{"id":"https://openalex.org/W7214748835","doi":"https://doi.org/10.48550/arxiv.2609.34681","title":"SOLAR: A State-Driven Online Learning Rate Scheduler for LLM Pretraining","display_name":"SOLAR: A State-Driven Online Learning Rate Scheduler for LLM Pretraining","publication_year":2026,"publication_date":"2026-09-28","ids":{"openalex":"https://openalex.org/W7214748835","doi":"https://doi.org/10.48550/arxiv.2609.34681"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2609.34681","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.34681","pdf_url":"https://arxiv.org/pdf/2609.34681","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","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2609.34681","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5061822049","display_name":"Qiulin Shang","orcid":"https://orcid.org/0009-0004-5727-9185","observed_orcids":["https://orcid.org/0009-0004-5727-9185"]},"institutions":[{"id":"https://openalex.org/I20231570","display_name":"Peking University","ror":"https://ror.org/02v51f717","country_code":"CN","type":"education","lineage":["https://openalex.org/I20231570"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Shang, Qiulin","raw_affiliation_strings":["Peking University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Peking University","institution_ids":["https://openalex.org/I20231570"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5133056842","display_name":"Binyu Wang","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I881766915","display_name":"Nanjing University","ror":"https://ror.org/01rxvg760","country_code":"CN","type":"education","lineage":["https://openalex.org/I881766915"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Wang, Binyu","raw_affiliation_strings":["Nanjing University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Nanjing University","institution_ids":["https://openalex.org/I881766915"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5153472322","display_name":"Yongqi Qiao","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I20231570","display_name":"Peking University","ror":"https://ror.org/02v51f717","country_code":"CN","type":"education","lineage":["https://openalex.org/I20231570"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Qiao, Yongqi","raw_affiliation_strings":["Peking University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Peking University","institution_ids":["https://openalex.org/I20231570"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5148319891","display_name":"Songde Rao","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I20231570","display_name":"Peking University","ror":"https://ror.org/02v51f717","country_code":"CN","type":"education","lineage":["https://openalex.org/I20231570"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Rao, Songde","raw_affiliation_strings":["Peking University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Peking University","institution_ids":["https://openalex.org/I20231570"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5021868496","display_name":"Zhoutong Wu","orcid":"https://orcid.org/0009-0003-4753-2805","observed_orcids":["https://orcid.org/0009-0003-4753-2805"]},"institutions":[{"id":"https://openalex.org/I20231570","display_name":"Peking University","ror":"https://ror.org/02v51f717","country_code":"CN","type":"education","lineage":["https://openalex.org/I20231570"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Wu, Zhoutong","raw_affiliation_strings":["Peking University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Peking University","institution_ids":["https://openalex.org/I20231570"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5100614598","display_name":"Kun Yuan","orcid":"https://orcid.org/0000-0001-8394-8187","observed_orcids":["https://orcid.org/0000-0001-8394-8187"]},"institutions":[{"id":"https://openalex.org/I20231570","display_name":"Peking University","ror":"https://ror.org/02v51f717","country_code":"CN","type":"education","lineage":["https://openalex.org/I20231570"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Yuan, Kun","raw_affiliation_strings":["Peking University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Peking University","institution_ids":["https://openalex.org/I20231570"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"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":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11612","display_name":"Stochastic Gradient Optimization Techniques","score":0.6420999765396118,"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/T11612","display_name":"Stochastic Gradient Optimization Techniques","score":0.6420999765396118,"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":0.10350000113248825,"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.09179999679327011,"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/large-language-model-pretraining","display_name":"large language model pretraining","score":0.8859999775886536},{"id":"https://openalex.org/keywords/learning-to-optimize","display_name":"learning to optimize","score":0.8429999947547913},{"id":"https://openalex.org/keywords/circuit-breakers","display_name":"circuit breakers","score":0.5699999928474426},{"id":"https://openalex.org/keywords/ppo","display_name":"PPO","score":0.3490000069141388},{"id":"https://openalex.org/keywords/perplexity","display_name":"perplexity","score":0.7319999933242798},{"id":"https://openalex.org/keywords/moe-models","display_name":"MoE models","score":0.503000020980835},{"id":"https://openalex.org/keywords/adamw","display_name":"AdamW","score":0.7080000042915344},{"id":"https://openalex.org/keywords/muon-optimizer","display_name":"Muon optimizer","score":0.75}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7070000171661377},{"id":"https://openalex.org/C127705205","wikidata":"https://www.wikidata.org/wiki/Q5748245","display_name":"Heuristics","level":2,"score":0.5593000054359436},{"id":"https://openalex.org/C206729178","wikidata":"https://www.wikidata.org/wiki/Q2271896","display_name":"Scheduling (production processes)","level":2,"score":0.5407999753952026},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.49900001287460327},{"id":"https://openalex.org/C12302492","wikidata":"https://www.wikidata.org/wiki/Q1752097","display_name":"Setpoint","level":2,"score":0.48739999532699585},{"id":"https://openalex.org/C68387754","wikidata":"https://www.wikidata.org/wiki/Q7271585","display_name":"Schedule","level":2,"score":0.45509999990463257},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.44359999895095825},{"id":"https://openalex.org/C62230096","wikidata":"https://www.wikidata.org/wiki/Q275969","display_name":"Crowdsourcing","level":2,"score":0.43459999561309814},{"id":"https://openalex.org/C155512373","wikidata":"https://www.wikidata.org/wiki/Q287450","display_name":"Residual","level":2,"score":0.40959998965263367},{"id":"https://openalex.org/C4554734","wikidata":"https://www.wikidata.org/wiki/Q593744","display_name":"Knowledge base","level":2,"score":0.38989999890327454},{"id":"https://openalex.org/C203479927","wikidata":"https://www.wikidata.org/wiki/Q5165939","display_name":"Controller (irrigation)","level":2,"score":0.3443000018596649},{"id":"https://openalex.org/C2986087404","wikidata":"https://www.wikidata.org/wiki/Q15946010","display_name":"Online learning","level":2,"score":0.32670000195503235},{"id":"https://openalex.org/C2780626000","wikidata":"https://www.wikidata.org/wiki/Q5936775","display_name":"Human-in-the-loop","level":2,"score":0.326200008392334},{"id":"https://openalex.org/C100279451","wikidata":"https://www.wikidata.org/wiki/Q372193","display_name":"Perplexity","level":3,"score":0.3181999921798706},{"id":"https://openalex.org/C2778755073","wikidata":"https://www.wikidata.org/wiki/Q10858537","display_name":"Scale (ratio)","level":2,"score":0.3165999948978424},{"id":"https://openalex.org/C79403827","wikidata":"https://www.wikidata.org/wiki/Q3988","display_name":"Real-time computing","level":1,"score":0.31439998745918274},{"id":"https://openalex.org/C3770464","wikidata":"https://www.wikidata.org/wiki/Q775963","display_name":"Smoothing","level":2,"score":0.31040000915527344},{"id":"https://openalex.org/C106189395","wikidata":"https://www.wikidata.org/wiki/Q176789","display_name":"Markov decision process","level":3,"score":0.30889999866485596},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.3066999912261963},{"id":"https://openalex.org/C196921405","wikidata":"https://www.wikidata.org/wiki/Q786431","display_name":"Online algorithm","level":2,"score":0.30219998955726624},{"id":"https://openalex.org/C159886148","wikidata":"https://www.wikidata.org/wiki/Q176645","display_name":"Markov process","level":2,"score":0.2849999964237213},{"id":"https://openalex.org/C175291020","wikidata":"https://www.wikidata.org/wiki/Q1156822","display_name":"Offset (computer science)","level":2,"score":0.27239999175071716},{"id":"https://openalex.org/C17098449","wikidata":"https://www.wikidata.org/wiki/Q176814","display_name":"Partially observable Markov decision process","level":4,"score":0.26010000705718994},{"id":"https://openalex.org/C22019652","wikidata":"https://www.wikidata.org/wiki/Q331309","display_name":"Overfitting","level":3,"score":0.2567000091075897}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2609.34681","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.34681","pdf_url":"https://arxiv.org/pdf/2609.34681","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.2609.34681","doi":"https://doi.org/10.48550/arxiv.2609.34681","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2609.34681","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2609.34681","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.34681","pdf_url":"https://arxiv.org/pdf/2609.34681","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":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7214748835.pdf","grobid_xml":"https://content.openalex.org/works/W7214748835.grobid-xml"},"referenced_works_count":19,"referenced_works":["https://openalex.org/W2261408131","https://openalex.org/W2619219755","https://openalex.org/W2736601468","https://openalex.org/W2749581528","https://openalex.org/W2903697572","https://openalex.org/W2973727699","https://openalex.org/W2974578743","https://openalex.org/W3118781290","https://openalex.org/W4225390052","https://openalex.org/W4229005866","https://openalex.org/W4322718191","https://openalex.org/W4384918448","https://openalex.org/W4389261255","https://openalex.org/W4390437677","https://openalex.org/W4396815229","https://openalex.org/W4401660192","https://openalex.org/W4415799054","https://openalex.org/W4417465430","https://openalex.org/W7131837684"],"related_works":["https://openalex.org/W7127913843","https://openalex.org/W7161805535","https://openalex.org/W7127646912","https://openalex.org/W7138853627","https://openalex.org/W7223093425","https://openalex.org/W4403662640","https://openalex.org/W4417473639","https://openalex.org/W6979318390","https://openalex.org/W4403964581","https://openalex.org/W7168244084"],"abstract_inverted_index":{"Learning-rate":[0],"(LR)":[1],"scheduling":[2,43],"plays":[3],"a":[4,52,81,91,95,139,192,217,237,246],"central":[5],"role":[6],"in":[7,32],"large":[8],"language":[9],"model":[10],"(LLM)":[11],"pretraining,":[12,150],"yet":[13],"current":[14],"practice":[15],"still":[16],"relies":[17],"heavily":[18],"on":[19,207,216],"hand-crafted":[20],"heuristics":[21],"such":[22],"as":[23,94,245],"Warmup-Cosine-Decay":[24],"and":[25,67,97,132,159,171,173,188,224],"Warmup-Stable-Decay.":[26],"Because":[27],"these":[28],"schedules":[29,158],"are":[30],"fixed":[31],"advance,":[33],"they":[34],"cannot":[35],"adapt":[36,120],"to":[37,47,62,110,119,168,178,198],"evolving":[38],"optimization":[39],"dynamics.":[40],"Online":[41,77],"learned":[42,248],"within":[44],"the":[45,68,111,117,121,125,208],"Learning":[46,78],"Optimize":[48],"(L2O)":[49],"framework":[50,83],"offers":[51],"dynamic":[53],"alternative,":[54],"but":[55],"remains":[56],"brittle":[57],"at":[58,113,226],"LLM":[59,252],"scale":[60],"due":[61],"noisy":[63],"signals,":[64],"delayed":[65],"feedback,":[66],"risk":[69],"of":[70],"catastrophic":[71],"divergence.":[72],"We":[73],"propose":[74],"SOLAR":[75,89,151,244],"(State-driven":[76],"rAte":[79],"scheduleR),":[80],"stabilized":[82],"for":[84,103,163,251],"reliable":[85],"online":[86,136],"LR":[87,122,161,249],"adaptation.":[88],"uses":[90],"base":[92,112,186],"schedule":[93],"reference":[96],"learns":[98],"bounded,":[99],"state-dependent":[100],"residual":[101,213],"corrections":[102],"individual":[104],"parameter":[105],"groups.":[106],"Each":[107],"correction":[108],"re-anchors":[109],"every":[114],"step,":[115],"allowing":[116],"policy":[118,214],"without":[123,230],"relearning":[124],"warmup-decay":[126],"profile.":[127],"A":[128,212],"lightweight":[129],"state":[130],"representation":[131],"progress-aware":[133],"reward":[134],"guide":[135],"learning,":[137],"while":[138,202],"Circuit-Breaker":[140],"restores":[141],"training":[142],"after":[143],"rare":[144],"unsafe":[145],"actions.":[146],"Across":[147],"autoregressive":[148],"language-model":[149],"improves":[152,191],"final":[153,200],"perplexity":[154],"over":[155],"tuned":[156],"static":[157],"automatic":[160],"tuners":[162],"dense":[164,228],"models":[165],"from":[166,196],"60M":[167,218],"1B,":[169],"AdamW":[170],"Muon,":[172],"two":[174,210],"MoE":[175],"settings":[176],"up":[177],"3B.":[179],"Matched":[180],"130M":[181],"controls":[182],"show":[183],"that":[184],"adding":[185],"anchoring":[187],"action":[189],"bounds":[190],"global":[193],"PPO":[194,232],"controller":[195,250],"27.09":[197],"23.74":[199],"PPL,":[201],"group-wise":[203],"control":[204],"reaches":[205],"22.87":[206],"same":[209],"seeds.":[211],"trained":[215],"proxy":[219],"can":[220],"also":[221],"be":[222],"frozen":[223],"reused":[225],"larger":[227],"scales":[229],"target":[231],"updates,":[233],"remaining":[234],"effective":[235],"across":[236],"fourfold":[238],"base-LR":[239],"range.":[240],"These":[241],"results":[242],"establish":[243],"practical":[247],"pretraining.":[253]},"counts_by_year":[],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2026-09-30T00:00:00"}