{"id":"https://openalex.org/W4417536763","doi":"https://doi.org/10.48550/arxiv.2506.15733","title":"$\\texttt{SPECS}$: Faster Test-Time Scaling through Speculative Drafts","display_name":"$\\texttt{SPECS}$: Faster Test-Time Scaling through Speculative Drafts","publication_year":2025,"publication_date":"2025-06-15","ids":{"openalex":"https://openalex.org/W4417536763","doi":"https://doi.org/10.48550/arxiv.2506.15733"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2506.15733","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2506.15733","pdf_url":"https://arxiv.org/pdf/2506.15733","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"},"volume":null,"type":"preprint","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2506.15733","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":null,"display_name":"Cemri, Mert","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Cemri, Mert","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Rajaraman, Nived","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Rajaraman, Nived","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Tiwari, Rishabh","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tiwari, Rishabh","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Liu, Xiaoxuan","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liu, Xiaoxuan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Keutzer, Kurt","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Keutzer, Kurt","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Stoica, Ion","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Stoica, Ion","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Ramchandran, Kannan","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ramchandran, Kannan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5008645615","display_name":"Ahmad Beirami","orcid":"https://orcid.org/0000-0002-1998-5271","observed_orcids":["https://orcid.org/0000-0002-1998-5271"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Beirami, Ahmad","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":null,"display_name":"Sun, Ziteng","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Sun, Ziteng","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":false,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.6288999915122986,"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":0.6288999915122986,"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":0.2827000021934509,"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/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.005400000140070915,"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/test-time-scaling","display_name":"test-time scaling","score":0.9390000104904175},{"id":"https://openalex.org/keywords/speculative-decoding","display_name":"speculative decoding","score":0.7519999742507935},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.8970000147819519},{"id":"https://openalex.org/keywords/reward-models","display_name":"reward models","score":0.4390000104904175},{"id":"https://openalex.org/keywords/beam-search","display_name":"beam search","score":0.7490000128746033},{"id":"https://openalex.org/keywords/mathematical-reasoning","display_name":"mathematical reasoning","score":0.5350000262260437}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7858999967575073},{"id":"https://openalex.org/C99844830","wikidata":"https://www.wikidata.org/wiki/Q102441924","display_name":"Scaling","level":2,"score":0.7114999890327454},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.6898999810218811},{"id":"https://openalex.org/C82876162","wikidata":"https://www.wikidata.org/wiki/Q17096504","display_name":"Latency (audio)","level":2,"score":0.5148000121116638},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.4066999852657318},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.38850000500679016},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.3569999933242798},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.325300008058548},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.31619998812675476},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.2994999885559082},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.2937999963760376},{"id":"https://openalex.org/C46637626","wikidata":"https://www.wikidata.org/wiki/Q6693015","display_name":"Low latency (capital markets)","level":2,"score":0.2809999883174896},{"id":"https://openalex.org/C8505890","wikidata":"https://www.wikidata.org/wiki/Q605095","display_name":"Budget constraint","level":2,"score":0.26409998536109924},{"id":"https://openalex.org/C3770464","wikidata":"https://www.wikidata.org/wiki/Q775963","display_name":"Smoothing","level":2,"score":0.26330000162124634},{"id":"https://openalex.org/C120936955","wikidata":"https://www.wikidata.org/wiki/Q2155640","display_name":"Empirical research","level":2,"score":0.2574000060558319}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2506.15733","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2506.15733","pdf_url":"https://arxiv.org/pdf/2506.15733","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.2506.15733","doi":"https://doi.org/10.48550/arxiv.2506.15733","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2506.15733","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":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2506.15733","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2506.15733","pdf_url":"https://arxiv.org/pdf/2506.15733","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"},"sustainable_development_goals":[],"sustainable_development_goals_aurora":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Scaling":[0],"test-time":[1,43,69],"compute":[2,28,53],"has":[3],"driven":[4],"the":[5,9,32,149],"recent":[6],"advances":[7],"in":[8],"reasoning":[10],"capabilities":[11],"of":[12,34,151],"large":[13],"language":[14],"models":[15],"(LLMs),":[16],"typically":[17],"by":[18,73,136],"allocating":[19],"additional":[20],"computation":[21],"for":[22,48],"more":[23],"thorough":[24],"exploration.":[25],"However,":[26],"increased":[27],"often":[29,56],"comes":[30],"at":[31],"expense":[33],"higher":[35],"user-facing":[36],"latency,":[37],"directly":[38],"impacting":[39],"user":[40],"experience.":[41],"Current":[42],"scaling":[44,70],"methods":[45],"primarily":[46],"optimize":[47],"accuracy":[49,132],"based":[50],"on":[51,119],"total":[52],"resources":[54],"(FLOPS),":[55],"overlooking":[57],"latency":[58,135],"constraints.":[59],"To":[60],"address":[61],"this":[62],"gap,":[63],"we":[64],"propose":[65],"$\\texttt{SPECS}$,":[66],"a":[67,77,94,99,113,152],"latency-aware":[68],"method":[71],"inspired":[72],"speculative":[74],"decoding.":[75],"$\\texttt{SPECS}$~uses":[76],"smaller,":[78],"faster":[79],"model":[80,97],"to":[81,138,148],"generate":[82],"candidate":[83],"sequences":[84],"efficiently,":[85],"and":[86,98,112,122],"evaluates":[87],"these":[88],"candidates":[89],"using":[90],"signals":[91],"from":[92],"both":[93],"larger":[95],"target":[96],"dedicated":[100],"reward":[101],"model.":[102],"We":[103],"introduce":[104],"new":[105],"integration":[106],"strategies,":[107],"including":[108],"reward-guided":[109],"soft":[110],"verification":[111],"reward-based":[114],"deferral":[115],"mechanism.":[116],"Empirical":[117],"results":[118],"MATH500,":[120],"AMC23":[121],"OlympiadBench":[123],"datasets":[124],"show":[125],"that":[126,144],"$\\texttt{SPECS}$~matches":[127],"or":[128],"surpasses":[129],"beam":[130,159],"search":[131],"while":[133],"reducing":[134],"up":[137],"$\\sim$19.1\\%.":[139],"Our":[140],"theoretical":[141],"analysis":[142],"shows":[143],"our":[145],"algorithm":[146],"converges":[147],"solution":[150],"KL-regularized":[153],"reinforcement":[154],"learning":[155],"objective":[156],"with":[157],"increasing":[158],"width.":[160]},"counts_by_year":[],"updated_date":"2026-10-14T16:39:01.820549","created_date":"2025-10-10T00:00:00"}