腾讯混元-TurboS：首个混合Transformer-Mamba MoE超大模型来了

AI寒武纪2025-03-11

腾讯混元团队发布了他们最新的力作——Hunyuan-TurboS，首个混合Transformer-Mamba MoE架构的超大模型大家都知道，传统Transformer模型在处理长文本时一直面临挑战，O(N²)的复杂度以及KV-Cache问题让长文本训练和推理效率大打折扣。而这次，混元-TurboS巧妙地融合了Mamba和Transformer两种架构的优势：• Mamba的高效长序列处理能力•...

网页链接

免责声明：本文观点仅代表作者个人观点，不构成本平台的投资建议，本平台不对文章信息准确性、完整性和及时性做出任何保证，亦不对因使用或信赖文章信息引发的任何损失承担责任。

精彩评论

我们需要你的真知灼见来填补这片空白

发表看法

{"i18n":{"language":"zh_CN"},"isChannel":false,"data":{"share":"https://www.laohunote.com/m/news/2518372738?lang=zh_CN&edition=full","thumbnail":"","is_english":false,"pubTime":"2025-03-11 07:47","share_image_url":"https://static.laohu8.com/e9f99090a1c2ed51c021029395664489","id":"2518372738","market":"hk","top_or_hot":-1,"title":"腾讯混元-TurboS：首个混合Transformer-Mamba MoE超大模型来了","media":"AI寒武纪","content":"<div>\n<p>腾讯混元团队发布了他们最新的力作——Hunyuan-TurboS，首个混合Transformer-Mamba MoE架构的超大模型大家都知道，传统Transformer模型在处理长文本时一直面临挑战，O(N²)的复杂度以及KV-Cache问题让长文本训练和推理效率大打折扣。 而这次，混元-TurboS巧妙地融合了Mamba和Transformer两种架构的优势：• Mamba的高效长序列处理能力•...</p>\n\n<a href=\"https://tech.ifeng.com/c/8hbYegR6fAG\">网页链接</a>\n\n</div>\n","source":"fenghuang_stock","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>腾讯混元-TurboS：首个混合Transformer-Mamba MoE超大模型来了</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腾讯混元-TurboS：首个混合Transformer-Mamba MoE超大模型来了\n</h2>\n\n<h4 class=\"meta\">\n\n\n2025-03-11 07:47 北京时间&nbsp;&nbsp;&nbsp;<a href=https://tech.ifeng.com/c/8hbYegR6fAG><strong>AI寒武纪</strong></a>\n\n\n</h4>\n\n</header>\n<article>\n<div>\n<p>腾讯混元团队发布了他们最新的力作——Hunyuan-TurboS，首个混合Transformer-Mamba MoE架构的超大模型大家都知道，传统Transformer模型在处理长文本时一直面临挑战，O(N²)的复杂度以及KV-Cache问题让长文本训练和推理效率大打折扣。 而这次，混元-TurboS巧妙地融合了Mamba和Transformer两种架构的优势：• Mamba的高效长序列处理能力•...</p>\n\n<a href=\"https://tech.ifeng.com/c/8hbYegR6fAG\">网页链接</a>\n\n</div>\n\n\n</article>\n</div>\n</body>\n</html>\n","isBrief":false,"type":0,"news_type":1,"symbol":"LU0287142896.SGD","symbol_name":"Fidelity China Focus A-SGD","start_time":0,"source_url":"https://tech.ifeng.com/c/8hbYegR6fAG","article_id":"2518372738","we_media_id":null,"thumbnails":[],"rights":null,"url":"https://stock-news.laohu8.com/highlight/detail?id=2518372738","pubTimestamp":1741650434,"columns":[],"sourceInfo":{"source_id":"fenghuang_stock","name":"凤凰网"},"weMediaInfo":null,"summary":"腾讯混元团队发布了他们最新的力作——Hunyuan-TurboS，首个混合Transformer-Mamba MoE架构的超大模型大家都知道，传统Transformer模型在处理长文本时一直面临挑战，O的复杂度以及KV-Cache问题让长文本训练和推理效率大打折扣。而这次，混元-TurboS巧妙地融合了Mamba和Transformer两种架构的优势： Mamba的高效长序列处理能力 Transformer强大的上下文理解能力强强联合，效果自然惊艳！官方数据显示，Hunyuan-TurboS在多个关键benchmark上表现出色： 数学、推理、对齐能力超越或持平 GPT-4o-0806、DeepSeek-V3 以及各类开源模型。","collect":0,"end_time":0,"defaultTopTitle":"ifeng.com","property":[],"viewcount":null,"language":"zh","relate_stocks":{"80700":"腾讯控股-R","LU0287142896.SGD":"Fidelity China Focus A-SGD","IE0031814852.USD":"FSSA GREATER CHINA GROWTH \"I\" (USD) ACC","LU0821914370.USD":"贝莱德亚洲成长领袖A2","SG9999001689.USD":"施罗德亚洲成长股票","LU2097828805.USD":"AZ EQUITY CHINA \"A-AZ\" (USD) ACC","LU1196710351.USD":"SCHRODER ISF EMERGING MARKETS MULTI-ASSET \"A\" (USD) INC MF","LU1883866441.USD":"AMUNDI FUNDS REAL ASSETS TARGET INCOME \"A2\" (USD) ACC","LU1993786604.SGD":"UBS (LUX) EQUITY SICAV ALL CHINA \"P\" (SGD) ACC","LU2251237306.USD":"TEMPLETON EMERGING MARKETS DYNAMIC INCOME \"A\" (USD) INC A","SG9999000327.SGD":"Schroder Asian Growth A Dis SGD","LU0577902298.EUR":"FULLERTON LUX FUNDS - ASIA GROWTH & INCOME EQUITIE \"I\" (EUR) ACC","LU1226287792.SGD":"UBS (LUX) KEY SEL S-CHINA ALLOCATION OPP \"P\" (SGDHDG) INC","BK1608":"元宇宙概念","LU0259732245.USD":"EASTSPRING INVESTMENTS DRAGON PEACOCK A","HK0000320264.USD":"TAIKANG KAITAI CHINA NEW OPPORTUNITIES FUND \"A\" (USD) ACC","00700":"腾讯控股","LU0251144936.SGD":"Fidelity Sustainable Asia Equity A-SGD","LU0048580855.USD":"富达大中华区A","HTCD.SI":"Tencent HK SDR 10to1","LU0047713382.USD":"BGF EMERGING MARKETS \"A2\" ACC","LU0819123356.HKD":"HSBC GIF MANAGED SOLUTIONS ASIA FOCUSED GROWTH \"AC\" (HKD) ACC","TCEHY":"腾讯控股ADR","LU1629966141.USD":"Aberdeen Standard SICAV I - Asia Pacific Multi Asset A Acc USD","LU2242644610.SGD":"Fidelity China Innovation A-ACC-SGD","LU0348788117.USD":"ALLIANZ EMERGING ASIA EQUITY \"A\" (USD) INC","LU1981816686.USD":"EASTSPRING INV ASIAN MULTI FACTOR EQUITY \"A\" (USD) ACC","SG9999004220.SGD":"Nikko AM Shenton Asia Dividend Equity Fund  SGD","LU1328615791.USD":"FULLERTON LUX FUNDS - ALL CHINA EQUITIES \"I\" (USD) ACC","SG9999002562.SGD":"LionGlobal Asia Pacific SGD","LU0051755006.USD":"摩根大通中国A (dist)","LU0096374516.USD":"CT (LUX) I ASIA EQUITY INCOME \"DU\" (USD) ACC","LU0072462343.USD":"贝莱德亚洲巨龙基金","LU0594300419.USD":"富达中国消费基金A","LU2237443978.SGD":"Aberdeen Standard SICAV I - Global Dynamic Dividend A Acc SGD-H","LU1366334578.USD":"ALLSPRING EMERGING MARKETS EQUITY INCOME \"A\" (USD) INC","LU1868837722.USD":"CT (LUX) I GLOBAL EMERGING MARKET EQUITIES \"2\" (USD) ACC","SG9999014914.USD":"UNITED GLOBAL QUALITY GROWTH (USDHDG) INC","LU1725190679.HKD":"SCHRODER ISF EMERGING MARKETS MULTI-ASSET \"A\" (HKD) INC MF","LU0192582467.USD":"SCHRODER ISF ASIAN EQUITY YIELD \"A\" (USD) INC MF","LU0149721374.USD":"HSBC GIF HONG KONG EQUITY \"AD\" INC","LU0880133367.SGD":"UBS (LUX) EQUITY FUND CHINA OPPORTUNITY USD \"P\" (SGD) ACC","BK4588":"碎股","LU0320764599.SGD":"FTIF - Templeton China A Acc SGD","LU1568876335.HKD":"ALLIANZ CHINA MULTI INCOME PLUS \"AMG\" (HKD) INC","LU0918141887.USD":"安联亚洲实际收益股票基金","LU0084288322.USD":"Natixis Asia Equity RD USD","LU0348814723.USD":"ALLIANZ TOTAL RETURN ASIAN EQUITY \"A\" (USD) INC  NC","LU1282649810.SGD":"Allianz Asian Multi Income Plus Cl AMg DIS H2-SGD","SG9999014906.USD":"大华全球优质成长基金Acc USD"},"translate_title":"Tencent Hunyuan-TurboS: The first hybrid Transformer-Mamba MoE super-large model is here","themeId":null,"isJumpTheme":false,"ttsUrl":null,"symbols_score_info":{"80700":0.6,"00700":1,"HTCD.SI":0.6,"TCEHY":1},"content_text":"腾讯混元团队发布了他们最新的力作——Hunyuan-TurboS，首个混合Transformer-Mamba MoE架构的超大模型大家都知道，传统Transformer模型在处理长文本时一直面临挑战，O(N²)的复杂度以及KV-Cache问题让长文本训练和推理效率大打折扣。 而这次，混元-TurboS巧妙地融合了Mamba和Transformer两种架构的优势：• Mamba的高效长序列处理能力• Transformer强大的上下文理解能力强强联合，效果自然惊艳！官方数据显示，Hunyuan-TurboS在多个关键benchmark上表现出色：• 数学、推理、对齐能力超越或持平 GPT-4o-0806、DeepSeek-V3 以及各类开源模型。• 知识能力同样突出，在 MMLU-Pro 等专业评测中也具备竞争力Hunyuan-TurboS的推理成本相比之前的Turbo模型降低了整整7倍！ 性能提升的同时，成本大幅下降为了进一步提升模型能力，混元团队还对TurboS进行了多项后训练优化：• 引入慢思考（Slow-thinking）集成，显著提升了模型在数学、编程和推理任务上的表现。• 通过精细化的指令调优，增强了模型的对齐性和Agent执行能力。• 进行了英语训练优化，从而提升模型的通用性能。除了模型架构和训练优化，混元团队还升级了奖励系统：• 采用基于规则的评分和一致性验证，保证模型输出的质量和可靠性。• 引入代码沙箱反馈，大幅提升模型在STEM领域的准确性。• 使用生成式奖励，优化模型在问答和创意性任务上的表现，并有效减少奖励作弊现象写在最后：Mamba 是一种新型的状态空间模型 (State Space Model, SSM)。 简单来说，你可以把它理解为一种更高效处理长序列数据的模型架构，它主要为了解决 Transformer 模型在处理长序列时效率低下的问题而提出的Mamba 是由 Albert Gu （卡耐基梅隆大学）和 Tri Dao（普林斯顿大学） 两位研究人员主导开发的Mamba 的创新之处在于它引入了“选择性状态空间 (Selective State Spaces)” 的概念。 简单来说，Mamba 模型中的 SSM 参数不再是固定的，而是会根据输入的内容动态变化。 这样一来，模型就能根据当前处理的token，有选择性地记住或遗忘信息，从而更好地进行内容相关的推理此外，Mamba 的设计也考虑了硬件效率，它在保持类似循环神经网络 (RNN) 的结构的同时，还能进行并行计算，这使得 Mamba 在推理速度上比 Transformer 更快（约快 5 倍）。 而且，Mamba 的计算复杂度是线性于序列长度的，这意味着处理更长的序列也不会显著增加计算负担。","kind":"news","is_publish_news":true,"is_publish_highlight":false,"is_publish_live":false,"is_publish_wemedia":false,"editions":null,"column":"","sentiment":"0","news_tag":"","news_rank":0,"isVideo":false,"video":null,"symbols":[],"gpt_button":0,"need_auth":false,"need_login_tip":false,"code":"91000000","status":"200"},"commentList":[],"isCommentEnd":true,"newsSizeData":{"likeSize":0,"commentSize":0,"repostSize":0,"favoriteSize":0,"likeStatus":false,"favoriteStatus":false},"APP":{"userAgent":"Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko; compatible; ClaudeBot/1.0; +claudebot@anthropic.com)","isDev":false,"isTTM":false,"isLaohu8Offline":false,"tenantId":"TBCN","deviceId":"web-server-community-laohu8-v3","version":"4.44.1","shortVersion":"4.44.1","platform":"web","vendor":"web","appName":"laohu8","isIOS":false,"isAndroid":false,"isTiger":false,"isTHS":false,"isWeiXin":false,"isWeiXinMini":false,"isWeiBo":false,"isQQ":false,"isBaiduSwan":false,"isBaiduBox":false,"isDingTalk":false,"isToutiao":false,"isOnePlus":false,"isHuaWei":false,"isXiaomi":false,"isXiaomiWebView":false,"isOppo":false,"isVivo":false,"isSamsung":false,"isMobile":false},"href":"/m/news/2518372738","isCrawlerRequest":true}