalex.org/keywords/regret-bounds","display_name":"regret bounds","score":0.6169999837875366},{"id":"https://openalex.org/keywords/large-language-model-inference","display_name":"large language model inference","score":0.5080000162124634},{"id":"https://openalex.org/keywords/inference-acceleration","display_name":"inference acceleration","score":0.5809999704360962},{"id":"https://openalex.org/keywords/draft-model","display_name":"draft model","score":0.5049999952316284}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7263000011444092},{"id":"https://openalex.org/C50817715","wikidata":"https://www.wikidata.org/wiki/Q79895177","display_name":"Regret","level":2,"score":0.6917999982833862},{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.6901000142097473},{"id":"https://openalex.org/C8642999","wikidata":"https://www.wikidata.org/wiki/Q4171168","display_name":"Hyperparameter","level":2,"score":0.608299970626831},{"id":"https://openalex.org/C81917197","wikidata":"https://www.wikidata.org/wiki/Q628760","display_name":"Selection (genetic algorithm)","level":2,"score":0.4909999966621399},{"id":"https://openalex.org/C2776261394","wikidata":"https://www.wikidata.org/wiki/Q315562","display_name":"Impossibility","level":2,"score":0.4611000120639801},{"id":"https://openalex.org/C55166926","wikidata":"https://www.wikidata.org/wiki/Q2892946","display_name":"Oracle","level":2,"score":0.45680001378059387},{"id":"https://openalex.org/C193969084","wikidata":"https://www.wikidata.org/wiki/Q7452500","display_name":"Sequential decoding","level":4,"score":0.45399999618530273},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.4456000030040741},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.43309998512268066},{"id":"https://openalex.org/C10485038","wikidata":"https://www.wikidata.org/wiki/Q48996162","display_name":"Hyperparameter optimization","level":3,"score":0.3946000039577484},{"id":"https://openalex.org/C77553402","wikidata":"https://www.wikidata.org/wiki/Q13222579","display_name":"Upper and lower bounds","level":2,"score":0.37470000982284546},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.36419999599456787},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.34279999136924744},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.33649998903274536},{"id":"https://openalex.org/C196921405","wikidata":"https://www.wikidata.org/wiki/Q786431","display_name":"Online algorithm","level":2,"score":0.3215999901294708},{"id":"https://openalex.org/C204397858","wikidata":"https://www.wikidata.org/wiki/Q4437907","display_name":"List decoding","level":5,"score":0.31209999322891235},{"id":"https://openalex.org/C157764524","wikidata":"https://www.wikidata.org/wiki/Q1383412","display_name":"Throughput","level":3,"score":0.2946999967098236},{"id":"https://openalex.org/C149629883","wikidata":"https://www.wikidata.org/wiki/Q660926","display_name":"Fraction (chemistry)","level":2,"score":0.2883000075817108},{"id":"https://openalex.org/C93959086","wikidata":"https://www.wikidata.org/wiki/Q6888345","display_name":"Model selection","level":2,"score":0.2777999937534332},{"id":"https://openalex.org/C159254197","wikidata":"https://www.wikidata.org/wiki/Q1144915","display_name":"Lexicographical order","level":2,"score":0.2759000062942505},{"id":"https://openalex.org/C176181172","wikidata":"https://www.wikidata.org/wiki/Q3490301","display_name":"Nondeterministic algorithm","level":2,"score":0.2669000029563904},{"id":"https://openalex.org/C5465570","wikidata":"https://www.wikidata.org/wiki/Q5326898","display_name":"Early stopping","level":3,"score":0.26269999146461487},{"id":"https://openalex.org/C90805587","wikidata":"https://www.wikidata.org/wiki/Q10944557","display_name":"Word (group theory)","level":2,"score":0.2621000111103058},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.26109999418258667},{"id":"https://openalex.org/C115537543","wikidata":"https://www.wikidata.org/wiki/Q165596","display_name":"Cache","level":2,"score":0.26010000705718994},{"id":"https://openalex.org/C141603448","wikidata":"https://www.wikidata.org/wiki/Q134830","display_name":"Prefix","level":2,"score":0.2572000026702881},{"id":"https://openalex.org/C160234255","wikidata":"https://www.wikidata.org/wiki/Q812535","display_name":"Bayesian inference","level":3,"score":0.2522999942302704},{"id":"https://openalex.org/C168773036","wikidata":"https://www.wikidata.org/wiki/Q264164","display_name":"Recursion (computer science)","level":2,"score":0.251800000667572},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.25130000710487366}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2505.15141","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2505.15141","pdf_url":"https://arxiv.org/pdf/2505.15141","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2505.15141","doi":"https://doi.org/10.48550/arxiv.2505.15141","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2505.15141","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2505.15141","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2505.15141","pdf_url":"https://arxiv.org/pdf/2505.15141","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[{"id":"https://openalex.org/F4320320671","display_name":"National Research Foundation","ror":"https://ror.org/05s0g1g46"},{"id":"https://openalex.org/F4320320709","display_name":"National Research Foundation Singapore","ror":"https://ror.org/03cpyc314"}],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4415328504.pdf","grobid_xml":"https://content.openalex.org/works/W4415328504.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4416245062","https://openalex.org/W7166476568","https://openalex.org/W7103976692","https://openalex.org/W7115569238","https://openalex.org/W4416070506","https://openalex.org/W4412945503","https://openalex.org/W4403346564","https://openalex.org/W7162255714","https://openalex.org/W4387596095","https://openalex.org/W7166842263"],"abstract_inverted_index":{"Speculative":[0],"decoding":[1,31,72,95],"has":[2],"emerged":[3],"as":[4,73,85],"a":[5,28,57,86,92,114],"popular":[6],"method":[7],"to":[8,48,62,151,169,177],"accelerate":[9],"the":[10,35,52,65,68,117,143,173,178],"inference":[11],"of":[12,34,67,113,146],"Large":[13],"Language":[14],"Models":[15],"(LLMs)":[16],"while":[17],"retaining":[18],"their":[19],"superior":[20],"text":[21,74],"generation":[22],"performance.":[23],"Previous":[24],"methods":[25],"either":[26],"adopt":[27],"fixed":[29],"speculative":[30,71,94],"configuration":[32,66],"regardless":[33],"prefix":[36],"tokens,":[37],"or":[38,45],"train":[39],"draft":[40],"models":[41],"in":[42,111,182],"an":[43,135],"offline":[44],"online":[46,59],"manner":[47],"align":[49],"them":[50],"with":[51,158,188],"context.":[53],"This":[54],"paper":[55],"proposes":[56],"training-free":[58],"learning":[60],"framework":[61,96],"adaptively":[63],"choose":[64],"hyperparameters":[69],"for":[70],"is":[75,140,148,175],"being":[76],"generated.":[77],"We":[78,121],"first":[79],"formulate":[80],"this":[81,124],"hyperparameter":[82,101,181],"selection":[83,102],"problem":[84,89],"Multi-Armed":[87],"Bandit":[88],"and":[90,105,109,129,160,172],"provide":[91],"general":[93],"BanditSpec.":[97],"Furthermore,":[98],"two":[99],"bandit-based":[100],"algorithms,":[103],"UCBSpec":[104,147],"EXP3Spec,":[106],"are":[107,166],"designed":[108],"analyzed":[110],"terms":[112],"novel":[115],"quantity,":[116],"stopping":[118],"time":[119],"regret.":[120],"upper":[122],"bound":[123],"regret":[125,144],"under":[126],"both":[127],"stochastic":[128],"adversarial":[130],"reward":[131],"settings.":[132],"By":[133],"deriving":[134],"information-theoretic":[136],"impossibility":[137],"result,":[138],"it":[139],"shown":[141],"that":[142,163],"performance":[145],"optimal":[149],"up":[150],"universal":[152],"constants.":[153],"Finally,":[154],"extensive":[155],"empirical":[156],"experiments":[157],"LLaMA3":[159],"Qwen2":[161],"demonstrate":[162],"our":[164],"algorithms":[165],"effective":[167],"compared":[168],"existing":[170],"methods,":[171],"throughput":[174],"close":[176],"oracle":[179],"best":[180],"simulated":[183],"real-life":[184],"LLM":[185],"serving":[186],"scenarios":[187],"diverse":[189],"input":[190],"prompts.":[191]},"counts_by_year":[],"updated_date":"2026-10-16T09:59:22.235173","created_date":"2025-10-19T00:00:00"}