{"id":"https://openalex.org/W7154220992","doi":"https://doi.org/10.48550/arxiv.2604.11554","title":"Relax: An Asynchronous Reinforcement Learning Engine for Omni-Modal Post-Training at Scale","display_name":"Relax: An Asynchronous Reinforcement Learning Engine for Omni-Modal Post-Training at Scale","publication_year":2026,"publication_date":"2026-04-13","ids":{"openalex":"https://openalex.org/W7154220992","doi":"https://doi.org/10.48550/arxiv.2604.11554"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2604.11554","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.11554","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.2604.11554","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5016293831","display_name":"Liujie Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Liujie","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5026699200","display_name":"Benzhe Ning","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ning, Benzhe","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5133594661","display_name":"Rui Yang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yang, Rui","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5133586764","display_name":"Xiaoyan Yu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yu, Xiaoyan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5133556867","display_name":"Jiaxing Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Jiaxing","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5133606645","display_name":"Lumeng Wu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wu, Lumeng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5054125653","display_name":"Jia Cong Liu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liu, Jia","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5133604018","display_name":"Minghao Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Minghao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5103275486","display_name":"Weihang Chen","orcid":"https://orcid.org/0000-0003-1380-7631"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, Weihang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5110118449","display_name":"Weiqi Hu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Hu, Weiqi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5133587577","display_name":"Lei Zhang","orcid":"https://orcid.org/0009-0006-7934-056X"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Lei","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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.3043000102043152,"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"}},"topics":[{"id":"https://openalex.org/T11714","display_name":"Multimodal Machine Learning Applications","score":0.3043000102043152,"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"}},{"id":"https://openalex.org/T10462","display_name":"Reinforcement Learning in Robotics","score":0.14749999344348907,"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":0.11739999800920486,"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/speedup","display_name":"Speedup","score":0.8705999851226807},{"id":"https://openalex.org/keywords/asynchronous-communication","display_name":"Asynchronous communication","score":0.723800003528595},{"id":"https://openalex.org/keywords/reinforcement-learning","display_name":"Reinforcement learning","score":0.7214999794960022},{"id":"https://openalex.org/keywords/robustness","display_name":"Robustness (evolution)","score":0.5489000082015991},{"id":"https://openalex.org/keywords/toolbox","display_name":"Toolbox","score":0.3693999946117401},{"id":"https://openalex.org/keywords/convergence","display_name":"Convergence (economics)","score":0.349700003862381},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.3467000126838684},{"id":"https://openalex.org/keywords/emulation","display_name":"Emulation","score":0.3165999948978424}],"concepts":[{"id":"https://openalex.org/C68339613","wikidata":"https://www.wikidata.org/wiki/Q1549489","display_name":"Speedup","level":2,"score":0.8705999851226807},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8133999705314636},{"id":"https://openalex.org/C151319957","wikidata":"https://www.wikidata.org/wiki/Q752739","display_name":"Asynchronous communication","level":2,"score":0.723800003528595},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.7214999794960022},{"id":"https://openalex.org/C63479239","wikidata":"https://www.wikidata.org/wiki/Q7353546","display_name":"Robustness (evolution)","level":3,"score":0.5489000082015991},{"id":"https://openalex.org/C120314980","wikidata":"https://www.wikidata.org/wiki/Q180634","display_name":"Distributed computing","level":1,"score":0.4083999991416931},{"id":"https://openalex.org/C2777655017","wikidata":"https://www.wikidata.org/wiki/Q1501161","display_name":"Toolbox","level":2,"score":0.3693999946117401},{"id":"https://openalex.org/C2777303404","wikidata":"https://www.wikidata.org/wiki/Q759757","display_name":"Convergence (economics)","level":2,"score":0.349700003862381},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.3467000126838684},{"id":"https://openalex.org/C149810388","wikidata":"https://www.wikidata.org/wiki/Q5374873","display_name":"Emulation","level":2,"score":0.3165999948978424},{"id":"https://openalex.org/C101468663","wikidata":"https://www.wikidata.org/wiki/Q1620158","display_name":"Modular design","level":2,"score":0.30239999294281006},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3003000020980835},{"id":"https://openalex.org/C147297375","wikidata":"https://www.wikidata.org/wiki/Q6674930","display_name":"Look-ahead","level":2,"score":0.28450000286102295},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.2770000100135803},{"id":"https://openalex.org/C34736171","wikidata":"https://www.wikidata.org/wiki/Q918333","display_name":"Preprocessor","level":2,"score":0.2745000123977661},{"id":"https://openalex.org/C13662910","wikidata":"https://www.wikidata.org/wiki/Q193139","display_name":"Trajectory","level":2,"score":0.27399998903274536},{"id":"https://openalex.org/C106189395","wikidata":"https://www.wikidata.org/wiki/Q176789","display_name":"Markov decision process","level":3,"score":0.27390000224113464},{"id":"https://openalex.org/C2778755073","wikidata":"https://www.wikidata.org/wiki/Q10858537","display_name":"Scale (ratio)","level":2,"score":0.2703000009059906},{"id":"https://openalex.org/C79403827","wikidata":"https://www.wikidata.org/wiki/Q3988","display_name":"Real-time computing","level":1,"score":0.265500009059906},{"id":"https://openalex.org/C172430144","wikidata":"https://www.wikidata.org/wiki/Q17111997","display_name":"Symmetric multiprocessor system","level":2,"score":0.2605000138282776},{"id":"https://openalex.org/C140745168","wikidata":"https://www.wikidata.org/wiki/Q1210082","display_name":"Tree traversal","level":2,"score":0.25929999351501465},{"id":"https://openalex.org/C205606062","wikidata":"https://www.wikidata.org/wiki/Q5249645","display_name":"Decoupling (probability)","level":2,"score":0.25540000200271606},{"id":"https://openalex.org/C165064840","wikidata":"https://www.wikidata.org/wiki/Q1321061","display_name":"Matching (statistics)","level":2,"score":0.2515999972820282}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2604.11554","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.11554","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.2604.11554","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.11554","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":{"Reinforcement":[0],"learning":[1],"(RL)":[2],"post-training":[3],"has":[4],"proven":[5],"effective":[6],"at":[7,40,232],"unlocking":[8],"reasoning,":[9],"self-reflection,":[10],"and":[11,24,42,85,115,141,169,219],"tool-use":[12],"capabilities":[13],"in":[14,201],"large":[15],"language":[16],"models.":[17],"As":[18],"models":[19,192],"extend":[20],"to":[21,88,179,198],"omni-modal":[22,211],"inputs":[23],"agentic":[25],"multi-turn":[26],"workflows,":[27],"RL":[28,58,102,212],"training":[29,59,125],"systems":[30],"face":[31],"three":[32,66],"interdependent":[33],"challenges:":[34],"heterogeneous":[35],"data":[36,83,129],"flows,":[37],"operational":[38],"robustness":[39],"scale,":[41],"the":[43,78,127,180,204],"staleness":[44,134],"--":[45,81,91],"throughput":[46],"tradeoff.":[47],"We":[48],"present":[49],"\\textbf{Relax}":[50],"(Reinforcement":[51],"Engine":[52],"Leveraging":[53],"Agentic":[54],"X-modality),":[55],"an":[56,71,106],"open-source":[57],"engine":[60],"that":[61,110],"addresses":[62],"these":[63],"challenges":[64],"through":[65],"co-designed":[67],"architectural":[68],"layers.":[69],"First,":[70],"\\emph{omni-native":[72],"architecture}":[73],"builds":[74],"multimodal":[75],"support":[76],"into":[77],"full":[79],"stack":[80],"from":[82],"preprocessing":[84],"modality-aware":[86],"parallelism":[87],"inference":[89],"generation":[90],"rather":[92],"than":[93],"retrofitting":[94],"it":[95],"onto":[96],"a":[97,132,147,162,170],"text-centric":[98],"pipeline.":[99],"Second,":[100],"each":[101],"role":[103],"runs":[104],"as":[105],"independent,":[107],"fault-isolated":[108],"service":[109],"can":[111],"be":[112],"scaled,":[113],"recovered,":[114],"upgraded":[116],"without":[117,227],"global":[118],"coordination.":[119],"Third,":[120],"service-level":[121],"decoupling":[122],"enables":[123],"asynchronous":[124,143],"via":[126],"TransferQueue":[128],"bus,":[130],"where":[131],"single":[133],"parameter":[135],"smoothly":[136],"interpolates":[137],"among":[138],"on-policy,":[139],"near-on-policy,":[140],"fully":[142,158],"execution.":[144],"Relax":[145,184,229],"achieves":[146],"1.20$\\times$":[148],"end-to-end":[149],"speedup":[150,164,172],"over":[151,165,222],"veRL":[152,202],"on":[153,167,173,214,225],"Qwen3-4B":[154,168],"on-policy":[155],"training.":[156],"Its":[157],"async":[159],"mode":[160],"delivers":[161],"1.76$\\times$":[163],"colocate":[166],"2.00$\\times$":[171],"Qwen3-Omni-30B,":[174],"while":[175],"all":[176],"modes":[177],"converge":[178],"same":[181,205],"reward":[182],"level.":[183],"supports":[185],"R3":[186],"(Rollout":[187],"Routing":[188],"Replay)~\\cite{ma2025r3}":[189],"for":[190],"MoE":[191],"with":[193],"only":[194],"1.9\\%":[195],"overhead,":[196],"compared":[197],"32\\%":[199],"degradation":[200],"under":[203],"configuration.":[206],"It":[207],"further":[208],"demonstrates":[209],"stable":[210],"convergence":[213],"Qwen3-Omni":[215],"across":[216],"image,":[217],"text,":[218],"audio,":[220],"sustaining":[221],"2{,}000":[223],"steps":[224],"video":[226],"degradation.":[228],"is":[230],"available":[231],"https://github.com/rednote-ai/Relax.":[233]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-04-15T00:00:00"}