CINXE.COM

<!doctype html><html><head> <script> var e=function(e,t,a){if(Math.ceil(100*Math.random())<=100*t){var o="491065",d="24814",n={ev_type:"batch",list:[{ev_type:"custom",payload:{name:"sdk_glue_load",type:"event",metrics:{},categories:{sdk_glue_load_status:e,sdk_glue_load_err_src:a,payload_bdms_aid:o,payload_bdms_page_id:d}},common:{context:{ctx_bdms_aid:o,ctx_bdms_page_id:d},bid:"web_bdms_cn",pid:window.location.pathname,view_id:"/_1",user_id:"",session_id:"0-a-1-2-c",release:"",env:"production",url:window.location.href,timestamp:+new Date,sdk_version:"1.6.1",sdk_name:"SDK_SLARDAR_WEB"}}]},i=new XMLHttpRequest;i.open("POST","https://mon.zijieapi.com/monitor_browser/collect/batch/?biz_id=web_bdms_cn",!0),i.setRequestHeader("Content-type","application/json"),i.send(JSON.stringify(n))}};e("before_load",.1,""),window.addEventListener("error",(function(t){var a=t.target||t.srcElement;a instanceof HTMLElement&&"SCRIPT"==a.nodeName&&(-1!=(a.src||"").indexOf("sdk-glue")&&e("load_error",1,a.src))}),!0);</script><script src="https://lf-headquarters-speed.yhgfb-cn-static.com/obj/rc-client-security/web/glue/1.0.0.38/sdk-glue.js"></script><script> ;(function (){ var sdkInfo = { csrf: { init: function (options) {window.secsdk.csrf.setOptions(options)}, isLoaded: function () { return !!window.secsdk }, srcList: ["https://lf1-cdn-tos.bytegoofy.com/obj/goofy/secsdk/secsdk-lastest.umd.js","https://lf3-cdn-tos.bytegoofy.com/obj/goofy/secsdk/secsdk-lastest.umd.js","https://lf6-cdn-tos.bytegoofy.com/obj/goofy/secsdk/secsdk-lastest.umd.js"] }, bdms: { init: function (options) {window.bdms.init(options)}, isLoaded: function () { return !!window.bdms }, srcList: ["https://lf-c-flwb.bytetos.com/obj/rc-client-security/web/stable/1.0.0.43/bdms.js","https://lf-headquarters-speed.yhgfb-cn-static.com/obj/rc-client-security/web/stable/1.0.0.43/bdms.js"], }, verifyCenter: { init: function (options) {window.verifySDK.init(options)}, isLoaded: function () { return !!window.verifySDK }, srcList: ["https://lf-rc1.yhgfb-cn-static.com/obj/rc-client-security/secsdk-captcha/2.28.11/captcha.js","https://lf-rc2.yhgfb-cn-static.com/obj/rc-client-security/secsdk-captcha/2.28.11/captcha.js"] }, }; var options = { bdms: {aid:491065,pageId:24814,paths:["/api/fe"],ddrt:3},self: {aid:491065,pageId:24814,} }; window._SdkGlueInit(options, sdkInfo); })()</script> <script>window.gfdatav1={"env":"prod","idc":"hl","ver":"1.0.1.918","canary":0,"envName":"prod","region":"cn","runtime":"workerV2","vdc":"hl","vregion":"China-North","extra":{"canaryType":null}}</script><script>window._SERVER_DATA={"router":{"baseUrl":"/","params":{}}}</script><meta charset="utf-8"><meta name="viewport" content="width=device-width,initial-scale=1,shrink-to-fit=no,viewport-fit=cover,minimum-scale=1,maximum-scale=1,user-scalable=no"><meta http-equiv="x-ua-compatible" content="ie=edge"><meta name="renderer" content="webkit"><meta name="layoutmode" content="standard"><meta name="imagemode" content="force"><meta name="wap-font-scale" content="no"><meta name="format-detection" content="telephone=no"><title data-react-helmet="true">小窗幽记机器学习 的个人主页 - 开发者社区 - 火山引擎</title><link rel="icon" href="//lf1-cdn-tos.bytegoofy.com/goofy/tech-fe/fav.png"/><link rel="apple-touch-icon" href="//lf1-cdn-tos.bytegoofy.com/goofy/tech-fe/logo193.png"/><meta name="referrer" content="always"><script>// Tea SDK 引导代码 (function (win, export_obj) { win['LogAnalyticsObject'] = export_obj; if (!win[export_obj]) { function _collect() { _collect.q.push(arguments); } _collect.q = _collect.q || []; win[export_obj] = _collect; } win[export_obj].l = +new Date(); })(window, 'collectEvent');</script><script>window.__assetPrefix__ = '//portal.volccdn.com/obj/volcfe-scm/deploy/volc_developer_/42325';</script><script>!function(){"use strict";var e,t,r,a,i,n,o,c={},d={};function f(e){var t=d[e];if(void 0!==t)return t.exports;var r=d[e]={id:e,loaded:!1,exports:{}};return c[e].call(r.exports,r,r.exports,f),r.loaded=!0,r.exports}f.m=c,e=[],f.O=function(t,r,a,i){if(!r){var n=1/0;for(s=0;s<e.length;s++){r=e[s][0],a=e[s][1],i=e[s][2];for(var o=!0,c=0;c<r.length;c++)(!1&i||n>=i)&&Object.keys(f.O).every((function(e){return f.O[e](r[c])}))?r.splice(c--,1):(o=!1,i<n&&(n=i));if(o){e.splice(s--,1);var d=a();void 0!==d&&(t=d)}}return t}i=i||0;for(var s=e.length;s>0&&e[s-1][2]>i;s--)e[s]=e[s-1];e[s]=[r,a,i]},f.n=function(e){var t=e&&e.__esModule?function(){return e.default}:function(){return e};return f.d(t,{a:t}),t},r=Object.getPrototypeOf?function(e){return Object.getPrototypeOf(e)}:function(e){return e.__proto__},f.t=function(e,a){if(1&a&&(e=this(e)),8&a)return e;if("object"==typeof e&&e){if(4&a&&e.__esModule)return e;if(16&a&&"function"==typeof e.then)return e}var i=Object.create(null);f.r(i);var n={};t=t||[null,r({}),r([]),r(r)];for(var o=2&a&&e;"object"==typeof o&&!~t.indexOf(o);o=r(o))Object.getOwnPropertyNames(o).forEach((function(t){n[t]=function(){return e[t]}}));return n.default=function(){return e},f.d(i,n),i},f.d=function(e,t){for(var r in t)f.o(t,r)&&!f.o(e,r)&&Object.defineProperty(e,r,{enumerable:!0,get:t[r]})},f.f={},f.e=function(e){return Promise.all(Object.keys(f.f).reduce((function(t,r){return f.f[r](e,t),t}),[]))},f.u=function(e){return"static/js/async/"+({572:"user/[id]/blog_moving/page",712:"user/[id]/draft/page",1225:"resource/[id]/page",1261:"user/edit/page",1687:"user/[id]/star/page",1826:"article/series/list/page",3180:"teams/[id]/articles/page",3487:"teams/page",4117:"mirror/page",4778:"articles/page",4786:"activity/[id]/page",4815:"user/[id]/qa/page",5400:"teams/[id]/page",5655:"teams/[id]/questions/page",5912:"user/[id]/page",6226:"questions/[id]/page",6242:"$",6480:"user/[id]/$",6772:"user/[id]/series/page",6827:"user/[id]/activity/page",6874:"articles/[id]/page",7281:"user/[id]/follow_list/page",7425:"resource/page",8079:"teams/[id]/layout",8138:"questions/page",8146:"user/$",8357:"user/page",8426:"article/series/detail/[id]/page",8528:"activities/[id]/page",8655:"team/[id]/page",8829:"activities/page",9130:"activity/page",9494:"team/page",9624:"user/[id]/articles/page",9768:"page",9844:"user/[id]/layout"}[e]||e)+"."+{407:"aefe6500",572:"6eb588a7",585:"728b06d1",637:"32464f61",712:"45e8a881",756:"56e71d33",885:"eca1e824",1008:"6f7bfd31",1094:"e7bbea48",1124:"830b6040",1225:"5cfeb49c",1261:"990d7081",1289:"bf78b8e3",1687:"77dfadf3",1826:"4456d7f6",2057:"66169876",2128:"b6a13c1c",2141:"048f1454",2417:"ae8d0598",2426:"48b8c954",2650:"28c0780d",3172:"b888ceda",3180:"034dce8a",3487:"e99070d3",3505:"fc596d93",3553:"27d193e6",4015:"0f26ce33",4117:"d449d71b",4433:"a0f3e45e",4778:"77241347",4786:"2be3f444",4815:"cb1df973",5092:"fc8ebe4d",5400:"2e7c6646",5435:"eccb96c2",5453:"b646a57c",5458:"9b8df362",5493:"c3044ac2",5505:"a3aa549d",5655:"49f52348",5777:"109859a7",5912:"77e2305e",5914:"1927ef78",5941:"59bcc29a",5998:"1d847129",6003:"48f9a706",6011:"22717107",6033:"cd1c2bf1",6097:"8d0cacad",6128:"b950f483",6226:"72327802",6242:"8664c2ba",6336:"0ce21484",6480:"07b5287b",6637:"e866599a",6705:"a19b8e49",6772:"f8f54b68",6827:"a50031a9",6874:"e60a41e0",7056:"d6e157d4",7281:"469e614f",7327:"9b4b3761",7425:"88d9a6e8",7474:"4dbf3b1c",7518:"648426dd",7542:"a4ca3948",7644:"a5ed476c",7832:"cc5e6442",7921:"8b6f440c",8079:"1a309ced",8138:"83a087b7",8146:"b9315b80",8357:"5acc5f96",8426:"d73a5ca5",8528:"cd81d041",8626:"72051b87",8655:"b76ed95c",8829:"95afcb7d",8913:"3652e416",9107:"8db20dc7",9129:"69cb86a2",9130:"eb82bf48",9404:"fa2c6b2f",9494:"ee37a04b",9624:"befc07a0",9768:"fbf929cc",9791:"22e5ab4e",9844:"e938f428"}[e]+".js"},f.miniCssF=function(e){return"static/css/async/"+({572:"user/[id]/blog_moving/page",712:"user/[id]/draft/page",1225:"resource/[id]/page",1261:"user/edit/page",1687:"user/[id]/star/page",1826:"article/series/list/page",3180:"teams/[id]/articles/page",3487:"teams/page",4117:"mirror/page",4778:"articles/page",4815:"user/[id]/qa/page",5400:"teams/[id]/page",5655:"teams/[id]/questions/page",5912:"user/[id]/page",6226:"questions/[id]/page",6242:"$",6772:"user/[id]/series/page",6827:"user/[id]/activity/page",6874:"articles/[id]/page",7281:"user/[id]/follow_list/page",7425:"resource/page",8079:"teams/[id]/layout",8138:"questions/page",8426:"article/series/detail/[id]/page",8528:"activities/[id]/page",8829:"activities/page",9624:"user/[id]/articles/page",9768:"page",9844:"user/[id]/layout"}[e]||e)+"."+{572:"5c7b055b",712:"72db1217",1225:"c771a8fa",1261:"556cf1b3",1687:"50ab220b",1826:"112c3666",2426:"01a02ce7",3180:"4fb2abd1",3487:"60e48adf",4117:"00b315f5",4778:"11c4f69f",4815:"06c5a7e1",5400:"43e4ce1a",5655:"4e016773",5912:"96b3cbfa",6226:"5c2bbaa5",6242:"6e3970d3",6772:"e6f14c09",6827:"68357391",6874:"c871667b",7056:"365859d0",7281:"9b1d94e8",7425:"95aa4a33",7542:"f7fe289f",8079:"bc06c4c6",8138:"aed8f20f",8426:"00a692ef",8528:"0ea227fe",8829:"95f92d4a",9107:"c4cafd96",9624:"96b3cbfa",9768:"394efaa3",9844:"9344d547"}[e]+".css"},f.g=function(){if("object"==typeof globalThis)return globalThis;try{return this||new Function("return this")()}catch(e){if("object"==typeof window)return window}}(),f.hmd=function(e){return(e=Object.create(e)).children||(e.children=[]),Object.defineProperty(e,"exports",{enumerable:!0,set:function(){throw new Error("ES Modules may not assign module.exports or exports.*, Use ESM export syntax, instead: "+e.id)}}),e},f.o=function(e,t){return Object.prototype.hasOwnProperty.call(e,t)},a={},i="volc-developer-fe:",f.l=function(e,t,r,n){if(a[e])a[e].push(t);else{var o,c;if(void 0!==r)for(var d=document.getElementsByTagName("script"),s=0;s<d.length;s++){var u=d[s];if(u.getAttribute("src")==e||u.getAttribute("data-webpack")==i+r){o=u;break}}o||(c=!0,(o=document.createElement("script")).charset="utf-8",o.timeout=120,f.nc&&o.setAttribute("nonce",f.nc),o.setAttribute("data-webpack",i+r),o.src=e,0!==o.src.indexOf(window.location.origin+"/")&&(o.crossOrigin="anonymous")),a[e]=[t];var l=function(t,r){o.onerror=o.onload=null,clearTimeout(p);var i=a[e];if(delete a[e],o.parentNode&&o.parentNode.removeChild(o),i&&i.forEach((function(e){return e(r)})),t)return t(r)},p=setTimeout(l.bind(null,void 0,{type:"timeout",target:o}),12e4);o.onerror=l.bind(null,o.onerror),o.onload=l.bind(null,o.onload),c&&document.head.appendChild(o)}},f.r=function(e){"undefined"!=typeof Symbol&&Symbol.toStringTag&&Object.defineProperty(e,Symbol.toStringTag,{value:"Module"}),Object.defineProperty(e,"__esModule",{value:!0})},f.nmd=function(e){return e.paths=[],e.children||(e.children=[]),e},f.p="//portal.volccdn.com/obj/volcfe-scm/deploy/volc_developer_/42325/",n=function(e){return new Promise((function(t,r){var a=f.miniCssF(e),i=f.p+a;if(function(e,t){for(var r=document.getElementsByTagName("link"),a=0;a<r.length;a++){var i=(o=r[a]).getAttribute("data-href")||o.getAttribute("href");if("stylesheet"===o.rel&&(i===e||i===t))return o}var n=document.getElementsByTagName("style");for(a=0;a<n.length;a++){var o;if((i=(o=n[a]).getAttribute("data-href"))===e||i===t)return o}}(a,i))return t();!function(e,t,r,a){var i=document.createElement("link");i.rel="stylesheet",i.type="text/css",i.onerror=i.onload=function(n){if(i.onerror=i.onload=null,"load"===n.type)r();else{var o=n&&("load"===n.type?"missing":n.type),c=n&&n.target&&n.target.href||t,d=new Error("Loading CSS chunk "+e+" failed.\n("+c+")");d.code="CSS_CHUNK_LOAD_FAILED",d.type=o,d.request=c,i.parentNode.removeChild(i),a(d)}},i.href=t,0!==i.href.indexOf(window.location.origin+"/")&&(i.crossOrigin="anonymous"),document.head.appendChild(i)}(e,i,t,r)}))},o={6272:0},f.f.miniCss=function(e,t){o[e]?t.push(o[e]):0!==o[e]&&{572:1,712:1,1225:1,1261:1,1687:1,1826:1,2426:1,3180:1,3487:1,4117:1,4778:1,4815:1,5400:1,5655:1,5912:1,6226:1,6242:1,6772:1,6827:1,6874:1,7056:1,7281:1,7425:1,7542:1,8079:1,8138:1,8426:1,8528:1,8829:1,9107:1,9624:1,9768:1,9844:1}[e]&&t.push(o[e]=n(e).then((function(){o[e]=0}),(function(t){throw delete o[e],t})))},function(){var e={6272:0};f.f.j=function(t,r){var a=f.o(e,t)?e[t]:void 0;if(0!==a)if(a)r.push(a[2]);else if(/^(2426|6272|9107)$/.test(t))e[t]=0;else{var i=new Promise((function(r,i){a=e[t]=[r,i]}));r.push(a[2]=i);var n=f.p+f.u(t),o=new Error;f.l(n,(function(r){if(f.o(e,t)&&(0!==(a=e[t])&&(e[t]=void 0),a)){var i=r&&("load"===r.type?"missing":r.type),n=r&&r.target&&r.target.src;o.message="Loading chunk "+t+" failed.\n("+i+": "+n+")",o.name="ChunkLoadError",o.type=i,o.request=n,a[1](o)}}),"chunk-"+t,t)}},f.O.j=function(t){return 0===e[t]};var t=function(t,r){var a,i,n=r[0],o=r[1],c=r[2],d=0;if(n.some((function(t){return 0!==e[t]}))){for(a in o)f.o(o,a)&&(f.m[a]=o[a]);if(c)var s=c(f)}for(t&&t(r);d<n.length;d++)i=n[d],f.o(e,i)&&e[i]&&e[i][0](),e[i]=0;return f.O(s)},r=self.__LOADABLE_LOADED_CHUNKS__=self.__LOADABLE_LOADED_CHUNKS__||[];r.forEach(t.bind(null,0)),r.push=t.bind(null,r.push.bind(r))}()}();</script><script defer="defer" src="//portal.volccdn.com/obj/volcfe-scm/deploy/volc_developer_/42325/static/js/lib-arco.a74ce714.js" crossorigin="anonymous"></script><script defer="defer" src="//portal.volccdn.com/obj/volcfe-scm/deploy/volc_developer_/42325/static/js/lib-polyfill.e32e6eae.js" crossorigin="anonymous"></script><script defer="defer" src="//portal.volccdn.com/obj/volcfe-scm/deploy/volc_developer_/42325/static/js/lib-lodash.66bf9186.js" crossorigin="anonymous"></script><script defer="defer" src="//portal.volccdn.com/obj/volcfe-scm/deploy/volc_developer_/42325/static/js/lib-react.fc946b50.js" crossorigin="anonymous"></script><script defer="defer" src="//portal.volccdn.com/obj/volcfe-scm/deploy/volc_developer_/42325/static/js/lib-router.91ad9bff.js" crossorigin="anonymous"></script><script defer="defer" src="//portal.volccdn.com/obj/volcfe-scm/deploy/volc_developer_/42325/static/js/4316.560dd7ea.js" crossorigin="anonymous"></script><script defer="defer" src="//portal.volccdn.com/obj/volcfe-scm/deploy/volc_developer_/42325/static/js/main.82580efb.js" crossorigin="anonymous"></script><link href="//portal.volccdn.com/obj/volcfe-scm/deploy/volc_developer_/42325/static/css/lib-arco.974890ef.css" rel="stylesheet" crossorigin="anonymous"><link href="//portal.volccdn.com/obj/volcfe-scm/deploy/volc_developer_/42325/static/css/4316.27e0ba1f.css" rel="stylesheet" crossorigin="anonymous"><link href="//portal.volccdn.com/obj/volcfe-scm/deploy/volc_developer_/42325/static/css/main.ff1f8027.css" rel="stylesheet" crossorigin="anonymous"><link href="//portal.volccdn.com/obj/volcfe-scm/deploy/volc_developer_/42325/static/css/lib-arco.974890ef.css" rel="stylesheet" /><link href="//portal.volccdn.com/obj/volcfe-scm/deploy/volc_developer_/42325/static/css/async/7542.f7fe289f.css" rel="stylesheet" /><link href="//portal.volccdn.com/obj/volcfe-scm/deploy/volc_developer_/42325/static/css/async/user/[id]/layout.9344d547.css" rel="stylesheet" /><link href="//portal.volccdn.com/obj/volcfe-scm/deploy/volc_developer_/42325/static/css/async/user/[id]/page.96b3cbfa.css" rel="stylesheet" /> <meta data-react-helmet="true" name="description" content="火山引擎开发者社区是面向开发者的开放型技术平台，聚合火山引擎产品团队内容沉淀，分享字节跳动快速发展过程中积累的技术资源，与开发者一起共同成长。"/><meta data-react-helmet="true" name="keywords" content="云原生、大数据、AI、数据库、移动开发、音视频、技术服务知识库、云基础、开源"/> </head><body><noscript>We're sorry but react app doesn't work properly without JavaScript enabled. Please enable it to continue.</noscript><div id="root"><div class="WFpVe"><div class="yYbKk"><div class="gcUkQ"><a href="https://www.volcengine.com" target="_blank"><img src="https://lf1-cdn-tos.bytegoofy.com/goofy/tech-fe/assets/vocl_logo_dark.c678a292.svg" draggable="false"/></a></div><div class="rpPns"><a href="https://developer.volcengine.com" target="_blank"><img src="https://lf1-cdn-tos.bytegoofy.com/goofy/tech-fe/assets/community-logo-text-icon.4b9f2115.svg" draggable="false"/></a></div><div class="Q44pY"><div class="yYwDU"><div class="arco-input-group-wrapper arco-input-group-wrapper-default PqPNy"><span class="arco-input-group"><span class="arco-input-inner-wrapper arco-input-inner-wrapper-has-prefix arco-input-inner-wrapper-default"><span class="arco-input-group-prefix"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" class="arco-icon arco-icon-search"><path d="M33.072 33.071c6.248-6.248 6.248-16.379 0-22.627-6.249-6.249-16.38-6.249-22.628 0-6.248 6.248-6.248 16.379 0 22.627 6.248 6.248 16.38 6.248 22.628 0Zm0 0 8.485 8.485"></path></svg></span><input placeholder="在社区搜文章/找答案" class="arco-input arco-input-size-default" value=""/></span></span></div><a href="https://www.volcengine.com/docs" target="blank">文档</a><a href="https://www.volcengine.com/beian" target="blank">备案</a><a href="https://console.volcengine.com/home" target="blank">控制台</a><a href="https://console.volcengine.com/auth/login?redirectURI=https%3A%2F%2Fdeveloper.volcengine.com%2Fuser%2F1977369294279187">登录</a><a href="https://console.volcengine.com/auth/signup?redirectURI=https%3A%2F%2Fdeveloper.volcengine.com%2Fuser%2F1977369294279187" class="arco-btn arco-btn-primary arco-btn-size-default arco-btn-shape-square arco-btn-link iqsRJ"><span>立即注册</span></a></div></div></div><div class="stbmn"><div class="ASi46"><div class="qZzL_"><a href="/" target="_blank" class="">首页</a><a href="/articles" target="_blank" class="">文章</a><a href="/questions" target="_blank" class="hide">问答</a><a href="/videos" target="_blank" class="hide">视频</a><a href="/activities" target="_blank" class="">活动</a><a href="/resource" target="_blank" class="hide">下载资源</a><a href="/teams" target="_blank" class="hide">团队号</a><a href="/mirror" target="_blank" class="">镜像站</a></div></div><div class="NugHr"><div class="CQeSf">发布</div></div></div></div><div class="hLFIT"><div style="position:relative"><div class="zL3em"><div class="_G5oh"><div class="IuHbi"><span class="xm9F8" style="width:88px;height:88px;vertical-align:top"><div style="width:88px;height:88px;font-size:44px;background-color:#3370ff;cursor:inherit;min-height:88px;min-width:88px" class="arco-avatar arco-avatar-circle HmXfH"><span class="arco-avatar-image"><img src="https://p26-passport.byteacctimg.com/img/user-avatar/1816ccc0ebc35f4a89457e0b4ffe692b~300x300.image" alt="小窗幽记机器学习"/></span></div></span></div><div class="VSMo8"><span class="LOr8L" style="line-height:38px">小窗幽记机器学习</span></div><span class="T1Ybx"></span></div></div><div class="zvN1c"><div class="arco-tabs arco-tabs-horizontal arco-tabs-line arco-tabs-top arco-tabs-size-default C40pP"><div class="arco-tabs-header-nav arco-tabs-header-nav-horizontal arco-tabs-header-nav-top arco-tabs-header-size-default arco-tabs-header-nav-line"><div class="arco-tabs-header-scroll"><div class="arco-tabs-header-wrapper"><div class="arco-tabs-header" style="transform:translateX(0px);-webkit-transform:translateX(0px);-ms-transform:translateX(0px);-moz-transform:translateX(0px);-o-transform:translateX(0px)"><div class="arco-tabs-header-title" role="tab" aria-selected="false" tabindex="0"><span class="arco-tabs-header-title-text">文章</span></div><div class="arco-tabs-header-title" role="tab" aria-selected="false" tabindex="0"><span class="arco-tabs-header-title-text">专栏</span></div><div class="arco-tabs-header-title" role="tab" aria-selected="false" tabindex="0"><span class="arco-tabs-header-title-text">问答</span></div><div class="arco-tabs-header-ink"></div></div></div></div></div></div><div><div class="article-list"><div class="zEp2T" style="cursor:auto"><div class="vuCON"><a href="/user/1977369294279187" target="_blank"><div style="width:20px;height:20px;font-size:10px" class="arco-avatar arco-avatar-circle"><span class="arco-avatar-image"><img src="https://p26-passport.byteacctimg.com/img/user-avatar/1816ccc0ebc35f4a89457e0b4ffe692b~300x300.image" alt="小窗幽记机器学习"/></span></div><div class="h_pFt"><span>小窗幽记机器学习</span></div></a></div><div class="eHpWb"><div class="BqxXc UG8D7"><div class="fHdNq"><div class="cSmO7"><a href="/articles/7440006190165639178" target="_blank">多模态大模型系列 | 18：Qwen2-VL(最新版)解读及其实战(精炼版)</a></div><div style="margin-left:16px;display:inline-block;vertical-align:text-bottom"></div></div><div class="aklXs"><span class="MppYW"><span class="Ji9yj"></span><span class="HIKQo">大模型</span></span><span class="KRApY">大模型</span><span class="KRApY">机器学习</span><span class="KRApY">数据库</span></div><div class="IsnSN"><div class="TRHtk">引言简介方法实验结果实战代码任务1：检测任务2：图片理解总结引言 =======梅子金黄杏子肥，麦花雪白菜花稀。日长篱落无人过，惟有蜻蜓蛱蝶飞。小伙伴们好，我是微信公众号《小窗幽记机器学习》的小编：卖热干面的小女孩。紧接前文：多模态大模型系列：Qwen-VL解读及其实战，今天这篇小作文主要介绍阿里在2024年9月份发布的视觉语言模型：Qwen2-VL。简介 =======这篇论文介绍了Qwen2-</div><img class="v4MkD" src="https://p6-volc-community-sign.byteimg.com/tos-cn-i-tlddhu82om/3725eb4f79e84b5081ccc75b49eda82e~tplv-tlddhu82om-image.image?=&amp;rk3s=8031ce6d&amp;x-expires=1732499851&amp;x-signature=6guhRzGSM%2FIT6MIICAQHPJuhvhI%3D"/></div></div><div class="nvyao"><img src="https://p6-volc-community-sign.byteimg.com/tos-cn-i-tlddhu82om/3725eb4f79e84b5081ccc75b49eda82e~tplv-tlddhu82om-image.image?=&amp;rk3s=8031ce6d&amp;x-expires=1732499851&amp;x-signature=6guhRzGSM%2FIT6MIICAQHPJuhvhI%3D"/></div></div></div><div class="B_zhD mdM_r"><div class="J8PPi"><div class="lugd4"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-eye"><path d="M24 37c6.627 0 12.627-4.333 18-13-5.373-8.667-11.373-13-18-13-6.627 0-12.627 4.333-18 13 5.373 8.667 11.373 13 18 13Z" clip-rule="evenodd"></path><path d="M29 24a5 5 0 1 1-10 0 5 5 0 0 1 10 0Z"></path></svg><span class="MRrY3">67</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-thumb-up"><path d="M7 17v26m35.17-21.394-5.948 18.697a1 1 0 0 1-.953.697H14V19h3l9.403-12.223a1 1 0 0 1 1.386-.196l2.535 1.87a6 6 0 0 1 2.044 6.974L31 19h9.265a2 2 0 0 1 1.906 2.606Z"></path></svg><span class="MRrY3">0</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-message"><path d="M15 20h18m-18 9h9M7 41h17.63C33.67 41 41 33.67 41 24.63V24c0-9.389-7.611-17-17-17S7 14.611 7 24v17Z"></path></svg><span class="MRrY3">0</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-star"><path d="M22.552 6.908a.5.5 0 0 1 .896 0l5.02 10.17a.5.5 0 0 0 .376.274l11.224 1.631a.5.5 0 0 1 .277.853l-8.122 7.916a.5.5 0 0 0-.143.443l1.917 11.178a.5.5 0 0 1-.726.527l-10.038-5.278a.5.5 0 0 0-.466 0L12.73 39.9a.5.5 0 0 1-.726-.527l1.918-11.178a.5.5 0 0 0-.144-.443l-8.122-7.916a.5.5 0 0 1 .278-.853l11.223-1.63a.5.5 0 0 0 .376-.274l5.02-10.17Z"></path></svg><span class="MRrY3">0</span></div><div class="opKvU"></div><span class="N_67f"><svg width="16" height="16" fill="none" xmlns="http://www.w3.org/2000/svg"><path fill-rule="evenodd" clip-rule="evenodd" d="M7.783 7.314a2 2 0 0 1 1.169 3.221l-1.888 2.332a2 2 0 1 1-3.109-2.518L4.967 9.1a1.431 1.431 0 0 1-.09-.372L4.725 7.28 2.919 9.51A3.333 3.333 0 1 0 8.1 13.705l1.888-2.332a3.333 3.333 0 0 0-1.304-5.173l-.9 1.113Z" fill="#939AA3"></path><path fill-rule="evenodd" clip-rule="evenodd" d="M8.58 9.291A2 2 0 0 1 7.411 6.07L9.3 3.74a2 2 0 1 1 3.11 2.517l-1.013 1.25c.046.116.077.241.09.372l.153 1.447 1.805-2.23a3.333 3.333 0 0 0-5.18-4.196L6.374 5.231a3.333 3.333 0 0 0 1.304 5.173l.901-1.113Z" fill="#939AA3"></path></svg></span></div></div><div class="zEp2T" style="cursor:auto"><div class="vuCON"><a href="/user/1977369294279187" target="_blank"><div style="width:20px;height:20px;font-size:10px" class="arco-avatar arco-avatar-circle"><span class="arco-avatar-image"><img src="https://p26-passport.byteacctimg.com/img/user-avatar/1816ccc0ebc35f4a89457e0b4ffe692b~300x300.image" alt="小窗幽记机器学习"/></span></div><div class="h_pFt"><span>小窗幽记机器学习</span></div></a></div><div class="eHpWb"><div class="BqxXc UG8D7"><div class="fHdNq"><div class="cSmO7"><a href="/articles/7440006140932718602" target="_blank">多模态大模型系列 | 19：微软提出LLM2CLIP，巧用LLM助力clip开启图文理解新篇章</a></div><div style="margin-left:16px;display:inline-block;vertical-align:text-bottom"></div></div><div class="aklXs"><span class="MppYW"><span class="Ji9yj"></span><span class="HIKQo">AI</span></span><span class="KRApY">大模型</span><span class="KRApY">向量数据库</span><span class="KRApY">机器学习</span></div><div class="IsnSN"><div class="TRHtk">引言简介方法实验结果总结引言 =======丹阳郭里送行舟，一别心知两地秋。CLIP是当今最重要的多模态基础模型之一，能将视觉和文本信息对齐到共享特征空间中。自然语言作为人类知识载体，为CLIP提供了丰富的监督信息，赋予了它强大的跨模态表示能力。随着大型语言模型(LLMs)的发展，越来越多的学者探讨如何利用LLMs提升多模态表示学习。LLMs强大的语言理解能力可以提升CLIP处理各类文本的能力，其</div><img class="v4MkD" src="https://p3-volc-community-sign.byteimg.com/tos-cn-i-tlddhu82om/09a6b96522eb415ebce8343797468c31~tplv-tlddhu82om-image.image?=&amp;rk3s=8031ce6d&amp;x-expires=1732499851&amp;x-signature=F%2FxomaNrkGQTl6oGp%2F43r029peM%3D"/></div></div><div class="nvyao"><img src="https://p3-volc-community-sign.byteimg.com/tos-cn-i-tlddhu82om/09a6b96522eb415ebce8343797468c31~tplv-tlddhu82om-image.image?=&amp;rk3s=8031ce6d&amp;x-expires=1732499851&amp;x-signature=F%2FxomaNrkGQTl6oGp%2F43r029peM%3D"/></div></div></div><div class="B_zhD mdM_r"><div class="J8PPi"><div class="lugd4"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-eye"><path d="M24 37c6.627 0 12.627-4.333 18-13-5.373-8.667-11.373-13-18-13-6.627 0-12.627 4.333-18 13 5.373 8.667 11.373 13 18 13Z" clip-rule="evenodd"></path><path d="M29 24a5 5 0 1 1-10 0 5 5 0 0 1 10 0Z"></path></svg><span class="MRrY3">8</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-thumb-up"><path d="M7 17v26m35.17-21.394-5.948 18.697a1 1 0 0 1-.953.697H14V19h3l9.403-12.223a1 1 0 0 1 1.386-.196l2.535 1.87a6 6 0 0 1 2.044 6.974L31 19h9.265a2 2 0 0 1 1.906 2.606Z"></path></svg><span class="MRrY3">0</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-message"><path d="M15 20h18m-18 9h9M7 41h17.63C33.67 41 41 33.67 41 24.63V24c0-9.389-7.611-17-17-17S7 14.611 7 24v17Z"></path></svg><span class="MRrY3">0</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-star"><path d="M22.552 6.908a.5.5 0 0 1 .896 0l5.02 10.17a.5.5 0 0 0 .376.274l11.224 1.631a.5.5 0 0 1 .277.853l-8.122 7.916a.5.5 0 0 0-.143.443l1.917 11.178a.5.5 0 0 1-.726.527l-10.038-5.278a.5.5 0 0 0-.466 0L12.73 39.9a.5.5 0 0 1-.726-.527l1.918-11.178a.5.5 0 0 0-.144-.443l-8.122-7.916a.5.5 0 0 1 .278-.853l11.223-1.63a.5.5 0 0 0 .376-.274l5.02-10.17Z"></path></svg><span class="MRrY3">0</span></div><div class="opKvU"></div><span class="N_67f"><svg width="16" height="16" fill="none" xmlns="http://www.w3.org/2000/svg"><path fill-rule="evenodd" clip-rule="evenodd" d="M7.783 7.314a2 2 0 0 1 1.169 3.221l-1.888 2.332a2 2 0 1 1-3.109-2.518L4.967 9.1a1.431 1.431 0 0 1-.09-.372L4.725 7.28 2.919 9.51A3.333 3.333 0 1 0 8.1 13.705l1.888-2.332a3.333 3.333 0 0 0-1.304-5.173l-.9 1.113Z" fill="#939AA3"></path><path fill-rule="evenodd" clip-rule="evenodd" d="M8.58 9.291A2 2 0 0 1 7.411 6.07L9.3 3.74a2 2 0 1 1 3.11 2.517l-1.013 1.25c.046.116.077.241.09.372l.153 1.447 1.805-2.23a3.333 3.333 0 0 0-5.18-4.196L6.374 5.231a3.333 3.333 0 0 0 1.304 5.173l.901-1.113Z" fill="#939AA3"></path></svg></span></div></div><div class="zEp2T" style="cursor:auto"><div class="vuCON"><a href="/user/1977369294279187" target="_blank"><div style="width:20px;height:20px;font-size:10px" class="arco-avatar arco-avatar-circle"><span class="arco-avatar-image"><img src="https://p26-passport.byteacctimg.com/img/user-avatar/1816ccc0ebc35f4a89457e0b4ffe692b~300x300.image" alt="小窗幽记机器学习"/></span></div><div class="h_pFt"><span>小窗幽记机器学习</span></div></a></div><div class="eHpWb"><div class="BqxXc UG8D7"><div class="fHdNq"><div class="cSmO7"><a href="/articles/7440006127816802341" target="_blank">多模态大模型系列 | 20：低延迟、零遗忘：语音多模态新范式-Freeze-Omni</a></div><div style="margin-left:16px;display:inline-block;vertical-align:text-bottom"></div></div><div class="aklXs"><span class="MppYW"><span class="Ji9yj"></span><span class="HIKQo">AI</span></span><span class="KRApY">大模型</span><span class="KRApY">智能语音交互</span><span class="KRApY">数据中台</span></div><div class="IsnSN"><div class="TRHtk">简介 ========语音交互是人机沟通的未来，但现有多模态大语言模型在保持模型原始智能的同时实现低延迟对话一直是一大挑战。Freeze-Omni通过创新的三阶段训练策略，在不微调大语言模型的前提下，实现了高效、智能的端到端语音对话，为多模态LLM研究开辟了新路径。Q1: 这篇文章想要解决什么问题？A1: 本文致力于解决大型语言模型(LLM)在语音交互中的两个关键挑战：如何在不破坏LLM原有智能的</div><img class="v4MkD" src="https://p3-volc-community-sign.byteimg.com/tos-cn-i-tlddhu82om/5e95a6ad0bba40c09a4d902429453fef~tplv-tlddhu82om-image.image?=&amp;rk3s=8031ce6d&amp;x-expires=1732499851&amp;x-signature=l3q4KE%2FOWSTPDkTzu77bV5zKjZc%3D"/></div></div><div class="nvyao"><img src="https://p3-volc-community-sign.byteimg.com/tos-cn-i-tlddhu82om/5e95a6ad0bba40c09a4d902429453fef~tplv-tlddhu82om-image.image?=&amp;rk3s=8031ce6d&amp;x-expires=1732499851&amp;x-signature=l3q4KE%2FOWSTPDkTzu77bV5zKjZc%3D"/></div></div></div><div class="B_zhD mdM_r"><div class="J8PPi"><div class="lugd4"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-eye"><path d="M24 37c6.627 0 12.627-4.333 18-13-5.373-8.667-11.373-13-18-13-6.627 0-12.627 4.333-18 13 5.373 8.667 11.373 13 18 13Z" clip-rule="evenodd"></path><path d="M29 24a5 5 0 1 1-10 0 5 5 0 0 1 10 0Z"></path></svg><span class="MRrY3">7</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-thumb-up"><path d="M7 17v26m35.17-21.394-5.948 18.697a1 1 0 0 1-.953.697H14V19h3l9.403-12.223a1 1 0 0 1 1.386-.196l2.535 1.87a6 6 0 0 1 2.044 6.974L31 19h9.265a2 2 0 0 1 1.906 2.606Z"></path></svg><span class="MRrY3">0</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-message"><path d="M15 20h18m-18 9h9M7 41h17.63C33.67 41 41 33.67 41 24.63V24c0-9.389-7.611-17-17-17S7 14.611 7 24v17Z"></path></svg><span class="MRrY3">0</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-star"><path d="M22.552 6.908a.5.5 0 0 1 .896 0l5.02 10.17a.5.5 0 0 0 .376.274l11.224 1.631a.5.5 0 0 1 .277.853l-8.122 7.916a.5.5 0 0 0-.143.443l1.917 11.178a.5.5 0 0 1-.726.527l-10.038-5.278a.5.5 0 0 0-.466 0L12.73 39.9a.5.5 0 0 1-.726-.527l1.918-11.178a.5.5 0 0 0-.144-.443l-8.122-7.916a.5.5 0 0 1 .278-.853l11.223-1.63a.5.5 0 0 0 .376-.274l5.02-10.17Z"></path></svg><span class="MRrY3">0</span></div><div class="opKvU"></div><span class="N_67f"><svg width="16" height="16" fill="none" xmlns="http://www.w3.org/2000/svg"><path fill-rule="evenodd" clip-rule="evenodd" d="M7.783 7.314a2 2 0 0 1 1.169 3.221l-1.888 2.332a2 2 0 1 1-3.109-2.518L4.967 9.1a1.431 1.431 0 0 1-.09-.372L4.725 7.28 2.919 9.51A3.333 3.333 0 1 0 8.1 13.705l1.888-2.332a3.333 3.333 0 0 0-1.304-5.173l-.9 1.113Z" fill="#939AA3"></path><path fill-rule="evenodd" clip-rule="evenodd" d="M8.58 9.291A2 2 0 0 1 7.411 6.07L9.3 3.74a2 2 0 1 1 3.11 2.517l-1.013 1.25c.046.116.077.241.09.372l.153 1.447 1.805-2.23a3.333 3.333 0 0 0-5.18-4.196L6.374 5.231a3.333 3.333 0 0 0 1.304 5.173l.901-1.113Z" fill="#939AA3"></path></svg></span></div></div><div class="zEp2T" style="cursor:auto"><div class="vuCON"><a href="/user/1977369294279187" target="_blank"><div style="width:20px;height:20px;font-size:10px" class="arco-avatar arco-avatar-circle"><span class="arco-avatar-image"><img src="https://p26-passport.byteacctimg.com/img/user-avatar/1816ccc0ebc35f4a89457e0b4ffe692b~300x300.image" alt="小窗幽记机器学习"/></span></div><div class="h_pFt"><span>小窗幽记机器学习</span></div></a></div><div class="eHpWb"><div class="BqxXc UG8D7"><div class="fHdNq"><div class="cSmO7"><a href="/articles/7438453529850675237" target="_blank">全面解读Apple Ferret-UI 2，屏幕多模态大模型(详解版)</a></div><div style="margin-left:16px;display:inline-block;vertical-align:text-bottom"></div></div><div class="aklXs"><span class="MppYW"><span class="Ji9yj"></span><span class="HIKQo">大模型</span></span><span class="KRApY">大模型</span><span class="KRApY">向量数据库</span><span class="KRApY">云通信</span></div><div class="IsnSN"><div class="TRHtk">引言简介FERRET-UI 2数据集构建模型架构实验实验设置实验结果消融研究结论引言 =======用户界面(UI)是人机交互的核心，随着智能设备和平台的多样化，UI的复杂性不断增加。 然而，现有的UI理解与交互方法在多平台环境中仍有局限。 Ferret-UI(You et al., 2024) 在UI指称与语义理解方面取得了进展，但其固定的分辨率和仅针对移动设备的限制，难以应对多平台的复杂性。 </div><img class="v4MkD" src="https://p3-volc-community-sign.byteimg.com/tos-cn-i-tlddhu82om/d820fe0d2f26415c8fb6f85f95f5df05~tplv-tlddhu82om-image.image?=&amp;rk3s=8031ce6d&amp;x-expires=1732499851&amp;x-signature=xQF4mQYw4jnCauJvqMG3NFuw%2B5I%3D"/></div></div><div class="nvyao"><img src="https://p3-volc-community-sign.byteimg.com/tos-cn-i-tlddhu82om/d820fe0d2f26415c8fb6f85f95f5df05~tplv-tlddhu82om-image.image?=&amp;rk3s=8031ce6d&amp;x-expires=1732499851&amp;x-signature=xQF4mQYw4jnCauJvqMG3NFuw%2B5I%3D"/></div></div></div><div class="B_zhD mdM_r"><div class="J8PPi"><div class="lugd4"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-eye"><path d="M24 37c6.627 0 12.627-4.333 18-13-5.373-8.667-11.373-13-18-13-6.627 0-12.627 4.333-18 13 5.373 8.667 11.373 13 18 13Z" clip-rule="evenodd"></path><path d="M29 24a5 5 0 1 1-10 0 5 5 0 0 1 10 0Z"></path></svg><span class="MRrY3">56</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-thumb-up"><path d="M7 17v26m35.17-21.394-5.948 18.697a1 1 0 0 1-.953.697H14V19h3l9.403-12.223a1 1 0 0 1 1.386-.196l2.535 1.87a6 6 0 0 1 2.044 6.974L31 19h9.265a2 2 0 0 1 1.906 2.606Z"></path></svg><span class="MRrY3">0</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-message"><path d="M15 20h18m-18 9h9M7 41h17.63C33.67 41 41 33.67 41 24.63V24c0-9.389-7.611-17-17-17S7 14.611 7 24v17Z"></path></svg><span class="MRrY3">0</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-star"><path d="M22.552 6.908a.5.5 0 0 1 .896 0l5.02 10.17a.5.5 0 0 0 .376.274l11.224 1.631a.5.5 0 0 1 .277.853l-8.122 7.916a.5.5 0 0 0-.143.443l1.917 11.178a.5.5 0 0 1-.726.527l-10.038-5.278a.5.5 0 0 0-.466 0L12.73 39.9a.5.5 0 0 1-.726-.527l1.918-11.178a.5.5 0 0 0-.144-.443l-8.122-7.916a.5.5 0 0 1 .278-.853l11.223-1.63a.5.5 0 0 0 .376-.274l5.02-10.17Z"></path></svg><span class="MRrY3">0</span></div><div class="opKvU"></div><span class="N_67f"><svg width="16" height="16" fill="none" xmlns="http://www.w3.org/2000/svg"><path fill-rule="evenodd" clip-rule="evenodd" d="M7.783 7.314a2 2 0 0 1 1.169 3.221l-1.888 2.332a2 2 0 1 1-3.109-2.518L4.967 9.1a1.431 1.431 0 0 1-.09-.372L4.725 7.28 2.919 9.51A3.333 3.333 0 1 0 8.1 13.705l1.888-2.332a3.333 3.333 0 0 0-1.304-5.173l-.9 1.113Z" fill="#939AA3"></path><path fill-rule="evenodd" clip-rule="evenodd" d="M8.58 9.291A2 2 0 0 1 7.411 6.07L9.3 3.74a2 2 0 1 1 3.11 2.517l-1.013 1.25c.046.116.077.241.09.372l.153 1.447 1.805-2.23a3.333 3.333 0 0 0-5.18-4.196L6.374 5.231a3.333 3.333 0 0 0 1.304 5.173l.901-1.113Z" fill="#939AA3"></path></svg></span></div></div><div class="zEp2T" style="cursor:auto"><div class="vuCON"><a href="/user/1977369294279187" target="_blank"><div style="width:20px;height:20px;font-size:10px" class="arco-avatar arco-avatar-circle"><span class="arco-avatar-image"><img src="https://p26-passport.byteacctimg.com/img/user-avatar/1816ccc0ebc35f4a89457e0b4ffe692b~300x300.image" alt="小窗幽记机器学习"/></span></div><div class="h_pFt"><span>小窗幽记机器学习</span></div></a></div><div class="eHpWb"><div class="BqxXc UG8D7"><div class="fHdNq"><div class="cSmO7"><a href="/articles/7437404826503217189" target="_blank">苹果发布Ferret-UI 2: 跨平台UI理解多模态大模型(精炼版)</a></div><div style="margin-left:16px;display:inline-block;vertical-align:text-bottom"></div></div><div class="aklXs"><span class="MppYW"><span class="Ji9yj"></span><span class="HIKQo">AI</span></span><span class="KRApY">大模型</span><span class="KRApY">向量数据库</span><span class="KRApY">云通信</span></div><div class="IsnSN"><div class="TRHtk">引言简介方法实验结果总结实战引言 =======在数字设备日益普及的今天，用户界面(UI)已经成为人机交互的核心桥梁。近期，苹果公司发布的Ferret-UI 2 凭借其多平台兼容性和自适应编码等创新特性，在通用UI理解能力方面取得了显著突破。本文将简要介绍其核心技术与应用效果等要点 。如果小伙伴们想深入了解技术细节 ，欢迎关注下一篇文章：《全面解读Ferret-UI 2屏幕多模态大模型（详解版）》</div><img class="v4MkD" src="https://p3-volc-community-sign.byteimg.com/tos-cn-i-tlddhu82om/587b8d7930ca4ca0837b5d5f8023a5ff~tplv-tlddhu82om-image.image?=&amp;rk3s=8031ce6d&amp;x-expires=1732499851&amp;x-signature=oV0B55eNhjTvlhRh4I8MOkfmYEs%3D"/></div></div><div class="nvyao"><img src="https://p3-volc-community-sign.byteimg.com/tos-cn-i-tlddhu82om/587b8d7930ca4ca0837b5d5f8023a5ff~tplv-tlddhu82om-image.image?=&amp;rk3s=8031ce6d&amp;x-expires=1732499851&amp;x-signature=oV0B55eNhjTvlhRh4I8MOkfmYEs%3D"/></div></div></div><div class="B_zhD mdM_r"><div class="J8PPi"><div class="lugd4"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-eye"><path d="M24 37c6.627 0 12.627-4.333 18-13-5.373-8.667-11.373-13-18-13-6.627 0-12.627 4.333-18 13 5.373 8.667 11.373 13 18 13Z" clip-rule="evenodd"></path><path d="M29 24a5 5 0 1 1-10 0 5 5 0 0 1 10 0Z"></path></svg><span class="MRrY3">32</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-thumb-up"><path d="M7 17v26m35.17-21.394-5.948 18.697a1 1 0 0 1-.953.697H14V19h3l9.403-12.223a1 1 0 0 1 1.386-.196l2.535 1.87a6 6 0 0 1 2.044 6.974L31 19h9.265a2 2 0 0 1 1.906 2.606Z"></path></svg><span class="MRrY3">0</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-message"><path d="M15 20h18m-18 9h9M7 41h17.63C33.67 41 41 33.67 41 24.63V24c0-9.389-7.611-17-17-17S7 14.611 7 24v17Z"></path></svg><span class="MRrY3">0</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-star"><path d="M22.552 6.908a.5.5 0 0 1 .896 0l5.02 10.17a.5.5 0 0 0 .376.274l11.224 1.631a.5.5 0 0 1 .277.853l-8.122 7.916a.5.5 0 0 0-.143.443l1.917 11.178a.5.5 0 0 1-.726.527l-10.038-5.278a.5.5 0 0 0-.466 0L12.73 39.9a.5.5 0 0 1-.726-.527l1.918-11.178a.5.5 0 0 0-.144-.443l-8.122-7.916a.5.5 0 0 1 .278-.853l11.223-1.63a.5.5 0 0 0 .376-.274l5.02-10.17Z"></path></svg><span class="MRrY3">0</span></div><div class="opKvU"></div><span class="N_67f"><svg width="16" height="16" fill="none" xmlns="http://www.w3.org/2000/svg"><path fill-rule="evenodd" clip-rule="evenodd" d="M7.783 7.314a2 2 0 0 1 1.169 3.221l-1.888 2.332a2 2 0 1 1-3.109-2.518L4.967 9.1a1.431 1.431 0 0 1-.09-.372L4.725 7.28 2.919 9.51A3.333 3.333 0 1 0 8.1 13.705l1.888-2.332a3.333 3.333 0 0 0-1.304-5.173l-.9 1.113Z" fill="#939AA3"></path><path fill-rule="evenodd" clip-rule="evenodd" d="M8.58 9.291A2 2 0 0 1 7.411 6.07L9.3 3.74a2 2 0 1 1 3.11 2.517l-1.013 1.25c.046.116.077.241.09.372l.153 1.447 1.805-2.23a3.333 3.333 0 0 0-5.18-4.196L6.374 5.231a3.333 3.333 0 0 0 1.304 5.173l.901-1.113Z" fill="#939AA3"></path></svg></span></div></div><div class="zEp2T" style="cursor:auto"><div class="vuCON"><a href="/user/1977369294279187" target="_blank"><div style="width:20px;height:20px;font-size:10px" class="arco-avatar arco-avatar-circle"><span class="arco-avatar-image"><img src="https://p26-passport.byteacctimg.com/img/user-avatar/1816ccc0ebc35f4a89457e0b4ffe692b~300x300.image" alt="小窗幽记机器学习"/></span></div><div class="h_pFt"><span>小窗幽记机器学习</span></div></a></div><div class="eHpWb"><div class="BqxXc UG8D7"><div class="fHdNq"><div class="cSmO7"><a href="/articles/7436341741323550771" target="_blank">Agent系列：AppAgent v2-屏幕智能Agent(详解版)</a></div><div style="margin-left:16px;display:inline-block;vertical-align:text-bottom"></div></div><div class="aklXs"><span class="MppYW"><span class="Ji9yj"></span><span class="HIKQo">移动开发</span></span><span class="KRApY">大模型</span><span class="KRApY">数据库</span><span class="KRApY">算法</span></div><div class="IsnSN"><div class="TRHtk">引言简介方法Agent 框架Agent 交互探索阶段部署阶段文档生成高级功能实验结果总结局限性未来工作引言 =======大语言模型（LLM）如 ChatGPT 和 GPT-4 显著提升了自然语言处理能力，并且推动了智能体在自主决策中的应用。最初，这些智能体专为基于文本的交互方式设计，展现了卓越的表现，包括记忆自适应性和多任务处理能力。然而，现实世界的应用程序不仅仅局限于文本输入，还涉及视觉和其他</div><img class="v4MkD" src="https://p3-volc-community-sign.byteimg.com/tos-cn-i-tlddhu82om/8d019bcbb7244532a63a8702460e7135~tplv-tlddhu82om-image.image?=&amp;rk3s=8031ce6d&amp;x-expires=1732499851&amp;x-signature=U4BghdJ3CO%2Bi5ZwckoqdAgiVOOs%3D"/></div></div><div class="nvyao"><img src="https://p3-volc-community-sign.byteimg.com/tos-cn-i-tlddhu82om/8d019bcbb7244532a63a8702460e7135~tplv-tlddhu82om-image.image?=&amp;rk3s=8031ce6d&amp;x-expires=1732499851&amp;x-signature=U4BghdJ3CO%2Bi5ZwckoqdAgiVOOs%3D"/></div></div></div><div class="B_zhD mdM_r"><div class="J8PPi"><div class="lugd4"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-eye"><path d="M24 37c6.627 0 12.627-4.333 18-13-5.373-8.667-11.373-13-18-13-6.627 0-12.627 4.333-18 13 5.373 8.667 11.373 13 18 13Z" clip-rule="evenodd"></path><path d="M29 24a5 5 0 1 1-10 0 5 5 0 0 1 10 0Z"></path></svg><span class="MRrY3">235</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-thumb-up"><path d="M7 17v26m35.17-21.394-5.948 18.697a1 1 0 0 1-.953.697H14V19h3l9.403-12.223a1 1 0 0 1 1.386-.196l2.535 1.87a6 6 0 0 1 2.044 6.974L31 19h9.265a2 2 0 0 1 1.906 2.606Z"></path></svg><span class="MRrY3">0</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-message"><path d="M15 20h18m-18 9h9M7 41h17.63C33.67 41 41 33.67 41 24.63V24c0-9.389-7.611-17-17-17S7 14.611 7 24v17Z"></path></svg><span class="MRrY3">0</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-star"><path d="M22.552 6.908a.5.5 0 0 1 .896 0l5.02 10.17a.5.5 0 0 0 .376.274l11.224 1.631a.5.5 0 0 1 .277.853l-8.122 7.916a.5.5 0 0 0-.143.443l1.917 11.178a.5.5 0 0 1-.726.527l-10.038-5.278a.5.5 0 0 0-.466 0L12.73 39.9a.5.5 0 0 1-.726-.527l1.918-11.178a.5.5 0 0 0-.144-.443l-8.122-7.916a.5.5 0 0 1 .278-.853l11.223-1.63a.5.5 0 0 0 .376-.274l5.02-10.17Z"></path></svg><span class="MRrY3">0</span></div><div class="opKvU"></div><span class="N_67f"><svg width="16" height="16" fill="none" xmlns="http://www.w3.org/2000/svg"><path fill-rule="evenodd" clip-rule="evenodd" d="M7.783 7.314a2 2 0 0 1 1.169 3.221l-1.888 2.332a2 2 0 1 1-3.109-2.518L4.967 9.1a1.431 1.431 0 0 1-.09-.372L4.725 7.28 2.919 9.51A3.333 3.333 0 1 0 8.1 13.705l1.888-2.332a3.333 3.333 0 0 0-1.304-5.173l-.9 1.113Z" fill="#939AA3"></path><path fill-rule="evenodd" clip-rule="evenodd" d="M8.58 9.291A2 2 0 0 1 7.411 6.07L9.3 3.74a2 2 0 1 1 3.11 2.517l-1.013 1.25c.046.116.077.241.09.372l.153 1.447 1.805-2.23a3.333 3.333 0 0 0-5.18-4.196L6.374 5.231a3.333 3.333 0 0 0 1.304 5.173l.901-1.113Z" fill="#939AA3"></path></svg></span></div></div><div class="zEp2T" style="cursor:auto"><div class="vuCON"><a href="/user/1977369294279187" target="_blank"><div style="width:20px;height:20px;font-size:10px" class="arco-avatar arco-avatar-circle"><span class="arco-avatar-image"><img src="https://p26-passport.byteacctimg.com/img/user-avatar/1816ccc0ebc35f4a89457e0b4ffe692b~300x300.image" alt="小窗幽记机器学习"/></span></div><div class="h_pFt"><span>小窗幽记机器学习</span></div></a></div><div class="eHpWb"><div class="BqxXc UG8D7"><div class="fHdNq"><div class="cSmO7"><a href="/articles/7436341741281935410" target="_blank">Agent系列：多模态智能体AppAgent v2助力AI手机(简化版)</a></div><div style="margin-left:16px;display:inline-block;vertical-align:text-bottom"></div></div><div class="aklXs"><span class="MppYW"><span class="Ji9yj"></span><span class="HIKQo">AI</span></span><span class="KRApY">智能应用</span><span class="KRApY">数据库</span><span class="KRApY">图像处理</span></div><div class="IsnSN"><div class="TRHtk">AppAgent v2是一种专为移动设备设计的多模态智能体框架。该框架能够在移动设备上导航，模拟用户交互，适应各种应用程序，具体通过解析器、文本和视觉描述来增强其灵活性。AppAgent v2 的操作分为探索和部署两个阶段。</div><img class="v4MkD" src="https://p6-volc-community-sign.byteimg.com/tos-cn-i-tlddhu82om/bc97023f97e84c5ca3671d20715a958c~tplv-tlddhu82om-image.image?=&amp;rk3s=8031ce6d&amp;x-expires=1732499851&amp;x-signature=E8%2BE7IPt%2Bl3%2BVFKie3%2BFj9uO9tM%3D"/></div></div><div class="nvyao"><img src="https://p6-volc-community-sign.byteimg.com/tos-cn-i-tlddhu82om/bc97023f97e84c5ca3671d20715a958c~tplv-tlddhu82om-image.image?=&amp;rk3s=8031ce6d&amp;x-expires=1732499851&amp;x-signature=E8%2BE7IPt%2Bl3%2BVFKie3%2BFj9uO9tM%3D"/></div></div></div><div class="B_zhD mdM_r"><div class="J8PPi"><div class="lugd4"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-eye"><path d="M24 37c6.627 0 12.627-4.333 18-13-5.373-8.667-11.373-13-18-13-6.627 0-12.627 4.333-18 13 5.373 8.667 11.373 13 18 13Z" clip-rule="evenodd"></path><path d="M29 24a5 5 0 1 1-10 0 5 5 0 0 1 10 0Z"></path></svg><span class="MRrY3">43</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-thumb-up"><path d="M7 17v26m35.17-21.394-5.948 18.697a1 1 0 0 1-.953.697H14V19h3l9.403-12.223a1 1 0 0 1 1.386-.196l2.535 1.87a6 6 0 0 1 2.044 6.974L31 19h9.265a2 2 0 0 1 1.906 2.606Z"></path></svg><span class="MRrY3">0</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-message"><path d="M15 20h18m-18 9h9M7 41h17.63C33.67 41 41 33.67 41 24.63V24c0-9.389-7.611-17-17-17S7 14.611 7 24v17Z"></path></svg><span class="MRrY3">0</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-star"><path d="M22.552 6.908a.5.5 0 0 1 .896 0l5.02 10.17a.5.5 0 0 0 .376.274l11.224 1.631a.5.5 0 0 1 .277.853l-8.122 7.916a.5.5 0 0 0-.143.443l1.917 11.178a.5.5 0 0 1-.726.527l-10.038-5.278a.5.5 0 0 0-.466 0L12.73 39.9a.5.5 0 0 1-.726-.527l1.918-11.178a.5.5 0 0 0-.144-.443l-8.122-7.916a.5.5 0 0 1 .278-.853l11.223-1.63a.5.5 0 0 0 .376-.274l5.02-10.17Z"></path></svg><span class="MRrY3">0</span></div><div class="opKvU"></div><span class="N_67f"><svg width="16" height="16" fill="none" xmlns="http://www.w3.org/2000/svg"><path fill-rule="evenodd" clip-rule="evenodd" d="M7.783 7.314a2 2 0 0 1 1.169 3.221l-1.888 2.332a2 2 0 1 1-3.109-2.518L4.967 9.1a1.431 1.431 0 0 1-.09-.372L4.725 7.28 2.919 9.51A3.333 3.333 0 1 0 8.1 13.705l1.888-2.332a3.333 3.333 0 0 0-1.304-5.173l-.9 1.113Z" fill="#939AA3"></path><path fill-rule="evenodd" clip-rule="evenodd" d="M8.58 9.291A2 2 0 0 1 7.411 6.07L9.3 3.74a2 2 0 1 1 3.11 2.517l-1.013 1.25c.046.116.077.241.09.372l.153 1.447 1.805-2.23a3.333 3.333 0 0 0-5.18-4.196L6.374 5.231a3.333 3.333 0 0 0 1.304 5.173l.901-1.113Z" fill="#939AA3"></path></svg></span></div></div><div class="zEp2T" style="cursor:auto"><div class="vuCON"><a href="/user/1977369294279187" target="_blank"><div style="width:20px;height:20px;font-size:10px" class="arco-avatar arco-avatar-circle"><span class="arco-avatar-image"><img src="https://p26-passport.byteacctimg.com/img/user-avatar/1816ccc0ebc35f4a89457e0b4ffe692b~300x300.image" alt="小窗幽记机器学习"/></span></div><div class="h_pFt"><span>小窗幽记机器学习</span></div></a></div><div class="eHpWb"><div class="BqxXc UG8D7"><div class="fHdNq"><div class="cSmO7"><a href="/articles/7433674203150155802" target="_blank">万字长文细说端侧大模型进展(下篇)：AutoGLM类Agent隐私安全有感</a></div><div style="margin-left:16px;display:inline-block;vertical-align:text-bottom"></div></div><div class="aklXs"><span class="MppYW"><span class="Ji9yj"></span><span class="HIKQo">大模型</span></span><span class="KRApY">大模型</span><span class="KRApY">机器学习</span><span class="KRApY">数据库</span></div><div class="IsnSN"><div class="TRHtk">引言 =======铺床凉满梧桐月，月在梧桐缺处明。小伙伴们好，我是微信公众号《小窗幽记机器学习》的小编：卖沙茶面的小女孩。最近，智谱AI推出的智能体AutoGLM引起了广泛关注。然而，随着测试的深入，用户对该产品将屏幕数据上传至云端所带来的隐私安全问题愈发担忧。因此，今天小编将基于近期一篇关于端侧大模型的综述文章，介绍该领域的一些最新进展。随着端侧大模型的发展，这些隐私安全问题有望得到有效缓解。</div><img class="v4MkD" src="https://p3-volc-community-sign.byteimg.com/tos-cn-i-tlddhu82om/888587e1302640638ba80b2ac0cefc10~tplv-tlddhu82om-image.image?=&amp;rk3s=8031ce6d&amp;x-expires=1732499851&amp;x-signature=QIOIFnRgLSElOGv1wviyI1qadX4%3D"/></div></div><div class="nvyao"><img src="https://p3-volc-community-sign.byteimg.com/tos-cn-i-tlddhu82om/888587e1302640638ba80b2ac0cefc10~tplv-tlddhu82om-image.image?=&amp;rk3s=8031ce6d&amp;x-expires=1732499851&amp;x-signature=QIOIFnRgLSElOGv1wviyI1qadX4%3D"/></div></div></div><div class="B_zhD mdM_r"><div class="J8PPi"><div class="lugd4"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-eye"><path d="M24 37c6.627 0 12.627-4.333 18-13-5.373-8.667-11.373-13-18-13-6.627 0-12.627 4.333-18 13 5.373 8.667 11.373 13 18 13Z" clip-rule="evenodd"></path><path d="M29 24a5 5 0 1 1-10 0 5 5 0 0 1 10 0Z"></path></svg><span class="MRrY3">310</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-thumb-up"><path d="M7 17v26m35.17-21.394-5.948 18.697a1 1 0 0 1-.953.697H14V19h3l9.403-12.223a1 1 0 0 1 1.386-.196l2.535 1.87a6 6 0 0 1 2.044 6.974L31 19h9.265a2 2 0 0 1 1.906 2.606Z"></path></svg><span class="MRrY3">0</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-message"><path d="M15 20h18m-18 9h9M7 41h17.63C33.67 41 41 33.67 41 24.63V24c0-9.389-7.611-17-17-17S7 14.611 7 24v17Z"></path></svg><span class="MRrY3">0</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-star"><path d="M22.552 6.908a.5.5 0 0 1 .896 0l5.02 10.17a.5.5 0 0 0 .376.274l11.224 1.631a.5.5 0 0 1 .277.853l-8.122 7.916a.5.5 0 0 0-.143.443l1.917 11.178a.5.5 0 0 1-.726.527l-10.038-5.278a.5.5 0 0 0-.466 0L12.73 39.9a.5.5 0 0 1-.726-.527l1.918-11.178a.5.5 0 0 0-.144-.443l-8.122-7.916a.5.5 0 0 1 .278-.853l11.223-1.63a.5.5 0 0 0 .376-.274l5.02-10.17Z"></path></svg><span class="MRrY3">0</span></div><div class="opKvU"></div><span class="N_67f"><svg width="16" height="16" fill="none" xmlns="http://www.w3.org/2000/svg"><path fill-rule="evenodd" clip-rule="evenodd" d="M7.783 7.314a2 2 0 0 1 1.169 3.221l-1.888 2.332a2 2 0 1 1-3.109-2.518L4.967 9.1a1.431 1.431 0 0 1-.09-.372L4.725 7.28 2.919 9.51A3.333 3.333 0 1 0 8.1 13.705l1.888-2.332a3.333 3.333 0 0 0-1.304-5.173l-.9 1.113Z" fill="#939AA3"></path><path fill-rule="evenodd" clip-rule="evenodd" d="M8.58 9.291A2 2 0 0 1 7.411 6.07L9.3 3.74a2 2 0 1 1 3.11 2.517l-1.013 1.25c.046.116.077.241.09.372l.153 1.447 1.805-2.23a3.333 3.333 0 0 0-5.18-4.196L6.374 5.231a3.333 3.333 0 0 0 1.304 5.173l.901-1.113Z" fill="#939AA3"></path></svg></span></div></div><div class="zEp2T" style="cursor:auto"><div class="vuCON"><a href="/user/1977369294279187" target="_blank"><div style="width:20px;height:20px;font-size:10px" class="arco-avatar arco-avatar-circle"><span class="arco-avatar-image"><img src="https://p26-passport.byteacctimg.com/img/user-avatar/1816ccc0ebc35f4a89457e0b4ffe692b~300x300.image" alt="小窗幽记机器学习"/></span></div><div class="h_pFt"><span>小窗幽记机器学习</span></div></a></div><div class="eHpWb"><div class="BqxXc UG8D7"><div class="fHdNq"><div class="cSmO7"><a href="/articles/7433674165921513523" target="_blank">万字长文细说端侧大模型进展(综述)</a></div><div style="margin-left:16px;display:inline-block;vertical-align:text-bottom"></div></div><div class="aklXs"><span class="MppYW"><span class="Ji9yj"></span><span class="HIKQo">大模型</span></span><span class="KRApY">大模型</span><span class="KRApY">数据安全</span><span class="KRApY">机器学习</span></div><div class="IsnSN"><div class="TRHtk">文本介绍端侧大模型的最新进展，助力端侧智能体Agent发展。分为上下两篇：上篇主要介绍端侧大模型的进展及其模型架构，下篇则聚焦于端侧大模型的模型压缩技术、加速和部署方案以及应用实例。</div><img class="v4MkD" src="https://p3-volc-community-sign.byteimg.com/tos-cn-i-tlddhu82om/288d54381cb54c8e9d6da7f22fcc9d16~tplv-tlddhu82om-image.image?=&amp;rk3s=8031ce6d&amp;x-expires=1732499851&amp;x-signature=osc8TohEehgQpQduxQNjRrE2b1o%3D"/></div></div><div class="nvyao"><img src="https://p3-volc-community-sign.byteimg.com/tos-cn-i-tlddhu82om/288d54381cb54c8e9d6da7f22fcc9d16~tplv-tlddhu82om-image.image?=&amp;rk3s=8031ce6d&amp;x-expires=1732499851&amp;x-signature=osc8TohEehgQpQduxQNjRrE2b1o%3D"/></div></div></div><div class="B_zhD mdM_r"><div class="J8PPi"><div class="lugd4"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-eye"><path d="M24 37c6.627 0 12.627-4.333 18-13-5.373-8.667-11.373-13-18-13-6.627 0-12.627 4.333-18 13 5.373 8.667 11.373 13 18 13Z" clip-rule="evenodd"></path><path d="M29 24a5 5 0 1 1-10 0 5 5 0 0 1 10 0Z"></path></svg><span class="MRrY3">1038</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-thumb-up"><path d="M7 17v26m35.17-21.394-5.948 18.697a1 1 0 0 1-.953.697H14V19h3l9.403-12.223a1 1 0 0 1 1.386-.196l2.535 1.87a6 6 0 0 1 2.044 6.974L31 19h9.265a2 2 0 0 1 1.906 2.606Z"></path></svg><span class="MRrY3">0</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-message"><path d="M15 20h18m-18 9h9M7 41h17.63C33.67 41 41 33.67 41 24.63V24c0-9.389-7.611-17-17-17S7 14.611 7 24v17Z"></path></svg><span class="MRrY3">0</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-star"><path d="M22.552 6.908a.5.5 0 0 1 .896 0l5.02 10.17a.5.5 0 0 0 .376.274l11.224 1.631a.5.5 0 0 1 .277.853l-8.122 7.916a.5.5 0 0 0-.143.443l1.917 11.178a.5.5 0 0 1-.726.527l-10.038-5.278a.5.5 0 0 0-.466 0L12.73 39.9a.5.5 0 0 1-.726-.527l1.918-11.178a.5.5 0 0 0-.144-.443l-8.122-7.916a.5.5 0 0 1 .278-.853l11.223-1.63a.5.5 0 0 0 .376-.274l5.02-10.17Z"></path></svg><span class="MRrY3">1</span></div><div class="opKvU"></div><span class="N_67f"><svg width="16" height="16" fill="none" xmlns="http://www.w3.org/2000/svg"><path fill-rule="evenodd" clip-rule="evenodd" d="M7.783 7.314a2 2 0 0 1 1.169 3.221l-1.888 2.332a2 2 0 1 1-3.109-2.518L4.967 9.1a1.431 1.431 0 0 1-.09-.372L4.725 7.28 2.919 9.51A3.333 3.333 0 1 0 8.1 13.705l1.888-2.332a3.333 3.333 0 0 0-1.304-5.173l-.9 1.113Z" fill="#939AA3"></path><path fill-rule="evenodd" clip-rule="evenodd" d="M8.58 9.291A2 2 0 0 1 7.411 6.07L9.3 3.74a2 2 0 1 1 3.11 2.517l-1.013 1.25c.046.116.077.241.09.372l.153 1.447 1.805-2.23a3.333 3.333 0 0 0-5.18-4.196L6.374 5.231a3.333 3.333 0 0 0 1.304 5.173l.901-1.113Z" fill="#939AA3"></path></svg></span></div></div><div class="zEp2T" style="cursor:auto"><div class="vuCON"><a href="/user/1977369294279187" target="_blank"><div style="width:20px;height:20px;font-size:10px" class="arco-avatar arco-avatar-circle"><span class="arco-avatar-image"><img src="https://p26-passport.byteacctimg.com/img/user-avatar/1816ccc0ebc35f4a89457e0b4ffe692b~300x300.image" alt="小窗幽记机器学习"/></span></div><div class="h_pFt"><span>小窗幽记机器学习</span></div></a></div><div class="eHpWb"><div class="BqxXc UG8D7"><div class="fHdNq"><div class="cSmO7"><a href="/articles/7433674165934096393" target="_blank">AI手机新纪元：AutoGLM开启后APP时代下的挑战与机遇</a></div><div style="margin-left:16px;display:inline-block;vertical-align:text-bottom"></div></div><div class="aklXs"><span class="MppYW"><span class="Ji9yj"></span><span class="HIKQo">AI</span></span><span class="KRApY">大模型</span><span class="KRApY">智能语音交互</span><span class="KRApY">数据安全</span></div><div class="IsnSN"><div class="TRHtk">山黛远，月波长，暮云秋影蘸潇湘。小伙伴们好，我是微信公众号《小窗幽记机器学习》的小编：卖热干面的小女孩。近日，国内知名AI公司智谱在中国计算机大会（CNCC）上发布的AutoGLM引发了业界广泛关注。这款突破性的AI智能体产品能通过语音指令理解用户意图，模拟人类操作手机，自动完成从网页浏览、商品购物到社交媒体互动等多种复杂任务。同期，Anthropic公司推出的Claude 3.5系列模型（Son</div><img class="v4MkD" src="https://p6-volc-community-sign.byteimg.com/tos-cn-i-tlddhu82om/870e42ff182049b8a337f1be96b5ba2e~tplv-tlddhu82om-image.image?=&amp;rk3s=8031ce6d&amp;x-expires=1732499851&amp;x-signature=Rb4%2Fn%2FNFZqv%2Bx0RGTF%2BChI51DQs%3D"/></div></div><div class="nvyao"><img src="https://p6-volc-community-sign.byteimg.com/tos-cn-i-tlddhu82om/870e42ff182049b8a337f1be96b5ba2e~tplv-tlddhu82om-image.image?=&amp;rk3s=8031ce6d&amp;x-expires=1732499851&amp;x-signature=Rb4%2Fn%2FNFZqv%2Bx0RGTF%2BChI51DQs%3D"/></div></div></div><div class="B_zhD mdM_r"><div class="J8PPi"><div class="lugd4"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-eye"><path d="M24 37c6.627 0 12.627-4.333 18-13-5.373-8.667-11.373-13-18-13-6.627 0-12.627 4.333-18 13 5.373 8.667 11.373 13 18 13Z" clip-rule="evenodd"></path><path d="M29 24a5 5 0 1 1-10 0 5 5 0 0 1 10 0Z"></path></svg><span class="MRrY3">65</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-thumb-up"><path d="M7 17v26m35.17-21.394-5.948 18.697a1 1 0 0 1-.953.697H14V19h3l9.403-12.223a1 1 0 0 1 1.386-.196l2.535 1.87a6 6 0 0 1 2.044 6.974L31 19h9.265a2 2 0 0 1 1.906 2.606Z"></path></svg><span class="MRrY3">0</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-message"><path d="M15 20h18m-18 9h9M7 41h17.63C33.67 41 41 33.67 41 24.63V24c0-9.389-7.611-17-17-17S7 14.611 7 24v17Z"></path></svg><span class="MRrY3">0</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-star"><path d="M22.552 6.908a.5.5 0 0 1 .896 0l5.02 10.17a.5.5 0 0 0 .376.274l11.224 1.631a.5.5 0 0 1 .277.853l-8.122 7.916a.5.5 0 0 0-.143.443l1.917 11.178a.5.5 0 0 1-.726.527l-10.038-5.278a.5.5 0 0 0-.466 0L12.73 39.9a.5.5 0 0 1-.726-.527l1.918-11.178a.5.5 0 0 0-.144-.443l-8.122-7.916a.5.5 0 0 1 .278-.853l11.223-1.63a.5.5 0 0 0 .376-.274l5.02-10.17Z"></path></svg><span class="MRrY3">0</span></div><div class="opKvU"></div><span class="N_67f"><svg width="16" height="16" fill="none" xmlns="http://www.w3.org/2000/svg"><path fill-rule="evenodd" clip-rule="evenodd" d="M7.783 7.314a2 2 0 0 1 1.169 3.221l-1.888 2.332a2 2 0 1 1-3.109-2.518L4.967 9.1a1.431 1.431 0 0 1-.09-.372L4.725 7.28 2.919 9.51A3.333 3.333 0 1 0 8.1 13.705l1.888-2.332a3.333 3.333 0 0 0-1.304-5.173l-.9 1.113Z" fill="#939AA3"></path><path fill-rule="evenodd" clip-rule="evenodd" d="M8.58 9.291A2 2 0 0 1 7.411 6.07L9.3 3.74a2 2 0 1 1 3.11 2.517l-1.013 1.25c.046.116.077.241.09.372l.153 1.447 1.805-2.23a3.333 3.333 0 0 0-5.18-4.196L6.374 5.231a3.333 3.333 0 0 0 1.304 5.173l.901-1.113Z" fill="#939AA3"></path></svg></span></div></div><div class="zEp2T" style="cursor:auto"><div class="vuCON"><a href="/user/1977369294279187" target="_blank"><div style="width:20px;height:20px;font-size:10px" class="arco-avatar arco-avatar-circle"><span class="arco-avatar-image"><img src="https://p26-passport.byteacctimg.com/img/user-avatar/1816ccc0ebc35f4a89457e0b4ffe692b~300x300.image" alt="小窗幽记机器学习"/></span></div><div class="h_pFt"><span>小窗幽记机器学习</span></div></a></div><div class="eHpWb"><div class="BqxXc UG8D7"><div class="fHdNq"><div class="cSmO7"><a href="/articles/7432210796606423078" target="_blank">万字长文梳理端侧大模型进展(上篇)：由AutoGLM类Agent隐私安全有感</a></div><div style="margin-left:16px;display:inline-block;vertical-align:text-bottom"></div></div><div class="aklXs"><span class="MppYW"><span class="Ji9yj"></span><span class="HIKQo">大模型</span></span><span class="KRApY">大模型</span><span class="KRApY">机器学习</span><span class="KRApY">算法</span></div><div class="IsnSN"><div class="TRHtk">文本介绍端侧大模型的最新进展，助力端侧智能体Agent发展。分为上下两篇：上篇主要介绍端侧大模型的进展及其模型架构，下篇则聚焦于端侧大模型的模型压缩技术、加速和部署方案以及应用实例。</div><img class="v4MkD" src="https://p3-volc-community-sign.byteimg.com/tos-cn-i-tlddhu82om/0623d2e339f64d839a3ae835e2b63b2b~tplv-tlddhu82om-image.image?=&amp;rk3s=8031ce6d&amp;x-expires=1732499851&amp;x-signature=Oni8DnU4xBjrEYCL44RA%2BPmj8eU%3D"/></div></div><div class="nvyao"><img src="https://p3-volc-community-sign.byteimg.com/tos-cn-i-tlddhu82om/0623d2e339f64d839a3ae835e2b63b2b~tplv-tlddhu82om-image.image?=&amp;rk3s=8031ce6d&amp;x-expires=1732499851&amp;x-signature=Oni8DnU4xBjrEYCL44RA%2BPmj8eU%3D"/></div></div></div><div class="B_zhD mdM_r"><div class="J8PPi"><div class="lugd4"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-eye"><path d="M24 37c6.627 0 12.627-4.333 18-13-5.373-8.667-11.373-13-18-13-6.627 0-12.627 4.333-18 13 5.373 8.667 11.373 13 18 13Z" clip-rule="evenodd"></path><path d="M29 24a5 5 0 1 1-10 0 5 5 0 0 1 10 0Z"></path></svg><span class="MRrY3">299</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-thumb-up"><path d="M7 17v26m35.17-21.394-5.948 18.697a1 1 0 0 1-.953.697H14V19h3l9.403-12.223a1 1 0 0 1 1.386-.196l2.535 1.87a6 6 0 0 1 2.044 6.974L31 19h9.265a2 2 0 0 1 1.906 2.606Z"></path></svg><span class="MRrY3">0</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-message"><path d="M15 20h18m-18 9h9M7 41h17.63C33.67 41 41 33.67 41 24.63V24c0-9.389-7.611-17-17-17S7 14.611 7 24v17Z"></path></svg><span class="MRrY3">0</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-star"><path d="M22.552 6.908a.5.5 0 0 1 .896 0l5.02 10.17a.5.5 0 0 0 .376.274l11.224 1.631a.5.5 0 0 1 .277.853l-8.122 7.916a.5.5 0 0 0-.143.443l1.917 11.178a.5.5 0 0 1-.726.527l-10.038-5.278a.5.5 0 0 0-.466 0L12.73 39.9a.5.5 0 0 1-.726-.527l1.918-11.178a.5.5 0 0 0-.144-.443l-8.122-7.916a.5.5 0 0 1 .278-.853l11.223-1.63a.5.5 0 0 0 .376-.274l5.02-10.17Z"></path></svg><span class="MRrY3">0</span></div><div class="opKvU"></div><span class="N_67f"><svg width="16" height="16" fill="none" xmlns="http://www.w3.org/2000/svg"><path fill-rule="evenodd" clip-rule="evenodd" d="M7.783 7.314a2 2 0 0 1 1.169 3.221l-1.888 2.332a2 2 0 1 1-3.109-2.518L4.967 9.1a1.431 1.431 0 0 1-.09-.372L4.725 7.28 2.919 9.51A3.333 3.333 0 1 0 8.1 13.705l1.888-2.332a3.333 3.333 0 0 0-1.304-5.173l-.9 1.113Z" fill="#939AA3"></path><path fill-rule="evenodd" clip-rule="evenodd" d="M8.58 9.291A2 2 0 0 1 7.411 6.07L9.3 3.74a2 2 0 1 1 3.11 2.517l-1.013 1.25c.046.116.077.241.09.372l.153 1.447 1.805-2.23a3.333 3.333 0 0 0-5.18-4.196L6.374 5.231a3.333 3.333 0 0 0 1.304 5.173l.901-1.113Z" fill="#939AA3"></path></svg></span></div></div><div class="zEp2T" style="cursor:auto"><div class="vuCON"><a href="/user/1977369294279187" target="_blank"><div style="width:20px;height:20px;font-size:10px" class="arco-avatar arco-avatar-circle"><span class="arco-avatar-image"><img src="https://p26-passport.byteacctimg.com/img/user-avatar/1816ccc0ebc35f4a89457e0b4ffe692b~300x300.image" alt="小窗幽记机器学习"/></span></div><div class="h_pFt"><span>小窗幽记机器学习</span></div></a></div><div class="eHpWb"><div class="BqxXc UG8D7"><div class="fHdNq"><div class="cSmO7"><a href="/articles/7428432985748078602" target="_blank">万字长文深度解读Movie Gen技术原理(5部曲)：图像&amp;视频联合生成模型 (2)</a></div><div style="margin-left:16px;display:inline-block;vertical-align:text-bottom"></div></div><div class="aklXs"><span class="MppYW"><span class="Ji9yj"></span><span class="HIKQo">音视频</span></span><span class="KRApY">大模型</span><span class="KRApY">机器学习</span><span class="KRApY">图像处理</span></div><div class="IsnSN"><div class="TRHtk">详细介绍Movie Gen中图像和视频的联合生成技术，包括:时间自编码器的设计与优化、基于流匹配的训练目标、联合生成的网络架构、文本嵌入和视觉-文本生成方法、空间上采样技术、模型扩展和训练效率优化等。此外，还详细介绍了预训练数据的准备过程.</div><img class="v4MkD" src="https://p3-volc-community-sign.byteimg.com/tos-cn-i-tlddhu82om/b9d1d1cee6e44e01bb9ddbddcc03f31e~tplv-tlddhu82om-image.image?=&amp;rk3s=8031ce6d&amp;x-expires=1732499851&amp;x-signature=IEySpvUDnvIyj6WlYo39r2px1uc%3D"/></div></div><div class="nvyao"><img src="https://p3-volc-community-sign.byteimg.com/tos-cn-i-tlddhu82om/b9d1d1cee6e44e01bb9ddbddcc03f31e~tplv-tlddhu82om-image.image?=&amp;rk3s=8031ce6d&amp;x-expires=1732499851&amp;x-signature=IEySpvUDnvIyj6WlYo39r2px1uc%3D"/></div></div></div><div class="B_zhD mdM_r"><div class="J8PPi"><div class="lugd4"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-eye"><path d="M24 37c6.627 0 12.627-4.333 18-13-5.373-8.667-11.373-13-18-13-6.627 0-12.627 4.333-18 13 5.373 8.667 11.373 13 18 13Z" clip-rule="evenodd"></path><path d="M29 24a5 5 0 1 1-10 0 5 5 0 0 1 10 0Z"></path></svg><span class="MRrY3">61</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-thumb-up"><path d="M7 17v26m35.17-21.394-5.948 18.697a1 1 0 0 1-.953.697H14V19h3l9.403-12.223a1 1 0 0 1 1.386-.196l2.535 1.87a6 6 0 0 1 2.044 6.974L31 19h9.265a2 2 0 0 1 1.906 2.606Z"></path></svg><span class="MRrY3">0</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-message"><path d="M15 20h18m-18 9h9M7 41h17.63C33.67 41 41 33.67 41 24.63V24c0-9.389-7.611-17-17-17S7 14.611 7 24v17Z"></path></svg><span class="MRrY3">0</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-star"><path d="M22.552 6.908a.5.5 0 0 1 .896 0l5.02 10.17a.5.5 0 0 0 .376.274l11.224 1.631a.5.5 0 0 1 .277.853l-8.122 7.916a.5.5 0 0 0-.143.443l1.917 11.178a.5.5 0 0 1-.726.527l-10.038-5.278a.5.5 0 0 0-.466 0L12.73 39.9a.5.5 0 0 1-.726-.527l1.918-11.178a.5.5 0 0 0-.144-.443l-8.122-7.916a.5.5 0 0 1 .278-.853l11.223-1.63a.5.5 0 0 0 .376-.274l5.02-10.17Z"></path></svg><span class="MRrY3">0</span></div><div class="opKvU"></div><span class="N_67f"><svg width="16" height="16" fill="none" xmlns="http://www.w3.org/2000/svg"><path fill-rule="evenodd" clip-rule="evenodd" d="M7.783 7.314a2 2 0 0 1 1.169 3.221l-1.888 2.332a2 2 0 1 1-3.109-2.518L4.967 9.1a1.431 1.431 0 0 1-.09-.372L4.725 7.28 2.919 9.51A3.333 3.333 0 1 0 8.1 13.705l1.888-2.332a3.333 3.333 0 0 0-1.304-5.173l-.9 1.113Z" fill="#939AA3"></path><path fill-rule="evenodd" clip-rule="evenodd" d="M8.58 9.291A2 2 0 0 1 7.411 6.07L9.3 3.74a2 2 0 1 1 3.11 2.517l-1.013 1.25c.046.116.077.241.09.372l.153 1.447 1.805-2.23a3.333 3.333 0 0 0-5.18-4.196L6.374 5.231a3.333 3.333 0 0 0 1.304 5.173l.901-1.113Z" fill="#939AA3"></path></svg></span></div></div><div class="zEp2T" style="cursor:auto"><div class="vuCON"><a href="/user/1977369294279187" target="_blank"><div style="width:20px;height:20px;font-size:10px" class="arco-avatar arco-avatar-circle"><span class="arco-avatar-image"><img src="https://p26-passport.byteacctimg.com/img/user-avatar/1816ccc0ebc35f4a89457e0b4ffe692b~300x300.image" alt="小窗幽记机器学习"/></span></div><div class="h_pFt"><span>小窗幽记机器学习</span></div></a></div><div class="eHpWb"><div class="BqxXc UG8D7"><div class="fHdNq"><div class="cSmO7"><a href="/articles/7428432985717735451" target="_blank">全面深入解读Movie Gen技术原理(5部曲)：个性化视频生成(3)</a></div><div style="margin-left:16px;display:inline-block;vertical-align:text-bottom"></div></div><div class="aklXs"><span class="MppYW"><span class="Ji9yj"></span><span class="HIKQo">AI</span></span><span class="KRApY">大模型</span><span class="KRApY">视频服务</span><span class="KRApY">机器学习</span></div><div class="IsnSN"><div class="TRHtk">引言简介生成个性化视频模型预训练预训练数据预训练方法监督微调评估结果总结最是人间留不住，朱颜辞镜花辞树。小伙伴们好，我是微信公众号《小窗幽记机器学习》的小编：缅A消费的小女孩。紧接此前Movie Gen解读系列：突发！Meta重磅发布Movie Gen入局视频生成赛道！全面深入解读Movie Gen技术原理(5部曲)：概述 (1)万字长文深度解读Movie Gen技术原理(5部曲)：图像视频联合生</div><img class="v4MkD" src="https://p6-volc-community-sign.byteimg.com/tos-cn-i-tlddhu82om/7a22739199dd46119a90c1c1be49a85e~tplv-tlddhu82om-image.image?=&amp;rk3s=8031ce6d&amp;x-expires=1732499851&amp;x-signature=9lP0K1jOf4mZQAlYIE7TeEpT2q0%3D"/></div></div><div class="nvyao"><img src="https://p6-volc-community-sign.byteimg.com/tos-cn-i-tlddhu82om/7a22739199dd46119a90c1c1be49a85e~tplv-tlddhu82om-image.image?=&amp;rk3s=8031ce6d&amp;x-expires=1732499851&amp;x-signature=9lP0K1jOf4mZQAlYIE7TeEpT2q0%3D"/></div></div></div><div class="B_zhD mdM_r"><div class="J8PPi"><div class="lugd4"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-eye"><path d="M24 37c6.627 0 12.627-4.333 18-13-5.373-8.667-11.373-13-18-13-6.627 0-12.627 4.333-18 13 5.373 8.667 11.373 13 18 13Z" clip-rule="evenodd"></path><path d="M29 24a5 5 0 1 1-10 0 5 5 0 0 1 10 0Z"></path></svg><span class="MRrY3">73</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-thumb-up"><path d="M7 17v26m35.17-21.394-5.948 18.697a1 1 0 0 1-.953.697H14V19h3l9.403-12.223a1 1 0 0 1 1.386-.196l2.535 1.87a6 6 0 0 1 2.044 6.974L31 19h9.265a2 2 0 0 1 1.906 2.606Z"></path></svg><span class="MRrY3">0</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-message"><path d="M15 20h18m-18 9h9M7 41h17.63C33.67 41 41 33.67 41 24.63V24c0-9.389-7.611-17-17-17S7 14.611 7 24v17Z"></path></svg><span class="MRrY3">0</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-star"><path d="M22.552 6.908a.5.5 0 0 1 .896 0l5.02 10.17a.5.5 0 0 0 .376.274l11.224 1.631a.5.5 0 0 1 .277.853l-8.122 7.916a.5.5 0 0 0-.143.443l1.917 11.178a.5.5 0 0 1-.726.527l-10.038-5.278a.5.5 0 0 0-.466 0L12.73 39.9a.5.5 0 0 1-.726-.527l1.918-11.178a.5.5 0 0 0-.144-.443l-8.122-7.916a.5.5 0 0 1 .278-.853l11.223-1.63a.5.5 0 0 0 .376-.274l5.02-10.17Z"></path></svg><span class="MRrY3">0</span></div><div class="opKvU"></div><span class="N_67f"><svg width="16" height="16" fill="none" xmlns="http://www.w3.org/2000/svg"><path fill-rule="evenodd" clip-rule="evenodd" d="M7.783 7.314a2 2 0 0 1 1.169 3.221l-1.888 2.332a2 2 0 1 1-3.109-2.518L4.967 9.1a1.431 1.431 0 0 1-.09-.372L4.725 7.28 2.919 9.51A3.333 3.333 0 1 0 8.1 13.705l1.888-2.332a3.333 3.333 0 0 0-1.304-5.173l-.9 1.113Z" fill="#939AA3"></path><path fill-rule="evenodd" clip-rule="evenodd" d="M8.58 9.291A2 2 0 0 1 7.411 6.07L9.3 3.74a2 2 0 1 1 3.11 2.517l-1.013 1.25c.046.116.077.241.09.372l.153 1.447 1.805-2.23a3.333 3.333 0 0 0-5.18-4.196L6.374 5.231a3.333 3.333 0 0 0 1.304 5.173l.901-1.113Z" fill="#939AA3"></path></svg></span></div></div><div class="zEp2T" style="cursor:auto"><div class="vuCON"><a href="/user/1977369294279187" target="_blank"><div style="width:20px;height:20px;font-size:10px" class="arco-avatar arco-avatar-circle"><span class="arco-avatar-image"><img src="https://p26-passport.byteacctimg.com/img/user-avatar/1816ccc0ebc35f4a89457e0b4ffe692b~300x300.image" alt="小窗幽记机器学习"/></span></div><div class="h_pFt"><span>小窗幽记机器学习</span></div></a></div><div class="eHpWb"><div class="BqxXc UG8D7"><div class="fHdNq"><div class="cSmO7"><a href="/articles/7428432985050841115" target="_blank">全面深入解读Movie Gen技术原理(5部曲)：概述 (1)</a></div><div style="margin-left:16px;display:inline-block;vertical-align:text-bottom"></div></div><div class="aklXs"><span class="MppYW"><span class="Ji9yj"></span><span class="HIKQo">音视频</span></span><span class="KRApY">大模型</span><span class="KRApY">视频服务</span><span class="KRApY">机器学习</span></div><div class="IsnSN"><div class="TRHtk">引言 =======小伙伴们好，我是微信公众号《小窗幽记机器学习》的小编：卖蚵仔煎的小男孩。紧接此前的文章：突发！Meta重磅发布Movie Gen入局视频生成赛道！，这几天临时搁置端侧大模型系列专题的深挖，先腾挪些时间阅读Meta官方发布的Movie Gen技术报告，从而基于官方一手资料 详细解读Movie Gen 模型。2024年10月4日，Meta发布其视频生成产品Movie Gen，对标</div><img class="v4MkD" src="https://p6-volc-community-sign.byteimg.com/tos-cn-i-tlddhu82om/04bfea696b754a7c941c5af6928f2dac~tplv-tlddhu82om-image.image?=&amp;rk3s=8031ce6d&amp;x-expires=1732499851&amp;x-signature=W%2Bi%2BMbQVi1EGMsIgKl7xRGwRGJM%3D"/></div></div><div class="nvyao"><img src="https://p6-volc-community-sign.byteimg.com/tos-cn-i-tlddhu82om/04bfea696b754a7c941c5af6928f2dac~tplv-tlddhu82om-image.image?=&amp;rk3s=8031ce6d&amp;x-expires=1732499851&amp;x-signature=W%2Bi%2BMbQVi1EGMsIgKl7xRGwRGJM%3D"/></div></div></div><div class="B_zhD mdM_r"><div class="J8PPi"><div class="lugd4"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-eye"><path d="M24 37c6.627 0 12.627-4.333 18-13-5.373-8.667-11.373-13-18-13-6.627 0-12.627 4.333-18 13 5.373 8.667 11.373 13 18 13Z" clip-rule="evenodd"></path><path d="M29 24a5 5 0 1 1-10 0 5 5 0 0 1 10 0Z"></path></svg><span class="MRrY3">64</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-thumb-up"><path d="M7 17v26m35.17-21.394-5.948 18.697a1 1 0 0 1-.953.697H14V19h3l9.403-12.223a1 1 0 0 1 1.386-.196l2.535 1.87a6 6 0 0 1 2.044 6.974L31 19h9.265a2 2 0 0 1 1.906 2.606Z"></path></svg><span class="MRrY3">0</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-message"><path d="M15 20h18m-18 9h9M7 41h17.63C33.67 41 41 33.67 41 24.63V24c0-9.389-7.611-17-17-17S7 14.611 7 24v17Z"></path></svg><span class="MRrY3">0</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-star"><path d="M22.552 6.908a.5.5 0 0 1 .896 0l5.02 10.17a.5.5 0 0 0 .376.274l11.224 1.631a.5.5 0 0 1 .277.853l-8.122 7.916a.5.5 0 0 0-.143.443l1.917 11.178a.5.5 0 0 1-.726.527l-10.038-5.278a.5.5 0 0 0-.466 0L12.73 39.9a.5.5 0 0 1-.726-.527l1.918-11.178a.5.5 0 0 0-.144-.443l-8.122-7.916a.5.5 0 0 1 .278-.853l11.223-1.63a.5.5 0 0 0 .376-.274l5.02-10.17Z"></path></svg><span class="MRrY3">0</span></div><div class="opKvU"></div><span class="N_67f"><svg width="16" height="16" fill="none" xmlns="http://www.w3.org/2000/svg"><path fill-rule="evenodd" clip-rule="evenodd" d="M7.783 7.314a2 2 0 0 1 1.169 3.221l-1.888 2.332a2 2 0 1 1-3.109-2.518L4.967 9.1a1.431 1.431 0 0 1-.09-.372L4.725 7.28 2.919 9.51A3.333 3.333 0 1 0 8.1 13.705l1.888-2.332a3.333 3.333 0 0 0-1.304-5.173l-.9 1.113Z" fill="#939AA3"></path><path fill-rule="evenodd" clip-rule="evenodd" d="M8.58 9.291A2 2 0 0 1 7.411 6.07L9.3 3.74a2 2 0 1 1 3.11 2.517l-1.013 1.25c.046.116.077.241.09.372l.153 1.447 1.805-2.23a3.333 3.333 0 0 0-5.18-4.196L6.374 5.231a3.333 3.333 0 0 0 1.304 5.173l.901-1.113Z" fill="#939AA3"></path></svg></span></div></div><div class="zEp2T" style="cursor:auto"><div class="vuCON"><a href="/user/1977369294279187" target="_blank"><div style="width:20px;height:20px;font-size:10px" class="arco-avatar arco-avatar-circle"><span class="arco-avatar-image"><img src="https://p26-passport.byteacctimg.com/img/user-avatar/1816ccc0ebc35f4a89457e0b4ffe692b~300x300.image" alt="小窗幽记机器学习"/></span></div><div class="h_pFt"><span>小窗幽记机器学习</span></div></a></div><div class="eHpWb"><div class="BqxXc UG8D7"><div class="fHdNq"><div class="cSmO7"><a href="/articles/7428432985701941275" target="_blank">突发！Meta重磅发布Movie Gen入局视频生成赛道！</a></div><div style="margin-left:16px;display:inline-block;vertical-align:text-bottom"></div></div><div class="aklXs"><span class="MppYW"><span class="Ji9yj"></span><span class="HIKQo">AI</span></span><span class="KRApY">视频服务</span><span class="KRApY">智能体验与创作</span><span class="KRApY">机器学习</span></div><div class="IsnSN"><div class="TRHtk">Meta于2024年10月4日首次推出 Meta Movie Gen，号称是迄今为止最先进的媒体基础模型。Movie Gen 由 Meta 的 AI 研究团队开发，在一系列功能上获取最先进的效果，包括 ： 文生视频、 创建个性化视频、精准 的视频编辑和 音频创作。无论是渴望在好莱坞闯出一片天地的新晋电影人，还是热衷于为观众制作视频的创作者，每个人都应该有机会使用能够增强创造力的工具。 Meta公司</div><img class="v4MkD" src="https://p6-volc-community-sign.byteimg.com/tos-cn-i-tlddhu82om/c7e68aab7a494f609feaede4bac53e74~tplv-tlddhu82om-image.image?=&amp;rk3s=8031ce6d&amp;x-expires=1732499851&amp;x-signature=1Bn%2Bl75pXsgPlOuwtZPOenQXx9g%3D"/></div></div><div class="nvyao"><img src="https://p6-volc-community-sign.byteimg.com/tos-cn-i-tlddhu82om/c7e68aab7a494f609feaede4bac53e74~tplv-tlddhu82om-image.image?=&amp;rk3s=8031ce6d&amp;x-expires=1732499851&amp;x-signature=1Bn%2Bl75pXsgPlOuwtZPOenQXx9g%3D"/></div></div></div><div class="B_zhD mdM_r"><div class="J8PPi"><div class="lugd4"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-eye"><path d="M24 37c6.627 0 12.627-4.333 18-13-5.373-8.667-11.373-13-18-13-6.627 0-12.627 4.333-18 13 5.373 8.667 11.373 13 18 13Z" clip-rule="evenodd"></path><path d="M29 24a5 5 0 1 1-10 0 5 5 0 0 1 10 0Z"></path></svg><span class="MRrY3">56</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-thumb-up"><path d="M7 17v26m35.17-21.394-5.948 18.697a1 1 0 0 1-.953.697H14V19h3l9.403-12.223a1 1 0 0 1 1.386-.196l2.535 1.87a6 6 0 0 1 2.044 6.974L31 19h9.265a2 2 0 0 1 1.906 2.606Z"></path></svg><span class="MRrY3">0</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-message"><path d="M15 20h18m-18 9h9M7 41h17.63C33.67 41 41 33.67 41 24.63V24c0-9.389-7.611-17-17-17S7 14.611 7 24v17Z"></path></svg><span class="MRrY3">0</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-star"><path d="M22.552 6.908a.5.5 0 0 1 .896 0l5.02 10.17a.5.5 0 0 0 .376.274l11.224 1.631a.5.5 0 0 1 .277.853l-8.122 7.916a.5.5 0 0 0-.143.443l1.917 11.178a.5.5 0 0 1-.726.527l-10.038-5.278a.5.5 0 0 0-.466 0L12.73 39.9a.5.5 0 0 1-.726-.527l1.918-11.178a.5.5 0 0 0-.144-.443l-8.122-7.916a.5.5 0 0 1 .278-.853l11.223-1.63a.5.5 0 0 0 .376-.274l5.02-10.17Z"></path></svg><span class="MRrY3">0</span></div><div class="opKvU"></div><span class="N_67f"><svg width="16" height="16" fill="none" xmlns="http://www.w3.org/2000/svg"><path fill-rule="evenodd" clip-rule="evenodd" d="M7.783 7.314a2 2 0 0 1 1.169 3.221l-1.888 2.332a2 2 0 1 1-3.109-2.518L4.967 9.1a1.431 1.431 0 0 1-.09-.372L4.725 7.28 2.919 9.51A3.333 3.333 0 1 0 8.1 13.705l1.888-2.332a3.333 3.333 0 0 0-1.304-5.173l-.9 1.113Z" fill="#939AA3"></path><path fill-rule="evenodd" clip-rule="evenodd" d="M8.58 9.291A2 2 0 0 1 7.411 6.07L9.3 3.74a2 2 0 1 1 3.11 2.517l-1.013 1.25c.046.116.077.241.09.372l.153 1.447 1.805-2.23a3.333 3.333 0 0 0-5.18-4.196L6.374 5.231a3.333 3.333 0 0 0 1.304 5.173l.901-1.113Z" fill="#939AA3"></path></svg></span></div></div><div class="zEp2T" style="cursor:auto"><div class="vuCON"><a href="/user/1977369294279187" target="_blank"><div style="width:20px;height:20px;font-size:10px" class="arco-avatar arco-avatar-circle"><span class="arco-avatar-image"><img src="https://p26-passport.byteacctimg.com/img/user-avatar/1816ccc0ebc35f4a89457e0b4ffe692b~300x300.image" alt="小窗幽记机器学习"/></span></div><div class="h_pFt"><span>小窗幽记机器学习</span></div></a></div><div class="eHpWb"><div class="BqxXc UG8D7"><div class="fHdNq"><div class="cSmO7"><a href="/articles/7428432810723164169" target="_blank">Sora已死？全面深入解读Movie Gen技术原理5部曲：4-精准视频编辑</a></div><div style="margin-left:16px;display:inline-block;vertical-align:text-bottom"></div></div><div class="aklXs"><span class="MppYW"><span class="Ji9yj"></span><span class="HIKQo">AI</span></span><span class="KRApY">视频服务</span><span class="KRApY">图像处理</span><span class="KRApY">机器学习</span></div><div class="IsnSN"><div class="TRHtk">引言简介模型架构改进单帧视频编辑训练多帧视频编辑训练反向翻译的视频编辑训练结果一年好景君须记，最是橙黄橘绿时。小伙伴们好，我是微信公众号《小窗幽记机器学习》的小编：缅A消费积极分子。紧接此前Movie Gen解读系列：突发！Meta重磅发布Movie Gen入局视频生成赛道！全面深入解读Movie Gen技术原理(5部曲)：1-概述全面深入解读Movie Gen技术原理(5部曲)：2-图像视频联合</div><img class="v4MkD" src="https://p6-volc-community-sign.byteimg.com/tos-cn-i-tlddhu82om/fd7f70f6024d4fc5b7cabdb1eb4be0dc~tplv-tlddhu82om-image.image?=&amp;rk3s=8031ce6d&amp;x-expires=1732499851&amp;x-signature=uKxPADgr18Jk3PVT3P8Qnys%2FVJ0%3D"/></div></div><div class="nvyao"><img src="https://p6-volc-community-sign.byteimg.com/tos-cn-i-tlddhu82om/fd7f70f6024d4fc5b7cabdb1eb4be0dc~tplv-tlddhu82om-image.image?=&amp;rk3s=8031ce6d&amp;x-expires=1732499851&amp;x-signature=uKxPADgr18Jk3PVT3P8Qnys%2FVJ0%3D"/></div></div></div><div class="B_zhD mdM_r"><div class="J8PPi"><div class="lugd4"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-eye"><path d="M24 37c6.627 0 12.627-4.333 18-13-5.373-8.667-11.373-13-18-13-6.627 0-12.627 4.333-18 13 5.373 8.667 11.373 13 18 13Z" clip-rule="evenodd"></path><path d="M29 24a5 5 0 1 1-10 0 5 5 0 0 1 10 0Z"></path></svg><span class="MRrY3">64</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-thumb-up"><path d="M7 17v26m35.17-21.394-5.948 18.697a1 1 0 0 1-.953.697H14V19h3l9.403-12.223a1 1 0 0 1 1.386-.196l2.535 1.87a6 6 0 0 1 2.044 6.974L31 19h9.265a2 2 0 0 1 1.906 2.606Z"></path></svg><span class="MRrY3">0</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-message"><path d="M15 20h18m-18 9h9M7 41h17.63C33.67 41 41 33.67 41 24.63V24c0-9.389-7.611-17-17-17S7 14.611 7 24v17Z"></path></svg><span class="MRrY3">0</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-star"><path d="M22.552 6.908a.5.5 0 0 1 .896 0l5.02 10.17a.5.5 0 0 0 .376.274l11.224 1.631a.5.5 0 0 1 .277.853l-8.122 7.916a.5.5 0 0 0-.143.443l1.917 11.178a.5.5 0 0 1-.726.527l-10.038-5.278a.5.5 0 0 0-.466 0L12.73 39.9a.5.5 0 0 1-.726-.527l1.918-11.178a.5.5 0 0 0-.144-.443l-8.122-7.916a.5.5 0 0 1 .278-.853l11.223-1.63a.5.5 0 0 0 .376-.274l5.02-10.17Z"></path></svg><span class="MRrY3">0</span></div><div class="opKvU"></div><span class="N_67f"><svg width="16" height="16" fill="none" xmlns="http://www.w3.org/2000/svg"><path fill-rule="evenodd" clip-rule="evenodd" d="M7.783 7.314a2 2 0 0 1 1.169 3.221l-1.888 2.332a2 2 0 1 1-3.109-2.518L4.967 9.1a1.431 1.431 0 0 1-.09-.372L4.725 7.28 2.919 9.51A3.333 3.333 0 1 0 8.1 13.705l1.888-2.332a3.333 3.333 0 0 0-1.304-5.173l-.9 1.113Z" fill="#939AA3"></path><path fill-rule="evenodd" clip-rule="evenodd" d="M8.58 9.291A2 2 0 0 1 7.411 6.07L9.3 3.74a2 2 0 1 1 3.11 2.517l-1.013 1.25c.046.116.077.241.09.372l.153 1.447 1.805-2.23a3.333 3.333 0 0 0-5.18-4.196L6.374 5.231a3.333 3.333 0 0 0 1.304 5.173l.901-1.113Z" fill="#939AA3"></path></svg></span></div></div><div class="zEp2T" style="cursor:auto"><div class="vuCON"><a href="/user/1977369294279187" target="_blank"><div style="width:20px;height:20px;font-size:10px" class="arco-avatar arco-avatar-circle"><span class="arco-avatar-image"><img src="https://p26-passport.byteacctimg.com/img/user-avatar/1816ccc0ebc35f4a89457e0b4ffe692b~300x300.image" alt="小窗幽记机器学习"/></span></div><div class="h_pFt"><span>小窗幽记机器学习</span></div></a></div><div class="eHpWb"><div class="BqxXc UG8D7"><div class="fHdNq"><div class="cSmO7"><a href="/articles/7424424401317134363" target="_blank">万字长文深度解读Movie Gen技术原理(5部曲)：图像&amp;视频联合生成模型 (2)</a></div><div style="margin-left:16px;display:inline-block;vertical-align:text-bottom"></div></div><div class="aklXs"><span class="MppYW"><span class="Ji9yj"></span><span class="HIKQo">音视频</span></span><span class="KRApY">大模型</span><span class="KRApY">图像处理</span><span class="KRApY">机器学习</span></div><div class="IsnSN"><div class="TRHtk">详细介绍Movie Gen中图像和视频的联合生成技术，包括:时间自编码器的设计与优化、基于流匹配的训练目标、联合生成的网络架构、文本嵌入和视觉-文本生成方法、空间上采样技术、模型扩展和训练效率优化等。此外，还详细介绍了预训练数据的准备过程.</div><img class="v4MkD" src="https://p6-volc-community-sign.byteimg.com/tos-cn-i-tlddhu82om/8bf15e423f33468f952d11f598f08d08~tplv-tlddhu82om-image.image?=&amp;rk3s=8031ce6d&amp;x-expires=1732499851&amp;x-signature=tuqqyNahu0rm4Pdn3SKC%2FoxkJJ8%3D"/></div></div><div class="nvyao"><img src="https://p6-volc-community-sign.byteimg.com/tos-cn-i-tlddhu82om/8bf15e423f33468f952d11f598f08d08~tplv-tlddhu82om-image.image?=&amp;rk3s=8031ce6d&amp;x-expires=1732499851&amp;x-signature=tuqqyNahu0rm4Pdn3SKC%2FoxkJJ8%3D"/></div></div></div><div class="B_zhD mdM_r"><div class="J8PPi"><div class="lugd4"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-eye"><path d="M24 37c6.627 0 12.627-4.333 18-13-5.373-8.667-11.373-13-18-13-6.627 0-12.627 4.333-18 13 5.373 8.667 11.373 13 18 13Z" clip-rule="evenodd"></path><path d="M29 24a5 5 0 1 1-10 0 5 5 0 0 1 10 0Z"></path></svg><span class="MRrY3">243</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-thumb-up"><path d="M7 17v26m35.17-21.394-5.948 18.697a1 1 0 0 1-.953.697H14V19h3l9.403-12.223a1 1 0 0 1 1.386-.196l2.535 1.87a6 6 0 0 1 2.044 6.974L31 19h9.265a2 2 0 0 1 1.906 2.606Z"></path></svg><span class="MRrY3">0</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-message"><path d="M15 20h18m-18 9h9M7 41h17.63C33.67 41 41 33.67 41 24.63V24c0-9.389-7.611-17-17-17S7 14.611 7 24v17Z"></path></svg><span class="MRrY3">0</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-star"><path d="M22.552 6.908a.5.5 0 0 1 .896 0l5.02 10.17a.5.5 0 0 0 .376.274l11.224 1.631a.5.5 0 0 1 .277.853l-8.122 7.916a.5.5 0 0 0-.143.443l1.917 11.178a.5.5 0 0 1-.726.527l-10.038-5.278a.5.5 0 0 0-.466 0L12.73 39.9a.5.5 0 0 1-.726-.527l1.918-11.178a.5.5 0 0 0-.144-.443l-8.122-7.916a.5.5 0 0 1 .278-.853l11.223-1.63a.5.5 0 0 0 .376-.274l5.02-10.17Z"></path></svg><span class="MRrY3">0</span></div><div class="opKvU"></div><span class="N_67f"><svg width="16" height="16" fill="none" xmlns="http://www.w3.org/2000/svg"><path fill-rule="evenodd" clip-rule="evenodd" d="M7.783 7.314a2 2 0 0 1 1.169 3.221l-1.888 2.332a2 2 0 1 1-3.109-2.518L4.967 9.1a1.431 1.431 0 0 1-.09-.372L4.725 7.28 2.919 9.51A3.333 3.333 0 1 0 8.1 13.705l1.888-2.332a3.333 3.333 0 0 0-1.304-5.173l-.9 1.113Z" fill="#939AA3"></path><path fill-rule="evenodd" clip-rule="evenodd" d="M8.58 9.291A2 2 0 0 1 7.411 6.07L9.3 3.74a2 2 0 1 1 3.11 2.517l-1.013 1.25c.046.116.077.241.09.372l.153 1.447 1.805-2.23a3.333 3.333 0 0 0-5.18-4.196L6.374 5.231a3.333 3.333 0 0 0 1.304 5.173l.901-1.113Z" fill="#939AA3"></path></svg></span></div></div><div class="zEp2T" style="cursor:auto"><div class="vuCON"><a href="/user/1977369294279187" target="_blank"><div style="width:20px;height:20px;font-size:10px" class="arco-avatar arco-avatar-circle"><span class="arco-avatar-image"><img src="https://p26-passport.byteacctimg.com/img/user-avatar/1816ccc0ebc35f4a89457e0b4ffe692b~300x300.image" alt="小窗幽记机器学习"/></span></div><div class="h_pFt"><span>小窗幽记机器学习</span></div></a></div><div class="eHpWb"><div class="BqxXc UG8D7"><div class="fHdNq"><div class="cSmO7"><a href="/articles/7423667140789551130" target="_blank">全面深入解读Movie Gen技术原理(5部曲)：概述 (1)</a></div><div style="margin-left:16px;display:inline-block;vertical-align:text-bottom"></div></div><div class="aklXs"><span class="MppYW"><span class="Ji9yj"></span><span class="HIKQo">音视频</span></span><span class="KRApY">大模型</span><span class="KRApY">视频服务</span><span class="KRApY">机器学习</span></div><div class="IsnSN"><div class="TRHtk">引言 =======小伙伴们好，我是微信公众号《小窗幽记机器学习》的小编：卖蚵仔煎的小男孩。紧接此前的文章：突发！Meta重磅发布Movie Gen入局视频生成赛道！，这几天临时搁置端侧大模型系列专题的深挖，先腾挪些时间阅读Meta官方发布的Movie Gen技术报告，从而基于官方一手资料 详细解读Movie Gen 模型。2024年10月4日，Meta发布其视频生成产品Movie Gen，对标</div><img class="v4MkD" src="https://p6-volc-community-sign.byteimg.com/tos-cn-i-tlddhu82om/3594d0cf6bfb4f71b0d55ea0ec3f4d61~tplv-tlddhu82om-image.image?=&amp;rk3s=8031ce6d&amp;x-expires=1732499851&amp;x-signature=2xe6MzRTBbA4i2wTZLGTXOD%2BR1k%3D"/></div></div><div class="nvyao"><img src="https://p6-volc-community-sign.byteimg.com/tos-cn-i-tlddhu82om/3594d0cf6bfb4f71b0d55ea0ec3f4d61~tplv-tlddhu82om-image.image?=&amp;rk3s=8031ce6d&amp;x-expires=1732499851&amp;x-signature=2xe6MzRTBbA4i2wTZLGTXOD%2BR1k%3D"/></div></div></div><div class="B_zhD mdM_r"><div class="J8PPi"><div class="lugd4"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-eye"><path d="M24 37c6.627 0 12.627-4.333 18-13-5.373-8.667-11.373-13-18-13-6.627 0-12.627 4.333-18 13 5.373 8.667 11.373 13 18 13Z" clip-rule="evenodd"></path><path d="M29 24a5 5 0 1 1-10 0 5 5 0 0 1 10 0Z"></path></svg><span class="MRrY3">264</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-thumb-up"><path d="M7 17v26m35.17-21.394-5.948 18.697a1 1 0 0 1-.953.697H14V19h3l9.403-12.223a1 1 0 0 1 1.386-.196l2.535 1.87a6 6 0 0 1 2.044 6.974L31 19h9.265a2 2 0 0 1 1.906 2.606Z"></path></svg><span class="MRrY3">0</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-message"><path d="M15 20h18m-18 9h9M7 41h17.63C33.67 41 41 33.67 41 24.63V24c0-9.389-7.611-17-17-17S7 14.611 7 24v17Z"></path></svg><span class="MRrY3">0</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-star"><path d="M22.552 6.908a.5.5 0 0 1 .896 0l5.02 10.17a.5.5 0 0 0 .376.274l11.224 1.631a.5.5 0 0 1 .277.853l-8.122 7.916a.5.5 0 0 0-.143.443l1.917 11.178a.5.5 0 0 1-.726.527l-10.038-5.278a.5.5 0 0 0-.466 0L12.73 39.9a.5.5 0 0 1-.726-.527l1.918-11.178a.5.5 0 0 0-.144-.443l-8.122-7.916a.5.5 0 0 1 .278-.853l11.223-1.63a.5.5 0 0 0 .376-.274l5.02-10.17Z"></path></svg><span class="MRrY3">0</span></div><div class="opKvU"></div><span class="N_67f"><svg width="16" height="16" fill="none" xmlns="http://www.w3.org/2000/svg"><path fill-rule="evenodd" clip-rule="evenodd" d="M7.783 7.314a2 2 0 0 1 1.169 3.221l-1.888 2.332a2 2 0 1 1-3.109-2.518L4.967 9.1a1.431 1.431 0 0 1-.09-.372L4.725 7.28 2.919 9.51A3.333 3.333 0 1 0 8.1 13.705l1.888-2.332a3.333 3.333 0 0 0-1.304-5.173l-.9 1.113Z" fill="#939AA3"></path><path fill-rule="evenodd" clip-rule="evenodd" d="M8.58 9.291A2 2 0 0 1 7.411 6.07L9.3 3.74a2 2 0 1 1 3.11 2.517l-1.013 1.25c.046.116.077.241.09.372l.153 1.447 1.805-2.23a3.333 3.333 0 0 0-5.18-4.196L6.374 5.231a3.333 3.333 0 0 0 1.304 5.173l.901-1.113Z" fill="#939AA3"></path></svg></span></div></div><div class="zEp2T" style="cursor:auto"><div class="vuCON"><a href="/user/1977369294279187" target="_blank"><div style="width:20px;height:20px;font-size:10px" class="arco-avatar arco-avatar-circle"><span class="arco-avatar-image"><img src="https://p26-passport.byteacctimg.com/img/user-avatar/1816ccc0ebc35f4a89457e0b4ffe692b~300x300.image" alt="小窗幽记机器学习"/></span></div><div class="h_pFt"><span>小窗幽记机器学习</span></div></a></div><div class="eHpWb"><div class="BqxXc UG8D7"><div class="fHdNq"><div class="cSmO7"><a href="/articles/7423250703340339251" target="_blank">端侧Agent系列 | 端侧AI Agent任务拆解大师如何助力AI手机？(详解版)</a></div><div style="margin-left:16px;display:inline-block;vertical-align:text-bottom"></div></div><div class="aklXs"><span class="MppYW"><span class="Ji9yj"></span><span class="HIKQo">大模型</span></span><span class="KRApY">大模型</span><span class="KRApY">机器学习</span><span class="KRApY">数据库</span></div><div class="IsnSN"><div class="TRHtk">本文主要介绍端侧Agent智能体如何实现任务规划和拆解，从而实现复杂场景下多指令任务的执行。Octo-planner，这是一个专为边缘设备而设计的用于规划任务的AI Agent框架。</div><img class="v4MkD" src="https://p3-volc-community-sign.byteimg.com/tos-cn-i-tlddhu82om/b3eab84269564e2594124a3dd817652b~tplv-tlddhu82om-image.image?=&amp;rk3s=8031ce6d&amp;x-expires=1732499851&amp;x-signature=DQuGWkDFw4secDwIHB%2BNf4n02PI%3D"/></div></div><div class="nvyao"><img src="https://p3-volc-community-sign.byteimg.com/tos-cn-i-tlddhu82om/b3eab84269564e2594124a3dd817652b~tplv-tlddhu82om-image.image?=&amp;rk3s=8031ce6d&amp;x-expires=1732499851&amp;x-signature=DQuGWkDFw4secDwIHB%2BNf4n02PI%3D"/></div></div></div><div class="B_zhD mdM_r"><div class="J8PPi"><div class="lugd4"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-eye"><path d="M24 37c6.627 0 12.627-4.333 18-13-5.373-8.667-11.373-13-18-13-6.627 0-12.627 4.333-18 13 5.373 8.667 11.373 13 18 13Z" clip-rule="evenodd"></path><path d="M29 24a5 5 0 1 1-10 0 5 5 0 0 1 10 0Z"></path></svg><span class="MRrY3">76</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-thumb-up"><path d="M7 17v26m35.17-21.394-5.948 18.697a1 1 0 0 1-.953.697H14V19h3l9.403-12.223a1 1 0 0 1 1.386-.196l2.535 1.87a6 6 0 0 1 2.044 6.974L31 19h9.265a2 2 0 0 1 1.906 2.606Z"></path></svg><span class="MRrY3">0</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-message"><path d="M15 20h18m-18 9h9M7 41h17.63C33.67 41 41 33.67 41 24.63V24c0-9.389-7.611-17-17-17S7 14.611 7 24v17Z"></path></svg><span class="MRrY3">0</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-star"><path d="M22.552 6.908a.5.5 0 0 1 .896 0l5.02 10.17a.5.5 0 0 0 .376.274l11.224 1.631a.5.5 0 0 1 .277.853l-8.122 7.916a.5.5 0 0 0-.143.443l1.917 11.178a.5.5 0 0 1-.726.527l-10.038-5.278a.5.5 0 0 0-.466 0L12.73 39.9a.5.5 0 0 1-.726-.527l1.918-11.178a.5.5 0 0 0-.144-.443l-8.122-7.916a.5.5 0 0 1 .278-.853l11.223-1.63a.5.5 0 0 0 .376-.274l5.02-10.17Z"></path></svg><span class="MRrY3">0</span></div><div class="opKvU"></div><span class="N_67f"><svg width="16" height="16" fill="none" xmlns="http://www.w3.org/2000/svg"><path fill-rule="evenodd" clip-rule="evenodd" d="M7.783 7.314a2 2 0 0 1 1.169 3.221l-1.888 2.332a2 2 0 1 1-3.109-2.518L4.967 9.1a1.431 1.431 0 0 1-.09-.372L4.725 7.28 2.919 9.51A3.333 3.333 0 1 0 8.1 13.705l1.888-2.332a3.333 3.333 0 0 0-1.304-5.173l-.9 1.113Z" fill="#939AA3"></path><path fill-rule="evenodd" clip-rule="evenodd" d="M8.58 9.291A2 2 0 0 1 7.411 6.07L9.3 3.74a2 2 0 1 1 3.11 2.517l-1.013 1.25c.046.116.077.241.09.372l.153 1.447 1.805-2.23a3.333 3.333 0 0 0-5.18-4.196L6.374 5.231a3.333 3.333 0 0 0 1.304 5.173l.901-1.113Z" fill="#939AA3"></path></svg></span></div></div><div class="zEp2T" style="cursor:auto"><div class="vuCON"><a href="/user/1977369294279187" target="_blank"><div style="width:20px;height:20px;font-size:10px" class="arco-avatar arco-avatar-circle"><span class="arco-avatar-image"><img src="https://p26-passport.byteacctimg.com/img/user-avatar/1816ccc0ebc35f4a89457e0b4ffe692b~300x300.image" alt="小窗幽记机器学习"/></span></div><div class="h_pFt"><span>小窗幽记机器学习</span></div></a></div><div class="eHpWb"><div class="BqxXc UG8D7"><div class="fHdNq"><div class="cSmO7"><a href="/articles/7423250703256633394" target="_blank">端侧大模型系列 | 端侧AI Agent任务拆解大师如何助力AI手机？(简短版)</a></div><div style="margin-left:16px;display:inline-block;vertical-align:text-bottom"></div></div><div class="aklXs"><span class="MppYW"><span class="Ji9yj"></span><span class="HIKQo">AI</span></span><span class="KRApY">大模型</span><span class="KRApY">机器学习</span><span class="KRApY">算法</span></div><div class="IsnSN"><div class="TRHtk">引言简介模型实验意义&amp;前景:总结今人不见古时月，今月曾经照古人。小伙伴们好，我是微信公众号《小窗幽记机器学习》的小编：卖沙茶面的小女孩。设想一下，你的智能手机不再只是&quot;聪明&quot;，而是真正的&quot;智能&quot;。它不仅能响应指令，还能预测需求、规划日程，并轻松完成复杂任务。这不是科幻小说，而是设备内置AI助手的新时代，即将到来。NexaAI提出的Octo-planner通过分离规划与执行过程，将先进的AI功能装进</div><img class="v4MkD" src="https://p6-volc-community-sign.byteimg.com/tos-cn-i-tlddhu82om/32ba7371fc44430c9cefa72605d1ec0c~tplv-tlddhu82om-image.image?=&amp;rk3s=8031ce6d&amp;x-expires=1732499851&amp;x-signature=ktB7tET0dt5auAklmQdwhRCi2h0%3D"/></div></div><div class="nvyao"><img src="https://p6-volc-community-sign.byteimg.com/tos-cn-i-tlddhu82om/32ba7371fc44430c9cefa72605d1ec0c~tplv-tlddhu82om-image.image?=&amp;rk3s=8031ce6d&amp;x-expires=1732499851&amp;x-signature=ktB7tET0dt5auAklmQdwhRCi2h0%3D"/></div></div></div><div class="B_zhD mdM_r"><div class="J8PPi"><div class="lugd4"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-eye"><path d="M24 37c6.627 0 12.627-4.333 18-13-5.373-8.667-11.373-13-18-13-6.627 0-12.627 4.333-18 13 5.373 8.667 11.373 13 18 13Z" clip-rule="evenodd"></path><path d="M29 24a5 5 0 1 1-10 0 5 5 0 0 1 10 0Z"></path></svg><span class="MRrY3">79</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-thumb-up"><path d="M7 17v26m35.17-21.394-5.948 18.697a1 1 0 0 1-.953.697H14V19h3l9.403-12.223a1 1 0 0 1 1.386-.196l2.535 1.87a6 6 0 0 1 2.044 6.974L31 19h9.265a2 2 0 0 1 1.906 2.606Z"></path></svg><span class="MRrY3">0</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-message"><path d="M15 20h18m-18 9h9M7 41h17.63C33.67 41 41 33.67 41 24.63V24c0-9.389-7.611-17-17-17S7 14.611 7 24v17Z"></path></svg><span class="MRrY3">0</span></div><div class="lugd4 jOV0E"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" style="font-size:16px" class="arco-icon arco-icon-star"><path d="M22.552 6.908a.5.5 0 0 1 .896 0l5.02 10.17a.5.5 0 0 0 .376.274l11.224 1.631a.5.5 0 0 1 .277.853l-8.122 7.916a.5.5 0 0 0-.143.443l1.917 11.178a.5.5 0 0 1-.726.527l-10.038-5.278a.5.5 0 0 0-.466 0L12.73 39.9a.5.5 0 0 1-.726-.527l1.918-11.178a.5.5 0 0 0-.144-.443l-8.122-7.916a.5.5 0 0 1 .278-.853l11.223-1.63a.5.5 0 0 0 .376-.274l5.02-10.17Z"></path></svg><span class="MRrY3">0</span></div><div class="opKvU"></div><span class="N_67f"><svg width="16" height="16" fill="none" xmlns="http://www.w3.org/2000/svg"><path fill-rule="evenodd" clip-rule="evenodd" d="M7.783 7.314a2 2 0 0 1 1.169 3.221l-1.888 2.332a2 2 0 1 1-3.109-2.518L4.967 9.1a1.431 1.431 0 0 1-.09-.372L4.725 7.28 2.919 9.51A3.333 3.333 0 1 0 8.1 13.705l1.888-2.332a3.333 3.333 0 0 0-1.304-5.173l-.9 1.113Z" fill="#939AA3"></path><path fill-rule="evenodd" clip-rule="evenodd" d="M8.58 9.291A2 2 0 0 1 7.411 6.07L9.3 3.74a2 2 0 1 1 3.11 2.517l-1.013 1.25c.046.116.077.241.09.372l.153 1.447 1.805-2.23a3.333 3.333 0 0 0-5.18-4.196L6.374 5.231a3.333 3.333 0 0 0 1.304 5.173l.901-1.113Z" fill="#939AA3"></path></svg></span></div></div><div class="J3FsF"><div class="arco-spin"><span class="arco-spin-icon"><svg fill="none" stroke="currentColor" stroke-width="4" viewBox="0 0 48 48" aria-hidden="true" focusable="false" class="arco-icon arco-icon-loading"><path d="M42 24c0 9.941-8.059 18-18 18S6 33.941 6 24 14.059 6 24 6"></path></svg></span></div></div></div></div></div></div></div></div><script id="__LOADABLE_REQUIRED_CHUNKS__" type="application/json">[2649,7542,6128,9844,2986,6033,5912]</script><script id="__LOADABLE_REQUIRED_CHUNKS___ext" type="application/json">{"namedChunks":["user/[id]/layout","user/[id]/page"]}</script><script src="//portal.volccdn.com/obj/volcfe-scm/deploy/volc_developer_/42325/static/js/lib-arco.a74ce714.js" ></script><script src="//portal.volccdn.com/obj/volcfe-scm/deploy/volc_developer_/42325/static/js/async/7542.a4ca3948.js" ></script><script src="//portal.volccdn.com/obj/volcfe-scm/deploy/volc_developer_/42325/static/js/async/6128.b950f483.js" ></script><script src="//portal.volccdn.com/obj/volcfe-scm/deploy/volc_developer_/42325/static/js/async/user/[id]/layout.e938f428.js" ></script><script src="//portal.volccdn.com/obj/volcfe-scm/deploy/volc_developer_/42325/static/js/lib-polyfill.e32e6eae.js" ></script><script src="//portal.volccdn.com/obj/volcfe-scm/deploy/volc_developer_/42325/static/js/async/6033.cd1c2bf1.js" ></script><script src="//portal.volccdn.com/obj/volcfe-scm/deploy/volc_developer_/42325/static/js/async/user/[id]/page.77e2305e.js" ></script><script>window._SSR_DATA = {"data":{},"context":{"request":{"params":{},"query":{},"pathname":"\u002Fuser\u002F1977369294279187","host":"developer.volcengine.com","url":"https:\u002F\u002Fdeveloper.volcengine.com\u002Fuser\u002F1977369294279187"}},"renderLevel":2}</script> <script>window._ROUTER_DATA = {"loaderData":{"layout":{"err_no":401,"err_msg":"NotAuthorized"},"user\u002Flayout":null,"user\u002F[id]\u002Flayout":{"userInfo":{"data":{"avatar":"https:\u002F\u002Fp26-passport.byteacctimg.com\u002Fimg\u002Fuser-avatar\u002F1816ccc0ebc35f4a89457e0b4ffe692b~300x300.image","employee_id":"1977369294279187","full_name":"小窗幽记机器学习","name":"小窗幽记机器学习","status":2,"xid":"1977369294279187","user_type":14,"volc_data":{}},"err_msg":"","err_no":0}},"user\u002F[id]\u002Fpage":{"articleInfo":{"err_no":0,"err_msg":"","data":[{"content":{"item_id":"7440006190165639178","item_type":2,"app_id":3569,"user_id":"1977369294279187","version":1,"status":2,"create_time":1732261431,"update_time":1732273256,"publish_time":1732261431,"name":"多模态大模型系列 | 18：Qwen2-VL(最新版)解读及其实战(精炼版)","abstract":"引言简介方法实验结果实战代码任务1：检测任务2：图片理解总结引言\n=======梅子金黄杏子肥，麦花雪白菜花稀。日长篱落无人过，惟有蜻蜓蛱蝶飞。小伙伴们好，我是微信公众号《小窗幽记机器学习》的小编：卖热干面的小女孩。紧接前文：多模态大模型系列：Qwen-VL解读及其实战，今天这篇小作文主要介绍阿里在2024年9月份发布的视觉语言模型：Qwen2-VL。简介\n=======这篇论文介绍了Qwen2-","cover_image":{"key":"","url":"https:\u002F\u002Fp6-volc-community-sign.byteimg.com\u002Ftos-cn-i-tlddhu82om\u002F3725eb4f79e84b5081ccc75b49eda82e~tplv-tlddhu82om-image.image?=&rk3s=8031ce6d&x-expires=1732499851&x-signature=6guhRzGSM%2FIT6MIICAQHPJuhvhI%3D","size":0,"mime_type":"","rid":""},"mime_type":"","content":"","resource":{},"extra":{"source":{"app_id":3569,"name":"","author":""},"html_content":""},"parent_id":"0","parent_type":0,"last_version":1,"last_status":2,"last_create_time":1732261431,"last_update_time":1732261431,"status_tags":[],"create_user_id":"0","item_source":"2","arcosite_id":"","text_url":""},"user":{"user_id":"1977369294279187","user_type":14,"name":"小窗幽记机器学习","avatar":{"key":"","url":"https:\u002F\u002Fp26-passport.byteacctimg.com\u002Fimg\u002Fuser-avatar\u002F1816ccc0ebc35f4a89457e0b4ffe692b~300x300.image","size":0,"mime_type":"","rid":""},"status":2},"categories":[{"category_id":"7306040281718063141","name":"大模型","parent_id":"0","create_time":1701917585,"update_time":1701917585,"op_user_id":"3945470841857399","abstract":"大模型技术干货","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"rank":0,"category_url":""}],"tags":[{"tag_id":"7307103161498632233","name":"大模型","create_time":1701317532,"update_time":1701917647,"op_user_id":"3945470841857399","create_user_id":"620546920289582","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2},{"tag_id":"7146205183897239566","name":"机器学习","create_time":1663855553,"update_time":1663855553,"op_user_id":"4015857634130296","create_user_id":"4015857634130296","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2},{"tag_id":"7146114517833875464","name":"数据库","create_time":1663834443,"update_time":1663834443,"op_user_id":"743694848765533","create_user_id":"743694848765533","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2}],"interact_status":{"is_like":false,"is_collect":false,"is_join":false,"like_count":0,"collect_count":0,"comment_count":0,"join_count":0,"evaluate_count":0,"evaluate_avg":0,"evaluate_self":0},"pool_status":{"is_top":false,"is_good":false,"is_recommend":false},"content_count":{"view_count":67,"child_count":0},"op_info":{"op_user":{},"op_time":0,"op_resource":""}},{"content":{"item_id":"7440006140932718602","item_type":2,"app_id":3569,"user_id":"1977369294279187","version":1,"status":2,"create_time":1732261420,"update_time":1732272892,"publish_time":1732261420,"name":"多模态大模型系列 | 19：微软提出LLM2CLIP，巧用LLM助力clip开启图文理解新篇章","abstract":"引言简介方法实验结果总结引言\n=======丹阳郭里送行舟，一别心知两地秋。CLIP是当今最重要的多模态基础模型之一，能将视觉和文本信息对齐到共享特征空间中。自然语言作为人类知识载体，为CLIP提供了丰富的监督信息，赋予了它强大的跨模态表示能力。随着大型语言模型(LLMs)的发展，越来越多的学者探讨如何利用LLMs提升多模态表示学习。LLMs强大的语言理解能力可以提升CLIP处理各类文本的能力，其","cover_image":{"key":"","url":"https:\u002F\u002Fp3-volc-community-sign.byteimg.com\u002Ftos-cn-i-tlddhu82om\u002F09a6b96522eb415ebce8343797468c31~tplv-tlddhu82om-image.image?=&rk3s=8031ce6d&x-expires=1732499851&x-signature=F%2FxomaNrkGQTl6oGp%2F43r029peM%3D","size":0,"mime_type":"","rid":""},"mime_type":"","content":"","resource":{},"extra":{"source":{"app_id":3569,"name":"","author":""},"html_content":""},"parent_id":"0","parent_type":0,"last_version":1,"last_status":2,"last_create_time":1732261420,"last_update_time":1732261420,"status_tags":[],"create_user_id":"0","item_source":"2","arcosite_id":"","text_url":""},"user":{"user_id":"1977369294279187","user_type":14,"name":"小窗幽记机器学习","avatar":{"key":"","url":"https:\u002F\u002Fp26-passport.byteacctimg.com\u002Fimg\u002Fuser-avatar\u002F1816ccc0ebc35f4a89457e0b4ffe692b~300x300.image","size":0,"mime_type":"","rid":""},"status":2},"categories":[{"category_id":"3","name":"AI","parent_id":"0","create_time":1637495002,"update_time":1637495249,"op_user_id":"0","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"rank":0,"category_url":""}],"tags":[{"tag_id":"7307103161498632233","name":"大模型","create_time":1701317532,"update_time":1701917647,"op_user_id":"3945470841857399","create_user_id":"620546920289582","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2},{"tag_id":"7309680367596732443","name":"向量数据库","create_time":1701917641,"update_time":1705378351,"op_user_id":"3945470841857399","create_user_id":"3945470841857399","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2},{"tag_id":"7146205183897239566","name":"机器学习","create_time":1663855553,"update_time":1663855553,"op_user_id":"4015857634130296","create_user_id":"4015857634130296","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2}],"interact_status":{"is_like":false,"is_collect":false,"is_join":false,"like_count":0,"collect_count":0,"comment_count":0,"join_count":0,"evaluate_count":0,"evaluate_avg":0,"evaluate_self":0},"pool_status":{"is_top":false,"is_good":false,"is_recommend":false},"content_count":{"view_count":8,"child_count":0},"op_info":{"op_user":{},"op_time":0,"op_resource":""}},{"content":{"item_id":"7440006127816802341","item_type":2,"app_id":3569,"user_id":"1977369294279187","version":1,"status":2,"create_time":1732261417,"update_time":1732272749,"publish_time":1732261417,"name":"多模态大模型系列 | 20：低延迟、零遗忘：语音多模态新范式-Freeze-Omni","abstract":"简介\n========语音交互是人机沟通的未来，但现有多模态大语言模型在保持模型原始智能的同时实现低延迟对话一直是一大挑战。Freeze-Omni通过创新的三阶段训练策略，在不微调大语言模型的前提下，实现了高效、智能的端到端语音对话，为多模态LLM研究开辟了新路径。Q1: 这篇文章想要解决什么问题？A1: 本文致力于解决大型语言模型(LLM)在语音交互中的两个关键挑战：如何在不破坏LLM原有智能的","cover_image":{"key":"","url":"https:\u002F\u002Fp3-volc-community-sign.byteimg.com\u002Ftos-cn-i-tlddhu82om\u002F5e95a6ad0bba40c09a4d902429453fef~tplv-tlddhu82om-image.image?=&rk3s=8031ce6d&x-expires=1732499851&x-signature=l3q4KE%2FOWSTPDkTzu77bV5zKjZc%3D","size":0,"mime_type":"","rid":""},"mime_type":"","content":"","resource":{},"extra":{"source":{"app_id":3569,"name":"","author":""},"html_content":""},"parent_id":"0","parent_type":0,"last_version":1,"last_status":2,"last_create_time":1732261417,"last_update_time":1732261417,"status_tags":[],"create_user_id":"0","item_source":"2","arcosite_id":"","text_url":""},"user":{"user_id":"1977369294279187","user_type":14,"name":"小窗幽记机器学习","avatar":{"key":"","url":"https:\u002F\u002Fp26-passport.byteacctimg.com\u002Fimg\u002Fuser-avatar\u002F1816ccc0ebc35f4a89457e0b4ffe692b~300x300.image","size":0,"mime_type":"","rid":""},"status":2},"categories":[{"category_id":"3","name":"AI","parent_id":"0","create_time":1637495002,"update_time":1637495249,"op_user_id":"0","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"rank":0,"category_url":""}],"tags":[{"tag_id":"7307103161498632233","name":"大模型","create_time":1701317532,"update_time":1701917647,"op_user_id":"3945470841857399","create_user_id":"620546920289582","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2},{"tag_id":"7246291206315966524","name":"智能语音交互","create_time":1687158646,"update_time":1687158646,"op_user_id":"7189883613045997626","create_user_id":"7189883613045997626","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2},{"tag_id":"7246291206278709303","name":"数据中台","create_time":1687158646,"update_time":1687158646,"op_user_id":"7189883613045997626","create_user_id":"7189883613045997626","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2}],"interact_status":{"is_like":false,"is_collect":false,"is_join":false,"like_count":0,"collect_count":0,"comment_count":0,"join_count":0,"evaluate_count":0,"evaluate_avg":0,"evaluate_self":0},"pool_status":{"is_top":false,"is_good":false,"is_recommend":false},"content_count":{"view_count":7,"child_count":0},"op_info":{"op_user":{},"op_time":0,"op_resource":""}},{"content":{"item_id":"7438453529850675237","item_type":2,"app_id":3569,"user_id":"1977369294279187","version":1,"status":2,"create_time":1731899925,"update_time":1731906243,"publish_time":1731899925,"name":"全面解读Apple Ferret-UI 2，屏幕多模态大模型(详解版)","abstract":"引言简介FERRET-UI 2数据集构建模型架构实验实验设置实验结果消融研究结论引言\n=======用户界面(UI)是人机交互的核心，随着智能设备和平台的多样化，UI的复杂性不断增加。\n然而，现有的UI理解与交互方法在多平台环境中仍有局限。\nFerret-UI(You et al., 2024) 在UI指称与语义理解方面取得了进展，但其固定的分辨率和仅针对移动设备的限制，难以应对多平台的复杂性。\n","cover_image":{"key":"","url":"https:\u002F\u002Fp3-volc-community-sign.byteimg.com\u002Ftos-cn-i-tlddhu82om\u002Fd820fe0d2f26415c8fb6f85f95f5df05~tplv-tlddhu82om-image.image?=&rk3s=8031ce6d&x-expires=1732499851&x-signature=xQF4mQYw4jnCauJvqMG3NFuw%2B5I%3D","size":0,"mime_type":"","rid":""},"mime_type":"","content":"","resource":{},"extra":{"source":{"app_id":3569,"name":"","author":""},"html_content":""},"parent_id":"0","parent_type":0,"last_version":1,"last_status":2,"last_create_time":1731899925,"last_update_time":1731899925,"status_tags":[],"create_user_id":"0","item_source":"2","arcosite_id":"","text_url":""},"user":{"user_id":"1977369294279187","user_type":14,"name":"小窗幽记机器学习","avatar":{"key":"","url":"https:\u002F\u002Fp26-passport.byteacctimg.com\u002Fimg\u002Fuser-avatar\u002F1816ccc0ebc35f4a89457e0b4ffe692b~300x300.image","size":0,"mime_type":"","rid":""},"status":2},"categories":[{"category_id":"7306040281718063141","name":"大模型","parent_id":"0","create_time":1701917585,"update_time":1701917585,"op_user_id":"3945470841857399","abstract":"大模型技术干货","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"rank":0,"category_url":""}],"tags":[{"tag_id":"7307103161498632233","name":"大模型","create_time":1701317532,"update_time":1701917647,"op_user_id":"3945470841857399","create_user_id":"620546920289582","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2},{"tag_id":"7309680367596732443","name":"向量数据库","create_time":1701917641,"update_time":1705378351,"op_user_id":"3945470841857399","create_user_id":"3945470841857399","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2},{"tag_id":"7246291206341132347","name":"云通信","create_time":1687158646,"update_time":1687158646,"op_user_id":"7189883613045997626","create_user_id":"7189883613045997626","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2}],"interact_status":{"is_like":false,"is_collect":false,"is_join":false,"like_count":0,"collect_count":0,"comment_count":0,"join_count":0,"evaluate_count":0,"evaluate_avg":0,"evaluate_self":0},"pool_status":{"is_top":false,"is_good":false,"is_recommend":false},"content_count":{"view_count":56,"child_count":0},"op_info":{"op_user":{},"op_time":0,"op_resource":""}},{"content":{"item_id":"7437404826503217189","item_type":2,"app_id":3569,"user_id":"1977369294279187","version":1,"status":2,"create_time":1731655755,"update_time":1731661905,"publish_time":1731655755,"name":"苹果发布Ferret-UI 2: 跨平台UI理解多模态大模型(精炼版)","abstract":"引言简介方法实验结果总结实战引言\n=======在数字设备日益普及的今天，用户界面(UI)已经成为人机交互的核心桥梁。近期，苹果公司发布的Ferret-UI 2 凭借其多平台兼容性和自适应编码等创新特性，在通用UI理解能力方面取得了显著突破。本文将简要介绍其核心技术与应用效果等要点 。如果小伙伴们想深入了解技术细节 ，欢迎关注下一篇文章：《全面解读Ferret-UI 2屏幕多模态大模型（详解版）》","cover_image":{"key":"","url":"https:\u002F\u002Fp3-volc-community-sign.byteimg.com\u002Ftos-cn-i-tlddhu82om\u002F587b8d7930ca4ca0837b5d5f8023a5ff~tplv-tlddhu82om-image.image?=&rk3s=8031ce6d&x-expires=1732499851&x-signature=oV0B55eNhjTvlhRh4I8MOkfmYEs%3D","size":0,"mime_type":"","rid":""},"mime_type":"","content":"","resource":{},"extra":{"source":{"app_id":3569,"name":"","author":""},"html_content":""},"parent_id":"0","parent_type":0,"last_version":1,"last_status":2,"last_create_time":1731655755,"last_update_time":1731655755,"status_tags":[],"create_user_id":"0","item_source":"2","arcosite_id":"","text_url":""},"user":{"user_id":"1977369294279187","user_type":14,"name":"小窗幽记机器学习","avatar":{"key":"","url":"https:\u002F\u002Fp26-passport.byteacctimg.com\u002Fimg\u002Fuser-avatar\u002F1816ccc0ebc35f4a89457e0b4ffe692b~300x300.image","size":0,"mime_type":"","rid":""},"status":2},"categories":[{"category_id":"3","name":"AI","parent_id":"0","create_time":1637495002,"update_time":1637495249,"op_user_id":"0","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"rank":0,"category_url":""}],"tags":[{"tag_id":"7307103161498632233","name":"大模型","create_time":1701317532,"update_time":1701917647,"op_user_id":"3945470841857399","create_user_id":"620546920289582","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2},{"tag_id":"7309680367596732443","name":"向量数据库","create_time":1701917641,"update_time":1705378351,"op_user_id":"3945470841857399","create_user_id":"3945470841857399","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2},{"tag_id":"7246291206341132347","name":"云通信","create_time":1687158646,"update_time":1687158646,"op_user_id":"7189883613045997626","create_user_id":"7189883613045997626","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2}],"interact_status":{"is_like":false,"is_collect":false,"is_join":false,"like_count":0,"collect_count":0,"comment_count":0,"join_count":0,"evaluate_count":0,"evaluate_avg":0,"evaluate_self":0},"pool_status":{"is_top":false,"is_good":false,"is_recommend":false},"content_count":{"view_count":32,"child_count":0},"op_info":{"op_user":{},"op_time":0,"op_resource":""}},{"content":{"item_id":"7436341741323550771","item_type":2,"app_id":3569,"user_id":"1977369294279187","version":1,"status":2,"create_time":1731408236,"update_time":1731409252,"publish_time":1731408236,"name":"Agent系列：AppAgent v2-屏幕智能Agent(详解版)","abstract":"引言简介方法Agent 框架Agent 交互探索阶段部署阶段文档生成高级功能实验结果总结局限性未来工作引言\n=======大语言模型（LLM）如 ChatGPT 和 GPT-4 显著提升了自然语言处理能力，并且推动了智能体在自主决策中的应用。最初，这些智能体专为基于文本的交互方式设计，展现了卓越的表现，包括记忆自适应性和多任务处理能力。然而，现实世界的应用程序不仅仅局限于文本输入，还涉及视觉和其他","cover_image":{"key":"","url":"https:\u002F\u002Fp3-volc-community-sign.byteimg.com\u002Ftos-cn-i-tlddhu82om\u002F8d019bcbb7244532a63a8702460e7135~tplv-tlddhu82om-image.image?=&rk3s=8031ce6d&x-expires=1732499851&x-signature=U4BghdJ3CO%2Bi5ZwckoqdAgiVOOs%3D","size":0,"mime_type":"","rid":""},"mime_type":"","content":"","resource":{},"extra":{"source":{"app_id":3569,"name":"","author":""},"html_content":""},"parent_id":"0","parent_type":0,"last_version":1,"last_status":2,"last_create_time":1731408236,"last_update_time":1731408236,"status_tags":[],"create_user_id":"0","item_source":"2","arcosite_id":"","text_url":""},"user":{"user_id":"1977369294279187","user_type":14,"name":"小窗幽记机器学习","avatar":{"key":"","url":"https:\u002F\u002Fp26-passport.byteacctimg.com\u002Fimg\u002Fuser-avatar\u002F1816ccc0ebc35f4a89457e0b4ffe692b~300x300.image","size":0,"mime_type":"","rid":""},"status":2},"categories":[{"category_id":"5","name":"移动开发","parent_id":"0","create_time":1637495002,"update_time":1637495249,"op_user_id":"0","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"rank":0,"category_url":""}],"tags":[{"tag_id":"7307103161498632233","name":"大模型","create_time":1701317532,"update_time":1701917647,"op_user_id":"3945470841857399","create_user_id":"620546920289582","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2},{"tag_id":"7146114517833875464","name":"数据库","create_time":1663834443,"update_time":1663834443,"op_user_id":"743694848765533","create_user_id":"743694848765533","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2},{"tag_id":"7146205183884656670","name":"算法","create_time":1663855553,"update_time":1663855553,"op_user_id":"4015857634130296","create_user_id":"4015857634130296","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2}],"interact_status":{"is_like":false,"is_collect":false,"is_join":false,"like_count":0,"collect_count":0,"comment_count":0,"join_count":0,"evaluate_count":0,"evaluate_avg":0,"evaluate_self":0},"pool_status":{"is_top":false,"is_good":false,"is_recommend":false},"content_count":{"view_count":235,"child_count":0},"op_info":{"op_user":{},"op_time":0,"op_resource":""}},{"content":{"item_id":"7436341741281935410","item_type":2,"app_id":3569,"user_id":"1977369294279187","version":1,"status":2,"create_time":1731408235,"update_time":1731408717,"publish_time":1731408235,"name":"Agent系列：多模态智能体AppAgent v2助力AI手机(简化版)","abstract":"AppAgent v2是一种专为移动设备设计的多模态智能体框架。该框架能够在移动设备上导航，模拟用户交互，适应各种应用程序，具体通过解析器、文本和视觉描述来增强其灵活性。AppAgent v2 的操作分为探索和部署两个阶段。","cover_image":{"key":"","url":"https:\u002F\u002Fp6-volc-community-sign.byteimg.com\u002Ftos-cn-i-tlddhu82om\u002Fbc97023f97e84c5ca3671d20715a958c~tplv-tlddhu82om-image.image?=&rk3s=8031ce6d&x-expires=1732499851&x-signature=E8%2BE7IPt%2Bl3%2BVFKie3%2BFj9uO9tM%3D","size":0,"mime_type":"","rid":""},"mime_type":"","content":"","resource":{},"extra":{"source":{"app_id":3569,"name":"","author":""},"html_content":""},"parent_id":"0","parent_type":0,"last_version":1,"last_status":2,"last_create_time":1731408235,"last_update_time":1731408235,"status_tags":[],"create_user_id":"0","item_source":"2","arcosite_id":"","text_url":""},"user":{"user_id":"1977369294279187","user_type":14,"name":"小窗幽记机器学习","avatar":{"key":"","url":"https:\u002F\u002Fp26-passport.byteacctimg.com\u002Fimg\u002Fuser-avatar\u002F1816ccc0ebc35f4a89457e0b4ffe692b~300x300.image","size":0,"mime_type":"","rid":""},"status":2},"categories":[{"category_id":"3","name":"AI","parent_id":"0","create_time":1637495002,"update_time":1637495249,"op_user_id":"0","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"rank":0,"category_url":""}],"tags":[{"tag_id":"7246291206312755255","name":"智能应用","create_time":1687158646,"update_time":1687158646,"op_user_id":"7189883613045997626","create_user_id":"7189883613045997626","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2},{"tag_id":"7146114517833875464","name":"数据库","create_time":1663834443,"update_time":1663834443,"op_user_id":"743694848765533","create_user_id":"743694848765533","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2},{"tag_id":"7146205408968278029","name":"图像处理","create_time":1663855605,"update_time":1663855605,"op_user_id":"4015857634130296","create_user_id":"4015857634130296","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2}],"interact_status":{"is_like":false,"is_collect":false,"is_join":false,"like_count":0,"collect_count":0,"comment_count":0,"join_count":0,"evaluate_count":0,"evaluate_avg":0,"evaluate_self":0},"pool_status":{"is_top":false,"is_good":false,"is_recommend":false},"content_count":{"view_count":43,"child_count":0},"op_info":{"op_user":{},"op_time":0,"op_resource":""}},{"content":{"item_id":"7433674203150155802","item_type":2,"app_id":3569,"user_id":"1977369294279187","version":1,"status":2,"create_time":1730787151,"update_time":1730798278,"publish_time":1730787151,"name":"万字长文细说端侧大模型进展(下篇)：AutoGLM类Agent隐私安全有感","abstract":"引言\n=======铺床凉满梧桐月，月在梧桐缺处明。小伙伴们好，我是微信公众号《小窗幽记机器学习》的小编：卖沙茶面的小女孩。最近，智谱AI推出的智能体AutoGLM引起了广泛关注。然而，随着测试的深入，用户对该产品将屏幕数据上传至云端所带来的隐私安全问题愈发担忧。因此，今天小编将基于近期一篇关于端侧大模型的综述文章，介绍该领域的一些最新进展。随着端侧大模型的发展，这些隐私安全问题有望得到有效缓解。","cover_image":{"key":"","url":"https:\u002F\u002Fp3-volc-community-sign.byteimg.com\u002Ftos-cn-i-tlddhu82om\u002F888587e1302640638ba80b2ac0cefc10~tplv-tlddhu82om-image.image?=&rk3s=8031ce6d&x-expires=1732499851&x-signature=QIOIFnRgLSElOGv1wviyI1qadX4%3D","size":0,"mime_type":"","rid":""},"mime_type":"","content":"","resource":{},"extra":{"source":{"app_id":3569,"name":"","author":""},"html_content":""},"parent_id":"0","parent_type":0,"last_version":1,"last_status":2,"last_create_time":1730787151,"last_update_time":1730787151,"status_tags":[],"create_user_id":"0","item_source":"2","arcosite_id":"","text_url":""},"user":{"user_id":"1977369294279187","user_type":14,"name":"小窗幽记机器学习","avatar":{"key":"","url":"https:\u002F\u002Fp26-passport.byteacctimg.com\u002Fimg\u002Fuser-avatar\u002F1816ccc0ebc35f4a89457e0b4ffe692b~300x300.image","size":0,"mime_type":"","rid":""},"status":2},"categories":[{"category_id":"7306040281718063141","name":"大模型","parent_id":"0","create_time":1701917585,"update_time":1701917585,"op_user_id":"3945470841857399","abstract":"大模型技术干货","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"rank":0,"category_url":""}],"tags":[{"tag_id":"7307103161498632233","name":"大模型","create_time":1701317532,"update_time":1701917647,"op_user_id":"3945470841857399","create_user_id":"620546920289582","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2},{"tag_id":"7146205183897239566","name":"机器学习","create_time":1663855553,"update_time":1663855553,"op_user_id":"4015857634130296","create_user_id":"4015857634130296","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2},{"tag_id":"7146114517833875464","name":"数据库","create_time":1663834443,"update_time":1663834443,"op_user_id":"743694848765533","create_user_id":"743694848765533","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2}],"interact_status":{"is_like":false,"is_collect":false,"is_join":false,"like_count":0,"collect_count":0,"comment_count":0,"join_count":0,"evaluate_count":0,"evaluate_avg":0,"evaluate_self":0},"pool_status":{"is_top":false,"is_good":false,"is_recommend":false},"content_count":{"view_count":310,"child_count":0},"op_info":{"op_user":{},"op_time":0,"op_resource":""}},{"content":{"item_id":"7433674165921513523","item_type":2,"app_id":3569,"user_id":"1977369294279187","version":1,"status":2,"create_time":1730787143,"update_time":1730790147,"publish_time":1730787143,"name":"万字长文细说端侧大模型进展(综述)","abstract":"文本介绍端侧大模型的最新进展，助力端侧智能体Agent发展。分为上下两篇：上篇主要介绍端侧大模型的进展及其模型架构，下篇则聚焦于端侧大模型的模型压缩技术、加速和部署方案以及应用实例。","cover_image":{"key":"","url":"https:\u002F\u002Fp3-volc-community-sign.byteimg.com\u002Ftos-cn-i-tlddhu82om\u002F288d54381cb54c8e9d6da7f22fcc9d16~tplv-tlddhu82om-image.image?=&rk3s=8031ce6d&x-expires=1732499851&x-signature=osc8TohEehgQpQduxQNjRrE2b1o%3D","size":0,"mime_type":"","rid":""},"mime_type":"","content":"","resource":{},"extra":{"source":{"app_id":3569,"name":"","author":""},"html_content":""},"parent_id":"0","parent_type":0,"last_version":1,"last_status":2,"last_create_time":1730787143,"last_update_time":1730787143,"status_tags":[],"create_user_id":"0","item_source":"2","arcosite_id":"","text_url":""},"user":{"user_id":"1977369294279187","user_type":14,"name":"小窗幽记机器学习","avatar":{"key":"","url":"https:\u002F\u002Fp26-passport.byteacctimg.com\u002Fimg\u002Fuser-avatar\u002F1816ccc0ebc35f4a89457e0b4ffe692b~300x300.image","size":0,"mime_type":"","rid":""},"status":2},"categories":[{"category_id":"7306040281718063141","name":"大模型","parent_id":"0","create_time":1701917585,"update_time":1701917585,"op_user_id":"3945470841857399","abstract":"大模型技术干货","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"rank":0,"category_url":""}],"tags":[{"tag_id":"7307103161498632233","name":"大模型","create_time":1701317532,"update_time":1701917647,"op_user_id":"3945470841857399","create_user_id":"620546920289582","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2},{"tag_id":"7246291206202900538","name":"数据安全","create_time":1687158646,"update_time":1687158646,"op_user_id":"7189883613045997626","create_user_id":"7189883613045997626","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2},{"tag_id":"7146205183897239566","name":"机器学习","create_time":1663855553,"update_time":1663855553,"op_user_id":"4015857634130296","create_user_id":"4015857634130296","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2}],"interact_status":{"is_like":false,"is_collect":false,"is_join":false,"like_count":0,"collect_count":1,"comment_count":0,"join_count":0,"evaluate_count":0,"evaluate_avg":0,"evaluate_self":0},"pool_status":{"is_top":false,"is_good":false,"is_recommend":false},"content_count":{"view_count":1038,"child_count":0},"op_info":{"op_user":{},"op_time":0,"op_resource":""}},{"content":{"item_id":"7433674165934096393","item_type":2,"app_id":3569,"user_id":"1977369294279187","version":1,"status":2,"create_time":1730787142,"update_time":1730789916,"publish_time":1730787142,"name":"AI手机新纪元：AutoGLM开启后APP时代下的挑战与机遇","abstract":"山黛远，月波长，暮云秋影蘸潇湘。小伙伴们好，我是微信公众号《小窗幽记机器学习》的小编：卖热干面的小女孩。近日，国内知名AI公司智谱在中国计算机大会（CNCC）上发布的AutoGLM引发了业界广泛关注。这款突破性的AI智能体产品能通过语音指令理解用户意图，模拟人类操作手机，自动完成从网页浏览、商品购物到社交媒体互动等多种复杂任务。同期，Anthropic公司推出的Claude 3.5系列模型（Son","cover_image":{"key":"","url":"https:\u002F\u002Fp6-volc-community-sign.byteimg.com\u002Ftos-cn-i-tlddhu82om\u002F870e42ff182049b8a337f1be96b5ba2e~tplv-tlddhu82om-image.image?=&rk3s=8031ce6d&x-expires=1732499851&x-signature=Rb4%2Fn%2FNFZqv%2Bx0RGTF%2BChI51DQs%3D","size":0,"mime_type":"","rid":""},"mime_type":"","content":"","resource":{},"extra":{"source":{"app_id":3569,"name":"","author":""},"html_content":""},"parent_id":"0","parent_type":0,"last_version":1,"last_status":2,"last_create_time":1730787142,"last_update_time":1730787142,"status_tags":[],"create_user_id":"0","item_source":"2","arcosite_id":"","text_url":""},"user":{"user_id":"1977369294279187","user_type":14,"name":"小窗幽记机器学习","avatar":{"key":"","url":"https:\u002F\u002Fp26-passport.byteacctimg.com\u002Fimg\u002Fuser-avatar\u002F1816ccc0ebc35f4a89457e0b4ffe692b~300x300.image","size":0,"mime_type":"","rid":""},"status":2},"categories":[{"category_id":"3","name":"AI","parent_id":"0","create_time":1637495002,"update_time":1637495249,"op_user_id":"0","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"rank":0,"category_url":""}],"tags":[{"tag_id":"7307103161498632233","name":"大模型","create_time":1701317532,"update_time":1701917647,"op_user_id":"3945470841857399","create_user_id":"620546920289582","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2},{"tag_id":"7246291206315966524","name":"智能语音交互","create_time":1687158646,"update_time":1687158646,"op_user_id":"7189883613045997626","create_user_id":"7189883613045997626","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2},{"tag_id":"7246291206202900538","name":"数据安全","create_time":1687158646,"update_time":1687158646,"op_user_id":"7189883613045997626","create_user_id":"7189883613045997626","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2}],"interact_status":{"is_like":false,"is_collect":false,"is_join":false,"like_count":0,"collect_count":0,"comment_count":0,"join_count":0,"evaluate_count":0,"evaluate_avg":0,"evaluate_self":0},"pool_status":{"is_top":false,"is_good":false,"is_recommend":false},"content_count":{"view_count":65,"child_count":0},"op_info":{"op_user":{},"op_time":0,"op_resource":""}},{"content":{"item_id":"7432210796606423078","item_type":2,"app_id":3569,"user_id":"1977369294279187","version":1,"status":2,"create_time":1730446426,"update_time":1730460586,"publish_time":1730446426,"name":"万字长文梳理端侧大模型进展(上篇)：由AutoGLM类Agent隐私安全有感","abstract":"文本介绍端侧大模型的最新进展，助力端侧智能体Agent发展。分为上下两篇：上篇主要介绍端侧大模型的进展及其模型架构，下篇则聚焦于端侧大模型的模型压缩技术、加速和部署方案以及应用实例。","cover_image":{"key":"","url":"https:\u002F\u002Fp3-volc-community-sign.byteimg.com\u002Ftos-cn-i-tlddhu82om\u002F0623d2e339f64d839a3ae835e2b63b2b~tplv-tlddhu82om-image.image?=&rk3s=8031ce6d&x-expires=1732499851&x-signature=Oni8DnU4xBjrEYCL44RA%2BPmj8eU%3D","size":0,"mime_type":"","rid":""},"mime_type":"","content":"","resource":{},"extra":{"source":{"app_id":3569,"name":"","author":""},"html_content":""},"parent_id":"0","parent_type":0,"last_version":1,"last_status":2,"last_create_time":1730446426,"last_update_time":1730446426,"status_tags":[],"create_user_id":"0","item_source":"2","arcosite_id":"","text_url":""},"user":{"user_id":"1977369294279187","user_type":14,"name":"小窗幽记机器学习","avatar":{"key":"","url":"https:\u002F\u002Fp26-passport.byteacctimg.com\u002Fimg\u002Fuser-avatar\u002F1816ccc0ebc35f4a89457e0b4ffe692b~300x300.image","size":0,"mime_type":"","rid":""},"status":2},"categories":[{"category_id":"7306040281718063141","name":"大模型","parent_id":"0","create_time":1701917585,"update_time":1701917585,"op_user_id":"3945470841857399","abstract":"大模型技术干货","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"rank":0,"category_url":""}],"tags":[{"tag_id":"7307103161498632233","name":"大模型","create_time":1701317532,"update_time":1701917647,"op_user_id":"3945470841857399","create_user_id":"620546920289582","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2},{"tag_id":"7146205183897239566","name":"机器学习","create_time":1663855553,"update_time":1663855553,"op_user_id":"4015857634130296","create_user_id":"4015857634130296","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2},{"tag_id":"7146205183884656670","name":"算法","create_time":1663855553,"update_time":1663855553,"op_user_id":"4015857634130296","create_user_id":"4015857634130296","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2}],"interact_status":{"is_like":false,"is_collect":false,"is_join":false,"like_count":0,"collect_count":0,"comment_count":0,"join_count":0,"evaluate_count":0,"evaluate_avg":0,"evaluate_self":0},"pool_status":{"is_top":false,"is_good":false,"is_recommend":false},"content_count":{"view_count":299,"child_count":0},"op_info":{"op_user":{},"op_time":0,"op_resource":""}},{"content":{"item_id":"7428432985748078602","item_type":2,"app_id":3569,"user_id":"1977369294279187","version":1,"status":2,"create_time":1729566836,"update_time":1729575447,"publish_time":1729566836,"name":"万字长文深度解读Movie Gen技术原理(5部曲)：图像&视频联合生成模型 (2)","abstract":"详细介绍Movie Gen中图像和视频的联合生成技术，包括:时间自编码器的设计与优化、基于流匹配的训练目标、联合生成的网络架构、文本嵌入和视觉-文本生成方法、空间上采样技术、模型扩展和训练效率优化等。此外，还详细介绍了预训练数据的准备过程.","cover_image":{"key":"","url":"https:\u002F\u002Fp3-volc-community-sign.byteimg.com\u002Ftos-cn-i-tlddhu82om\u002Fb9d1d1cee6e44e01bb9ddbddcc03f31e~tplv-tlddhu82om-image.image?=&rk3s=8031ce6d&x-expires=1732499851&x-signature=IEySpvUDnvIyj6WlYo39r2px1uc%3D","size":0,"mime_type":"","rid":""},"mime_type":"","content":"","resource":{},"extra":{"source":{"app_id":3569,"name":"","author":""},"html_content":""},"parent_id":"0","parent_type":0,"last_version":1,"last_status":2,"last_create_time":1729566836,"last_update_time":1729566836,"status_tags":[],"create_user_id":"0","item_source":"2","arcosite_id":"","text_url":""},"user":{"user_id":"1977369294279187","user_type":14,"name":"小窗幽记机器学习","avatar":{"key":"","url":"https:\u002F\u002Fp26-passport.byteacctimg.com\u002Fimg\u002Fuser-avatar\u002F1816ccc0ebc35f4a89457e0b4ffe692b~300x300.image","size":0,"mime_type":"","rid":""},"status":2},"categories":[{"category_id":"6","name":"音视频","parent_id":"0","create_time":1637495002,"update_time":1637495249,"op_user_id":"0","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"rank":0,"category_url":""}],"tags":[{"tag_id":"7307103161498632233","name":"大模型","create_time":1701317532,"update_time":1701917647,"op_user_id":"3945470841857399","create_user_id":"620546920289582","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2},{"tag_id":"7146205183897239566","name":"机器学习","create_time":1663855553,"update_time":1663855553,"op_user_id":"4015857634130296","create_user_id":"4015857634130296","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2},{"tag_id":"7146205408968278029","name":"图像处理","create_time":1663855605,"update_time":1663855605,"op_user_id":"4015857634130296","create_user_id":"4015857634130296","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2}],"interact_status":{"is_like":false,"is_collect":false,"is_join":false,"like_count":0,"collect_count":0,"comment_count":0,"join_count":0,"evaluate_count":0,"evaluate_avg":0,"evaluate_self":0},"pool_status":{"is_top":false,"is_good":false,"is_recommend":false},"content_count":{"view_count":61,"child_count":0},"op_info":{"op_user":{},"op_time":0,"op_resource":""}},{"content":{"item_id":"7428432985717735451","item_type":2,"app_id":3569,"user_id":"1977369294279187","version":1,"status":2,"create_time":1729566836,"update_time":1729575442,"publish_time":1729566836,"name":"全面深入解读Movie Gen技术原理(5部曲)：个性化视频生成(3)","abstract":"引言简介生成个性化视频模型预训练预训练数据预训练方法监督微调评估结果总结最是人间留不住，朱颜辞镜花辞树。小伙伴们好，我是微信公众号《小窗幽记机器学习》的小编：缅A消费的小女孩。紧接此前Movie Gen解读系列：突发！Meta重磅发布Movie Gen入局视频生成赛道！全面深入解读Movie Gen技术原理(5部曲)：概述 (1)万字长文深度解读Movie Gen技术原理(5部曲)：图像视频联合生","cover_image":{"key":"","url":"https:\u002F\u002Fp6-volc-community-sign.byteimg.com\u002Ftos-cn-i-tlddhu82om\u002F7a22739199dd46119a90c1c1be49a85e~tplv-tlddhu82om-image.image?=&rk3s=8031ce6d&x-expires=1732499851&x-signature=9lP0K1jOf4mZQAlYIE7TeEpT2q0%3D","size":0,"mime_type":"","rid":""},"mime_type":"","content":"","resource":{},"extra":{"source":{"app_id":3569,"name":"","author":""},"html_content":""},"parent_id":"0","parent_type":0,"last_version":1,"last_status":2,"last_create_time":1729566836,"last_update_time":1729566836,"status_tags":[],"create_user_id":"0","item_source":"2","arcosite_id":"","text_url":""},"user":{"user_id":"1977369294279187","user_type":14,"name":"小窗幽记机器学习","avatar":{"key":"","url":"https:\u002F\u002Fp26-passport.byteacctimg.com\u002Fimg\u002Fuser-avatar\u002F1816ccc0ebc35f4a89457e0b4ffe692b~300x300.image","size":0,"mime_type":"","rid":""},"status":2},"categories":[{"category_id":"3","name":"AI","parent_id":"0","create_time":1637495002,"update_time":1637495249,"op_user_id":"0","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"rank":0,"category_url":""}],"tags":[{"tag_id":"7307103161498632233","name":"大模型","create_time":1701317532,"update_time":1701917647,"op_user_id":"3945470841857399","create_user_id":"620546920289582","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2},{"tag_id":"7246291206253068347","name":"视频服务","create_time":1687158646,"update_time":1687158646,"op_user_id":"7189883613045997626","create_user_id":"7189883613045997626","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2},{"tag_id":"7146205183897239566","name":"机器学习","create_time":1663855553,"update_time":1663855553,"op_user_id":"4015857634130296","create_user_id":"4015857634130296","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2}],"interact_status":{"is_like":false,"is_collect":false,"is_join":false,"like_count":0,"collect_count":0,"comment_count":0,"join_count":0,"evaluate_count":0,"evaluate_avg":0,"evaluate_self":0},"pool_status":{"is_top":false,"is_good":false,"is_recommend":false},"content_count":{"view_count":73,"child_count":0},"op_info":{"op_user":{},"op_time":0,"op_resource":""}},{"content":{"item_id":"7428432985050841115","item_type":2,"app_id":3569,"user_id":"1977369294279187","version":1,"status":2,"create_time":1729566835,"update_time":1729575339,"publish_time":1729566835,"name":"全面深入解读Movie Gen技术原理(5部曲)：概述 (1)","abstract":"引言\n=======小伙伴们好，我是微信公众号《小窗幽记机器学习》的小编：卖蚵仔煎的小男孩。紧接此前的文章：突发！Meta重磅发布Movie Gen入局视频生成赛道！，这几天临时搁置端侧大模型系列专题的深挖，先腾挪些时间阅读Meta官方发布的Movie Gen技术报告，从而基于官方一手资料 详细解读Movie Gen 模型。2024年10月4日，Meta发布其视频生成产品Movie Gen，对标","cover_image":{"key":"","url":"https:\u002F\u002Fp6-volc-community-sign.byteimg.com\u002Ftos-cn-i-tlddhu82om\u002F04bfea696b754a7c941c5af6928f2dac~tplv-tlddhu82om-image.image?=&rk3s=8031ce6d&x-expires=1732499851&x-signature=W%2Bi%2BMbQVi1EGMsIgKl7xRGwRGJM%3D","size":0,"mime_type":"","rid":""},"mime_type":"","content":"","resource":{},"extra":{"source":{"app_id":3569,"name":"","author":""},"html_content":""},"parent_id":"0","parent_type":0,"last_version":1,"last_status":2,"last_create_time":1729566835,"last_update_time":1729566835,"status_tags":[],"create_user_id":"0","item_source":"2","arcosite_id":"","text_url":""},"user":{"user_id":"1977369294279187","user_type":14,"name":"小窗幽记机器学习","avatar":{"key":"","url":"https:\u002F\u002Fp26-passport.byteacctimg.com\u002Fimg\u002Fuser-avatar\u002F1816ccc0ebc35f4a89457e0b4ffe692b~300x300.image","size":0,"mime_type":"","rid":""},"status":2},"categories":[{"category_id":"6","name":"音视频","parent_id":"0","create_time":1637495002,"update_time":1637495249,"op_user_id":"0","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"rank":0,"category_url":""}],"tags":[{"tag_id":"7307103161498632233","name":"大模型","create_time":1701317532,"update_time":1701917647,"op_user_id":"3945470841857399","create_user_id":"620546920289582","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2},{"tag_id":"7246291206253068347","name":"视频服务","create_time":1687158646,"update_time":1687158646,"op_user_id":"7189883613045997626","create_user_id":"7189883613045997626","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2},{"tag_id":"7146205183897239566","name":"机器学习","create_time":1663855553,"update_time":1663855553,"op_user_id":"4015857634130296","create_user_id":"4015857634130296","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2}],"interact_status":{"is_like":false,"is_collect":false,"is_join":false,"like_count":0,"collect_count":0,"comment_count":0,"join_count":0,"evaluate_count":0,"evaluate_avg":0,"evaluate_self":0},"pool_status":{"is_top":false,"is_good":false,"is_recommend":false},"content_count":{"view_count":64,"child_count":0},"op_info":{"op_user":{},"op_time":0,"op_resource":""}},{"content":{"item_id":"7428432985701941275","item_type":2,"app_id":3569,"user_id":"1977369294279187","version":1,"status":2,"create_time":1729566834,"update_time":1729575309,"publish_time":1729566834,"name":"突发！Meta重磅发布Movie Gen入局视频生成赛道！","abstract":"Meta于2024年10月4日首次推出 Meta Movie Gen，号称是迄今为止最先进的媒体基础模型。Movie Gen 由 Meta 的 AI 研究团队开发，在一系列功能上获取最先进的效果，包括\n：\n文生视频、\n创建个性化视频、精准\n的视频编辑和\n音频创作。无论是渴望在好莱坞闯出一片天地的新晋电影人，还是热衷于为观众制作视频的创作者，每个人都应该有机会使用能够增强创造力的工具。\nMeta公司","cover_image":{"key":"","url":"https:\u002F\u002Fp6-volc-community-sign.byteimg.com\u002Ftos-cn-i-tlddhu82om\u002Fc7e68aab7a494f609feaede4bac53e74~tplv-tlddhu82om-image.image?=&rk3s=8031ce6d&x-expires=1732499851&x-signature=1Bn%2Bl75pXsgPlOuwtZPOenQXx9g%3D","size":0,"mime_type":"","rid":""},"mime_type":"","content":"","resource":{},"extra":{"source":{"app_id":3569,"name":"","author":""},"html_content":""},"parent_id":"0","parent_type":0,"last_version":1,"last_status":2,"last_create_time":1729566834,"last_update_time":1729566834,"status_tags":[],"create_user_id":"0","item_source":"2","arcosite_id":"","text_url":""},"user":{"user_id":"1977369294279187","user_type":14,"name":"小窗幽记机器学习","avatar":{"key":"","url":"https:\u002F\u002Fp26-passport.byteacctimg.com\u002Fimg\u002Fuser-avatar\u002F1816ccc0ebc35f4a89457e0b4ffe692b~300x300.image","size":0,"mime_type":"","rid":""},"status":2},"categories":[{"category_id":"3","name":"AI","parent_id":"0","create_time":1637495002,"update_time":1637495249,"op_user_id":"0","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"rank":0,"category_url":""}],"tags":[{"tag_id":"7246291206253068347","name":"视频服务","create_time":1687158646,"update_time":1687158646,"op_user_id":"7189883613045997626","create_user_id":"7189883613045997626","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2},{"tag_id":"7246291206308560952","name":"智能体验与创作","create_time":1687158646,"update_time":1687158646,"op_user_id":"7189883613045997626","create_user_id":"7189883613045997626","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2},{"tag_id":"7146205183897239566","name":"机器学习","create_time":1663855553,"update_time":1663855553,"op_user_id":"4015857634130296","create_user_id":"4015857634130296","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2}],"interact_status":{"is_like":false,"is_collect":false,"is_join":false,"like_count":0,"collect_count":0,"comment_count":0,"join_count":0,"evaluate_count":0,"evaluate_avg":0,"evaluate_self":0},"pool_status":{"is_top":false,"is_good":false,"is_recommend":false},"content_count":{"view_count":56,"child_count":0},"op_info":{"op_user":{},"op_time":0,"op_resource":""}},{"content":{"item_id":"7428432810723164169","item_type":2,"app_id":3569,"user_id":"1977369294279187","version":1,"status":2,"create_time":1729566795,"update_time":1729570778,"publish_time":1729566795,"name":"Sora已死？全面深入解读Movie Gen技术原理5部曲：4-精准视频编辑","abstract":"引言简介模型架构改进单帧视频编辑训练多帧视频编辑训练反向翻译的视频编辑训练结果一年好景君须记，最是橙黄橘绿时。小伙伴们好，我是微信公众号《小窗幽记机器学习》的小编：缅A消费积极分子。紧接此前Movie Gen解读系列：突发！Meta重磅发布Movie Gen入局视频生成赛道！全面深入解读Movie Gen技术原理(5部曲)：1-概述全面深入解读Movie Gen技术原理(5部曲)：2-图像视频联合","cover_image":{"key":"","url":"https:\u002F\u002Fp6-volc-community-sign.byteimg.com\u002Ftos-cn-i-tlddhu82om\u002Ffd7f70f6024d4fc5b7cabdb1eb4be0dc~tplv-tlddhu82om-image.image?=&rk3s=8031ce6d&x-expires=1732499851&x-signature=uKxPADgr18Jk3PVT3P8Qnys%2FVJ0%3D","size":0,"mime_type":"","rid":""},"mime_type":"","content":"","resource":{},"extra":{"source":{"app_id":3569,"name":"","author":""},"html_content":""},"parent_id":"0","parent_type":0,"last_version":1,"last_status":2,"last_create_time":1729566795,"last_update_time":1729566795,"status_tags":[],"create_user_id":"0","item_source":"2","arcosite_id":"","text_url":""},"user":{"user_id":"1977369294279187","user_type":14,"name":"小窗幽记机器学习","avatar":{"key":"","url":"https:\u002F\u002Fp26-passport.byteacctimg.com\u002Fimg\u002Fuser-avatar\u002F1816ccc0ebc35f4a89457e0b4ffe692b~300x300.image","size":0,"mime_type":"","rid":""},"status":2},"categories":[{"category_id":"3","name":"AI","parent_id":"0","create_time":1637495002,"update_time":1637495249,"op_user_id":"0","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"rank":0,"category_url":""}],"tags":[{"tag_id":"7246291206253068347","name":"视频服务","create_time":1687158646,"update_time":1687158646,"op_user_id":"7189883613045997626","create_user_id":"7189883613045997626","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2},{"tag_id":"7146205408968278029","name":"图像处理","create_time":1663855605,"update_time":1663855605,"op_user_id":"4015857634130296","create_user_id":"4015857634130296","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2},{"tag_id":"7146205183897239566","name":"机器学习","create_time":1663855553,"update_time":1663855553,"op_user_id":"4015857634130296","create_user_id":"4015857634130296","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2}],"interact_status":{"is_like":false,"is_collect":false,"is_join":false,"like_count":0,"collect_count":0,"comment_count":0,"join_count":0,"evaluate_count":0,"evaluate_avg":0,"evaluate_self":0},"pool_status":{"is_top":false,"is_good":false,"is_recommend":false},"content_count":{"view_count":64,"child_count":0},"op_info":{"op_user":{},"op_time":0,"op_resource":""}},{"content":{"item_id":"7424424401317134363","item_type":2,"app_id":3569,"user_id":"1977369294279187","version":1,"status":2,"create_time":1728633515,"update_time":1728635761,"publish_time":1728633515,"name":"万字长文深度解读Movie Gen技术原理(5部曲)：图像&视频联合生成模型 (2)","abstract":"详细介绍Movie Gen中图像和视频的联合生成技术，包括:时间自编码器的设计与优化、基于流匹配的训练目标、联合生成的网络架构、文本嵌入和视觉-文本生成方法、空间上采样技术、模型扩展和训练效率优化等。此外，还详细介绍了预训练数据的准备过程.","cover_image":{"key":"","url":"https:\u002F\u002Fp6-volc-community-sign.byteimg.com\u002Ftos-cn-i-tlddhu82om\u002F8bf15e423f33468f952d11f598f08d08~tplv-tlddhu82om-image.image?=&rk3s=8031ce6d&x-expires=1732499851&x-signature=tuqqyNahu0rm4Pdn3SKC%2FoxkJJ8%3D","size":0,"mime_type":"","rid":""},"mime_type":"","content":"","resource":{},"extra":{"source":{"app_id":3569,"name":"","author":""},"html_content":""},"parent_id":"0","parent_type":0,"last_version":1,"last_status":2,"last_create_time":1728633515,"last_update_time":1728633515,"status_tags":[],"create_user_id":"0","item_source":"2","arcosite_id":"","text_url":""},"user":{"user_id":"1977369294279187","user_type":14,"name":"小窗幽记机器学习","avatar":{"key":"","url":"https:\u002F\u002Fp26-passport.byteacctimg.com\u002Fimg\u002Fuser-avatar\u002F1816ccc0ebc35f4a89457e0b4ffe692b~300x300.image","size":0,"mime_type":"","rid":""},"status":2},"categories":[{"category_id":"6","name":"音视频","parent_id":"0","create_time":1637495002,"update_time":1637495249,"op_user_id":"0","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"rank":0,"category_url":""}],"tags":[{"tag_id":"7307103161498632233","name":"大模型","create_time":1701317532,"update_time":1701917647,"op_user_id":"3945470841857399","create_user_id":"620546920289582","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2},{"tag_id":"7146205408968278029","name":"图像处理","create_time":1663855605,"update_time":1663855605,"op_user_id":"4015857634130296","create_user_id":"4015857634130296","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2},{"tag_id":"7146205183897239566","name":"机器学习","create_time":1663855553,"update_time":1663855553,"op_user_id":"4015857634130296","create_user_id":"4015857634130296","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2}],"interact_status":{"is_like":false,"is_collect":false,"is_join":false,"like_count":0,"collect_count":0,"comment_count":0,"join_count":0,"evaluate_count":0,"evaluate_avg":0,"evaluate_self":0},"pool_status":{"is_top":false,"is_good":false,"is_recommend":false},"content_count":{"view_count":243,"child_count":0},"op_info":{"op_user":{},"op_time":0,"op_resource":""}},{"content":{"item_id":"7423667140789551130","item_type":2,"app_id":3569,"user_id":"1977369294279187","version":1,"status":2,"create_time":1728457201,"update_time":1728459645,"publish_time":1728457201,"name":"全面深入解读Movie Gen技术原理(5部曲)：概述 (1)","abstract":"引言\n=======小伙伴们好，我是微信公众号《小窗幽记机器学习》的小编：卖蚵仔煎的小男孩。紧接此前的文章：突发！Meta重磅发布Movie Gen入局视频生成赛道！，这几天临时搁置端侧大模型系列专题的深挖，先腾挪些时间阅读Meta官方发布的Movie Gen技术报告，从而基于官方一手资料 详细解读Movie Gen 模型。2024年10月4日，Meta发布其视频生成产品Movie Gen，对标","cover_image":{"key":"","url":"https:\u002F\u002Fp6-volc-community-sign.byteimg.com\u002Ftos-cn-i-tlddhu82om\u002F3594d0cf6bfb4f71b0d55ea0ec3f4d61~tplv-tlddhu82om-image.image?=&rk3s=8031ce6d&x-expires=1732499851&x-signature=2xe6MzRTBbA4i2wTZLGTXOD%2BR1k%3D","size":0,"mime_type":"","rid":""},"mime_type":"","content":"","resource":{},"extra":{"source":{"app_id":3569,"name":"","author":""},"html_content":""},"parent_id":"0","parent_type":0,"last_version":1,"last_status":2,"last_create_time":1728457200,"last_update_time":1728457201,"status_tags":[],"create_user_id":"0","item_source":"2","arcosite_id":"","text_url":""},"user":{"user_id":"1977369294279187","user_type":14,"name":"小窗幽记机器学习","avatar":{"key":"","url":"https:\u002F\u002Fp26-passport.byteacctimg.com\u002Fimg\u002Fuser-avatar\u002F1816ccc0ebc35f4a89457e0b4ffe692b~300x300.image","size":0,"mime_type":"","rid":""},"status":2},"categories":[{"category_id":"6","name":"音视频","parent_id":"0","create_time":1637495002,"update_time":1637495249,"op_user_id":"0","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"rank":0,"category_url":""}],"tags":[{"tag_id":"7307103161498632233","name":"大模型","create_time":1701317532,"update_time":1701917647,"op_user_id":"3945470841857399","create_user_id":"620546920289582","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2},{"tag_id":"7246291206253068347","name":"视频服务","create_time":1687158646,"update_time":1687158646,"op_user_id":"7189883613045997626","create_user_id":"7189883613045997626","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2},{"tag_id":"7146205183897239566","name":"机器学习","create_time":1663855553,"update_time":1663855553,"op_user_id":"4015857634130296","create_user_id":"4015857634130296","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2}],"interact_status":{"is_like":false,"is_collect":false,"is_join":false,"like_count":0,"collect_count":0,"comment_count":0,"join_count":0,"evaluate_count":0,"evaluate_avg":0,"evaluate_self":0},"pool_status":{"is_top":false,"is_good":false,"is_recommend":false},"content_count":{"view_count":264,"child_count":0},"op_info":{"op_user":{},"op_time":0,"op_resource":""}},{"content":{"item_id":"7423250703340339251","item_type":2,"app_id":3569,"user_id":"1977369294279187","version":1,"status":2,"create_time":1728360241,"update_time":1728364719,"publish_time":1728360241,"name":"端侧Agent系列 | 端侧AI Agent任务拆解大师如何助力AI手机？(详解版)","abstract":"本文主要介绍端侧Agent智能体如何实现任务规划和拆解，从而实现复杂场景下多指令任务的执行。Octo-planner，这是一个专为边缘设备而设计的用于规划任务的AI Agent框架。","cover_image":{"key":"","url":"https:\u002F\u002Fp3-volc-community-sign.byteimg.com\u002Ftos-cn-i-tlddhu82om\u002Fb3eab84269564e2594124a3dd817652b~tplv-tlddhu82om-image.image?=&rk3s=8031ce6d&x-expires=1732499851&x-signature=DQuGWkDFw4secDwIHB%2BNf4n02PI%3D","size":0,"mime_type":"","rid":""},"mime_type":"","content":"","resource":{},"extra":{"source":{"app_id":3569,"name":"","author":""},"html_content":""},"parent_id":"0","parent_type":0,"last_version":1,"last_status":2,"last_create_time":1728360241,"last_update_time":1728360241,"status_tags":[],"create_user_id":"0","item_source":"2","arcosite_id":"","text_url":""},"user":{"user_id":"1977369294279187","user_type":14,"name":"小窗幽记机器学习","avatar":{"key":"","url":"https:\u002F\u002Fp26-passport.byteacctimg.com\u002Fimg\u002Fuser-avatar\u002F1816ccc0ebc35f4a89457e0b4ffe692b~300x300.image","size":0,"mime_type":"","rid":""},"status":2},"categories":[{"category_id":"7306040281718063141","name":"大模型","parent_id":"0","create_time":1701917585,"update_time":1701917585,"op_user_id":"3945470841857399","abstract":"大模型技术干货","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"rank":0,"category_url":""}],"tags":[{"tag_id":"7307103161498632233","name":"大模型","create_time":1701317532,"update_time":1701917647,"op_user_id":"3945470841857399","create_user_id":"620546920289582","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2},{"tag_id":"7146205183897239566","name":"机器学习","create_time":1663855553,"update_time":1663855553,"op_user_id":"4015857634130296","create_user_id":"4015857634130296","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2},{"tag_id":"7146114517833875464","name":"数据库","create_time":1663834443,"update_time":1663834443,"op_user_id":"743694848765533","create_user_id":"743694848765533","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2}],"interact_status":{"is_like":false,"is_collect":false,"is_join":false,"like_count":0,"collect_count":0,"comment_count":0,"join_count":0,"evaluate_count":0,"evaluate_avg":0,"evaluate_self":0},"pool_status":{"is_top":false,"is_good":false,"is_recommend":false},"content_count":{"view_count":76,"child_count":0},"op_info":{"op_user":{},"op_time":0,"op_resource":""}},{"content":{"item_id":"7423250703256633394","item_type":2,"app_id":3569,"user_id":"1977369294279187","version":1,"status":2,"create_time":1728360241,"update_time":1728364690,"publish_time":1728360241,"name":"端侧大模型系列 | 端侧AI Agent任务拆解大师如何助力AI手机？(简短版)","abstract":"引言简介模型实验意义&前景:总结今人不见古时月，今月曾经照古人。小伙伴们好，我是微信公众号《小窗幽记机器学习》的小编：卖沙茶面的小女孩。设想一下，你的智能手机不再只是\"聪明\"，而是真正的\"智能\"。它不仅能响应指令，还能预测需求、规划日程，并轻松完成复杂任务。这不是科幻小说，而是设备内置AI助手的新时代，即将到来。NexaAI提出的Octo-planner通过分离规划与执行过程，将先进的AI功能装进","cover_image":{"key":"","url":"https:\u002F\u002Fp6-volc-community-sign.byteimg.com\u002Ftos-cn-i-tlddhu82om\u002F32ba7371fc44430c9cefa72605d1ec0c~tplv-tlddhu82om-image.image?=&rk3s=8031ce6d&x-expires=1732499851&x-signature=ktB7tET0dt5auAklmQdwhRCi2h0%3D","size":0,"mime_type":"","rid":""},"mime_type":"","content":"","resource":{},"extra":{"source":{"app_id":3569,"name":"","author":""},"html_content":""},"parent_id":"0","parent_type":0,"last_version":1,"last_status":2,"last_create_time":1728360241,"last_update_time":1728360241,"status_tags":[],"create_user_id":"0","item_source":"2","arcosite_id":"","text_url":""},"user":{"user_id":"1977369294279187","user_type":14,"name":"小窗幽记机器学习","avatar":{"key":"","url":"https:\u002F\u002Fp26-passport.byteacctimg.com\u002Fimg\u002Fuser-avatar\u002F1816ccc0ebc35f4a89457e0b4ffe692b~300x300.image","size":0,"mime_type":"","rid":""},"status":2},"categories":[{"category_id":"3","name":"AI","parent_id":"0","create_time":1637495002,"update_time":1637495249,"op_user_id":"0","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"rank":0,"category_url":""}],"tags":[{"tag_id":"7307103161498632233","name":"大模型","create_time":1701317532,"update_time":1701917647,"op_user_id":"3945470841857399","create_user_id":"620546920289582","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2},{"tag_id":"7146205183897239566","name":"机器学习","create_time":1663855553,"update_time":1663855553,"op_user_id":"4015857634130296","create_user_id":"4015857634130296","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2},{"tag_id":"7146205183884656670","name":"算法","create_time":1663855553,"update_time":1663855553,"op_user_id":"4015857634130296","create_user_id":"4015857634130296","abstract":"","cover_image":{"key":"","url":"","size":0,"mime_type":"","rid":""},"status":2}],"interact_status":{"is_like":false,"is_collect":false,"is_join":false,"like_count":0,"collect_count":0,"comment_count":0,"join_count":0,"evaluate_count":0,"evaluate_avg":0,"evaluate_self":0},"pool_status":{"is_top":false,"is_good":false,"is_recommend":false},"content_count":{"view_count":79,"child_count":0},"op_info":{"op_user":{},"op_time":0,"op_resource":""}}],"has_more":true,"cursor":"1728360241-84657"}}},"errors":null}</script> <script> const __ENV = 'prod'; function slardarInit(w, d, u, b, n, pc, ga, ae, po, s, p, e, t, pp) { pc = 'precollect'; ga = 'getAttribute'; ae = 'addEventListener'; po = 'PerformanceObserver'; s = function (m) { p = [].slice.call(arguments); p.push(Date.now(), location.href); (m == pc ? s.p.a : s.q).push(p); }; s.q = []; s.p = { a: [] }; w[n] = s; e = document.createElement('script'); e.src = u + '?bid=' + b + '&globalName=' + n; e.crossOrigin = u.indexOf('sdk-web') > 0 ? 'anonymous' : 'use-credentials'; d.getElementsByTagName('head')[0].appendChild(e); if (ae in w) { s.pcErr = function (e) { e = e || w.event; t = e.target || e.srcElement; if (t instanceof Element || t instanceof HTMLElement) { if (t[ga]('integrity')) { w[n](pc, 'sri', t[ga]('href') || t[ga]('src')); } else { w[n](pc, 'st', { tagName: t.tagName, url: t[ga]('href') || t[ga]('src'), }); } } else { w[n](pc, 'err', e.error || e.message); } }; s.pcRej = function (e) { e = e || w.event; w[n](pc, 'err', e.reason || (e.detail && e.detail.reason)); }; w[ae]('error', s.pcErr, true); w[ae]('unhandledrejection', s.pcRej, true); } if ('PerformanceLongTaskTiming' in w) { pp = s.pp = { entries: [] }; pp.observer = new PerformanceObserver(function (l) { pp.entries = pp.entries.concat(l.getEntries()); }); pp.observer.observe({ entryTypes: ['longtask', 'largest-contentful-paint', 'layout-shift'], }); } } slardarInit( window, document, 'https://lf3-short.ibytedapm.com/slardar/fe/sdk-web/browser.cn.js', 'eps_platform_fe_tech', 'Slardar', ); const env = __ENV === 'prod' ? 'production' : 'development'; window.Slardar('init', { bid: 'eps_platform_fe_tech', env, release: '1.0.1.918', // 区分上报版本 }); window.Slardar('start'); </script> <!-- <script src="/sdk/mermaid/mermaid@9.3.0.min.js"></script> <script> window.mermaid.init({ noteMargin: 10 }, '.language-mermaid'); </script> --> <script type="module"> import mermaid from 'https://cdn.jsdelivr.net/npm/mermaid@9/dist/mermaid.esm.min.mjs'; mermaid.init({ noteMargin: 10 }, '.language-mermaid'); </script> <script type="text/javascript" src="https://res2.wx.qq.com/open/js/jweixin-1.6.0.js" ></script> </body></html>

Pages： 1 2 3 4 5 6 7 8 9 10