5ep9lzv 发表于 2024-9-28 19:12:44

腾讯文档AI助手技术架构设计剖析


    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">▼<span style="color: black;">近期</span>直播超级多，</span><span style="color: black;"><strong style="color: blue;">预约</strong></span><span style="color: black;">保你有收获</span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><img src="https://mmbiz.qpic.cn/mmbiz_png/9TPn66HT933Hw2PyQ8c2q9sT9SlSg0Sq06icoAMYHqLsaMKMSSHHGyrapgU20uGa0dwqcyRK2dJ5EdfvzScvhiaw/640?wx_fmt=png&amp;from=appmsg&amp;wxfrom=5&amp;wx_lazy=1&amp;wx_co=1&amp;tp=webp" style="width: 50%; margin-bottom: 20px;"></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;">&nbsp;—1</span></strong></span><span style="color: black;">—</span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;">文档 AI 助手总体技术架构剖析</span></strong></span></strong></span></strong></span></strong></p><strong style="color: blue;"><span style="color: black;">腾讯文档</span></strong><span style="color: black;">（https://docs.qq.com/）相信<span style="color: black;">大众</span>都<span style="color: black;">运用</span>过，在大模型的新时代，腾讯文档<span style="color: black;">亦</span>推出了 AI 大模型助手应用，如下图所示：</span><span style="color: black;"><img src="https://mmbiz.qpic.cn/mmbiz_png/9TPn66HT931zSAIEAFIqa0CII23O1k5044ialJXDmbrcTEuth1L4QGzy7Y4icl2RERBG0KiaoxbHEcia4wrqsvoezw/640?wx_fmt=png&amp;from=appmsg&amp;tp=webp&amp;wxfrom=5&amp;wx_lazy=1&amp;wx_co=1" style="width: 50%; margin-bottom: 20px;"></span><span style="color: black;">腾讯文档的 AI 大模型助手总体架构如下图所示，<span style="color: black;">包含</span>6大模块：AICopilot、AIServer、AIAgent、AIEngine、AIOperation、AIExtension。</span>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><img src="https://mmbiz.qpic.cn/mmbiz_jpg/9TPn66HT931zSAIEAFIqa0CII23O1k50Rw2SgKsicKk8Cj2U1iaI80mB6jDEABPPrKzo8eTxs3y8xaufRKhkjMag/640?wx_fmt=jpeg&amp;from=appmsg&amp;tp=webp&amp;wxfrom=5&amp;wx_lazy=1&amp;wx_co=1" style="width: 50%; margin-bottom: 20px;"></p><strong style="color: blue;"><span style="color: black;">AICopilot 模块</span></strong><span style="color: black;">：<span style="color: black;">供给</span> AI 侧边栏对话功能，负责意图识别、对话管理、缓存及存档等功能。</span><strong style="color: blue;"><span style="color: black;">AIServer 模块</span></strong><span style="color: black;">：<span style="color: black;">供给</span>各类别定制化的浮层助手服务。</span><strong style="color: blue;"><span style="color: black;">AIAgent 模块</span></strong><span style="color: black;">：<span style="color: black;">做为</span> AI 智能代理，集成并<span style="color: black;">供给</span>各类别的文档处理工具，由上层服务调用识别意图后驱动。</span><strong style="color: blue;"><span style="color: black;">AIEngine 模块</span></strong><span style="color: black;">：<span style="color: black;">做为</span>文档 AI 引擎，统一抽象并封装各项 AI 能力（<span style="color: black;">例如</span>：文生文、文生图、语音转写、语音识别、图像识别、嵌入式 AI 等），实现能力间无感切换。</span><strong style="color: blue;"><span style="color: black;">AIOperation 模块</span></strong><span style="color: black;">：负责文档 AI 灰度发布策略、隐私<span style="color: black;">守护</span><span style="color: black;">办法</span>以及运营操作。</span><strong style="color: blue;"><span style="color: black;">AIExtension 模块</span></strong><span style="color: black;">：扩展 AI 服务，支持AI应用落地所需的支持能力，<span style="color: black;">例如</span>：文本搜索、<span style="color: black;">照片</span>搜索、Python 执行环境等。</span>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;">&nbsp;—2</strong></span><span style="color: black;">—</span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;"><span style="color: black;">文档问答场景技术架构剖析</span></strong></p><span style="color: black;">文档<span style="color: black;">制品</span>的关键能力在于有效传达信息，其中，运用 AI 大模型进行信息问答是重要应用场景，尤其针对 Word、PPT、Sheet、思维导图、数据收集表及知识库等多种内容形态的问题解答。</span><span style="color: black;">构建文档 AI 大模型应用的核心挑战在于<span style="color: black;">创立</span><span style="color: black;">基本</span>的问答系统架构。<span style="color: black;">解决</span>这一<span style="color: black;">困难</span>的关键，在于<span style="color: black;">怎样</span>使 AI 大模型<span style="color: black;">精细</span><span style="color: black;">把握</span>并理解各类文档的<span style="color: black;">行业</span>知识内容。</span>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><img src="https://mmbiz.qpic.cn/mmbiz_jpg/9TPn66HT931zSAIEAFIqa0CII23O1k50Zkv4BP62DwNffMOINsGvibec08XOIutLjQrpuD0iagALKXIib5QPeozCQ/640?wx_fmt=jpeg&amp;from=appmsg&amp;tp=webp&amp;wxfrom=5&amp;wx_lazy=1&amp;wx_co=1" style="width: 50%; margin-bottom: 20px;"></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">一般</span>有<strong style="color: blue;">两种<span style="color: black;">处理</span><span style="color: black;">方法</span></strong>
    </p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">：<span style="color: black;">行业</span>知识<span style="color: black;">经过</span>微调（Fine-tuning）记忆到大模型中、<span style="color: black;">经过</span> Prompt 的方式把<span style="color: black;">行业</span>知识即时给到大模型。</p>

    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">用户文档信息本质上是用户个人数据的整合，<span style="color: black;">重点</span>用于个性化服务。<span style="color: black;">因为</span>用户文档常更新且注重时效性，<span style="color: black;">没法</span>每次变更都重新训练模型；<span style="color: black;">同期</span>出于隐私<span style="color: black;">守护</span>原则，用户数据<span style="color: black;">不可</span>用于模型训练。<span style="color: black;">因此呢</span>，针对每位用户单独训练模型的<span style="color: black;">方法</span>并不现实可行。</span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">因此呢</span><span style="color: black;">选择</span><strong style="color: blue;"><span style="color: black;">第<span style="color: black;">二种</span></span></strong><strong style="color: blue;"><span style="color: black;"> RAG <span style="color: black;">加强</span>的<span style="color: black;">方法</span></span></strong>。</span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><img src="https://mmbiz.qpic.cn/mmbiz_jpg/9TPn66HT931zSAIEAFIqa0CII23O1k50qO6P7xYdYsHhib9X5c5mJnZdD95E4V1GM99QXxFCkkD3hQqibIjshk0A/640?wx_fmt=jpeg&amp;from=appmsg&amp;tp=webp&amp;wxfrom=5&amp;wx_lazy=1&amp;wx_co=1" style="width: 50%; margin-bottom: 20px;"></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">RAG 检索<span style="color: black;">加强</span>生成的技术<span style="color: black;">方法</span>由以下<span style="color: black;">模块串联完成</span>：</span></p><strong style="color: blue;"><span style="color: black;">第<span style="color: black;">1、</span>文档加载</span></strong><span style="color: black;">：定义统一的 Document 数据模型，将实现默认典型的数据源加载实现，业务方<span style="color: black;">亦</span><span style="color: black;">能够</span><span style="color: black;">按照</span>接口自定义实现<span style="color: black;">自己</span>所需文档数据源。</span><strong style="color: blue;"><span style="color: black;">第<span style="color: black;">2、</span>文档分片</span></strong><span style="color: black;">：大模型上下文<span style="color: black;">体积</span>有<span style="color: black;">必定</span>限制，需要将<span style="color: black;">海量</span>数据进行分割操作。</span><strong style="color: blue;"><span style="color: black;">第<span style="color: black;">3、</span>文档 Embedding</span></strong><span style="color: black;">：Embedding 过程将对应文本向量化，以<span style="color: black;">供给</span>更好的语义表达。</span><strong style="color: blue;"><span style="color: black;">第<span style="color: black;">4、</span>文档向量存储</span></strong><span style="color: black;">：<span style="color: black;">运用</span>向量数据库存储文档向量数据。</span><strong style="color: blue;"><span style="color: black;">第<span style="color: black;">5、</span>文档召回</span></strong><span style="color: black;">：<span style="color: black;">按照</span>用户输入的问题召回和问题最<span style="color: black;">关联</span>的文档信息。</span><strong style="color: blue;"><span style="color: black;">第<span style="color: black;">6、</span>问题解答</span></strong><span style="color: black;">：<span style="color: black;">按照</span>召回文档资料 + 用户输入问题<span style="color: black;">供给</span>给大模型进行知识问答。</span><span style="color: black;">为<span style="color: black;">处理</span>以下两种场景，在原有架构上规划进行<strong style="color: blue;"><span style="color: black;">进一步的升级</span></strong>。</span><strong style="color: blue;"><span style="color: black;">第<span style="color: black;">1、</span><span style="color: black;">处理</span>元数据问答</span><span style="color: black;">、总结、非总结类问题</span></strong><span style="color: black;">。</span><strong style="color: blue;"><span style="color: black;">第<span style="color: black;">2、</span><span style="color: black;">处理</span><span style="color: black;">触及</span>多模态文档的问答</span></strong><span style="color: black;">。</span>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><img src="https://mmbiz.qpic.cn/mmbiz_jpg/9TPn66HT931zSAIEAFIqa0CII23O1k50a3ACMSOHYxG0mjMzljpPic9TvfMZujubFgicDb2lwJFMwOraM5nGx0CQ/640?wx_fmt=jpeg&amp;from=appmsg&amp;tp=webp&amp;wxfrom=5&amp;wx_lazy=1&amp;wx_co=1" style="width: 50%; margin-bottom: 20px;"></p><span style="color: black;">为了<span style="color: black;">帮忙</span><span style="color: black;">朋友</span>们彻底<span style="color: black;">把握</span>大模型&nbsp;<span style="color: black;">Agent 智能体</span>、知识库、向量数据库、 RAG、知识图谱的<strong style="color: blue;"><span style="color: black;">应用<span style="color: black;">研发</span>、<span style="color: black;">安排</span>、生产化</span></strong>，今天我会开4场直播和<span style="color: black;">朋友</span>们深度剖析，请<span style="color: black;">朋友</span>们点击以下</span><span style="color: black;"><strong style="color: blue;">预约按钮免费预约</strong></span><span style="color: black;">。</span>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">参考：https://mp.weixin.qq.com/s/MNY6647V4hPByNzghyDUfQ</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;">&nbsp;—3</strong></span><span style="color: black;">—</span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;">!送！</span></strong></span><strong style="color: blue;"><span style="color: black;">AI大模型<span style="color: black;">研发</span>直播课程</span></strong></p><span style="color: black;">大模型的技术体系非常<span style="color: black;">繁杂</span>，即使有了知识图谱和学习路线后，快速<span style="color: black;">把握</span>并<span style="color: black;">不易</span>，<span style="color: black;">咱们</span>打造了大模型应用技术的系列直播课程，<span style="color: black;">包含</span>：</span><strong style="color: blue;">通用大模型技术架构原理、大模型 Agent 应用<span style="color: black;">研发</span>、企业私有大模型<span style="color: black;">研发</span>、向量数据库、大模型应用治理、大模型应用行业落地案例</strong><span style="color: black;">等6项核心技能，<span style="color: black;">帮忙</span><span style="color: black;">朋友</span>们快速<span style="color: black;">把握</span> AI 大模型的技能。</span><span style="color: black;"><strong style="color: blue;">&nbsp;

nykek5i 发表于 2024-10-6 14:00:57

软文发布论坛开幕式圆满成功。 http://www.fok120.com
页: [1]
查看完整版本: 腾讯文档AI助手技术架构设计剖析