[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"detail-sidebar-cat-0-en-105":3,"doc-seo-503758-105":59,"doc-detail-503758-en":130},{"code":4,"msg":5,"data":6},0,"success",[7,13,18,23,28,33,38,43,48,51,55],{"id":8,"doc_module":4,"doc_module_name":9,"category_name":10,"show_sort_weight":11,"slug":12},1,"Document","Story & Novel",90,"story-novel",{"id":14,"doc_module":4,"doc_module_name":9,"category_name":15,"show_sort_weight":16,"slug":17},2,"Literature",80,"literature",{"id":19,"doc_module":4,"doc_module_name":9,"category_name":20,"show_sort_weight":21,"slug":22},4,"Exam",70,"exam",{"id":24,"doc_module":4,"doc_module_name":9,"category_name":25,"show_sort_weight":26,"slug":27},5,"Comic",60,"comic",{"id":29,"doc_module":4,"doc_module_name":9,"category_name":30,"show_sort_weight":31,"slug":32},6,"Technology",50,"technology",{"id":34,"doc_module":4,"doc_module_name":9,"category_name":35,"show_sort_weight":36,"slug":37},7,"Healthcare",40,"healthcare",{"id":39,"doc_module":4,"doc_module_name":9,"category_name":40,"show_sort_weight":41,"slug":42},8,"Research & Report",30,"research-report",{"id":44,"doc_module":4,"doc_module_name":9,"category_name":45,"show_sort_weight":46,"slug":47},9,"Religion & Spirituality",20,"religion-spirituality",{"id":46,"doc_module":4,"doc_module_name":9,"category_name":49,"show_sort_weight":46,"slug":50},"World Cup","world-cup",{"id":52,"doc_module":4,"doc_module_name":9,"category_name":53,"show_sort_weight":52,"slug":54},10,"Lifestyle","lifestyle",{"id":56,"doc_module":4,"doc_module_name":9,"category_name":57,"show_sort_weight":24,"slug":58},19,"General","general",{"code":4,"msg":60,"data":61},"ok",{"site_id":62,"language":63,"slug":64,"title":65,"keywords":66,"description":67,"schema_data":68,"social_meta":123,"head_meta":125,"extra_data":127,"updated_unix":129},105,"en","weave-efficient-co-scheduling-for-disaggregated-rl-post-training","WEAVE - Efficient Co-Scheduling for Disaggregated RL Post-Training","","Rollout–training disaggregation is becoming the standard architecture for reinforcement learning (RL) post-training by separating memory-bound rollout and compute-bound training across dedicated clusters to improve hardware utilization. On-policy algorithms require strict synchronization, causing dependency bubbles where one cluster must idle. WEAVE reclaims these bubbles via cross-cluster orchestration using co-execution groups with locality domains, a two-tier scheduler, and warm-start residency constraints. Evaluated on 328 H20 and 328 H800 GPUs, WEAVE attains 100% SLO and improves cost efficiency by 1.84× versus standard disaggregation and 1.38× over co-located baselines.",{"@graph":69,"@context":122},[70,84,105],{"@type":71,"itemListElement":72},"BreadcrumbList",[73,77,79,82],{"item":74,"name":75,"@type":76,"position":8},"https://docshare.wps.com","Home","ListItem",{"item":78,"name":9,"@type":76,"position":14},"https://docshare.wps.com/document/",{"item":80,"name":40,"@type":76,"position":81},"https://docshare.wps.com/document/research-report/",3,{"item":83,"name":65,"@type":76,"position":19},"https://docshare.wps.com/document/weave-efficient-co-scheduling-for-disaggregated-rl-post-training/503758/",{"url":83,"name":65,"@type":85,"image":86,"author":91,"headline":65,"publisher":94,"fileFormat":97,"inLanguage":63,"description":67,"dateModified":98,"datePublished":99,"encodingFormat":97,"isAccessibleForFree":100,"interactionStatistic":101},"DigitalDocument",{"url":87,"@type":88,"width":89,"height":90},"https://docshare.wps.com/thumbnails/weave-efficient-co-scheduling-for-disaggregated-rl-post-training/503758.png","ImageObject",300,407,{"name":92,"@type":93},"Genevieve","Person",{"url":74,"name":95,"@type":96},"DocShare","Organization","application/pdf","2026-10-17","2026-10-03",true,{"@type":102,"interactionType":103,"userInteractionCount":29},"InteractionCounter",{"@type":104},"ViewAction",{"@type":106,"mainEntity":107},"FAQPage",[108,114,118],{"name":109,"@type":110,"acceptedAnswer":111},"What problem does WEAVE address in disaggregated RL post-training?","Question",{"text":112,"@type":113},"WEAVE targets dependency bubbles caused by strict synchronization in on-policy algorithms, which force one cluster to idle while the dependent phase runs on another cluster.","Answer",{"name":115,"@type":110,"acceptedAnswer":116},"How does WEAVE reduce dependency bubbles?",{"text":117,"@type":113},"WEAVE uses cross-cluster orchestration through co-execution groups that tightly weave rollout and training phases across two resource pools, time-multiplexing both pools while preserving on-policy dependencies.",{"name":119,"@type":110,"acceptedAnswer":120},"What scheduling mechanism does WEAVE use?",{"text":121,"@type":113},"WEAVE employs a two-tier scheduling architecture: an inter-group scheduler for job placement using conservative stochastic planning, and an intra-group scheduler that orchestrates a provably optimal round-robin schedule with locality domains.","https://schema.org",{"og:url":83,"og:type":124,"og:title":65,"og:site_name":95,"og:description":67},"article",{"robots":126,"canonical":83},"index,follow",{"doc_id":128,"site_id":62},503758,1791295695,{"code":4,"msg":5,"data":131},{"doc_id":128,"user_id":132,"nickname":92,"user_avatar":133,"doc_module":4,"category_id":39,"category_name":40,"doc_title":65,"doc_description":67,"doc_content":134,"file_id":135,"file_url":136,"file_type":137,"file_size":138,"view_count":29,"is_deleted":4,"is_public":8,"is_downloadable":8,"audit_status":8,"page_count":46,"language":139,"language_code":63,"site_id":62,"html_lang":63,"table_of_contents":140,"faqs":141,"seo_title":142,"seo_description":67,"update_tm":143,"read_time":31},1374391974585,"https://ap-avatar.wpscdn.com/davatar_276721f389ce27ea32af1340a28f341c","Weave: Efficient Co-Scheduling for Disaggregated  \nRL Post-Training  \nTianyuan Wu and Lunxi Cao, Hong Kong University of Science and Technology; Yining Wei, University of Illinois Urbana–Champaign; Wei Gao, Yuheng Zhao, and Dakai An, Hong Kong University of Science and Technology; Shaopan Xiong, Zhiqiang Lv, Ju Huang, Siran Yang, Yinghao Yu, Jiamang Wang, and Lin Qu, Alibaba Group; Wei Wang, Hong Kong University of Science and Technology  \n[https://www.usenix.org/conference/osdi26/presentation/wu-tianyuan](https://www.usenix.org/conference/osdi26/presentation/wu-tianyuan)  \nThis paper is included in the Proceedings of the 20th USENIX Symposium on Operating Systems Design and Implementation .  \nJuly 13–15, 2026 • Seattle, WA, USA  \nISBN 978-1-939133-55-7  \nOpen access to the Proceedings of the 20th USENIX Symposium on Operating Systems Design and Implementation is sponsored by  \nWEAVE: Efficient Co-Scheduling for Disaggregated RL Post-Training  \nTianyuan Wu†, Lunxi Cao†, Yining Wei‡, Wei Gao†, Yuheng Zhao†, Dakai An†, Shaopan Xiong§, Zhiqiang Lv§, Ju Huang§, Siran Yang§, Yinghao Yu§, Jiamang Wang§, Lin Qu§, Wei Wang†† Hong Kong University of Science and Technology, ‡ UIUC,§Alibaba Group  \nAbstract  \nRollout–training disaggregation is emerging as the standard architecture for Reinforcement Learning (RL) post-training, where memory-bound rollout and compute-bound training are physically disaggregated onto purpose-built clusters to maximize hardware efficiency. However, the strict synchronization required by on-policy algorithms introduces severe dependency bubbles, forcing one cluster to idle while the dependent phase is running on the other. We present WEAVE, a cluster scheduling framework that reclaims these bubbles through cross-cluster orchestration. WEAVE is built on the insight that the structural idleness of one job can be effectively utilized by the active phase of another. To realize this, we introduce the co-execution group abstraction, which partitions the cluster into isolated locality domains. This abstraction enables a twotier scheduling architecture: an inter-group scheduler that optimizes job placement using conservative stochastic planning, and an intra-group scheduler that orchestrates a provably optimal round-robin schedule. The group abstraction also imposesa residency constraint, ensuring that massive model states remain cached in host memory to enable “warm-start” context switching. We evaluate WEAVE on a production-scale testbed with 328 H20 and 328 H800 GPUs. WEAVE improves cost efficiency by 1.84× over standard disaggregation and 1.38 × over state-of-the-art co-located baselines, all while achieving 100% SLO attainment.  \n1 Introduction  \nThe development of Large Language Model (LLM) is shifting from pre-training to Reinforcement Learning (RL) posttraining [1, 3, 7], which unlocks reasoning capabilities in complex domains such as mathematics [2], coding [33], and tool use [9, 54] . To achieve optimal performance and model stability, production deployments have converged on synchronous, on-policy algorithms [7, 37, 41] that prescribe a strict, iterative learning loop through three phases, each with a distinct resource bottleneck: (1) rollout, a memory-bandwidth-bound inference stage that generates token trajectories; (2) training, a compute-intensive stage that updates parameters from rewards; and (3) synchronization, a network-bound stage that propagates updated weights back to inference workers.  \nThis paper targets the deployment that production onpolicy RL post-training has converged on—a disaggregated  \nNaïve Disaggregation  \n Dependency Bubbles  \nRollout Pool Training Pool  \n|  | Train-A ① |  |\n| --- | --- | --- |\n|  |  |  |\n\nSolo iteration time (􀀡!\"\\#\" )  \n\n| | \u003Cbr>Rollout-A ② |\n| --- | --- |\n|  |  |\n\nSync-A  \nDisaggregation & Co-Scheduling  \nA meta-iteration of Co-execution Group {A, B}  \n\n|  |\n| --- |\n| Train-A ② |\n\nRollout Pool  \nTraining Pool  \nFigure 1: Comparison between existing disaggregated","cbCaiuWMIcWzYG13","https://ap.wps.com/l/cbCaiuWMIcWzYG13","pdf",2565249,"English","# Abstract\n# 1 Introduction\n## Rollout–training disaggregation and on-policy synchronization\n## Dependency bubbles and prior approaches\n## WEAVE framework and co-execution groups","[{\"question\":\"What problem does WEAVE address in disaggregated RL post-training?\",\"answer\":\"WEAVE targets dependency bubbles caused by strict synchronization in on-policy algorithms, which force one cluster to idle while the dependent phase runs on another cluster.\"},{\"question\":\"How does WEAVE reduce dependency bubbles?\",\"answer\":\"WEAVE uses cross-cluster orchestration through co-execution groups that tightly weave rollout and training phases across two resource pools, time-multiplexing both pools while preserving on-policy dependencies.\"},{\"question\":\"What scheduling mechanism does WEAVE use?\",\"answer\":\"WEAVE employs a two-tier scheduling architecture: an inter-group scheduler for job placement using conservative stochastic planning, and an intra-group scheduler that orchestrates a provably optimal round-robin schedule with locality domains.\"}]","WEAVE - Efficient Co-Scheduling for Disaggregated RL Post-Training | PDF",1791060239]