{"id":"https://openalex.org/W4416034735","doi":"https://doi.org/10.18653/v1/2025.findings-emnlp.551","title":"Accelerating LLM Reasoning via Early Rejection with Partial Reward Modeling","display_name":"Accelerating LLM Reasoning via Early Rejection with Partial Reward Modeling","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4416034735","doi":"https://doi.org/10.18653/v1/2025.findings-emnlp.551"},"language":null,"primary_location":{"id":"doi:10.18653/v1/2025.findings-emnlp.551","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-emnlp.551","pdf_url":"https://aclanthology.org/2025.findings-emnlp.551.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: EMNLP 2025","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2025.findings-emnlp.551.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5120308670","display_name":"Seyyed Saeid Cheshmi","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Seyyed Saeid Cheshmi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5063807294","display_name":"Azal Ahmad Khan","orcid":"https://orcid.org/0009-0000-9435-5328"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Azal Ahmad Khan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100458479","display_name":"Xinran Wang","orcid":"https://orcid.org/0000-0002-3975-1667"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xinran Wang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Zirui Liu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zirui Liu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5120308671","display_name":"Ali Anwar","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ali Anwar","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":"10433","last_page":"10447"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11010","display_name":"Logic, Reasoning, and Knowledge","score":0.40560001134872437,"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/T11010","display_name":"Logic, Reasoning, and Knowledge","score":0.40560001134872437,"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/T10142","display_name":"Formal Methods in Verification","score":0.10490000247955322,"subfield":{"id":"https://openalex.org/subfields/1703","display_name":"Computational Theory and Mathematics"},"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/T11303","display_name":"Bayesian Modeling and Causal Inference","score":0.05530000105500221,"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/constraint","display_name":"Constraint (computer-aided design)","score":0.24860000610351562},{"id":"https://openalex.org/keywords/key","display_name":"Key (lock)","score":0.24199999868869781},{"id":"https://openalex.org/keywords/action","display_name":"Action (physics)","score":0.2371000051498413},{"id":"https://openalex.org/keywords/interpretability","display_name":"Interpretability","score":0.22840000689029694},{"id":"https://openalex.org/keywords/feature","display_name":"Feature (linguistics)","score":0.22660000622272491}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5521000027656555},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.45590001344680786},{"id":"https://openalex.org/C2776036281","wikidata":"https://www.wikidata.org/wiki/Q48769818","display_name":"Constraint (computer-aided design)","level":2,"score":0.24860000610351562},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.24199999868869781},{"id":"https://openalex.org/C2780791683","wikidata":"https://www.wikidata.org/wiki/Q846785","display_name":"Action (physics)","level":2,"score":0.2371000051498413},{"id":"https://openalex.org/C2781067378","wikidata":"https://www.wikidata.org/wiki/Q17027399","display_name":"Interpretability","level":2,"score":0.22840000689029694},{"id":"https://openalex.org/C2776401178","wikidata":"https://www.wikidata.org/wiki/Q12050496","display_name":"Feature (linguistics)","level":2,"score":0.22660000622272491},{"id":"https://openalex.org/C98045186","wikidata":"https://www.wikidata.org/wiki/Q205663","display_name":"Process (computing)","level":2,"score":0.226500004529953},{"id":"https://openalex.org/C112972136","wikidata":"https://www.wikidata.org/wiki/Q7595718","display_name":"Stability (learning theory)","level":2,"score":0.22220000624656677},{"id":"https://openalex.org/C165064840","wikidata":"https://www.wikidata.org/wiki/Q1321061","display_name":"Matching (statistics)","level":2,"score":0.21960000693798065}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2025.findings-emnlp.551","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-emnlp.551","pdf_url":"https://aclanthology.org/2025.findings-emnlp.551.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: EMNLP 2025","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2025.findings-emnlp.551","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-emnlp.551","pdf_url":"https://aclanthology.org/2025.findings-emnlp.551.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: EMNLP 2025","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[],"funders":[{"id":"https://openalex.org/F4320332195","display_name":"Samsung","ror":"https://ror.org/04w3jy968"}],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4416034735.pdf","grobid_xml":"https://content.openalex.org/works/W4416034735.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Large":[0],"Language":[1],"Models":[2,41],"(LLMs)":[3],"are":[4,100,116,204],"increasingly":[5],"relied":[6],"upon":[7],"for":[8,123,193],"solving":[9],"complex":[10],"reasoning":[11,30,47,114,168,198],"tasks":[12],"in":[13,65,177,199],"domains":[14],"such":[15],"as":[16],"mathematics,":[17],"logic,":[18],"and":[19,151,160,202],"multi-step":[20],"question":[21],"answering.A":[22],"growing":[23],"line":[24],"of":[25,63,85,91,118,142,197],"work":[26],"seeks":[27],"to":[28,44,77,111,174],"improve":[29],"quality":[31],"by":[32,137,153],"scaling":[33],"inference":[34,178],"time":[35],"compute":[36],"particularly":[37],"through":[38],"Process":[39],"Reward":[40,103],"(PRMs),":[42],"used":[43,75],"reward":[45,165],"the":[46,83,96,107,140,195],"at":[48,206],"intermediate":[49,129],"steps.While":[50],"effective,":[51],"these":[52],"methods":[53],"introduce":[54,95],"substantial":[55],"computational":[56],"overhead,":[57],"especially":[58],"when":[59],"generating":[60],"large":[61],"numbers":[62],"solutions":[64],"parallel.In":[66],"this":[67,133],"paper,":[68],"we":[69],"investigate":[70],"whether":[71],"PRMs":[72,99],"can":[73],"be":[74],"midgeneration":[76],"provide":[78],"early":[79,125,187],"signals":[80],"that":[81,98,106,139,186],"enable":[82],"rejection":[84,126,188],"suboptimal":[86],"candidates":[87],"before":[88],"full":[89],"generation":[90,149],"step":[92,115],"is":[93,189],"complete.We":[94],"hypothesis":[97,134],"also":[101],"Partial":[102],"Models,":[104],"meaning":[105],"scores":[108],"they":[109],"assign":[110],"partially":[112],"completed":[113],"predictive":[117],"final":[119,161,182],"output":[120],"quality.This":[121],"allows":[122],"principled":[124],"based":[127],"on":[128],"token-level":[130],"signals.We":[131],"support":[132],"both":[135],"theoretically,":[136],"proving":[138],"risk":[141],"discarding":[143],"optimal":[144],"beams":[145],"decreases":[146],"exponentially":[147],"with":[148],"length":[150],"empirically,":[152],"demonstrating":[154],"a":[155,190],"strong":[156],"correlation":[157],"between":[158],"partial":[159],"rewards":[162],"across":[163],"multiple":[164],"models.On":[166],"math":[167],"benchmarks,":[169],"our":[170],"method":[171],"achieves":[172],"up":[173],"1.4-9":[175],"reduction":[176],"FLOPs":[179],"without":[180],"degrading":[181],"performance.These":[183],"results":[184],"suggest":[185],"powerful":[191],"mechanism":[192],"improving":[194],"compute-efficiency":[196],"LLMs.The":[200],"code":[201],"implementation":[203],"available":[205],"https://github.com/scheshmi/":[207],"accelerated-reasoning-ER-PRM.":[208]},"counts_by_year":[],"updated_date":"2026-07-14T23:27:15.235271","created_date":"2025-11-08T00:00:00"}