{"id":"https://openalex.org/W4394907048","doi":"https://doi.org/10.48550/arxiv.2404.10136","title":"Language Model Cascades: Token-level uncertainty and beyond","display_name":"Language Model Cascades: Token-level uncertainty and beyond","publication_year":2024,"publication_date":"2024-04-15","ids":{"openalex":"https://openalex.org/W4394907048","doi":"https://doi.org/10.48550/arxiv.2404.10136"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2404.10136","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2404.10136","pdf_url":"https://arxiv.org/pdf/2404.10136","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","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2404.10136","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":null,"display_name":"Gupta, Neha","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I1291425158","display_name":"Google (United States)","ror":"https://ror.org/00njsd438","country_code":"US","type":"company","lineage":["https://openalex.org/I1291425158","https://openalex.org/I4210128969"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Gupta, Neha","raw_affiliation_strings":["Google Research , New York"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Google Research , New York","institution_ids":["https://openalex.org/I1291425158"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5077482746","display_name":"Harikrishna Narasimhan","orcid":"https://orcid.org/0009-0007-0476-6635","observed_orcids":["https://orcid.org/0009-0007-0476-6635"]},"institutions":[{"id":"https://openalex.org/I1291425158","display_name":"Google (United States)","ror":"https://ror.org/00njsd438","country_code":"US","type":"company","lineage":["https://openalex.org/I1291425158","https://openalex.org/I4210128969"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Narasimhan, Harikrishna","raw_affiliation_strings":["Google Research , New York"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Google Research , New York","institution_ids":["https://openalex.org/I1291425158"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5072755496","display_name":"Wittawat Jitkrittum","orcid":"https://orcid.org/0000-0002-9400-9262","observed_orcids":["https://orcid.org/0000-0002-9400-9262"]},"institutions":[{"id":"https://openalex.org/I1291425158","display_name":"Google (United States)","ror":"https://ror.org/00njsd438","country_code":"US","type":"company","lineage":["https://openalex.org/I1291425158","https://openalex.org/I4210128969"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Jitkrittum, Wittawat","raw_affiliation_strings":["Google Research , New York"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Google Research , New York","institution_ids":["https://openalex.org/I1291425158"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5079621556","display_name":"Ankit Singh Rawat","orcid":"https://orcid.org/0000-0001-9790-6500","observed_orcids":["https://orcid.org/0000-0001-9790-6500"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Rawat, Ankit Singh","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5049656925","display_name":"Aditya Krishna Menon","orcid":"https://orcid.org/0000-0002-1352-7106","observed_orcids":["https://orcid.org/0000-0002-1352-7106"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Menon, Aditya Krishna","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5018301052","display_name":"Sanjiv Kumar","orcid":"https://orcid.org/0000-0002-4080-1414","observed_orcids":["https://orcid.org/0000-0002-4080-1414"]},"institutions":[{"id":"https://openalex.org/I1291425158","display_name":"Google (United States)","ror":"https://ror.org/00njsd438","country_code":"US","type":"company","lineage":["https://openalex.org/I1291425158","https://openalex.org/I4210128969"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Kumar, Sanjiv","raw_affiliation_strings":["Google Research , New York"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Google Research , New York","institution_ids":["https://openalex.org/I1291425158"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":68,"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/T10028","display_name":"Topic Modeling","score":0.6304000020027161,"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/T10028","display_name":"Topic Modeling","score":0.6304000020027161,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.328900009393692,"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":0.01810000091791153,"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/length-bias","display_name":"length bias","score":0.8939999938011169},{"id":"https://openalex.org/keywords/cost-quality-trade-off","display_name":"cost-quality trade-off","score":0.9309999942779541},{"id":"https://openalex.org/keywords/generative-language-models","display_name":"generative language models","score":0.7440000176429749},{"id":"https://openalex.org/keywords/flan-t5","display_name":"Flan-T5","score":0.7739999890327454},{"id":"https://openalex.org/keywords/uncertainty-aggregation","display_name":"uncertainty aggregation","score":0.6209999918937683}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6122575998306274},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.6113085746765137},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.44855552911758423},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.40746259689331055},{"id":"https://openalex.org/C199360897","wikidata":"https://www.wikidata.org/wiki/Q9143","display_name":"Programming language","level":1,"score":0.32686182856559753},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.32363080978393555},{"id":"https://openalex.org/C31258907","wikidata":"https://www.wikidata.org/wiki/Q1301371","display_name":"Computer network","level":1,"score":0.14403840899467468}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2404.10136","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2404.10136","pdf_url":"https://arxiv.org/pdf/2404.10136","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.2404.10136","doi":"https://doi.org/10.48550/arxiv.2404.10136","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2404.10136","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:2404.10136","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2404.10136","pdf_url":"https://arxiv.org/pdf/2404.10136","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":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4394907048.pdf","grobid_xml":"https://content.openalex.org/works/W4394907048.grobid-xml"},"referenced_works_count":31,"referenced_works":["https://openalex.org/W2118377301","https://openalex.org/W2164598857","https://openalex.org/W2473344385","https://openalex.org/W2612431505","https://openalex.org/W2735574368","https://openalex.org/W2885185669","https://openalex.org/W2889317799","https://openalex.org/W2890152612","https://openalex.org/W2963096510","https://openalex.org/W2963341956","https://openalex.org/W2963748441","https://openalex.org/W2971302374","https://openalex.org/W3000779003","https://openalex.org/W3016254928","https://openalex.org/W3157005959","https://openalex.org/W3194676777","https://openalex.org/W3205308903","https://openalex.org/W4224075666","https://openalex.org/W4281552875","https://openalex.org/W4285139078","https://openalex.org/W4285225959","https://openalex.org/W4307079201","https://openalex.org/W4320858367","https://openalex.org/W4321472057","https://openalex.org/W4385573392","https://openalex.org/W4386150385","https://openalex.org/W4387163503","https://openalex.org/W4387271845","https://openalex.org/W4388685627","https://openalex.org/W4388748173","https://openalex.org/W4402671533"],"related_works":["https://openalex.org/W4399354026","https://openalex.org/W7160877418","https://openalex.org/W7171664955","https://openalex.org/W4416037319","https://openalex.org/W7160629695","https://openalex.org/W4406548552","https://openalex.org/W7167628135","https://openalex.org/W7166589956","https://openalex.org/W4402670143","https://openalex.org/W4399198692"],"abstract_inverted_index":{"Recent":[0],"advances":[1],"in":[2,11,183,255],"language":[3,228],"models":[4],"(LMs)":[5],"have":[6],"led":[7],"to":[8,30,53,114,175,194],"significant":[9],"improvements":[10],"quality":[12],"on":[13,71,140,223],"complex":[14],"NLP":[15],"tasks,":[16,117],"but":[17],"at":[18],"the":[19,54,58,107,119,130,160,177,240,247,256],"expense":[20],"of":[21,97,110,150,162,198,226,246],"increased":[22],"inference":[23],"costs.":[24],"Cascading":[25],"offers":[26],"a":[27,37,47,80,94,148,195,224],"simple":[28,196,218],"strategy":[29],"achieve":[31],"more":[32],"favorable":[33],"cost-quality":[34,258],"tradeoffs:":[35],"here,":[36],"small":[38],"model":[39,242,249],"is":[40,83,144,165],"invoked":[41],"for":[42,64,85,100,154],"most":[43],"\"easy\"":[44],"instances,":[45],"while":[46],"few":[48],"\"hard\"":[49],"instances":[50],"are":[51,62],"deferred":[52],"large":[55],"model.":[56],"While":[57],"principles":[59],"underpinning":[60],"cascading":[61],"well-studied":[63],"classification":[65],"tasks":[66],"-":[67,79],"with":[68,230],"deferral":[69,98,212],"based":[70,139],"predicted":[72,111,120,190],"class":[73,112],"uncertainty":[74,113,151,180,192,208],"favored":[75],"theoretically":[76],"and":[77,158,243],"practically":[78],"similar":[81],"understanding":[82],"lacking":[84],"generative":[86,115,184],"LM":[87,101,116],"tasks.":[88],"In":[89],"this":[90,126,171],"work,":[91],"we":[92,173,203],"initiate":[93],"systematic":[95],"study":[96],"rules":[99,213],"cascades.":[102],"We":[103,123,186,233],"begin":[104],"by":[105],"examining":[106],"natural":[108,227],"extension":[109],"namely,":[118],"sequence":[121,149,191],"uncertainty.":[122],"show":[124,204,235],"that":[125,188,205,236],"measure":[127],"suffers":[128],"from":[129,239],"length":[131],"bias":[132],"problem,":[133],"either":[134],"over-":[135],"or":[136],"under-emphasizing":[137],"outputs":[138],"their":[141],"lengths.":[142],"This":[143],"because":[145],"LMs":[146],"produce":[147],"values,":[152],"one":[153],"each":[155],"output":[156,163],"token;":[157],"moreover,":[159],"number":[161],"tokens":[164],"variable":[166],"across":[167],"examples.":[168],"To":[169],"mitigate":[170],"issue,":[172],"propose":[174],"exploit":[176],"richer":[178],"token-level":[179,207],"information":[181],"implicit":[182],"LMs.":[185],"argue":[187],"naive":[189],"corresponds":[193],"aggregation":[197,219],"these":[199],"uncertainties.":[200],"By":[201],"contrast,":[202],"incorporating":[206,237],"through":[209],"learned":[210],"post-hoc":[211],"can":[214,250],"significantly":[215],"outperform":[216],"such":[217],"strategies,":[220],"via":[221],"experiments":[222],"range":[225],"benchmarks":[229],"FLAN-T5":[231],"models.":[232],"further":[234],"embeddings":[238],"smaller":[241],"intermediate":[244],"layers":[245],"larger":[248],"give":[251],"an":[252],"additional":[253],"boost":[254],"overall":[257],"tradeoff.":[259]},"counts_by_year":[{"year":2026,"cited_by_count":31},{"year":2025,"cited_by_count":23},{"year":2024,"cited_by_count":14}],"updated_date":"2026-11-10T11:39:43.021295","created_date":"2024-04-18T00:00:00"}