{"id":"https://openalex.org/W7139918783","doi":"https://doi.org/10.48550/arxiv.2603.18411","title":"TARo: Token-level Adaptive Routing for LLM Test-time Alignment","display_name":"TARo: Token-level Adaptive Routing for LLM Test-time Alignment","publication_year":2026,"publication_date":"2026-03-19","ids":{"openalex":"https://openalex.org/W7139918783","doi":"https://doi.org/10.48550/arxiv.2603.18411"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2603.18411","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.18411","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2603.18411","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5037312803","display_name":"Arushi Rai","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Rai, Arushi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5130226322","display_name":"Qiang Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Qiang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5130238707","display_name":"Hanqing Zeng","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zeng, Hanqing","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5130246644","display_name":"Yunkai Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Yunkai","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5029382909","display_name":"Dipesh Tamboli","orcid":"https://orcid.org/0000-0003-2765-5763"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tamboli, Dipesh","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5130214198","display_name":"Xiangjun Fan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Fan, Xiangjun","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5130225952","display_name":"Zhuokai Zhao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhao, Zhuokai","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":null,"display_name":"Zhang, Lizhu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Lizhu","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/T13702","display_name":"Machine Learning in Healthcare","score":0.47040000557899475,"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/T13702","display_name":"Machine Learning in Healthcare","score":0.47040000557899475,"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/T11636","display_name":"Artificial Intelligence in Healthcare and Education","score":0.18649999797344208,"subfield":{"id":"https://openalex.org/subfields/2718","display_name":"Health Informatics"},"field":{"id":"https://openalex.org/fields/27","display_name":"Medicine"},"domain":{"id":"https://openalex.org/domains/4","display_name":"Health Sciences"}},{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.14920000731945038,"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/boosting","display_name":"Boosting (machine learning)","score":0.7232999801635742},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.5952000021934509},{"id":"https://openalex.org/keywords/consistency","display_name":"Consistency (knowledge bases)","score":0.5662000179290771},{"id":"https://openalex.org/keywords/router","display_name":"Router","score":0.49889999628067017},{"id":"https://openalex.org/keywords/routing","display_name":"Routing (electronic design automation)","score":0.4528000056743622},{"id":"https://openalex.org/keywords/base","display_name":"Base (topology)","score":0.39430001378059387},{"id":"https://openalex.org/keywords/bridge","display_name":"Bridge (graph theory)","score":0.3783000111579895}],"concepts":[{"id":"https://openalex.org/C46686674","wikidata":"https://www.wikidata.org/wiki/Q466303","display_name":"Boosting (machine learning)","level":2,"score":0.7232999801635742},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6984000205993652},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.5952000021934509},{"id":"https://openalex.org/C2776436953","wikidata":"https://www.wikidata.org/wiki/Q5163215","display_name":"Consistency (knowledge bases)","level":2,"score":0.5662000179290771},{"id":"https://openalex.org/C2775896111","wikidata":"https://www.wikidata.org/wiki/Q642560","display_name":"Router","level":2,"score":0.49889999628067017},{"id":"https://openalex.org/C74172769","wikidata":"https://www.wikidata.org/wiki/Q1446839","display_name":"Routing (electronic design automation)","level":2,"score":0.4528000056743622},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4494999945163727},{"id":"https://openalex.org/C42058472","wikidata":"https://www.wikidata.org/wiki/Q810214","display_name":"Base (topology)","level":2,"score":0.39430001378059387},{"id":"https://openalex.org/C100776233","wikidata":"https://www.wikidata.org/wiki/Q2532492","display_name":"Bridge (graph theory)","level":2,"score":0.3783000111579895},{"id":"https://openalex.org/C2781249084","wikidata":"https://www.wikidata.org/wiki/Q908656","display_name":"Preference","level":2,"score":0.37470000982284546},{"id":"https://openalex.org/C22367795","wikidata":"https://www.wikidata.org/wiki/Q7625208","display_name":"Structured prediction","level":2,"score":0.37439998984336853},{"id":"https://openalex.org/C4554734","wikidata":"https://www.wikidata.org/wiki/Q593744","display_name":"Knowledge base","level":2,"score":0.3617999851703644},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.32850000262260437},{"id":"https://openalex.org/C48044578","wikidata":"https://www.wikidata.org/wiki/Q727490","display_name":"Scalability","level":2,"score":0.32710000872612},{"id":"https://openalex.org/C24856439","wikidata":"https://www.wikidata.org/wiki/Q352483","display_name":"Adaptive routing","level":5,"score":0.31769999861717224},{"id":"https://openalex.org/C195344581","wikidata":"https://www.wikidata.org/wiki/Q2555318","display_name":"Automated reasoning","level":2,"score":0.298799991607666},{"id":"https://openalex.org/C81917197","wikidata":"https://www.wikidata.org/wiki/Q628760","display_name":"Selection (genetic algorithm)","level":2,"score":0.29840001463890076}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2603.18411","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.18411","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2603.18411","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.18411","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Large":[0],"language":[1],"models":[2,62],"(LLMs)":[3],"exhibit":[4],"strong":[5],"reasoning":[6,52,99,121],"capabilities":[7],"but":[8,25],"typically":[9],"require":[10],"expensive":[11],"post-training":[12],"to":[13,67,88,103,133,144],"reach":[14],"high":[15],"performance.":[16],"Recent":[17],"test-time":[18,113,139],"alignment":[19,32,114,140],"methods":[20],"offer":[21],"a":[22,75],"lightweight":[23],"alternative,":[24],"have":[26],"been":[27],"explored":[28],"mainly":[29],"for":[30],"preference":[31,142],"rather":[33],"than":[34],"reasoning.":[35,147],"To":[36],"bridge":[37],"this":[38],"gap,":[39],"we":[40,58],"propose,":[41],"Token-level":[42],"Adaptive":[43],"Routing":[44],"(TARo),":[45],"which":[46],"steers":[47],"frozen":[48],"LLMs":[49],"toward":[50],"structured":[51],"entirely":[53],"at":[54],"inference":[55],"time.":[56],"Specifically,":[57],"first":[59],"train":[60],"reward":[61,86],"on":[63],"step-wise":[64],"mathematical":[65],"traces":[66],"capture":[68],"fine-grained":[69],"logical":[70],"consistency":[71],"signals,":[72],"then":[73],"introduce":[74],"learnable":[76],"token-level":[77,112],"router":[78],"that":[79,95],"automatically":[80],"controls":[81],"the":[82,85,89],"guidance":[83],"of":[84],"model":[87,107],"base":[90,106],"model.":[91],"Extensive":[92],"experiments":[93],"show":[94],"TARo":[96,128],"significantly":[97],"improves":[98],"performance":[100],"by":[101],"up":[102],"+22.4%":[104],"over":[105,110],"and":[108,123],"+8.4%":[109],"existing":[111],"methods,":[115],"while":[116],"also":[117,129],"boosting":[118],"out-of-distribution":[119],"clinical":[120],"(MedXpertQA)":[122],"instruction":[124],"following":[125],"(AlpacaEval).":[126],"Furthermore,":[127],"generalizes":[130],"from":[131,141],"small":[132],"large":[134],"backbones":[135],"without":[136],"retraining,":[137],"extending":[138],"optimization":[143],"robust,":[145],"cross-domain":[146]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-03-21T00:00:00"}