+关注
普罗米M修斯
暂无个人介绍
IP属地:浙江
23
关注
0
粉丝
0
主题
0
勋章
主贴
热门
普罗米M修斯
2021-01-28
$摩根大通(JPM)$
$花旗(C)$
$富国银行(WFC)$
$美国银行(BAC)$
四大行哪个基本面更好?
普罗米M修斯
2023-05-18
$纳斯达克(.IXIC)$
这是疯了嘛
普罗米M修斯
2021-01-28
$苹果(AAPL)$
盘后跌了这么多,有妖?
普罗米M修斯
2023-04-13
人类之光
重磅!微软开源Deep Speed Chat,人人拥有ChatGPT!
普罗米M修斯
2021-02-25
$债券20+美公债指数ETF-iShares(TLT)$
这波 会跌到多少
普罗米M修斯
2021-01-26
$苹果(AAPL)$
杀期权?[贱笑]
普罗米M修斯
2021-01-15
$富国银行(WFC)$
赌一波财报,short put 准备接货
去老虎APP查看更多动态
{"i18n":{"language":"zh_CN"},"userPageInfo":{"id":"3546901181241979","uuid":"3546901181241979","gmtCreate":1583808799918,"gmtModify":1610678151627,"name":"普罗米M修斯","pinyin":"plmmxspuluomimxiusi","introduction":"","introductionEn":"","signature":"","avatar":"https://static.tigerbbs.com/1abedaa616802bfcb0f72a8650637a6b","hat":null,"hatId":null,"hatName":null,"vip":1,"status":2,"fanSize":0,"headSize":23,"tweetSize":7,"questionSize":0,"limitLevel":999,"accountStatus":4,"level":{"id":2,"name":"无畏虎","nameTw":"無畏虎","represent":"初生牛犊","factor":"发布3条非转发主帖,1条获得他人回复或点赞","iconColor":"3C9E83","bgColor":"A2F1D9"},"themeCounts":0,"badgeCounts":0,"badges":[],"moderator":false,"superModerator":false,"manageSymbols":null,"badgeLevel":null,"boolIsFan":false,"boolIsHead":false,"favoriteSize":0,"symbols":null,"coverImage":null,"realNameVerified":null,"userBadges":[{"badgeId":"e50ce593bb40487ebfb542ca54f6a561-3","templateUuid":"e50ce593bb40487ebfb542ca54f6a561","name":"偶像虎友","description":"加入老虎社区1500天","bigImgUrl":"https://static.tigerbbs.com/8b40ae7da5bf081a1c84df14bf9e6367","smallImgUrl":"https://static.tigerbbs.com/f160eceddd7c284a8e1136557615cfad","grayImgUrl":"https://static.tigerbbs.com/11792805c468334a9b31c39f95a41c6a","redirectLinkEnabled":0,"redirectLink":null,"hasAllocated":1,"isWearing":0,"stamp":null,"stampPosition":0,"hasStamp":0,"allocationCount":1,"allocatedDate":"2024.04.19","exceedPercentage":null,"individualDisplayEnabled":0,"backgroundColor":null,"fontColor":null,"individualDisplaySort":0,"categoryType":1001},{"badgeId":"976c19eed35f4cd78f17501c2e99ef37-1","templateUuid":"976c19eed35f4cd78f17501c2e99ef37","name":"博闻投资者","description":"累计交易超过10只正股","bigImgUrl":"https://static.tigerbbs.com/e74cc24115c4fbae6154ec1b1041bf47","smallImgUrl":"https://static.tigerbbs.com/d48265cbfd97c57f9048db29f22227b0","grayImgUrl":"https://static.tigerbbs.com/76c6d6898b073c77e1c537ebe9ac1c57","redirectLinkEnabled":0,"redirectLink":null,"hasAllocated":1,"isWearing":0,"stamp":null,"stampPosition":0,"hasStamp":0,"allocationCount":1,"allocatedDate":"2022.01.30","exceedPercentage":null,"individualDisplayEnabled":0,"backgroundColor":null,"fontColor":null,"individualDisplaySort":0,"categoryType":1102},{"badgeId":"228c86a078844d74991fff2b7ab2428d-3","templateUuid":"228c86a078844d74991fff2b7ab2428d","name":"投资合伙人虎","description":"证券账户累计交易金额达到100万美元","bigImgUrl":"https://static.tigerbbs.com/fbeac6bb240db7da8b972e5183d050ba","smallImgUrl":"https://static.tigerbbs.com/436cdf80292b99f0a992e78750ac4e3a","grayImgUrl":"https://static.tigerbbs.com/506a259a7b456f037592c3b23c779599","redirectLinkEnabled":0,"redirectLink":null,"hasAllocated":1,"isWearing":0,"stamp":null,"stampPosition":0,"hasStamp":0,"allocationCount":1,"allocatedDate":"2022.01.11","exceedPercentage":"93.52%","individualDisplayEnabled":0,"backgroundColor":null,"fontColor":null,"individualDisplaySort":0,"categoryType":1101},{"badgeId":"518b5610c3e8410da5cfad115e4b0f5a-1","templateUuid":"518b5610c3e8410da5cfad115e4b0f5a","name":"实盘交易者","description":"完成一笔实盘交易","bigImgUrl":"https://static.tigerbbs.com/2e08a1cc2087a1de93402c2c290fa65b","smallImgUrl":"https://static.tigerbbs.com/4504a6397ce1137932d56e5f4ce27166","grayImgUrl":"https://static.tigerbbs.com/4b22c79415b4cd6e3d8ebc4a0fa32604","redirectLinkEnabled":0,"redirectLink":null,"hasAllocated":1,"isWearing":0,"stamp":null,"stampPosition":0,"hasStamp":0,"allocationCount":1,"allocatedDate":"2021.12.21","exceedPercentage":null,"individualDisplayEnabled":0,"backgroundColor":null,"fontColor":null,"individualDisplaySort":0,"categoryType":1100},{"badgeId":"35ec162348d5460f88c959321e554969-3","templateUuid":"35ec162348d5460f88c959321e554969","name":"传说交易员","description":"证券或期货账户累计交易次数达到300次","bigImgUrl":"https://static.tigerbbs.com/656db16598a0b8f21429e10d6c1cb033","smallImgUrl":"https://static.tigerbbs.com/03f10910d4dd9234f9b5702a3342193a","grayImgUrl":"https://static.tigerbbs.com/0c767e35268feb729d50d3fa9a386c5a","redirectLinkEnabled":0,"redirectLink":null,"hasAllocated":1,"isWearing":0,"stamp":null,"stampPosition":0,"hasStamp":0,"allocationCount":1,"allocatedDate":"2021.12.21","exceedPercentage":"93.86%","individualDisplayEnabled":0,"backgroundColor":null,"fontColor":null,"individualDisplaySort":0,"categoryType":1100}],"userBadgeCount":5,"currentWearingBadge":null,"individualDisplayBadges":null,"crmLevel":7,"crmLevelSwitch":1,"location":"浙江","starInvestorFollowerNum":0,"starInvestorFlag":false,"starInvestorOrderShareNum":0,"subscribeStarInvestorNum":3,"ror":null,"winRationPercentage":null,"showRor":false,"investmentPhilosophy":null,"starInvestorSubscribeFlag":false},"baikeInfo":{},"tab":"hot","tweets":[{"id":652887543,"gmtCreate":1684420023883,"gmtModify":1684420024918,"author":{"id":"3546901181241979","authorId":"3546901181241979","name":"普罗米M修斯","avatar":"https://static.tigerbbs.com/1abedaa616802bfcb0f72a8650637a6b","crmLevel":7,"crmLevelSwitch":1,"followedFlag":false,"idStr":"3546901181241979","authorIdStr":"3546901181241979"},"themes":[],"htmlText":"<a href=\"https://laohu8.com/S/.IXIC\">$纳斯达克(.IXIC)$ </a>这是疯了嘛","listText":"<a href=\"https://laohu8.com/S/.IXIC\">$纳斯达克(.IXIC)$ </a>这是疯了嘛","text":"$纳斯达克(.IXIC)$ 这是疯了嘛","images":[],"top":1,"highlighted":1,"essential":1,"paper":1,"likeSize":1,"commentSize":0,"repostSize":0,"link":"https://laohu8.com/post/652887543","isVote":1,"tweetType":1,"viewCount":1116,"authorTweetTopStatus":1,"verified":2,"comments":[],"imageCount":0,"langContent":"CN","totalScore":0},{"id":651066718,"gmtCreate":1681358584334,"gmtModify":1681358586034,"author":{"id":"3546901181241979","authorId":"3546901181241979","name":"普罗米M修斯","avatar":"https://static.tigerbbs.com/1abedaa616802bfcb0f72a8650637a6b","crmLevel":7,"crmLevelSwitch":1,"followedFlag":false,"idStr":"3546901181241979","authorIdStr":"3546901181241979"},"themes":[],"htmlText":"人类之光","listText":"人类之光","text":"人类之光","images":[],"top":1,"highlighted":1,"essential":1,"paper":1,"likeSize":0,"commentSize":0,"repostSize":0,"link":"https://laohu8.com/post/651066718","repostId":"2327971959","repostType":4,"repost":{"id":"2327971959","kind":"news","pubTimestamp":1681346445,"share":"https://www.laohu8.com/m/news/2327971959?lang=&edition=full","pubTime":"2023-04-13 08:40","market":"us","language":"zh","title":"重磅!微软开源Deep Speed Chat,人人拥有ChatGPT!","url":"https://stock-news.laohu8.com/highlight/detail?id=2327971959","media":"AIGC开放社区","summary":"用户通过Deep Speed Chat提供的“傻瓜式”操作,能以最短的时间、最高效的成本训练类ChatGPT大语言模型,这标志着一个人手一个ChatGPT的时代要来了。4月12日,微软宣布开源了Dee","content":"<html><head></head><body><blockquote>用户通过Deep Speed Chat提供的“傻瓜式”操作,能以最短的时间、最高效的成本训练类ChatGPT大语言模型,这标志着一个人手一个ChatGPT的时代要来了。</blockquote><p>4月12日,<a href=\"https://laohu8.com/S/MSFT\">微软</a>宣布开源了Deep Speed Chat,帮助用户轻松训练类ChatGPT等大语言模型,使得人人都能拥有自己的ChatGPT!(开源地址:https://github.com/microsoft/DeepSpeed) </p><p>据悉,Deep Speed Chat是基于微软Deep Speed深度学习优化库开发而成,具备训练、强化推理等功能,还使用了RLHF(人工反馈机制的强化学习)技术,可将训练速度提升15倍以上,成本却大幅度降低。例如,一个130亿参数的类ChatGPT模型,只需1.25小时就能完成训练。</p><p>简单来说,<strong>用户通过Deep Speed Chat提供的“傻瓜式”操作,能以最短的时间、最高效的成本训练类ChatGPT大语言模型,这标志着一个人手一个ChatGPT的时代要来了。</strong></p><h2>微软为什么开源Deep Speed Chat</h2><p>ChatGPT只用了几个月的时间便席卷全球,成为史上用户增长速度最快的消费级应用,对世界的经济发展、科技研发起到了重要的推动作用。很多企业、个人用户纷纷想参与到这场AI技术变革中,但由于庞大的资源、资金投入只能望而却步。</p><p>即便开源领域推出了LLaMa、Alpaca、Vicuna、Databricks-Dolly等优秀的类ChatGPT项目,<strong>仍缺乏一个支持端到端的基于RLHF的规模化系统,使得训练强大的类ChatGPT模型非常艰难</strong>。</p><p>例如,使用上述开源项目训练一个67亿参数的类ChatGPT模型,通常需要昂贵的多卡至多节点的GPU集群,但这些资源对大多数科学家、小型企业、个人开发者而言难以获取。</p><p><strong>即便是拥有这样的资源,现有的开源系统的训练效率通常还不到,这些机器所能达到的最大效率的5%</strong>。</p><p>现在,通过Deep Speed Chat解决资源、算法等难题,轻松、高效的训练数千亿参数的最先进的类ChatGPT模型。</p><h2>Deep Speed Chat技术简单介绍</h2><p>ChatGPT模型的训练是基于InstructGPT论文中的RLHF方式,这使得现有深度学习系统在训练类ChatGPT模型时存在种种局限。现在,通过Deep Speed Chat可以突破这些训练瓶颈,达到最佳效果。</p><p><strong>Deep Speed Chat拥有强化推理、RLHF模块、RLHF系统三大核心功能。</strong></p><p><strong>简化 ChatGPT 类型模型的训练和强化推理:</strong>只需一个脚本即可实现多个训练步骤,包括使用Huggingface 预训练的模型、使用 DeepSpeed-RLHF 系统运行 InstructGPT 训练的所有三个步骤,生成属于自己的类ChatGPT模型。此外,还提供了一个易于使用的推理API,用于在模型训练后测试对话性能。</p><p><strong>Deep Speed-RLHF 模块:</strong>DeepSpeed-RLHF 复刻了InstructGPT 论文中的训练模式,并确保包括:a) 监督微调(SFT),b) 奖励模型微调和 c) 基于人类反馈的强化学习(RLHF)在内的三个步骤与其对应。此外,还提供了数据抽象和混合功能,以支持用户使用多个不同来源的数据源进行训练。</p><p><strong>DeepSpeed-RLHF系统:</strong>将DeepSpeed的训练和推理能力整合到统一的混合引擎(DeepSpeed Hybrid Engine or DeepSpeed-HE)中用于RLHF 训练。DeepSpeed-HE 能够在RLHF中无缝地在推理和训练模式之间切换,使其能够利用来自 DeepSpeed-Inference 的各种优化。</p><p>例如,张量并行计算和高性能CUDA算子进行语言生成,同时对训练部分还能从 ZeRO- 和 LoRA-based 内存优化策略中受益。DeepSpeed-HE可以自动在 RLHF 的不同阶段进行<a href=\"https://laohu8.com/S/5RE.SI\">智能</a>的内存管理和数据缓存。</p><p>此外,Deep Speed Chat在成本、效率、扩展性等方面非常优秀。</p><p><strong>高效且节省成本:</strong>在Azure云上只需9小时即可训练一个OPT-13B模型,只需18小时即可训练一个 OPT-30B模型。这两种训练分别花费不到300美元和600 美元。</p><p><strong>卓越扩展性:</strong>能够支持训练拥有数千亿参数的模型,并在多节点多 GPU 系统上展现出卓越的扩展性。因此,即使是一个拥有 130 亿参数的模型,也只需 1.25 小时就能完成训练。而对于拥1750亿参数的模型,使用Deep Speed Chat进行训练也只需不到一天的时间。</p><p>注意:在Azure上训练时由于GPU不同,所以训练时间和费用也各不相同。</p><h2>全民ChatGPT时代来了</h2><p>做个不太恰当的比喻,微软此次开源Deep Speed Chat就像当年的互联网,为了提升发展速率降低了台式机的价格,使得普通用户也能拥有自己的电脑。</p><p>只有参与的用户多了,整个行业生态才能快速壮大发展,所以,微软此次开源对整个ChatGPT生态发展起到了至关重要的推动作用,使得人人都能拥有自己的ChatGPT。</p><p>目前,ChatGPT处于初级发展阶段,存在安全隐患、数据隐私、道德等问题。相信随着参与用户的增多,这些难题将会被有效的解决,从而为全球经济发展做出贡献。</p></body></html>","source":"lsy1680749794970","collect":0,"html":"<!DOCTYPE html>\n<html>\n<head>\n<meta http-equiv=\"Content-Type\" content=\"text/html; charset=utf-8\" />\n<meta name=\"viewport\" content=\"width=device-width,initial-scale=1.0,minimum-scale=1.0,maximum-scale=1.0,user-scalable=no\"/>\n<meta name=\"format-detection\" content=\"telephone=no,email=no,address=no\" />\n<title>重磅!微软开源Deep Speed Chat,人人拥有ChatGPT!</title>\n<style type=\"text/css\">\na,abbr,acronym,address,applet,article,aside,audio,b,big,blockquote,body,canvas,caption,center,cite,code,dd,del,details,dfn,div,dl,dt,\nem,embed,fieldset,figcaption,figure,footer,form,h1,h2,h3,h4,h5,h6,header,hgroup,html,i,iframe,img,ins,kbd,label,legend,li,mark,menu,nav,\nobject,ol,output,p,pre,q,ruby,s,samp,section,small,span,strike,strong,sub,summary,sup,table,tbody,td,tfoot,th,thead,time,tr,tt,u,ul,var,video{ font:inherit;margin:0;padding:0;vertical-align:baseline;border:0 }\nbody{ font-size:16px; line-height:1.5; color:#999; background:transparent; }\n.wrapper{ overflow:hidden;word-break:break-all;padding:10px; }\nh1,h2{ font-weight:normal; line-height:1.35; margin-bottom:.6em; }\nh3,h4,h5,h6{ line-height:1.35; margin-bottom:1em; }\nh1{ font-size:24px; }\nh2{ font-size:20px; }\nh3{ font-size:18px; }\nh4{ font-size:16px; }\nh5{ font-size:14px; }\nh6{ font-size:12px; }\np,ul,ol,blockquote,dl,table{ margin:1.2em 0; }\nul,ol{ margin-left:2em; }\nul{ list-style:disc; }\nol{ list-style:decimal; }\nli,li p{ margin:10px 0;}\nimg{ max-width:100%;display:block;margin:0 auto 1em; }\nblockquote{ color:#B5B2B1; border-left:3px solid #aaa; padding:1em; }\nstrong,b{font-weight:bold;}\nem,i{font-style:italic;}\ntable{ width:100%;border-collapse:collapse;border-spacing:1px;margin:1em 0;font-size:.9em; }\nth,td{ padding:5px;text-align:left;border:1px solid #aaa; }\nth{ font-weight:bold;background:#5d5d5d; }\n.symbol-link{font-weight:bold;}\n/* header{ border-bottom:1px solid #494756; } */\n.title{ margin:0 0 8px;line-height:1.3;color:#ddd; }\n.meta {color:#5e5c6d;font-size:13px;margin:0 0 .5em; }\na{text-decoration:none; color:#2a4b87;}\n.meta .head { display: inline-block; overflow: hidden}\n.head .h-thumb { width: 30px; height: 30px; margin: 0; padding: 0; border-radius: 50%; float: left;}\n.head .h-content { margin: 0; padding: 0 0 0 9px; float: left;}\n.head .h-name {font-size: 13px; color: #eee; margin: 0;}\n.head .h-time {font-size: 11px; color: #7E829C; margin: 0;line-height: 11px;}\n.small {font-size: 12.5px; display: inline-block; transform: scale(0.9); -webkit-transform: scale(0.9); transform-origin: left; -webkit-transform-origin: left;}\n.smaller {font-size: 12.5px; display: inline-block; transform: scale(0.8); -webkit-transform: scale(0.8); transform-origin: left; -webkit-transform-origin: left;}\n.bt-text {font-size: 12px;margin: 1.5em 0 0 0}\n.bt-text p {margin: 0}\n</style>\n</head>\n<body>\n<div class=\"wrapper\">\n<header>\n<h2 class=\"title\">\n重磅!微软开源Deep Speed Chat,人人拥有ChatGPT!\n</h2>\n\n<h4 class=\"meta\">\n\n\n2023-04-13 08:40 北京时间 <a href=https://mp.weixin.qq.com/s?__biz=Mzg3Mzg5MjY3Nw%3D%3D&mid=2247486633&idx=1&sn=086d517293533ee3bf6e1c8177bc4127&chksm=ced85283f9afdb95175af01642084bdf013d11dc55a6297da4397aba22a80a2afa98f6a12851&mpshare=1&scene=23&sharer_sharetime=1681345548401&sharer_shareid=00a55b671777cf0e253d4693000ead51#rd><strong>AIGC开放社区</strong></a>\n\n\n</h4>\n\n</header>\n<article>\n<div>\n<p>用户通过Deep Speed Chat提供的“傻瓜式”操作,能以最短的时间、最高效的成本训练类ChatGPT大语言模型,这标志着一个人手一个ChatGPT的时代要来了。4月12日,微软宣布开源了Deep Speed Chat,帮助用户轻松训练类ChatGPT等大语言模型,使得人人都能拥有自己的ChatGPT!(开源地址:https://github.com/microsoft/DeepSpeed)...</p>\n\n<a href=\"https://mp.weixin.qq.com/s?__biz=Mzg3Mzg5MjY3Nw%3D%3D&mid=2247486633&idx=1&sn=086d517293533ee3bf6e1c8177bc4127&chksm=ced85283f9afdb95175af01642084bdf013d11dc55a6297da4397aba22a80a2afa98f6a12851&mpshare=1&scene=23&sharer_sharetime=1681345548401&sharer_shareid=00a55b671777cf0e253d4693000ead51#rd\">Web Link</a>\n\n</div>\n\n\n</article>\n</div>\n</body>\n</html>\n","type":0,"thumbnail":"https://static.tigerbbs.com/3b52764921272100b8f755870954f0a5","relate_stocks":{"LU0080751232.USD":"富达环球多元动力基金A","LU0061474960.USD":"天利环球焦点基金AU Acc","BK4548":"巴美列捷福持仓","IE00BSNM7G36.USD":"NEUBERGER BERMAN SYSTEMATIC GLOBAL SUSTAINABLE VALUE \"A\" (USD) ACC","LU0079474960.USD":"联博美国增长基金A","LU0234572021.USD":"高盛美国核心股票组合Acc","BK4516":"特朗普概念","BK4528":"SaaS概念","IE00BZ1G4Q59.USD":"LEGG MASON CLEARBRIDGE US EQUITY SUSTAINABILITY LEADER \"A\"(USD) INC (A)","BK4554":"元宇宙及AR概念","LU0289739343.SGD":"SUSTAINABLE GLOBAL THEMATIC PORTFOLIO \"A\" (SGD) ACC","BK4532":"文艺复兴科技持仓","LU0011850046.USD":"贝莱德全球长线股票 A2 USD","IE00BWXC8680.SGD":"PINEBRIDGE US LARGE CAP RESEARCH ENHANCED \"A5\" (SGD) ACC","LU0061474705.USD":"THREADNEEDLE (LUX) GLOBAL DYNAMIC REAL RETURN \"AU\" (USD) ACC","LU0097036916.USD":"贝莱德美国增长A2 USD","BK4534":"瑞士信贷持仓","BK4585":"ETF&股票定投概念","MSFT":"微软","BK4576":"AR","LU0198837287.USD":"UBS (LUX) EQUITY SICAV - USA GROWTH \"P\" (USD) ACC","IE00B1BXHZ80.USD":"Legg Mason ClearBridge - US Appreciation A Acc USD","BK4533":"AQR资本管理(全球第二大对冲基金)","LU0289961442.SGD":"SUSTAINABLE GLOBAL THEMATIC PORTFOLIO \"AX\" (SGD) ACC","LU0061475181.USD":"THREADNEEDLE (LUX) AMERICAN \"AU\" (USD) ACC","IE00BFSS8Q28.SGD":"Janus Henderson Balanced A Inc SGD-H","BK4525":"远程办公概念","BK4535":"淡马锡持仓","IE00BKVL7J92.USD":"Legg Mason ClearBridge - US Equity Sustainability Leaders A Acc USD","BK4577":"网络游戏","LU0127658192.USD":"EASTSPRING INVESTMENTS GLOBAL TECHNOLOGY \"A\" (USD) ACC","IE00BJTD4N35.SGD":"Neuberger Berman US Long Short Equity A1 Acc SGD-H","BK4527":"明星科技股","BK4538":"云计算","LU0289941410.SGD":"AB FCP I Dynamic Diversified AX SGD","IE00B7KXQ091.USD":"Janus Henderson Balanced A Inc USD","IE00BLSP4239.USD":"Legg Mason ClearBridge - Tactical Dividend Income A Mdis USD Plus","BK4588":"碎股","IE00BLSP4452.SGD":"Legg Mason ClearBridge - Tactical Dividend Income A Mdis SGD-H Plus","IE00BFSS7M15.SGD":"Janus Henderson Balanced A Acc SGD-H","BK4503":"景林资产持仓","IE00B775SV38.USD":"NEUBERGER BERMAN US MULTICAP OPPORTUNITIES \"A\" (USD) ACC","LU0170899867.USD":"EASTSPRING INVESTMENTS WORLD VALUE EQUITY \"A\" (USD) ACC","LU0109391861.USD":"富兰克林美国机遇基金A Acc","IE0004445239.USD":"JANUS HENDERSON US FORTY \"A2\" (USD) ACC","IE00B19Z9505.USD":"美盛-美国大盘成长股A Acc","IE00BJJMRX11.SGD":"Janus Henderson Balanced A Acc SGD","IE00BJTD4V19.USD":"NEUBERGER BERMAN US LONG SHORT EQUITY \"A1\" (USD) ACC","LU0053666078.USD":"摩根大通基金-美国股票A(离岸)美元","BK4581":"高盛持仓","LU0082616367.USD":"摩根大通美国科技A(dist)"},"source_url":"https://mp.weixin.qq.com/s?__biz=Mzg3Mzg5MjY3Nw%3D%3D&mid=2247486633&idx=1&sn=086d517293533ee3bf6e1c8177bc4127&chksm=ced85283f9afdb95175af01642084bdf013d11dc55a6297da4397aba22a80a2afa98f6a12851&mpshare=1&scene=23&sharer_sharetime=1681345548401&sharer_shareid=00a55b671777cf0e253d4693000ead51#rd","is_english":false,"share_image_url":"https://static.laohu8.com/e9f99090a1c2ed51c021029395664489","article_id":"2327971959","content_text":"用户通过Deep Speed Chat提供的“傻瓜式”操作,能以最短的时间、最高效的成本训练类ChatGPT大语言模型,这标志着一个人手一个ChatGPT的时代要来了。4月12日,微软宣布开源了Deep Speed Chat,帮助用户轻松训练类ChatGPT等大语言模型,使得人人都能拥有自己的ChatGPT!(开源地址:https://github.com/microsoft/DeepSpeed) 据悉,Deep Speed Chat是基于微软Deep Speed深度学习优化库开发而成,具备训练、强化推理等功能,还使用了RLHF(人工反馈机制的强化学习)技术,可将训练速度提升15倍以上,成本却大幅度降低。例如,一个130亿参数的类ChatGPT模型,只需1.25小时就能完成训练。简单来说,用户通过Deep Speed Chat提供的“傻瓜式”操作,能以最短的时间、最高效的成本训练类ChatGPT大语言模型,这标志着一个人手一个ChatGPT的时代要来了。微软为什么开源Deep Speed ChatChatGPT只用了几个月的时间便席卷全球,成为史上用户增长速度最快的消费级应用,对世界的经济发展、科技研发起到了重要的推动作用。很多企业、个人用户纷纷想参与到这场AI技术变革中,但由于庞大的资源、资金投入只能望而却步。即便开源领域推出了LLaMa、Alpaca、Vicuna、Databricks-Dolly等优秀的类ChatGPT项目,仍缺乏一个支持端到端的基于RLHF的规模化系统,使得训练强大的类ChatGPT模型非常艰难。例如,使用上述开源项目训练一个67亿参数的类ChatGPT模型,通常需要昂贵的多卡至多节点的GPU集群,但这些资源对大多数科学家、小型企业、个人开发者而言难以获取。即便是拥有这样的资源,现有的开源系统的训练效率通常还不到,这些机器所能达到的最大效率的5%。现在,通过Deep Speed Chat解决资源、算法等难题,轻松、高效的训练数千亿参数的最先进的类ChatGPT模型。Deep Speed Chat技术简单介绍ChatGPT模型的训练是基于InstructGPT论文中的RLHF方式,这使得现有深度学习系统在训练类ChatGPT模型时存在种种局限。现在,通过Deep Speed Chat可以突破这些训练瓶颈,达到最佳效果。Deep Speed Chat拥有强化推理、RLHF模块、RLHF系统三大核心功能。简化 ChatGPT 类型模型的训练和强化推理:只需一个脚本即可实现多个训练步骤,包括使用Huggingface 预训练的模型、使用 DeepSpeed-RLHF 系统运行 InstructGPT 训练的所有三个步骤,生成属于自己的类ChatGPT模型。此外,还提供了一个易于使用的推理API,用于在模型训练后测试对话性能。Deep Speed-RLHF 模块:DeepSpeed-RLHF 复刻了InstructGPT 论文中的训练模式,并确保包括:a) 监督微调(SFT),b) 奖励模型微调和 c) 基于人类反馈的强化学习(RLHF)在内的三个步骤与其对应。此外,还提供了数据抽象和混合功能,以支持用户使用多个不同来源的数据源进行训练。DeepSpeed-RLHF系统:将DeepSpeed的训练和推理能力整合到统一的混合引擎(DeepSpeed Hybrid Engine or DeepSpeed-HE)中用于RLHF 训练。DeepSpeed-HE 能够在RLHF中无缝地在推理和训练模式之间切换,使其能够利用来自 DeepSpeed-Inference 的各种优化。例如,张量并行计算和高性能CUDA算子进行语言生成,同时对训练部分还能从 ZeRO- 和 LoRA-based 内存优化策略中受益。DeepSpeed-HE可以自动在 RLHF 的不同阶段进行智能的内存管理和数据缓存。此外,Deep Speed Chat在成本、效率、扩展性等方面非常优秀。高效且节省成本:在Azure云上只需9小时即可训练一个OPT-13B模型,只需18小时即可训练一个 OPT-30B模型。这两种训练分别花费不到300美元和600 美元。卓越扩展性:能够支持训练拥有数千亿参数的模型,并在多节点多 GPU 系统上展现出卓越的扩展性。因此,即使是一个拥有 130 亿参数的模型,也只需 1.25 小时就能完成训练。而对于拥1750亿参数的模型,使用Deep Speed Chat进行训练也只需不到一天的时间。注意:在Azure上训练时由于GPU不同,所以训练时间和费用也各不相同。全民ChatGPT时代来了做个不太恰当的比喻,微软此次开源Deep Speed Chat就像当年的互联网,为了提升发展速率降低了台式机的价格,使得普通用户也能拥有自己的电脑。只有参与的用户多了,整个行业生态才能快速壮大发展,所以,微软此次开源对整个ChatGPT生态发展起到了至关重要的推动作用,使得人人都能拥有自己的ChatGPT。目前,ChatGPT处于初级发展阶段,存在安全隐患、数据隐私、道德等问题。相信随着参与用户的增多,这些难题将会被有效的解决,从而为全球经济发展做出贡献。","news_type":1},"isVote":1,"tweetType":1,"viewCount":1135,"authorTweetTopStatus":1,"verified":2,"comments":[],"imageCount":0,"langContent":"CN","totalScore":0},{"id":368078252,"gmtCreate":1614268606276,"gmtModify":1614268606276,"author":{"id":"3546901181241979","authorId":"3546901181241979","name":"普罗米M修斯","avatar":"https://static.tigerbbs.com/1abedaa616802bfcb0f72a8650637a6b","crmLevel":7,"crmLevelSwitch":1,"followedFlag":false,"idStr":"3546901181241979","authorIdStr":"3546901181241979"},"themes":[],"htmlText":"<a target=\"_blank\" href=\"https://laohu8.com/S/TLT\">$债券20+美公债指数ETF-iShares(TLT)$</a> 这波 会跌到多少","listText":"<a target=\"_blank\" href=\"https://laohu8.com/S/TLT\">$债券20+美公债指数ETF-iShares(TLT)$</a> 这波 会跌到多少","text":"$债券20+美公债指数ETF-iShares(TLT)$ 这波 会跌到多少","images":[],"top":1,"highlighted":1,"essential":1,"paper":1,"likeSize":0,"commentSize":0,"repostSize":0,"link":"https://laohu8.com/post/368078252","isVote":1,"tweetType":1,"viewCount":1797,"authorTweetTopStatus":1,"verified":2,"comments":[],"imageCount":0,"langContent":"CN","totalScore":0},{"id":311566596,"gmtCreate":1611815207115,"gmtModify":1703754167976,"author":{"id":"3546901181241979","authorId":"3546901181241979","name":"普罗米M修斯","avatar":"https://static.tigerbbs.com/1abedaa616802bfcb0f72a8650637a6b","crmLevel":7,"crmLevelSwitch":1,"followedFlag":false,"idStr":"3546901181241979","authorIdStr":"3546901181241979"},"themes":[],"htmlText":"<a target=\"_blank\" href=\"https://laohu8.com/S/JPM\">$摩根大通(JPM)$</a> <a target=\"_blank\" href=\"https://laohu8.com/S/C\">$花旗(C)$</a> <a target=\"_blank\" href=\"https://laohu8.com/S/WFC\">$富国银行(WFC)$</a> <a target=\"_blank\" href=\"https://laohu8.com/S/BAC\">$美国银行(BAC)$</a> 四大行哪个基本面更好?","listText":"<a target=\"_blank\" href=\"https://laohu8.com/S/JPM\">$摩根大通(JPM)$</a> <a target=\"_blank\" href=\"https://laohu8.com/S/C\">$花旗(C)$</a> <a target=\"_blank\" href=\"https://laohu8.com/S/WFC\">$富国银行(WFC)$</a> <a target=\"_blank\" href=\"https://laohu8.com/S/BAC\">$美国银行(BAC)$</a> 四大行哪个基本面更好?","text":"$摩根大通(JPM)$ $花旗(C)$ $富国银行(WFC)$ $美国银行(BAC)$ 四大行哪个基本面更好?","images":[],"top":1,"highlighted":1,"essential":1,"paper":1,"likeSize":4,"commentSize":0,"repostSize":0,"link":"https://laohu8.com/post/311566596","isVote":1,"tweetType":1,"viewCount":2337,"authorTweetTopStatus":1,"verified":2,"comments":[],"imageCount":0,"langContent":"CN","totalScore":0},{"id":311272737,"gmtCreate":1611806488129,"gmtModify":1703754083077,"author":{"id":"3546901181241979","authorId":"3546901181241979","name":"普罗米M修斯","avatar":"https://static.tigerbbs.com/1abedaa616802bfcb0f72a8650637a6b","crmLevel":7,"crmLevelSwitch":1,"followedFlag":false,"idStr":"3546901181241979","authorIdStr":"3546901181241979"},"themes":[],"htmlText":"<a target=\"_blank\" href=\"https://laohu8.com/S/AAPL\">$苹果(AAPL)$</a> 盘后跌了这么多,有妖?","listText":"<a target=\"_blank\" href=\"https://laohu8.com/S/AAPL\">$苹果(AAPL)$</a> 盘后跌了这么多,有妖?","text":"$苹果(AAPL)$ 盘后跌了这么多,有妖?","images":[],"top":1,"highlighted":1,"essential":1,"paper":1,"likeSize":1,"commentSize":0,"repostSize":0,"link":"https://laohu8.com/post/311272737","isVote":1,"tweetType":1,"viewCount":962,"authorTweetTopStatus":1,"verified":2,"comments":[],"imageCount":0,"langContent":"CN","totalScore":0},{"id":319490398,"gmtCreate":1611592120514,"gmtModify":1703751493213,"author":{"id":"3546901181241979","authorId":"3546901181241979","name":"普罗米M修斯","avatar":"https://static.tigerbbs.com/1abedaa616802bfcb0f72a8650637a6b","crmLevel":7,"crmLevelSwitch":1,"followedFlag":false,"idStr":"3546901181241979","authorIdStr":"3546901181241979"},"themes":[],"htmlText":"<a target=\"_blank\" href=\"https://laohu8.com/S/AAPL\">$苹果(AAPL)$</a> 杀期权?[贱笑]","listText":"<a target=\"_blank\" href=\"https://laohu8.com/S/AAPL\">$苹果(AAPL)$</a> 杀期权?[贱笑]","text":"$苹果(AAPL)$ 杀期权?[贱笑]","images":[],"top":1,"highlighted":1,"essential":1,"paper":1,"likeSize":0,"commentSize":0,"repostSize":0,"link":"https://laohu8.com/post/319490398","isVote":1,"tweetType":1,"viewCount":1089,"authorTweetTopStatus":1,"verified":2,"comments":[],"imageCount":0,"langContent":"CN","totalScore":0},{"id":334055853,"gmtCreate":1610676587872,"gmtModify":1703746159231,"author":{"id":"3546901181241979","authorId":"3546901181241979","name":"普罗米M修斯","avatar":"https://static.tigerbbs.com/1abedaa616802bfcb0f72a8650637a6b","crmLevel":7,"crmLevelSwitch":1,"followedFlag":false,"idStr":"3546901181241979","authorIdStr":"3546901181241979"},"themes":[],"htmlText":"<a target=\"_blank\" href=\"https://laohu8.com/S/WFC\">$富国银行(WFC)$</a> 赌一波财报,short put 准备接货","listText":"<a target=\"_blank\" href=\"https://laohu8.com/S/WFC\">$富国银行(WFC)$</a> 赌一波财报,short put 准备接货","text":"$富国银行(WFC)$ 赌一波财报,short put 准备接货","images":[],"top":1,"highlighted":1,"essential":1,"paper":1,"likeSize":0,"commentSize":0,"repostSize":0,"link":"https://laohu8.com/post/334055853","isVote":1,"tweetType":1,"viewCount":1336,"authorTweetTopStatus":1,"verified":2,"comments":[],"imageCount":0,"langContent":"CN","totalScore":0}],"hots":[{"id":311566596,"gmtCreate":1611815207115,"gmtModify":1703754167976,"author":{"id":"3546901181241979","authorId":"3546901181241979","name":"普罗米M修斯","avatar":"https://static.tigerbbs.com/1abedaa616802bfcb0f72a8650637a6b","crmLevel":7,"crmLevelSwitch":1,"followedFlag":false,"idStr":"3546901181241979","authorIdStr":"3546901181241979"},"themes":[],"htmlText":"<a target=\"_blank\" href=\"https://laohu8.com/S/JPM\">$摩根大通(JPM)$</a> <a target=\"_blank\" href=\"https://laohu8.com/S/C\">$花旗(C)$</a> <a target=\"_blank\" href=\"https://laohu8.com/S/WFC\">$富国银行(WFC)$</a> <a target=\"_blank\" href=\"https://laohu8.com/S/BAC\">$美国银行(BAC)$</a> 四大行哪个基本面更好?","listText":"<a target=\"_blank\" href=\"https://laohu8.com/S/JPM\">$摩根大通(JPM)$</a> <a target=\"_blank\" href=\"https://laohu8.com/S/C\">$花旗(C)$</a> <a target=\"_blank\" href=\"https://laohu8.com/S/WFC\">$富国银行(WFC)$</a> <a target=\"_blank\" href=\"https://laohu8.com/S/BAC\">$美国银行(BAC)$</a> 四大行哪个基本面更好?","text":"$摩根大通(JPM)$ $花旗(C)$ $富国银行(WFC)$ $美国银行(BAC)$ 四大行哪个基本面更好?","images":[],"top":1,"highlighted":1,"essential":1,"paper":1,"likeSize":4,"commentSize":0,"repostSize":0,"link":"https://laohu8.com/post/311566596","isVote":1,"tweetType":1,"viewCount":2337,"authorTweetTopStatus":1,"verified":2,"comments":[],"imageCount":0,"langContent":"CN","totalScore":0},{"id":652887543,"gmtCreate":1684420023883,"gmtModify":1684420024918,"author":{"id":"3546901181241979","authorId":"3546901181241979","name":"普罗米M修斯","avatar":"https://static.tigerbbs.com/1abedaa616802bfcb0f72a8650637a6b","crmLevel":7,"crmLevelSwitch":1,"followedFlag":false,"idStr":"3546901181241979","authorIdStr":"3546901181241979"},"themes":[],"htmlText":"<a href=\"https://laohu8.com/S/.IXIC\">$纳斯达克(.IXIC)$ </a>这是疯了嘛","listText":"<a href=\"https://laohu8.com/S/.IXIC\">$纳斯达克(.IXIC)$ </a>这是疯了嘛","text":"$纳斯达克(.IXIC)$ 这是疯了嘛","images":[],"top":1,"highlighted":1,"essential":1,"paper":1,"likeSize":1,"commentSize":0,"repostSize":0,"link":"https://laohu8.com/post/652887543","isVote":1,"tweetType":1,"viewCount":1116,"authorTweetTopStatus":1,"verified":2,"comments":[],"imageCount":0,"langContent":"CN","totalScore":0},{"id":311272737,"gmtCreate":1611806488129,"gmtModify":1703754083077,"author":{"id":"3546901181241979","authorId":"3546901181241979","name":"普罗米M修斯","avatar":"https://static.tigerbbs.com/1abedaa616802bfcb0f72a8650637a6b","crmLevel":7,"crmLevelSwitch":1,"followedFlag":false,"idStr":"3546901181241979","authorIdStr":"3546901181241979"},"themes":[],"htmlText":"<a target=\"_blank\" href=\"https://laohu8.com/S/AAPL\">$苹果(AAPL)$</a> 盘后跌了这么多,有妖?","listText":"<a target=\"_blank\" href=\"https://laohu8.com/S/AAPL\">$苹果(AAPL)$</a> 盘后跌了这么多,有妖?","text":"$苹果(AAPL)$ 盘后跌了这么多,有妖?","images":[],"top":1,"highlighted":1,"essential":1,"paper":1,"likeSize":1,"commentSize":0,"repostSize":0,"link":"https://laohu8.com/post/311272737","isVote":1,"tweetType":1,"viewCount":962,"authorTweetTopStatus":1,"verified":2,"comments":[],"imageCount":0,"langContent":"CN","totalScore":0},{"id":651066718,"gmtCreate":1681358584334,"gmtModify":1681358586034,"author":{"id":"3546901181241979","authorId":"3546901181241979","name":"普罗米M修斯","avatar":"https://static.tigerbbs.com/1abedaa616802bfcb0f72a8650637a6b","crmLevel":7,"crmLevelSwitch":1,"followedFlag":false,"idStr":"3546901181241979","authorIdStr":"3546901181241979"},"themes":[],"htmlText":"人类之光","listText":"人类之光","text":"人类之光","images":[],"top":1,"highlighted":1,"essential":1,"paper":1,"likeSize":0,"commentSize":0,"repostSize":0,"link":"https://laohu8.com/post/651066718","repostId":"2327971959","repostType":4,"repost":{"id":"2327971959","kind":"news","pubTimestamp":1681346445,"share":"https://www.laohu8.com/m/news/2327971959?lang=&edition=full","pubTime":"2023-04-13 08:40","market":"us","language":"zh","title":"重磅!微软开源Deep Speed Chat,人人拥有ChatGPT!","url":"https://stock-news.laohu8.com/highlight/detail?id=2327971959","media":"AIGC开放社区","summary":"用户通过Deep Speed Chat提供的“傻瓜式”操作,能以最短的时间、最高效的成本训练类ChatGPT大语言模型,这标志着一个人手一个ChatGPT的时代要来了。4月12日,微软宣布开源了Dee","content":"<html><head></head><body><blockquote>用户通过Deep Speed Chat提供的“傻瓜式”操作,能以最短的时间、最高效的成本训练类ChatGPT大语言模型,这标志着一个人手一个ChatGPT的时代要来了。</blockquote><p>4月12日,<a href=\"https://laohu8.com/S/MSFT\">微软</a>宣布开源了Deep Speed Chat,帮助用户轻松训练类ChatGPT等大语言模型,使得人人都能拥有自己的ChatGPT!(开源地址:https://github.com/microsoft/DeepSpeed) </p><p>据悉,Deep Speed Chat是基于微软Deep Speed深度学习优化库开发而成,具备训练、强化推理等功能,还使用了RLHF(人工反馈机制的强化学习)技术,可将训练速度提升15倍以上,成本却大幅度降低。例如,一个130亿参数的类ChatGPT模型,只需1.25小时就能完成训练。</p><p>简单来说,<strong>用户通过Deep Speed Chat提供的“傻瓜式”操作,能以最短的时间、最高效的成本训练类ChatGPT大语言模型,这标志着一个人手一个ChatGPT的时代要来了。</strong></p><h2>微软为什么开源Deep Speed Chat</h2><p>ChatGPT只用了几个月的时间便席卷全球,成为史上用户增长速度最快的消费级应用,对世界的经济发展、科技研发起到了重要的推动作用。很多企业、个人用户纷纷想参与到这场AI技术变革中,但由于庞大的资源、资金投入只能望而却步。</p><p>即便开源领域推出了LLaMa、Alpaca、Vicuna、Databricks-Dolly等优秀的类ChatGPT项目,<strong>仍缺乏一个支持端到端的基于RLHF的规模化系统,使得训练强大的类ChatGPT模型非常艰难</strong>。</p><p>例如,使用上述开源项目训练一个67亿参数的类ChatGPT模型,通常需要昂贵的多卡至多节点的GPU集群,但这些资源对大多数科学家、小型企业、个人开发者而言难以获取。</p><p><strong>即便是拥有这样的资源,现有的开源系统的训练效率通常还不到,这些机器所能达到的最大效率的5%</strong>。</p><p>现在,通过Deep Speed Chat解决资源、算法等难题,轻松、高效的训练数千亿参数的最先进的类ChatGPT模型。</p><h2>Deep Speed Chat技术简单介绍</h2><p>ChatGPT模型的训练是基于InstructGPT论文中的RLHF方式,这使得现有深度学习系统在训练类ChatGPT模型时存在种种局限。现在,通过Deep Speed Chat可以突破这些训练瓶颈,达到最佳效果。</p><p><strong>Deep Speed Chat拥有强化推理、RLHF模块、RLHF系统三大核心功能。</strong></p><p><strong>简化 ChatGPT 类型模型的训练和强化推理:</strong>只需一个脚本即可实现多个训练步骤,包括使用Huggingface 预训练的模型、使用 DeepSpeed-RLHF 系统运行 InstructGPT 训练的所有三个步骤,生成属于自己的类ChatGPT模型。此外,还提供了一个易于使用的推理API,用于在模型训练后测试对话性能。</p><p><strong>Deep Speed-RLHF 模块:</strong>DeepSpeed-RLHF 复刻了InstructGPT 论文中的训练模式,并确保包括:a) 监督微调(SFT),b) 奖励模型微调和 c) 基于人类反馈的强化学习(RLHF)在内的三个步骤与其对应。此外,还提供了数据抽象和混合功能,以支持用户使用多个不同来源的数据源进行训练。</p><p><strong>DeepSpeed-RLHF系统:</strong>将DeepSpeed的训练和推理能力整合到统一的混合引擎(DeepSpeed Hybrid Engine or DeepSpeed-HE)中用于RLHF 训练。DeepSpeed-HE 能够在RLHF中无缝地在推理和训练模式之间切换,使其能够利用来自 DeepSpeed-Inference 的各种优化。</p><p>例如,张量并行计算和高性能CUDA算子进行语言生成,同时对训练部分还能从 ZeRO- 和 LoRA-based 内存优化策略中受益。DeepSpeed-HE可以自动在 RLHF 的不同阶段进行<a href=\"https://laohu8.com/S/5RE.SI\">智能</a>的内存管理和数据缓存。</p><p>此外,Deep Speed Chat在成本、效率、扩展性等方面非常优秀。</p><p><strong>高效且节省成本:</strong>在Azure云上只需9小时即可训练一个OPT-13B模型,只需18小时即可训练一个 OPT-30B模型。这两种训练分别花费不到300美元和600 美元。</p><p><strong>卓越扩展性:</strong>能够支持训练拥有数千亿参数的模型,并在多节点多 GPU 系统上展现出卓越的扩展性。因此,即使是一个拥有 130 亿参数的模型,也只需 1.25 小时就能完成训练。而对于拥1750亿参数的模型,使用Deep Speed Chat进行训练也只需不到一天的时间。</p><p>注意:在Azure上训练时由于GPU不同,所以训练时间和费用也各不相同。</p><h2>全民ChatGPT时代来了</h2><p>做个不太恰当的比喻,微软此次开源Deep Speed Chat就像当年的互联网,为了提升发展速率降低了台式机的价格,使得普通用户也能拥有自己的电脑。</p><p>只有参与的用户多了,整个行业生态才能快速壮大发展,所以,微软此次开源对整个ChatGPT生态发展起到了至关重要的推动作用,使得人人都能拥有自己的ChatGPT。</p><p>目前,ChatGPT处于初级发展阶段,存在安全隐患、数据隐私、道德等问题。相信随着参与用户的增多,这些难题将会被有效的解决,从而为全球经济发展做出贡献。</p></body></html>","source":"lsy1680749794970","collect":0,"html":"<!DOCTYPE html>\n<html>\n<head>\n<meta http-equiv=\"Content-Type\" content=\"text/html; charset=utf-8\" />\n<meta name=\"viewport\" content=\"width=device-width,initial-scale=1.0,minimum-scale=1.0,maximum-scale=1.0,user-scalable=no\"/>\n<meta name=\"format-detection\" content=\"telephone=no,email=no,address=no\" />\n<title>重磅!微软开源Deep Speed Chat,人人拥有ChatGPT!</title>\n<style type=\"text/css\">\na,abbr,acronym,address,applet,article,aside,audio,b,big,blockquote,body,canvas,caption,center,cite,code,dd,del,details,dfn,div,dl,dt,\nem,embed,fieldset,figcaption,figure,footer,form,h1,h2,h3,h4,h5,h6,header,hgroup,html,i,iframe,img,ins,kbd,label,legend,li,mark,menu,nav,\nobject,ol,output,p,pre,q,ruby,s,samp,section,small,span,strike,strong,sub,summary,sup,table,tbody,td,tfoot,th,thead,time,tr,tt,u,ul,var,video{ font:inherit;margin:0;padding:0;vertical-align:baseline;border:0 }\nbody{ font-size:16px; line-height:1.5; color:#999; background:transparent; }\n.wrapper{ overflow:hidden;word-break:break-all;padding:10px; }\nh1,h2{ font-weight:normal; line-height:1.35; margin-bottom:.6em; }\nh3,h4,h5,h6{ line-height:1.35; margin-bottom:1em; }\nh1{ font-size:24px; }\nh2{ font-size:20px; }\nh3{ font-size:18px; }\nh4{ font-size:16px; }\nh5{ font-size:14px; }\nh6{ font-size:12px; }\np,ul,ol,blockquote,dl,table{ margin:1.2em 0; }\nul,ol{ margin-left:2em; }\nul{ list-style:disc; }\nol{ list-style:decimal; }\nli,li p{ margin:10px 0;}\nimg{ max-width:100%;display:block;margin:0 auto 1em; }\nblockquote{ color:#B5B2B1; border-left:3px solid #aaa; padding:1em; }\nstrong,b{font-weight:bold;}\nem,i{font-style:italic;}\ntable{ width:100%;border-collapse:collapse;border-spacing:1px;margin:1em 0;font-size:.9em; }\nth,td{ padding:5px;text-align:left;border:1px solid #aaa; }\nth{ font-weight:bold;background:#5d5d5d; }\n.symbol-link{font-weight:bold;}\n/* header{ border-bottom:1px solid #494756; } */\n.title{ margin:0 0 8px;line-height:1.3;color:#ddd; }\n.meta {color:#5e5c6d;font-size:13px;margin:0 0 .5em; }\na{text-decoration:none; color:#2a4b87;}\n.meta .head { display: inline-block; overflow: hidden}\n.head .h-thumb { width: 30px; height: 30px; margin: 0; padding: 0; border-radius: 50%; float: left;}\n.head .h-content { margin: 0; padding: 0 0 0 9px; float: left;}\n.head .h-name {font-size: 13px; color: #eee; margin: 0;}\n.head .h-time {font-size: 11px; color: #7E829C; margin: 0;line-height: 11px;}\n.small {font-size: 12.5px; display: inline-block; transform: scale(0.9); -webkit-transform: scale(0.9); transform-origin: left; -webkit-transform-origin: left;}\n.smaller {font-size: 12.5px; display: inline-block; transform: scale(0.8); -webkit-transform: scale(0.8); transform-origin: left; -webkit-transform-origin: left;}\n.bt-text {font-size: 12px;margin: 1.5em 0 0 0}\n.bt-text p {margin: 0}\n</style>\n</head>\n<body>\n<div class=\"wrapper\">\n<header>\n<h2 class=\"title\">\n重磅!微软开源Deep Speed Chat,人人拥有ChatGPT!\n</h2>\n\n<h4 class=\"meta\">\n\n\n2023-04-13 08:40 北京时间 <a href=https://mp.weixin.qq.com/s?__biz=Mzg3Mzg5MjY3Nw%3D%3D&mid=2247486633&idx=1&sn=086d517293533ee3bf6e1c8177bc4127&chksm=ced85283f9afdb95175af01642084bdf013d11dc55a6297da4397aba22a80a2afa98f6a12851&mpshare=1&scene=23&sharer_sharetime=1681345548401&sharer_shareid=00a55b671777cf0e253d4693000ead51#rd><strong>AIGC开放社区</strong></a>\n\n\n</h4>\n\n</header>\n<article>\n<div>\n<p>用户通过Deep Speed Chat提供的“傻瓜式”操作,能以最短的时间、最高效的成本训练类ChatGPT大语言模型,这标志着一个人手一个ChatGPT的时代要来了。4月12日,微软宣布开源了Deep Speed Chat,帮助用户轻松训练类ChatGPT等大语言模型,使得人人都能拥有自己的ChatGPT!(开源地址:https://github.com/microsoft/DeepSpeed)...</p>\n\n<a href=\"https://mp.weixin.qq.com/s?__biz=Mzg3Mzg5MjY3Nw%3D%3D&mid=2247486633&idx=1&sn=086d517293533ee3bf6e1c8177bc4127&chksm=ced85283f9afdb95175af01642084bdf013d11dc55a6297da4397aba22a80a2afa98f6a12851&mpshare=1&scene=23&sharer_sharetime=1681345548401&sharer_shareid=00a55b671777cf0e253d4693000ead51#rd\">Web Link</a>\n\n</div>\n\n\n</article>\n</div>\n</body>\n</html>\n","type":0,"thumbnail":"https://static.tigerbbs.com/3b52764921272100b8f755870954f0a5","relate_stocks":{"LU0080751232.USD":"富达环球多元动力基金A","LU0061474960.USD":"天利环球焦点基金AU Acc","BK4548":"巴美列捷福持仓","IE00BSNM7G36.USD":"NEUBERGER BERMAN SYSTEMATIC GLOBAL SUSTAINABLE VALUE \"A\" (USD) ACC","LU0079474960.USD":"联博美国增长基金A","LU0234572021.USD":"高盛美国核心股票组合Acc","BK4516":"特朗普概念","BK4528":"SaaS概念","IE00BZ1G4Q59.USD":"LEGG MASON CLEARBRIDGE US EQUITY SUSTAINABILITY LEADER \"A\"(USD) INC (A)","BK4554":"元宇宙及AR概念","LU0289739343.SGD":"SUSTAINABLE GLOBAL THEMATIC PORTFOLIO \"A\" (SGD) ACC","BK4532":"文艺复兴科技持仓","LU0011850046.USD":"贝莱德全球长线股票 A2 USD","IE00BWXC8680.SGD":"PINEBRIDGE US LARGE CAP RESEARCH ENHANCED \"A5\" (SGD) ACC","LU0061474705.USD":"THREADNEEDLE (LUX) GLOBAL DYNAMIC REAL RETURN \"AU\" (USD) ACC","LU0097036916.USD":"贝莱德美国增长A2 USD","BK4534":"瑞士信贷持仓","BK4585":"ETF&股票定投概念","MSFT":"微软","BK4576":"AR","LU0198837287.USD":"UBS (LUX) EQUITY SICAV - USA GROWTH \"P\" (USD) ACC","IE00B1BXHZ80.USD":"Legg Mason ClearBridge - US Appreciation A Acc USD","BK4533":"AQR资本管理(全球第二大对冲基金)","LU0289961442.SGD":"SUSTAINABLE GLOBAL THEMATIC PORTFOLIO \"AX\" (SGD) ACC","LU0061475181.USD":"THREADNEEDLE (LUX) AMERICAN \"AU\" (USD) ACC","IE00BFSS8Q28.SGD":"Janus Henderson Balanced A Inc SGD-H","BK4525":"远程办公概念","BK4535":"淡马锡持仓","IE00BKVL7J92.USD":"Legg Mason ClearBridge - US Equity Sustainability Leaders A Acc USD","BK4577":"网络游戏","LU0127658192.USD":"EASTSPRING INVESTMENTS GLOBAL TECHNOLOGY \"A\" (USD) ACC","IE00BJTD4N35.SGD":"Neuberger Berman US Long Short Equity A1 Acc SGD-H","BK4527":"明星科技股","BK4538":"云计算","LU0289941410.SGD":"AB FCP I Dynamic Diversified AX SGD","IE00B7KXQ091.USD":"Janus Henderson Balanced A Inc USD","IE00BLSP4239.USD":"Legg Mason ClearBridge - Tactical Dividend Income A Mdis USD Plus","BK4588":"碎股","IE00BLSP4452.SGD":"Legg Mason ClearBridge - Tactical Dividend Income A Mdis SGD-H Plus","IE00BFSS7M15.SGD":"Janus Henderson Balanced A Acc SGD-H","BK4503":"景林资产持仓","IE00B775SV38.USD":"NEUBERGER BERMAN US MULTICAP OPPORTUNITIES \"A\" (USD) ACC","LU0170899867.USD":"EASTSPRING INVESTMENTS WORLD VALUE EQUITY \"A\" (USD) ACC","LU0109391861.USD":"富兰克林美国机遇基金A Acc","IE0004445239.USD":"JANUS HENDERSON US FORTY \"A2\" (USD) ACC","IE00B19Z9505.USD":"美盛-美国大盘成长股A Acc","IE00BJJMRX11.SGD":"Janus Henderson Balanced A Acc SGD","IE00BJTD4V19.USD":"NEUBERGER BERMAN US LONG SHORT EQUITY \"A1\" (USD) ACC","LU0053666078.USD":"摩根大通基金-美国股票A(离岸)美元","BK4581":"高盛持仓","LU0082616367.USD":"摩根大通美国科技A(dist)"},"source_url":"https://mp.weixin.qq.com/s?__biz=Mzg3Mzg5MjY3Nw%3D%3D&mid=2247486633&idx=1&sn=086d517293533ee3bf6e1c8177bc4127&chksm=ced85283f9afdb95175af01642084bdf013d11dc55a6297da4397aba22a80a2afa98f6a12851&mpshare=1&scene=23&sharer_sharetime=1681345548401&sharer_shareid=00a55b671777cf0e253d4693000ead51#rd","is_english":false,"share_image_url":"https://static.laohu8.com/e9f99090a1c2ed51c021029395664489","article_id":"2327971959","content_text":"用户通过Deep Speed Chat提供的“傻瓜式”操作,能以最短的时间、最高效的成本训练类ChatGPT大语言模型,这标志着一个人手一个ChatGPT的时代要来了。4月12日,微软宣布开源了Deep Speed Chat,帮助用户轻松训练类ChatGPT等大语言模型,使得人人都能拥有自己的ChatGPT!(开源地址:https://github.com/microsoft/DeepSpeed) 据悉,Deep Speed Chat是基于微软Deep Speed深度学习优化库开发而成,具备训练、强化推理等功能,还使用了RLHF(人工反馈机制的强化学习)技术,可将训练速度提升15倍以上,成本却大幅度降低。例如,一个130亿参数的类ChatGPT模型,只需1.25小时就能完成训练。简单来说,用户通过Deep Speed Chat提供的“傻瓜式”操作,能以最短的时间、最高效的成本训练类ChatGPT大语言模型,这标志着一个人手一个ChatGPT的时代要来了。微软为什么开源Deep Speed ChatChatGPT只用了几个月的时间便席卷全球,成为史上用户增长速度最快的消费级应用,对世界的经济发展、科技研发起到了重要的推动作用。很多企业、个人用户纷纷想参与到这场AI技术变革中,但由于庞大的资源、资金投入只能望而却步。即便开源领域推出了LLaMa、Alpaca、Vicuna、Databricks-Dolly等优秀的类ChatGPT项目,仍缺乏一个支持端到端的基于RLHF的规模化系统,使得训练强大的类ChatGPT模型非常艰难。例如,使用上述开源项目训练一个67亿参数的类ChatGPT模型,通常需要昂贵的多卡至多节点的GPU集群,但这些资源对大多数科学家、小型企业、个人开发者而言难以获取。即便是拥有这样的资源,现有的开源系统的训练效率通常还不到,这些机器所能达到的最大效率的5%。现在,通过Deep Speed Chat解决资源、算法等难题,轻松、高效的训练数千亿参数的最先进的类ChatGPT模型。Deep Speed Chat技术简单介绍ChatGPT模型的训练是基于InstructGPT论文中的RLHF方式,这使得现有深度学习系统在训练类ChatGPT模型时存在种种局限。现在,通过Deep Speed Chat可以突破这些训练瓶颈,达到最佳效果。Deep Speed Chat拥有强化推理、RLHF模块、RLHF系统三大核心功能。简化 ChatGPT 类型模型的训练和强化推理:只需一个脚本即可实现多个训练步骤,包括使用Huggingface 预训练的模型、使用 DeepSpeed-RLHF 系统运行 InstructGPT 训练的所有三个步骤,生成属于自己的类ChatGPT模型。此外,还提供了一个易于使用的推理API,用于在模型训练后测试对话性能。Deep Speed-RLHF 模块:DeepSpeed-RLHF 复刻了InstructGPT 论文中的训练模式,并确保包括:a) 监督微调(SFT),b) 奖励模型微调和 c) 基于人类反馈的强化学习(RLHF)在内的三个步骤与其对应。此外,还提供了数据抽象和混合功能,以支持用户使用多个不同来源的数据源进行训练。DeepSpeed-RLHF系统:将DeepSpeed的训练和推理能力整合到统一的混合引擎(DeepSpeed Hybrid Engine or DeepSpeed-HE)中用于RLHF 训练。DeepSpeed-HE 能够在RLHF中无缝地在推理和训练模式之间切换,使其能够利用来自 DeepSpeed-Inference 的各种优化。例如,张量并行计算和高性能CUDA算子进行语言生成,同时对训练部分还能从 ZeRO- 和 LoRA-based 内存优化策略中受益。DeepSpeed-HE可以自动在 RLHF 的不同阶段进行智能的内存管理和数据缓存。此外,Deep Speed Chat在成本、效率、扩展性等方面非常优秀。高效且节省成本:在Azure云上只需9小时即可训练一个OPT-13B模型,只需18小时即可训练一个 OPT-30B模型。这两种训练分别花费不到300美元和600 美元。卓越扩展性:能够支持训练拥有数千亿参数的模型,并在多节点多 GPU 系统上展现出卓越的扩展性。因此,即使是一个拥有 130 亿参数的模型,也只需 1.25 小时就能完成训练。而对于拥1750亿参数的模型,使用Deep Speed Chat进行训练也只需不到一天的时间。注意:在Azure上训练时由于GPU不同,所以训练时间和费用也各不相同。全民ChatGPT时代来了做个不太恰当的比喻,微软此次开源Deep Speed Chat就像当年的互联网,为了提升发展速率降低了台式机的价格,使得普通用户也能拥有自己的电脑。只有参与的用户多了,整个行业生态才能快速壮大发展,所以,微软此次开源对整个ChatGPT生态发展起到了至关重要的推动作用,使得人人都能拥有自己的ChatGPT。目前,ChatGPT处于初级发展阶段,存在安全隐患、数据隐私、道德等问题。相信随着参与用户的增多,这些难题将会被有效的解决,从而为全球经济发展做出贡献。","news_type":1},"isVote":1,"tweetType":1,"viewCount":1135,"authorTweetTopStatus":1,"verified":2,"comments":[],"imageCount":0,"langContent":"CN","totalScore":0},{"id":368078252,"gmtCreate":1614268606276,"gmtModify":1614268606276,"author":{"id":"3546901181241979","authorId":"3546901181241979","name":"普罗米M修斯","avatar":"https://static.tigerbbs.com/1abedaa616802bfcb0f72a8650637a6b","crmLevel":7,"crmLevelSwitch":1,"followedFlag":false,"idStr":"3546901181241979","authorIdStr":"3546901181241979"},"themes":[],"htmlText":"<a target=\"_blank\" href=\"https://laohu8.com/S/TLT\">$债券20+美公债指数ETF-iShares(TLT)$</a> 这波 会跌到多少","listText":"<a target=\"_blank\" href=\"https://laohu8.com/S/TLT\">$债券20+美公债指数ETF-iShares(TLT)$</a> 这波 会跌到多少","text":"$债券20+美公债指数ETF-iShares(TLT)$ 这波 会跌到多少","images":[],"top":1,"highlighted":1,"essential":1,"paper":1,"likeSize":0,"commentSize":0,"repostSize":0,"link":"https://laohu8.com/post/368078252","isVote":1,"tweetType":1,"viewCount":1797,"authorTweetTopStatus":1,"verified":2,"comments":[],"imageCount":0,"langContent":"CN","totalScore":0},{"id":319490398,"gmtCreate":1611592120514,"gmtModify":1703751493213,"author":{"id":"3546901181241979","authorId":"3546901181241979","name":"普罗米M修斯","avatar":"https://static.tigerbbs.com/1abedaa616802bfcb0f72a8650637a6b","crmLevel":7,"crmLevelSwitch":1,"followedFlag":false,"idStr":"3546901181241979","authorIdStr":"3546901181241979"},"themes":[],"htmlText":"<a target=\"_blank\" href=\"https://laohu8.com/S/AAPL\">$苹果(AAPL)$</a> 杀期权?[贱笑]","listText":"<a target=\"_blank\" href=\"https://laohu8.com/S/AAPL\">$苹果(AAPL)$</a> 杀期权?[贱笑]","text":"$苹果(AAPL)$ 杀期权?[贱笑]","images":[],"top":1,"highlighted":1,"essential":1,"paper":1,"likeSize":0,"commentSize":0,"repostSize":0,"link":"https://laohu8.com/post/319490398","isVote":1,"tweetType":1,"viewCount":1089,"authorTweetTopStatus":1,"verified":2,"comments":[],"imageCount":0,"langContent":"CN","totalScore":0},{"id":334055853,"gmtCreate":1610676587872,"gmtModify":1703746159231,"author":{"id":"3546901181241979","authorId":"3546901181241979","name":"普罗米M修斯","avatar":"https://static.tigerbbs.com/1abedaa616802bfcb0f72a8650637a6b","crmLevel":7,"crmLevelSwitch":1,"followedFlag":false,"idStr":"3546901181241979","authorIdStr":"3546901181241979"},"themes":[],"htmlText":"<a target=\"_blank\" href=\"https://laohu8.com/S/WFC\">$富国银行(WFC)$</a> 赌一波财报,short put 准备接货","listText":"<a target=\"_blank\" href=\"https://laohu8.com/S/WFC\">$富国银行(WFC)$</a> 赌一波财报,short put 准备接货","text":"$富国银行(WFC)$ 赌一波财报,short put 准备接货","images":[],"top":1,"highlighted":1,"essential":1,"paper":1,"likeSize":0,"commentSize":0,"repostSize":0,"link":"https://laohu8.com/post/334055853","isVote":1,"tweetType":1,"viewCount":1336,"authorTweetTopStatus":1,"verified":2,"comments":[],"imageCount":0,"langContent":"CN","totalScore":0}],"lives":[]}