25892589 root : "/" ,
25902590 debug : "env" ,
25912591 default : { "avatar" :"https://unpkg.com/volantis-static@0.0.1761982841160/media/placeholder/avatar/round/3442075.svg" , "link" :"https://unpkg.com/volantis-static@0.0.1761982841160/media/placeholder/link/8f277b4ee0ecd.svg" , "cover" :"https://unpkg.com/volantis-static@0.0.1761982841160/media/placeholder/cover/76b86c0226ffd.svg" , "image" :"https://unpkg.com/volantis-static@0.0.1761982841160/media/placeholder/image/2659360.svg" , "topic" :"https://gcore.jsdelivr.net/gh/cdn-x/placeholder@1.0.12/image/10433048.png" , "project" :"https://gcore.jsdelivr.net/gh/cdn-x/placeholder@1.0.12/image/2779789.png" , "banner" :"https://gcore.jsdelivr.net/gh/cdn-x/placeholder@1.0.12/banner/books.jpg" , "loading" :"https://api.iconify.design/eos-icons:three-dots-loading.svg?color=%231cd0fd" , "image_onerror" :"data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' width='2rem' height='2rem' viewBox='0 0 24 24'%3E%3C!-- Icon from Solar by 480 Design - https://creativecommons.org/licenses/by/4.0/ --%3E%3Cpath fill='%23F44336' d='M22 12.698c-.002 1.47-.013 2.718-.096 3.743c-.097 1.19-.296 2.184-.74 3.009a4.2 4.2 0 0 1-.73.983c-.833.833-1.893 1.21-3.237 1.39C15.884 22 14.2 22 12.053 22h-.106c-2.148 0-3.83 0-5.144-.177c-1.343-.18-2.404-.557-3.236-1.39c-.738-.738-1.12-1.656-1.322-2.795c-.2-1.12-.236-2.512-.243-4.241Q1.999 12.737 2 12v-.054c0-2.148 0-3.83.177-5.144c.18-1.343.557-2.404 1.39-3.236s1.893-1.21 3.236-1.39c1.168-.157 2.67-.175 4.499-.177a.697.697 0 1 1 0 1.396c-1.855.002-3.234.018-4.313.163c-1.189.16-1.906.464-2.436.994S3.72 5.8 3.56 6.99C3.397 8.2 3.395 9.788 3.395 12v.784l.932-.814a2.14 2.14 0 0 1 2.922.097l3.99 3.99a1.86 1.86 0 0 0 2.385.207l.278-.195a2.79 2.79 0 0 1 3.471.209l2.633 2.37c.265-.557.423-1.288.507-2.32c.079-.972.09-2.152.091-3.63a.698.698 0 0 1 1.396 0' opacity='.5'/%3E%3Cpath fill='%23F44336' fill-rule='evenodd' d='M17.5 11c-2.121 0-3.182 0-3.841-.659S13 8.621 13 6.5s0-3.182.659-3.841S15.379 2 17.5 2s3.182 0 3.841.659S22 4.379 22 6.5s0 3.182-.659 3.841S19.621 11 17.5 11m-1.47-7.03a.75.75 0 1 0-1.06 1.06l1.47 1.47l-1.47 1.47a.75.75 0 0 0 1.06 1.06l1.47-1.47l1.47 1.47a.75.75 0 1 0 1.06-1.06L18.56 6.5l1.47-1.47a.75.75 0 0 0-1.06-1.06L17.5 5.44z' clip-rule='evenodd'/%3E%3C/svg%3E" } ,
2592- lastupdate : new Date ( 1787153701077 ) ,
2592+ lastupdate : new Date ( 1787204923456 ) ,
25932593 scroll_behavior : "smooth" ,
25942594 cdn : {
25952595 jquery : "https://unpkg.com/jquery@3.7.1/dist/jquery.min.js" ,
30403040<!-- 文章预览卡片容器,使用schema.org的Article类型进行结构化数据标记 -->
30413041< div class ="post post-v3 white-box reveal shadow floatable blur " itemscope itemtype ="http://schema.org/Article ">
30423042 <!-- 标记文章的主要页面链接 -->
3043- < link itemprop ="mainEntityOfPage " href ="https://floatlibai.github.io/2026/08/hello-world / ">
3043+ < link itemprop ="mainEntityOfPage " href ="https://floatlibai.github.io/2026/08/rl-async-pause-resume / ">
30443044
30453045 <!-- 隐藏的发布者信息,使用schema.org的Organization类型 -->
30463046 < span hidden itemprop ="publisher " itemscope itemtype ="http://schema.org/Organization ">
30473047 < meta itemprop ="name " content ="floatlibai's blog ">
30483048 </ span >
30493049 < span hidden itemprop ="post " itemscope itemtype ="http://schema.org/CreativeWork ">
3050- < meta itemprop ="name " content ="Hello World ">
3050+ < meta itemprop ="name " content ="异步rl:pause/resume ">
30513051 < meta itemprop ="description " content ="floatlibai github page blog ">
30523052 </ span >
30533053
30673067 <!-- 文章标题区域,使用schema.org的name和headline属性 -->
30683068
30693069 < h2 class ="article-title " itemprop ="name headline ">
3070- < a href ="/2026/08/hello-world / " itemprop ="url ">
3071- Hello World
3070+ < a href ="/2026/08/rl-async-pause-resume / " itemprop ="url ">
3071+ 异步rl:pause/resume
30723072 </ a >
30733073 </ h2 >
30743074
@@ -3078,8 +3078,8 @@ <h2 class="article-title" itemprop="name headline">
30783078 <!-- 若既没有摘要也没有描述,则使用文章内容 -->
30793079
30803080 <!-- 主题启用自动摘要时,显示截断后的内容(200字符) -->
3081- Welcome to Hexo! This is your very first post. Check documentation for more info. If you get any problems when using Hexo, you can find the answer in troubleshooting or you can ask me on GitHub.
3082- Qu ...
3081+ 在标准 RL 训练循环中,生成(generation)和训练(training)顺序执行:策略模型先生成 rollout 样本,然后在这些样本上进行训练,往复循环。在生成阶段,训练侧处于空闲状态,反之亦然。
3082+ 所谓 one-off pipelining 方法将生成阶段和训练阶段拆分为两个并行的协程,使模型能够一边生成新样本,一边对先前生成的数据进行训练,带来更高的 GPU 利用率和更大的训练吞 ...
30833083
30843084
30853085 </ div >
@@ -3100,22 +3100,32 @@ <h2 class="article-title" itemprop="name headline">
31003100
31013101 <!-- 显示文章更新时间,使用schema.org的dateModified属性 -->
31023102 < time class ="update-time " itemprop ="dateModified "
3103- datetime ="2026-08-19T22:52:37 +08:00 "
3104- title ="2026年8月19日 ">
3105- 2026-08-19
3103+ datetime ="2026-08-20T13:48:26 +08:00 "
3104+ title ="2026年8月20日 ">
3105+ 2026-08-20
31063106 </ time >
31073107
31083108 <!-- 显示文章分类(如果需要显示) -->
31093109
31103110
31113111 <!-- 显示文章标签(如果需要显示) -->
31123112
3113+ < span class ="dot "> </ span >
3114+ < a class ="category-link " href ="/tags/Infra/ "> Infra</ a >
3115+
3116+ <!-- 标签的结构化数据 -->
3117+
3118+ < span hidden itemprop ="about " itemscope itemtype ="http://schema.org/Thing ">
3119+ < a href ="/tags/Infra/ " itemprop ="url "> < span itemprop ="name "> Infra</ span > </ a >
3120+ </ span >
3121+
3122+
31133123 </ div >
31143124
31153125 <!-- 阅读更多按钮 -->
31163126
31173127 < div >
3118- < a class ='readmore ' href ='/2026/08/hello-world / '>
3128+ < a class ='readmore ' href ='/2026/08/rl-async-pause-resume / '>
31193129 阅读全文
31203130 </ a >
31213131 </ div >
@@ -3218,7 +3228,7 @@ <h2 class="article-title" itemprop="name headline">
32183228
32193229
32203230 < div class ='content '>
3221- < widget class ="widget-wrapper recent post-list "> < div class ="widget-body fs14 "> < a class ="item title " href ="/2026/08/hello-world / "> < span class ="title "> Hello World </ span > </ a > </ div > </ widget >
3231+ < widget class ="widget-wrapper recent post-list "> < div class ="widget-body fs14 "> < a class ="item title " href ="/2026/08/rl-async-pause-resume / "> < span class ="title "> 异步rl:pause/resume </ span > </ a > </ div > </ widget >
32223232 </ div >
32233233</ section >
32243234
@@ -3337,7 +3347,6 @@ <h2 class="article-title" itemprop="name headline">
33373347 <!-- Custom Files footer end-->
33383348 </ footer >
33393349
3340-
33413350 < a id ="s-top " class ="fa-solid fa-arrow-up fa-fw " title ="top "> </ a >
33423351 </ div >
33433352 </ div >
0 commit comments