{"id":"https://openalex.org/W4417517258","doi":"https://doi.org/10.52202/085713-1608","title":"S-GRPO: Early Exit via Reinforcement Learning in Reasoning Models","display_name":"S-GRPO: Early Exit via Reinforcement Learning in Reasoning Models","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4417517258","doi":"https://doi.org/10.52202/085713-1608"},"language":"en","primary_location":{"id":"doi:10.52202/085713-1608","doi":"https://doi.org/10.52202/085713-1608","is_oa":false,"landing_page_url":"https://doi.org/10.52202/085713-1608","pdf_url":null,"source":{"id":"https://openalex.org/S4363606243","display_name":"neural information processing systems","issn_l":"1049-5258","issn":["1049-5258"],"is_oa":false,"is_in_doaj":false,"is_core":true,"listed_in":["cwts-core","norway-2"],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Advances in Neural Information Processing Systems 38","raw_type":"proceedings-article"},"volume":null,"type":"conference-paper","is_traditional":true,"indexed_in":["arxiv","crossref","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2505.07686","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5073609014","display_name":"Muzhi Dai","orcid":"https://orcid.org/0009-0004-6070-8470","observed_orcids":["https://orcid.org/0009-0004-6070-8470"]},"institutions":[{"id":"https://openalex.org/I99065089","display_name":"Tsinghua University","ror":"https://ror.org/03cve4549","country_code":"CN","type":"education","lineage":["https://openalex.org/I99065089"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Muzhi Dai","raw_affiliation_strings":["Tsinghua University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Tsinghua University","institution_ids":["https://openalex.org/I99065089"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5034004918","display_name":"Chenxu Yang","orcid":"https://orcid.org/0000-0003-2523-7090","observed_orcids":["https://orcid.org/0000-0003-2523-7090"]},"institutions":[{"id":"https://openalex.org/I19820366","display_name":"Chinese Academy of Sciences","ror":"https://ror.org/034t30j35","country_code":"CN","type":"government","lineage":["https://openalex.org/I19820366"]},{"id":"https://openalex.org/I4210156404","display_name":"Institute of Information Engineering","ror":"https://ror.org/04r53se39","country_code":"CN","type":"facility","lineage":["https://openalex.org/I19820366","https://openalex.org/I4210156404"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Chenxu Yang","raw_affiliation_strings":["Institute of Information Enginering, Chinese Academy of Sciences"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Institute of Information Enginering, Chinese Academy of Sciences","institution_ids":["https://openalex.org/I19820366","https://openalex.org/I4210156404"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5013535642","display_name":"Qingyi Si","orcid":"https://orcid.org/0000-0001-8433-0215","observed_orcids":["https://orcid.org/0000-0001-8433-0215"]},"institutions":[{"id":"https://openalex.org/I4210103986","display_name":"Jingdong (China)","ror":"https://ror.org/01dkjkq64","country_code":"CN","type":"company","lineage":["https://openalex.org/I4210103986"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Qingyi Si","raw_affiliation_strings":["JD.com"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"JD.com","institution_ids":["https://openalex.org/I4210103986"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":4,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":29.9606,"has_fulltext":true,"cited_by_count":85,"citation_normalized_percentile":{"value":0.99718051,"is_in_top_1_percent":true,"is_in_top_10_percent":true},"cited_by_percentile_year":{"min":88,"max":100},"biblio":{"volume":null,"issue":null,"first_page":"53735","last_page":"53761"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.5981000065803528,"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":0.5981000065803528,"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":0.0820000022649765,"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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.03229999914765358,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"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/early-exit","display_name":"early exit","score":0.7799999713897705},{"id":"https://openalex.org/keywords/chain-of-thought-generation","display_name":"chain-of-thought generation","score":0.9649999737739563},{"id":"https://openalex.org/keywords/reasoning-models","display_name":"reasoning models","score":0.8339999914169312},{"id":"https://openalex.org/keywords/reinforcement-learning","display_name":"reinforcement learning","score":0.4690000116825104},{"id":"https://openalex.org/keywords/test-time-scaling","display_name":"test-time scaling","score":0.6309999823570251},{"id":"https://openalex.org/keywords/sequence-length-reduction","display_name":"sequence length reduction","score":0.6990000009536743},{"id":"https://openalex.org/keywords/qwen3","display_name":"Qwen3","score":0.6150000095367432},{"id":"https://openalex.org/keywords/deepseek","display_name":"DeepSeek","score":0.550000011920929},{"id":"https://openalex.org/keywords/mathematical-reasoning","display_name":"mathematical reasoning","score":0.48899999260902405}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6552000045776367},{"id":"https://openalex.org/C2777735758","wikidata":"https://www.wikidata.org/wiki/Q817765","display_name":"Path (computing)","level":2,"score":0.5752000212669373},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.5536999702453613},{"id":"https://openalex.org/C152124472","wikidata":"https://www.wikidata.org/wiki/Q1204361","display_name":"Redundancy (engineering)","level":2,"score":0.5454999804496765},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5220999717712402},{"id":"https://openalex.org/C89288958","wikidata":"https://www.wikidata.org/wiki/Q7301504","display_name":"Reasoning system","level":2,"score":0.453900009393692},{"id":"https://openalex.org/C192209626","wikidata":"https://www.wikidata.org/wiki/Q190909","display_name":"Focus (optics)","level":2,"score":0.4507000148296356},{"id":"https://openalex.org/C86827895","wikidata":"https://www.wikidata.org/wiki/Q7098582","display_name":"Opportunistic reasoning","level":4,"score":0.37439998984336853},{"id":"https://openalex.org/C2778112365","wikidata":"https://www.wikidata.org/wiki/Q3511065","display_name":"Sequence (biology)","level":2,"score":0.36239999532699585},{"id":"https://openalex.org/C103057564","wikidata":"https://www.wikidata.org/wiki/Q4751139","display_name":"Analytic reasoning","level":3,"score":0.3447999954223633},{"id":"https://openalex.org/C37335422","wikidata":"https://www.wikidata.org/wiki/Q6888134","display_name":"Model-based reasoning","level":3,"score":0.33239999413490295},{"id":"https://openalex.org/C107848011","wikidata":"https://www.wikidata.org/wiki/Q4680756","display_name":"Adaptive reasoning","level":4,"score":0.3255999982357025},{"id":"https://openalex.org/C99844830","wikidata":"https://www.wikidata.org/wiki/Q102441924","display_name":"Scaling","level":2,"score":0.32519999146461487},{"id":"https://openalex.org/C83725634","wikidata":"https://www.wikidata.org/wiki/Q7268699","display_name":"Qualitative reasoning","level":2,"score":0.32429999113082886},{"id":"https://openalex.org/C159032336","wikidata":"https://www.wikidata.org/wiki/Q2488768","display_name":"Non-monotonic logic","level":2,"score":0.3028999865055084},{"id":"https://openalex.org/C36964233","wikidata":"https://www.wikidata.org/wiki/Q7920942","display_name":"Verbal reasoning","level":3,"score":0.2849000096321106},{"id":"https://openalex.org/C195344581","wikidata":"https://www.wikidata.org/wiki/Q2555318","display_name":"Automated reasoning","level":2,"score":0.28299999237060547},{"id":"https://openalex.org/C188147891","wikidata":"https://www.wikidata.org/wiki/Q147638","display_name":"Cognitive science","level":1,"score":0.2824000120162964},{"id":"https://openalex.org/C183521366","wikidata":"https://www.wikidata.org/wiki/Q7256422","display_name":"Psychology of reasoning","level":4,"score":0.28220000863075256},{"id":"https://openalex.org/C111335779","wikidata":"https://www.wikidata.org/wiki/Q3454686","display_name":"Reduction (mathematics)","level":2,"score":0.27469998598098755},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.2727999985218048},{"id":"https://openalex.org/C40506919","wikidata":"https://www.wikidata.org/wiki/Q7452469","display_name":"Sequence learning","level":2,"score":0.271699994802475},{"id":"https://openalex.org/C166052673","wikidata":"https://www.wikidata.org/wiki/Q83021","display_name":"Empirical evidence","level":2,"score":0.2506999969482422},{"id":"https://openalex.org/C20162079","wikidata":"https://www.wikidata.org/wiki/Q1151406","display_name":"Case-based reasoning","level":2,"score":0.25049999356269836}],"mesh":[],"locations_count":3,"locations":[{"id":"doi:10.52202/085713-1608","doi":"https://doi.org/10.52202/085713-1608","is_oa":false,"landing_page_url":"https://doi.org/10.52202/085713-1608","pdf_url":null,"source":{"id":"https://openalex.org/S4363606243","display_name":"neural information processing systems","issn_l":"1049-5258","issn":["1049-5258"],"is_oa":false,"is_in_doaj":false,"is_core":true,"listed_in":["cwts-core","norway-2"],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Advances in Neural Information Processing Systems 38","raw_type":"proceedings-article"},{"id":"pmh:oai:arXiv.org:2505.07686","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2505.07686","pdf_url":"https://arxiv.org/pdf/2505.07686","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.2505.07686","doi":"https://doi.org/10.48550/arxiv.2505.07686","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2505.07686","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:2505.07686","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2505.07686","pdf_url":"https://arxiv.org/pdf/2505.07686","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/W4417517258.pdf","grobid_xml":"https://content.openalex.org/works/W4417517258.grobid-xml"},"referenced_works_count":28,"referenced_works":["https://openalex.org/W2736601468","https://openalex.org/W2964121744","https://openalex.org/W4223908421","https://openalex.org/W4389520310","https://openalex.org/W4391124001","https://openalex.org/W4400518576","https://openalex.org/W4403884270","https://openalex.org/W4406772739","https://openalex.org/W4406774573","https://openalex.org/W4407091778","https://openalex.org/W4407231747","https://openalex.org/W4407308832","https://openalex.org/W4407424296","https://openalex.org/W4407570327","https://openalex.org/W4407764148","https://openalex.org/W4412944599","https://openalex.org/W4412944632","https://openalex.org/W4412945706","https://openalex.org/W4415080454","https://openalex.org/W4415158403","https://openalex.org/W4415797828","https://openalex.org/W4416036941","https://openalex.org/W4416113410","https://openalex.org/W4416365805","https://openalex.org/W4417221683","https://openalex.org/W6910671462","https://openalex.org/W6967130704","https://openalex.org/W7196953391"],"related_works":["https://openalex.org/W4414633696","https://openalex.org/W7166827583","https://openalex.org/W4415036968","https://openalex.org/W7166825786","https://openalex.org/W7134844655","https://openalex.org/W7083815555","https://openalex.org/W7214056961","https://openalex.org/W7129768706","https://openalex.org/W7131126100","https://openalex.org/W7137986045"],"abstract_inverted_index":{"As":[0],"Test-Time":[1],"Scaling":[2],"emerges":[3],"as":[4,204],"an":[5],"active":[6],"research":[7],"focus":[8],"in":[9,49,103,114,219],"the":[10,57,68,93,131,154,158,168],"large":[11],"language":[12],"model":[13,169],"community,":[14],"advanced":[15],"post-training":[16],"methods":[17],"increasingly":[18],"emphasize":[19],"extending":[20],"chain-of-thought":[21],"(CoT)":[22],"generation":[23],"length,":[24],"thereby":[25,99],"enhancing":[26],"reasoning":[27,33,40,72,97,112,122,159,194],"capabilities":[28],"to":[29,90,133,163,170],"approach":[30],"Deepseek":[31],"R1-like":[32],"models.":[34],"However,":[35],"recent":[36],"studies":[37],"reveal":[38],"that":[39,87,188],"models":[41,89],"(even":[42],"Qwen3)":[43],"consistently":[44],"exhibit":[45],"excessive":[46],"thought":[47],"redundancy":[48],"CoT":[50,104],"generation.":[51,105],"This":[52,74,165],"overthinking":[53],"issue":[54],"arises":[55],"from":[56,130,161],"inherent":[58],"limitations":[59],"of":[60,70,95],"conventional":[61],"outcome-reward":[62],"reinforcement":[63,84],"learning,":[64],"which":[65,108],"systematically":[66],"overlooks":[67],"regulation":[69],"intermediate":[71,96],"processes.":[73],"paper":[75],"introduces":[76],"Serial-Group":[77],"Decaying-Reward":[78],"Policy":[79],"Optimization":[80],"(S-GRPO),":[81],"a":[82,146,216],"novel":[83],"learning":[85],"paradigm":[86],"enables":[88],"implicitly":[91],"evaluate":[92],"sufficiency":[94],"steps,":[98],"facilitating":[100],"early":[101,180],"exit":[102,134,155],"Unlike":[106],"GRPO,":[107],"samples":[109,120],"multiple":[110,127],"possible":[111],"paths":[113],"parallel":[115],"(parallel":[116],"group),":[117],"S-GRPO":[118,189,214],"only":[119],"one":[121],"path":[123,132,160],"and":[124,136,174,198,211],"serially":[125],"selects":[126],"temporal":[128],"positions":[129,156],"thinking":[135,181],"directly":[137],"generate":[138],"answers":[139,144],"(serial":[140],"group).":[141],"For":[142],"correct":[143],"within":[145],"serial":[147],"group,":[148],"rewards":[149],"gradually":[150],"decrease":[151],"based":[152],"on":[153],"along":[157],"front":[162],"back.":[164],"design":[166],"encourages":[167],"produce":[171],"more":[172],"accurate":[173],"concise":[175],"thoughts,":[176],"while":[177,225],"also":[178],"incentivizing":[179],"termination":[182],"when":[183],"appropriate.":[184],"Empirical":[185],"evaluations":[186],"demonstrate":[187],"is":[190],"compatible":[191],"with":[192],"state-of-the-art":[193],"models,":[195],"including":[196],"Qwen3":[197],"Deepseek-distill.":[199],"Across":[200],"diverse":[201],"benchmarks":[202],"such":[203],"GSM8K,":[205],"AIME":[206],"2024,":[207],"AMC":[208],"2023,":[209],"MATH-500,":[210],"GPQA":[212],"Diamond,":[213],"achieves":[215],"substantial":[217],"reduction":[218],"sequence":[220],"length":[221],"(35.4%":[222],"-":[223,231],"61.1%)":[224],"simultaneously":[226],"improving":[227],"accuracy":[228],"(absolute":[229],"0.72%":[230],"6.08%).":[232]},"counts_by_year":[{"year":2026,"cited_by_count":48},{"year":2025,"cited_by_count":36},{"year":2024,"cited_by_count":1}],"updated_date":"2026-11-10T11:39:43.021295","created_date":"2025-10-10T00:00:00"}