zengli
/
coding-agent


			
							1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132113311341135113611371138113911401141114211431144114511461147114811491150115111521153115411551156115711581159116011611162116311641165116611671168116911701171117211731174117511761177117811791180118111821183118411851186118711881189119011911192119311941195119611971198119912001201120212031204120512061207120812091210121112121213121412151216121712181219122012211222122312241225122612271228122912301231123212331234123512361237123812391240124112421243124412451246124712481249125012511252125312541255125612571258125912601261126212631264126512661267126812691270127112721273127412751276127712781279128012811282128312841285128612871288128912901291129212931294129512961297129812991300130113021303130413051306130713081309131013111312131313141315131613171318131913201321132213231324132513261327132813291330133113321333133413351336133713381339134013411342134313441345134613471348134913501351135213531354135513561357135813591360136113621363136413651366136713681369137013711372137313741375137613771378137913801381138213831384138513861387138813891390139113921393139413951396139713981399140014011402140314041405140614071408140914101411141214131414141514161417141814191420142114221423142414251426142714281429143014311432143314341435143614371438143914401441144214431444144514461447144814491450145114521453145414551456145714581459146014611462146314641465146614671468146914701471147214731474147514761477147814791480148114821483148414851486148714881489149014911492149314941495149614971498149915001501150215031504150515061507150815091510151115121513151415151516151715181519152015211522152315241525152615271528152915301531153215331534153515361537153815391540154115421543154415451546154715481549155015511552155315541555155615571558155915601561156215631564156515661567156815691570157115721573157415751576157715781579158015811582158315841585158615871588158915901591159215931594159515961597159815991600160116021603160416051606160716081609161016111612161316141615161616171618161916201621162216231624162516261627162816291630163116321633163416351636163716381639164016411642164316441645164616471648164916501651165216531654165516561657165816591660166116621663166416651666166716681669167016711672167316741675167616771678167916801681168216831684168516861687168816891690169116921693169416951696169716981699170017011702170317041705170617071708170917101711171217131714171517161717171817191720172117221723172417251726172717281729173017311732173317341735173617371738173917401741174217431744174517461747174817491750175117521753175417551756175717581759176017611762176317641765176617671768176917701771177217731774177517761777177817791780178117821783178417851786178717881789179017911792179317941795179617971798179918001801180218031804180518061807180818091810181118121813181418151816181718181819182018211822182318241825182618271828182918301831183218331834183518361837183818391840184118421843184418451846184718481849185018511852185318541855185618571858185918601861186218631864186518661867186818691870187118721873187418751876187718781879188018811882188318841885188618871888188918901891189218931894189518961897189818991900190119021903190419051906190719081909191019111912191319141915191619171918191919201921192219231924192519261927192819291930193119321933193419351936193719381939194019411942194319441945194619471948194919501951195219531954195519561957195819591960196119621963196419651966196719681969197019711972197319741975197619771978197919801981198219831984198519861987198819891990199119921993199419951996199719981999200020012002200320042005200620072008200920102011201220132014201520162017201820192020202120222023202420252026202720282029203020312032203320342035203620372038203920402041204220432044204520462047204820492050205120522053205420552056205720582059206020612062206320642065206620672068206920702071207220732074207520762077207820792080208120822083208420852086208720882089209020912092209320942095209620972098209921002101210221032104210521062107210821092110211121122113211421152116211721182119212021212122212321242125212621272128212921302131213221332134213521362137213821392140214121422143214421452146214721482149215021512152215321542155215621572158215921602161216221632164216521662167216821692170217121722173217421752176217721782179218021812182218321842185218621872188218921902191219221932194219521962197219821992200220122022203220422052206220722082209221022112212221322142215221622172218221922202221222222232224222522262227222822292230223122322233223422352236223722382239224022412242224322442245224622472248224922502251225222532254225522562257225822592260226122622263226422652266226722682269227022712272227322742275227622772278227922802281228222832284228522862287228822892290229122922293229422952296229722982299230023012302230323042305230623072308230923102311231223132314231523162317231823192320232123222323232423252326232723282329233023312332233323342335233623372338233923402341234223432344234523462347234823492350235123522353235423552356235723582359236023612362236323642365236623672368236923702371237223732374237523762377237823792380238123822383238423852386238723882389239023912392239323942395239623972398239924002401240224032404240524062407240824092410241124122413241424152416241724182419242024212422242324242425242624272428242924302431243224332434243524362437243824392440244124422443244424452446244724482449245024512452245324542455245624572458245924602461246224632464246524662467246824692470247124722473247424752476247724782479248024812482248324842485248624872488248924902491249224932494249524962497249824992500250125022503250425052506250725082509251025112512251325142515251625172518251925202521252225232524252525262527252825292530253125322533253425352536253725382539254025412542254325442545254625472548254925502551255225532554255525562557255825592560256125622563256425652566256725682569257025712572257325742575257625772578257925802581258225832584258525862587258825892590259125922593259425952596259725982599260026012602260326042605260626072608260926102611261226132614261526162617261826192620262126222623262426252626262726282629263026312632263326342635263626372638263926402641264226432644264526462647264826492650265126522653265426552656265726582659266026612662266326642665266626672668266926702671267226732674267526762677267826792680268126822683268426852686268726882689269026912692269326942695269626972698269927002701270227032704270527062707270827092710271127122713271427152716271727182719272027212722272327242725272627272728272927302731273227332734273527362737273827392740274127422743274427452746274727482749275027512752275327542755275627572758275927602761276227632764276527662767276827692770277127722773277427752776277727782779278027812782278327842785278627872788278927902791279227932794279527962797279827992800280128022803280428052806280728082809281028112812281328142815281628172818281928202821282228232824282528262827282828292830283128322833283428352836283728382839284028412842284328442845284628472848284928502851285228532854285528562857285828592860286128622863286428652866286728682869287028712872287328742875287628772878287928802881288228832884288528862887288828892890289128922893289428952896289728982899290029012902290329042905290629072908290929102911291229132914291529162917291829192920292129222923292429252926292729282929293029312932293329342935293629372938293929402941294229432944294529462947294829492950295129522953295429552956295729582959296029612962296329642965296629672968296929702971297229732974297529762977297829792980298129822983298429852986298729882989299029912992299329942995299629972998299930003001300230033004300530063007300830093010301130123013301430153016301730183019302030213022302330243025302630273028302930303031303230333034303530363037303830393040304130423043304430453046304730483049305030513052305330543055305630573058305930603061306230633064306530663067306830693070307130723073307430753076307730783079308030813082308330843085308630873088308930903091309230933094309530963097309830993100310131023103310431053106310731083109311031113112311331143115311631173118311931203121312231233124312531263127312831293130313131323133313431353136313731383139314031413142314331443145314631473148314931503151315231533154315531563157315831593160316131623163316431653166316731683169317031713172317331743175317631773178317931803181318231833184318531863187318831893190319131923193319431953196319731983199320032013202320332043205320632073208320932103211321232133214321532163217321832193220322132223223322432253226322732283229323032313232323332343235323632373238323932403241324232433244
							
<!doctype html>
<html lang="zh" class="no-js">
  <head>
    
      <meta charset="utf-8">
      <meta name="viewport" content="width=device-width,initial-scale=1">
      
      
        <link rel="prev" href="2.1%20Ollama%E5%A4%A7%E6%A8%A1%E5%9E%8B%E9%AB%98%E6%95%88%E7%AE%A1%E7%90%86%E5%B7%A5%E5%85%B7.html">
      
      
        <link rel="next" href="2.3%20%E5%90%91%E9%87%8F%E7%B4%A2%E5%BC%95%EF%BC%9AHNSW%E3%80%81IVF%E4%B8%8EIVF_PQ.html">
      
      
      <link rel="icon" href="../img/AI.jpg">
      <meta name="generator" content="mkdocs-1.6.1, mkdocs-material-9.6.21">
    
    
        <title>2.2 LangChain基础知识.md - 企业知识库RAG智能问答项目V1.0</title>
      
    
      <link rel="stylesheet" href="../assets/stylesheets/main.2a3383ac.min.css">
      
      
        <link rel="preconnect" href="https://fonts.gstatic.com" crossorigin>
        <link rel="stylesheet" href="https://fonts.googleapis.com/css?family=Roboto:300,300i,400,400i,700,700i%7CRoboto+Mono:400,400i,700,700i&display=fallback">
        <style>:root{--md-text-font:"Roboto";--md-code-font:"Roboto Mono"}</style>
      
    
      <link rel="stylesheet" href="../stylesheets/extra.css">
    
    <script>__md_scope=new URL("..",location),__md_hash=e=>[...e].reduce(((e,_)=>(e<<5)-e+_.charCodeAt(0)),0),__md_get=(e,_=localStorage,t=__md_scope)=>JSON.parse(_.getItem(t.pathname+"."+e)),__md_set=(e,_,t=localStorage,a=__md_scope)=>{try{t.setItem(a.pathname+"."+e,JSON.stringify(_))}catch(e){}}</script>
    
      
  </head>
  
  
    <body dir="ltr">
  
    
    <input class="md-toggle" data-md-toggle="drawer" type="checkbox" id="__drawer" autocomplete="off">
    <input class="md-toggle" data-md-toggle="search" type="checkbox" id="__search" autocomplete="off">
    <label class="md-overlay" for="__drawer"></label>
    <div data-md-component="skip">
      
        
        <a href="#_1" class="md-skip">
          跳转至
        </a>
      
    </div>
    <div data-md-component="announce">
      
    </div>
    
    
<header class="md-header md-header--shadow" data-md-component="header">
  <nav class="md-header__inner md-grid" aria-label="页眉">
    <a href="../index.html" title="企业知识库RAG智能问答项目V1.0" class="md-header__button md-logo" aria-label="企业知识库RAG智能问答项目V1.0" data-md-component="logo">
      
  <img src="../img/AI.jpg" alt="logo">

    </a>
    <label class="md-header__button md-icon" for="__drawer">
      
      <svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24"><path d="M3 6h18v2H3zm0 5h18v2H3zm0 5h18v2H3z"/></svg>
    </label>
    <div class="md-header__title" data-md-component="header-title">
      <div class="md-header__ellipsis">
        <div class="md-header__topic">
          <span class="md-ellipsis">
            企业知识库RAG智能问答项目V1.0
          </span>
        </div>
        <div class="md-header__topic" data-md-component="header-topic">
          <span class="md-ellipsis">
            
              2.2 LangChain基础知识.md
            
          </span>
        </div>
      </div>
    </div>
    
    
      <script>var palette=__md_get("__palette");if(palette&&palette.color){if("(prefers-color-scheme)"===palette.color.media){var media=matchMedia("(prefers-color-scheme: light)"),input=document.querySelector(media.matches?"[data-md-color-media='(prefers-color-scheme: light)']":"[data-md-color-media='(prefers-color-scheme: dark)']");palette.color.media=input.getAttribute("data-md-color-media"),palette.color.scheme=input.getAttribute("data-md-color-scheme"),palette.color.primary=input.getAttribute("data-md-color-primary"),palette.color.accent=input.getAttribute("data-md-color-accent")}for(var[key,value]of Object.entries(palette.color))document.body.setAttribute("data-md-color-"+key,value)}</script>
    
    
        <label class="md-header__button md-icon" for="__search">
          
          <svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24"><path d="M9.5 3A6.5 6.5 0 0 1 16 9.5c0 1.61-.59 3.09-1.56 4.23l.27.27h.79l5 5-1.5 1.5-5-5v-.79l-.27-.27A6.52 6.52 0 0 1 9.5 16 6.5 6.5 0 0 1 3 9.5 6.5 6.5 0 0 1 9.5 3m0 2C7 5 5 7 5 9.5S7 14 9.5 14 14 12 14 9.5 12 5 9.5 5"/></svg>
        </label>
        <div class="md-search" data-md-component="search" role="dialog">
  <label class="md-search__overlay" for="__search"></label>
  <div class="md-search__inner" role="search">
    <form class="md-search__form" name="search">
      <input type="text" class="md-search__input" name="query" aria-label="搜索" placeholder="搜索" autocapitalize="off" autocorrect="off" autocomplete="off" spellcheck="false" data-md-component="search-query" required>
      <label class="md-search__icon md-icon" for="__search">
        
        <svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24"><path d="M9.5 3A6.5 6.5 0 0 1 16 9.5c0 1.61-.59 3.09-1.56 4.23l.27.27h.79l5 5-1.5 1.5-5-5v-.79l-.27-.27A6.52 6.52 0 0 1 9.5 16 6.5 6.5 0 0 1 3 9.5 6.5 6.5 0 0 1 9.5 3m0 2C7 5 5 7 5 9.5S7 14 9.5 14 14 12 14 9.5 12 5 9.5 5"/></svg>
        
        <svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24"><path d="M20 11v2H8l5.5 5.5-1.42 1.42L4.16 12l7.92-7.92L13.5 5.5 8 11z"/></svg>
      </label>
      <nav class="md-search__options" aria-label="查找">
        
        <button type="reset" class="md-search__icon md-icon" title="清空当前内容" aria-label="清空当前内容" tabindex="-1">
          
          <svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24"><path d="M19 6.41 17.59 5 12 10.59 6.41 5 5 6.41 10.59 12 5 17.59 6.41 19 12 13.41 17.59 19 19 17.59 13.41 12z"/></svg>
        </button>
      </nav>
      
    </form>
    <div class="md-search__output">
      <div class="md-search__scrollwrap" tabindex="0" data-md-scrollfix>
        <div class="md-search-result" data-md-component="search-result">
          <div class="md-search-result__meta">
            正在初始化搜索引擎
          </div>
          <ol class="md-search-result__list" role="presentation"></ol>
        </div>
      </div>
    </div>
  </div>
</div>
      
    
  </nav>
  
</header>
    
    <div class="md-container" data-md-component="container">
      
      
      <main class="md-main" data-md-component="main">
        <div class="md-main__inner md-grid">
          
            
              <div class="md-sidebar md-sidebar--primary" data-md-component="sidebar" data-md-type="navigation" >
                <div class="md-sidebar__scrollwrap">
                  <div class="md-sidebar__inner">
                    

<nav class="md-nav md-nav--primary" aria-label="导航栏" data-md-level="0">
  <label class="md-nav__title" for="__drawer">
    <a href="../index.html" title="企业知识库RAG智能问答项目V1.0" class="md-nav__button md-logo" aria-label="企业知识库RAG智能问答项目V1.0" data-md-component="logo">
      
  <img src="../img/AI.jpg" alt="logo">

    </a>
    企业知识库RAG智能问答项目V1.0
  </label>
  
  <ul class="md-nav__list" data-md-scrollfix>
    
      
    <li class="md-nav__item">
      <a href="../index.html" class="md-nav__link">
        
  
  <span class="md-ellipsis">
    知识库内容
    
  </span>
  

      </a>
    </li>
  

    <li class="md-nav__item md-nav__item--nested">
      
        
        <input class="md-nav__toggle md-toggle " type="checkbox" id="__nav_2" >
        
          
          <label class="md-nav__link" for="__nav_2" id="__nav_2_label" tabindex="0">
            
  
  <span class="md-ellipsis">
    ① 项目概述
    
  </span>
  

            <span class="md-nav__icon md-icon"></span>
          </label>
        
        <nav class="md-nav" data-md-level="1" aria-labelledby="__nav_2_label" aria-expanded="false">
          <label class="md-nav__title" for="__nav_2">
            <span class="md-nav__icon md-icon"></span>
            ① 项目概述
          </label>
          <ul class="md-nav__list" data-md-scrollfix>
            
              
    <li class="md-nav__item">
      <a href="../%E2%91%A0%20%E9%A1%B9%E7%9B%AE%E6%A6%82%E8%BF%B0/1.1%20%E9%A1%B9%E7%9B%AE%E6%9E%B6%E6%9E%84.html" class="md-nav__link">
        
  
  <span class="md-ellipsis">
    1.1 项目架构
    
  </span>
  

      </a>
    </li>
  

    <li class="md-nav__item">
      <a href="../%E2%91%A0%20%E9%A1%B9%E7%9B%AE%E6%A6%82%E8%BF%B0/1.2%20%E9%A1%B9%E7%9B%AE%E7%8E%AF%E5%A2%83%28window%29.html" class="md-nav__link">
        
  
  <span class="md-ellipsis">
    1.2 项目环境(window)
    
  </span>
  

      </a>
    </li>
  

    <li class="md-nav__item">
      <a href="../%E2%91%A0%20%E9%A1%B9%E7%9B%AE%E6%A6%82%E8%BF%B0/1.3%20%E9%A1%B9%E7%9B%AE%E7%8E%AF%E5%A2%83%28mac%29.html" class="md-nav__link">
        
  
  <span class="md-ellipsis">
    1.3 项目环境(mac)
    
  </span>
  

      </a>
    </li>
  

          </ul>
        </nav>
      
    </li>
  

    <li class="md-nav__item md-nav__item--active md-nav__item--nested">
      
        
        <input class="md-nav__toggle md-toggle " type="checkbox" id="__nav_3" checked>
        
          
          <label class="md-nav__link" for="__nav_3" id="__nav_3_label" tabindex="0">
            
  
  <span class="md-ellipsis">
    ② 项目工具
    
  </span>
  

            <span class="md-nav__icon md-icon"></span>
          </label>
        
        <nav class="md-nav" data-md-level="1" aria-labelledby="__nav_3_label" aria-expanded="true">
          <label class="md-nav__title" for="__nav_3">
            <span class="md-nav__icon md-icon"></span>
            ② 项目工具
          </label>
          <ul class="md-nav__list" data-md-scrollfix>
              
                
    <li class="md-nav__item">
      <a href="2.1%20Ollama%E5%A4%A7%E6%A8%A1%E5%9E%8B%E9%AB%98%E6%95%88%E7%AE%A1%E7%90%86%E5%B7%A5%E5%85%B7.html" class="md-nav__link">
        
  
  <span class="md-ellipsis">
    2.1 Ollama大模型高效管理工具
    
  </span>
  

      </a>
    </li>
  

    <li class="md-nav__item md-nav__item--active">
      <a href="2.2%20LangChain%E5%9F%BA%E7%A1%80%E7%9F%A5%E8%AF%86.md.html" class="md-nav__link md-nav__link--active">
        
  
  <span class="md-ellipsis">
    2.2 LangChain基础知识.md
    
  </span>
  

      </a>
    </li>
  

    <li class="md-nav__item">
      <a href="2.3%20%E5%90%91%E9%87%8F%E7%B4%A2%E5%BC%95%EF%BC%9AHNSW%E3%80%81IVF%E4%B8%8EIVF_PQ.html" class="md-nav__link">
        
  
  <span class="md-ellipsis">
    2.3 向量索引：HNSW、IVF与IVF_PQ
    
  </span>
  

      </a>
    </li>
  

    <li class="md-nav__item">
      <a href="2.4%20Milvus%E5%90%91%E9%87%8F%E6%95%B0%E6%8D%AE%E5%BA%93.html" class="md-nav__link">
        
  
  <span class="md-ellipsis">
    2.4 Milvus向量数据库
    
  </span>
  

      </a>
    </li>
  

          </ul>
        </nav>
      
    </li>
  

    <li class="md-nav__item md-nav__item--nested">
      
        
        <input class="md-nav__toggle md-toggle " type="checkbox" id="__nav_4" >
        
          
          <label class="md-nav__link" for="__nav_4" id="__nav_4_label" tabindex="0">
            
  
  <span class="md-ellipsis">
    ③ 基于Mysql库的问答系统
    
  </span>
  

            <span class="md-nav__icon md-icon"></span>
          </label>
        
        <nav class="md-nav" data-md-level="1" aria-labelledby="__nav_4_label" aria-expanded="false">
          <label class="md-nav__title" for="__nav_4">
            <span class="md-nav__icon md-icon"></span>
            ③ 基于Mysql库的问答系统
          </label>
          <ul class="md-nav__list" data-md-scrollfix>
            
              
    <li class="md-nav__item">
      <a href="../%E2%91%A2%20%E5%9F%BA%E4%BA%8EMysql%E5%BA%93%E7%9A%84%E9%97%AE%E7%AD%94%E7%B3%BB%E7%BB%9F/3.1%20Python%E6%97%A5%E5%BF%97%E4%BB%8B%E7%BB%8D%E4%B8%8E%E5%BA%94%E7%94%A8.html" class="md-nav__link">
        
  
  <span class="md-ellipsis">
    3.1 Python日志介绍与应用
    
  </span>
  

      </a>
    </li>
  

    <li class="md-nav__item">
      <a href="../%E2%91%A2%20%E5%9F%BA%E4%BA%8EMysql%E5%BA%93%E7%9A%84%E9%97%AE%E7%AD%94%E7%B3%BB%E7%BB%9F/3.2%20Redis%E6%95%B0%E6%8D%AE%E5%BA%93%E7%AE%80%E4%BB%8B%E4%B8%8E%E4%BD%BF%E7%94%A8.html" class="md-nav__link">
        
  
  <span class="md-ellipsis">
    3.2 Redis数据库简介与使用
    
  </span>
  

      </a>
    </li>
  

    <li class="md-nav__item">
      <a href="../%E2%91%A2%20%E5%9F%BA%E4%BA%8EMysql%E5%BA%93%E7%9A%84%E9%97%AE%E7%AD%94%E7%B3%BB%E7%BB%9F/3.3%20%E5%9F%BA%E4%BA%8EMysql%E6%95%B0%E6%8D%AE%E5%BA%93%E5%AE%9E%E7%8E%B0%E9%97%AE%E7%AD%94%E7%B3%BB%E7%BB%9F.html" class="md-nav__link">
        
  
  <span class="md-ellipsis">
    3.3 基于Mysql数据库实现问答系统
    
  </span>
  

      </a>
    </li>
  

    <li class="md-nav__item">
      <a href="../%E2%91%A2%20%E5%9F%BA%E4%BA%8EMysql%E5%BA%93%E7%9A%84%E9%97%AE%E7%AD%94%E7%B3%BB%E7%BB%9F/3.4%20BM25%E7%AE%97%E6%B3%95%E7%AE%80%E4%BB%8B%E4%B8%8E%E5%BA%94%E7%94%A8.html" class="md-nav__link">
        
  
  <span class="md-ellipsis">
    3.4 BM25算法简介与应用
    
  </span>
  

      </a>
    </li>
  

          </ul>
        </nav>
      
    </li>
  

    <li class="md-nav__item md-nav__item--nested">
      
        
        <input class="md-nav__toggle md-toggle " type="checkbox" id="__nav_5" >
        
          
          <label class="md-nav__link" for="__nav_5" id="__nav_5_label" tabindex="0">
            
  
  <span class="md-ellipsis">
    ④ 基于Milvus库的问答系统
    
  </span>
  

            <span class="md-nav__icon md-icon"></span>
          </label>
        
        <nav class="md-nav" data-md-level="1" aria-labelledby="__nav_5_label" aria-expanded="false">
          <label class="md-nav__title" for="__nav_5">
            <span class="md-nav__icon md-icon"></span>
            ④ 基于Milvus库的问答系统
          </label>
          <ul class="md-nav__list" data-md-scrollfix>
            
              
    <li class="md-nav__item">
      <a href="../%E2%91%A3%20%E5%9F%BA%E4%BA%8EMilvus%E5%BA%93%E7%9A%84%E9%97%AE%E7%AD%94%E7%B3%BB%E7%BB%9F/4.1%20%E6%95%B4%E4%BD%93%E6%9E%B6%E6%9E%84%E4%B8%8E%E5%B7%A5%E7%A8%8B%E6%B5%81%E7%A8%8B.html" class="md-nav__link">
        
  
  <span class="md-ellipsis">
    4.1 整体架构与工程流程
    
  </span>
  

      </a>
    </li>
  

    <li class="md-nav__item">
      <a href="../%E2%91%A3%20%E5%9F%BA%E4%BA%8EMilvus%E5%BA%93%E7%9A%84%E9%97%AE%E7%AD%94%E7%B3%BB%E7%BB%9F/4.2%20%E5%9F%BA%E7%A1%80%E6%A8%A1%E5%9D%97.html" class="md-nav__link">
        
  
  <span class="md-ellipsis">
    4.2 基础模块
    
  </span>
  

      </a>
    </li>
  

    <li class="md-nav__item">
      <a href="../%E2%91%A3%20%E5%9F%BA%E4%BA%8EMilvus%E5%BA%93%E7%9A%84%E9%97%AE%E7%AD%94%E7%B3%BB%E7%BB%9F/4.3%20%E6%96%87%E6%A1%A3%E5%A4%84%E7%90%86%E6%A8%A1%E5%9D%97.html" class="md-nav__link">
        
  
  <span class="md-ellipsis">
    4.3 文档处理模块
    
  </span>
  

      </a>
    </li>
  

    <li class="md-nav__item">
      <a href="../%E2%91%A3%20%E5%9F%BA%E4%BA%8EMilvus%E5%BA%93%E7%9A%84%E9%97%AE%E7%AD%94%E7%B3%BB%E7%BB%9F/4.4%20%E5%90%91%E9%87%8F%E5%AD%98%E5%82%A8.html" class="md-nav__link">
        
  
  <span class="md-ellipsis">
    4.4 向量存储
    
  </span>
  

      </a>
    </li>
  

    <li class="md-nav__item">
      <a href="../%E2%91%A3%20%E5%9F%BA%E4%BA%8EMilvus%E5%BA%93%E7%9A%84%E9%97%AE%E7%AD%94%E7%B3%BB%E7%BB%9F/4.5%20prompts%E8%AE%BE%E8%AE%A1%E4%B8%8Equery%E6%84%8F%E5%9B%BE%E8%AF%86%E5%88%AB.html" class="md-nav__link">
        
  
  <span class="md-ellipsis">
    4.5 prompts设计与query意图识别
    
  </span>
  

      </a>
    </li>
  

    <li class="md-nav__item">
      <a href="../%E2%91%A3%20%E5%9F%BA%E4%BA%8EMilvus%E5%BA%93%E7%9A%84%E9%97%AE%E7%AD%94%E7%B3%BB%E7%BB%9F/4.6%20%E6%A3%80%E7%B4%A2%E7%AD%96%E7%95%A5%E4%B8%8ERAG%E7%B3%BB%E7%BB%9F%E8%AE%BE%E8%AE%A1.html" class="md-nav__link">
        
  
  <span class="md-ellipsis">
    4.6 检索策略与RAG系统设计
    
  </span>
  

      </a>
    </li>
  

    <li class="md-nav__item">
      <a href="../%E2%91%A3%20%E5%9F%BA%E4%BA%8EMilvus%E5%BA%93%E7%9A%84%E9%97%AE%E7%AD%94%E7%B3%BB%E7%BB%9F/4.7%20RAG%E7%B3%BB%E7%BB%9F%E8%BF%90%E8%A1%8C.html" class="md-nav__link">
        
  
  <span class="md-ellipsis">
    4.7 RAG系统运行
    
  </span>
  

      </a>
    </li>
  

    <li class="md-nav__item">
      <a href="../%E2%91%A3%20%E5%9F%BA%E4%BA%8EMilvus%E5%BA%93%E7%9A%84%E9%97%AE%E7%AD%94%E7%B3%BB%E7%BB%9F/4.8%20%E6%96%87%E6%A1%A3%E8%A7%A3%E6%9E%90%E5%B7%A5%E5%85%B7%28%E6%89%A9%E5%B1%95%E8%B5%84%E6%96%99%29.html" class="md-nav__link">
        
  
  <span class="md-ellipsis">
    4.8 文档解析工具(扩展资料)
    
  </span>
  

      </a>
    </li>
  

    <li class="md-nav__item">
      <a href="../%E2%91%A3%20%E5%9F%BA%E4%BA%8EMilvus%E5%BA%93%E7%9A%84%E9%97%AE%E7%AD%94%E7%B3%BB%E7%BB%9F/4.9%20RAG%E4%B8%AD%E7%9A%84Query%E6%94%B9%E5%86%99%28%E6%89%A9%E5%B1%95%E8%B5%84%E6%96%99%29.html" class="md-nav__link">
        
  
  <span class="md-ellipsis">
    4.9 RAG中的Query改写(扩展资料)
    
  </span>
  

      </a>
    </li>
  

    <li class="md-nav__item">
      <a href="../%E2%91%A3%20%E5%9F%BA%E4%BA%8EMilvus%E5%BA%93%E7%9A%84%E9%97%AE%E7%AD%94%E7%B3%BB%E7%BB%9F/%E6%95%B0%E6%8D%AE%E9%9B%86%E7%94%9F%E6%88%90%E4%B8%8E%E4%BC%98%E5%8C%96%28%E6%89%A9%E5%B1%95%E8%B5%84%E6%96%99%29.html" class="md-nav__link">
        
  
  <span class="md-ellipsis">
    数据集生成与优化(扩展资料)
    
  </span>
  

      </a>
    </li>
  

          </ul>
        </nav>
      
    </li>
  

    <li class="md-nav__item md-nav__item--nested">
      
        
        <input class="md-nav__toggle md-toggle " type="checkbox" id="__nav_6" >
        
          
          <label class="md-nav__link" for="__nav_6" id="__nav_6_label" tabindex="0">
            
  
  <span class="md-ellipsis">
    ⑤ RAG系统评估
    
  </span>
  

            <span class="md-nav__icon md-icon"></span>
          </label>
        
        <nav class="md-nav" data-md-level="1" aria-labelledby="__nav_6_label" aria-expanded="false">
          <label class="md-nav__title" for="__nav_6">
            <span class="md-nav__icon md-icon"></span>
            ⑤ RAG系统评估
          </label>
          <ul class="md-nav__list" data-md-scrollfix>
            
              
    <li class="md-nav__item">
      <a href="../%E2%91%A4%20RAG%E7%B3%BB%E7%BB%9F%E8%AF%84%E4%BC%B0/5.1%20RAG%E7%B3%BB%E7%BB%9F%E8%AF%84%E4%BC%B0%E5%B7%A5%E5%85%B7.html" class="md-nav__link">
        
  
  <span class="md-ellipsis">
    5.1 RAG系统评估工具
    
  </span>
  

      </a>
    </li>
  

    <li class="md-nav__item">
      <a href="../%E2%91%A4%20RAG%E7%B3%BB%E7%BB%9F%E8%AF%84%E4%BC%B0/5.2%20%E5%AE%9E%E7%8E%B0EnterpriseKB-RAG%E7%B3%BB%E7%BB%9F%E7%9A%84%E8%AF%84%E4%BC%B0.html" class="md-nav__link">
        
  
  <span class="md-ellipsis">
    5.2 实现企业知识库RAG系统的评估
    
  </span>
  

      </a>
    </li>
  

    <li class="md-nav__item">
      <a href="../%E2%91%A4%20RAG%E7%B3%BB%E7%BB%9F%E8%AF%84%E4%BC%B0/5.3%20RAG%E8%AF%84%E4%BC%B0%E6%95%B0%E6%8D%AE%E9%9B%86%E7%9A%84%E6%9E%84%E9%80%A0%28%E6%89%A9%E5%B1%95%29.html" class="md-nav__link">
        
  
  <span class="md-ellipsis">
    5.3 RAG评估数据集的构造(扩展)
    
  </span>
  

      </a>
    </li>
  

          </ul>
        </nav>
      
    </li>
  

    <li class="md-nav__item md-nav__item--nested">
      
        
        <input class="md-nav__toggle md-toggle " type="checkbox" id="__nav_7" >
        
          
          <label class="md-nav__link" for="__nav_7" id="__nav_7_label" tabindex="0">
            
  
  <span class="md-ellipsis">
    ⑥ 融合Mysql的RAG系统
    
  </span>
  

            <span class="md-nav__icon md-icon"></span>
          </label>
        
        <nav class="md-nav" data-md-level="1" aria-labelledby="__nav_7_label" aria-expanded="false">
          <label class="md-nav__title" for="__nav_7">
            <span class="md-nav__icon md-icon"></span>
            ⑥ 融合Mysql的RAG系统
          </label>
          <ul class="md-nav__list" data-md-scrollfix>
            
              
    <li class="md-nav__item">
      <a href="../%E2%91%A5%20%E8%9E%8D%E5%90%88Mysql%E7%9A%84RAG%E7%B3%BB%E7%BB%9F/6.1%20%E8%9E%8D%E5%90%88FQA%E5%92%8C%E7%9F%A5%E8%AF%86%E5%BA%93%E6%9F%A5%E8%AF%A2.html" class="md-nav__link">
        
  
  <span class="md-ellipsis">
    6.1 融合FQA和知识库查询
    
  </span>
  

      </a>
    </li>
  

    <li class="md-nav__item">
      <a href="../%E2%91%A5%20%E8%9E%8D%E5%90%88Mysql%E7%9A%84RAG%E7%B3%BB%E7%BB%9F/6.2%20%E8%9E%8D%E5%90%88FAQ%20%E5%92%8C%E7%9F%A5%E8%AF%86%E5%BA%93%E6%9F%A5%E8%AF%A2%EF%BC%88%E4%BC%98%E5%8C%96%E7%89%88%EF%BC%89.html" class="md-nav__link">
        
  
  <span class="md-ellipsis">
    6.2 融合FAQ 和知识库查询（优化版）
    
  </span>
  

      </a>
    </li>
  

    <li class="md-nav__item">
      <a href="../%E2%91%A5%20%E8%9E%8D%E5%90%88Mysql%E7%9A%84RAG%E7%B3%BB%E7%BB%9F/6.3%20%E5%9F%BA%E4%BA%8E%20FastAPI%20%E7%9A%84%20FAQ%20%E4%B8%8E%E7%9F%A5%E8%AF%86%E5%BA%93%E6%9F%A5%E8%AF%A2%E6%8E%A5%E5%8F%A3.html" class="md-nav__link">
        
  
  <span class="md-ellipsis">
    6.3 基于 FastAPI 的 FAQ 与知识库查询接口
    
  </span>
  

      </a>
    </li>
  

    <li class="md-nav__item">
      <a href="../%E2%91%A5%20%E8%9E%8D%E5%90%88Mysql%E7%9A%84RAG%E7%B3%BB%E7%BB%9F/6.4%20FastAPI%E5%92%8CFlask%E7%9A%84%E5%8C%BA%E5%88%AB%E5%92%8C%E8%81%94%E7%B3%BB%28%E6%89%A9%E5%B1%95%E8%B5%84%E6%96%99%29.html" class="md-nav__link">
        
  
  <span class="md-ellipsis">
    6.4 FastAPI和Flask的区别和联系(扩展资料)
    
  </span>
  

      </a>
    </li>
  

          </ul>
        </nav>
      
    </li>
  

    <li class="md-nav__item md-nav__item--nested">
      
        
        <input class="md-nav__toggle md-toggle " type="checkbox" id="__nav_8" >
        
          
          <label class="md-nav__link" for="__nav_8" id="__nav_8_label" tabindex="0">
            
  
  <span class="md-ellipsis">
    ⑦ 企业级生产环境部署
    
  </span>
  

            <span class="md-nav__icon md-icon"></span>
          </label>
        
        <nav class="md-nav" data-md-level="1" aria-labelledby="__nav_8_label" aria-expanded="false">
          <label class="md-nav__title" for="__nav_8">
            <span class="md-nav__icon md-icon"></span>
            ⑦ 企业级生产环境部署
          </label>
          <ul class="md-nav__list" data-md-scrollfix>
            
              
    <li class="md-nav__item">
      <a href="../%E2%91%A6%20%E4%BC%81%E4%B8%9A%E7%BA%A7%E7%94%9F%E4%BA%A7%E7%8E%AF%E5%A2%83%E9%83%A8%E7%BD%B2/7.1%20Docker%E7%9A%84%E5%8E%9F%E7%90%86%E5%92%8C%E5%9F%BA%E6%9C%AC%E4%BD%BF%E7%94%A8.html" class="md-nav__link">
        
  
  <span class="md-ellipsis">
    7.1 Docker的原理和基本使用
    
  </span>
  

      </a>
    </li>
  

    <li class="md-nav__item">
      <a href="../%E2%91%A6%20%E4%BC%81%E4%B8%9A%E7%BA%A7%E7%94%9F%E4%BA%A7%E7%8E%AF%E5%A2%83%E9%83%A8%E7%BD%B2/7.2%20%E6%9C%8D%E5%8A%A1%E9%83%A8%E7%BD%B2.html" class="md-nav__link">
        
  
  <span class="md-ellipsis">
    7.2 服务部署
    
  </span>
  

      </a>
    </li>
  

          </ul>
        </nav>
      
    </li>
  

  </ul>
</nav>
                  </div>
                </div>
              </div>
            
            
              <div class="md-sidebar md-sidebar--secondary" data-md-component="sidebar" data-md-type="toc" >
                <div class="md-sidebar__scrollwrap">
                  <div class="md-sidebar__inner">
                    

<nav class="md-nav md-nav--secondary" aria-label="目录">
  
  
    <label class="md-nav__title" for="__toc">
      <span class="md-nav__icon md-icon"></span>
      目录
    </label>
    <ul class="md-nav__list" data-md-component="toc" data-md-scrollfix>
      
        <li class="md-nav__item">
  <a href="#_1" class="md-nav__link">
    <span class="md-ellipsis">
      学习目标
    </span>
  </a>
  
</li>
      
        <li class="md-nav__item">
  <a href="#_2" class="md-nav__link">
    <span class="md-ellipsis">
      一、简述
    </span>
  </a>
  
    <nav class="md-nav" aria-label="一、简述">
      <ul class="md-nav__list">
        
          <li class="md-nav__item">
  <a href="#1-langchain" class="md-nav__link">
    <span class="md-ellipsis">
      1 什么是LangChain
    </span>
  </a>
  
</li>
        
          <li class="md-nav__item">
  <a href="#2" class="md-nav__link">
    <span class="md-ellipsis">
      2 主要组件
    </span>
  </a>
  
</li>
        
          <li class="md-nav__item">
  <a href="#3-langchain" class="md-nav__link">
    <span class="md-ellipsis">
      3 LangChain核心包
    </span>
  </a>
  
</li>
        
          <li class="md-nav__item">
  <a href="#4" class="md-nav__link">
    <span class="md-ellipsis">
      4 环境准备
    </span>
  </a>
  
</li>
        
      </ul>
    </nav>
  
</li>
      
        <li class="md-nav__item">
  <a href="#models" class="md-nav__link">
    <span class="md-ellipsis">
      二、Models
    </span>
  </a>
  
    <nav class="md-nav" aria-label="二、Models">
      <ul class="md-nav__list">
        
          <li class="md-nav__item">
  <a href="#1-llms" class="md-nav__link">
    <span class="md-ellipsis">
      1 LLMs
    </span>
  </a>
  
</li>
        
          <li class="md-nav__item">
  <a href="#2-chat-models" class="md-nav__link">
    <span class="md-ellipsis">
      2 Chat Models
    </span>
  </a>
  
</li>
        
          <li class="md-nav__item">
  <a href="#3-embeddings-models" class="md-nav__link">
    <span class="md-ellipsis">
      3 Embeddings Models
    </span>
  </a>
  
</li>
        
      </ul>
    </nav>
  
</li>
      
        <li class="md-nav__item">
  <a href="#prompts" class="md-nav__link">
    <span class="md-ellipsis">
      三、Prompts
    </span>
  </a>
  
    <nav class="md-nav" aria-label="三、Prompts">
      <ul class="md-nav__list">
        
          <li class="md-nav__item">
  <a href="#1-prompt" class="md-nav__link">
    <span class="md-ellipsis">
      1 通用prompt
    </span>
  </a>
  
</li>
        
          <li class="md-nav__item">
  <a href="#2-chatprompts" class="md-nav__link">
    <span class="md-ellipsis">
      2 ChatPrompts
    </span>
  </a>
  
</li>
        
      </ul>
    </nav>
  
</li>
      
        <li class="md-nav__item">
  <a href="#chains" class="md-nav__link">
    <span class="md-ellipsis">
      四、Chains
    </span>
  </a>
  
</li>
      
        <li class="md-nav__item">
  <a href="#output_parsers" class="md-nav__link">
    <span class="md-ellipsis">
      五、output_parsers
    </span>
  </a>
  
    <nav class="md-nav" aria-label="五、output_parsers">
      <ul class="md-nav__list">
        
          <li class="md-nav__item">
  <a href="#1" class="md-nav__link">
    <span class="md-ellipsis">
      1 字符串解析器
    </span>
  </a>
  
</li>
        
          <li class="md-nav__item">
  <a href="#2_1" class="md-nav__link">
    <span class="md-ellipsis">
      2 列表解析器
    </span>
  </a>
  
</li>
        
          <li class="md-nav__item">
  <a href="#3-json" class="md-nav__link">
    <span class="md-ellipsis">
      3 JSON解析器
    </span>
  </a>
  
</li>
        
          <li class="md-nav__item">
  <a href="#4-pydantic" class="md-nav__link">
    <span class="md-ellipsis">
      4 Pydantic解析器
    </span>
  </a>
  
</li>
        
          <li class="md-nav__item">
  <a href="#5" class="md-nav__link">
    <span class="md-ellipsis">
      5 自定义解析器
    </span>
  </a>
  
</li>
        
      </ul>
    </nav>
  
</li>
      
        <li class="md-nav__item">
  <a href="#memory" class="md-nav__link">
    <span class="md-ellipsis">
      六、Memory
    </span>
  </a>
  
    <nav class="md-nav" aria-label="六、Memory">
      <ul class="md-nav__list">
        
          <li class="md-nav__item">
  <a href="#1-short-term-memory" class="md-nav__link">
    <span class="md-ellipsis">
      1 短期记忆（Short-term Memory）
    </span>
  </a>
  
</li>
        
          <li class="md-nav__item">
  <a href="#2_2" class="md-nav__link">
    <span class="md-ellipsis">
      2 长期记忆
    </span>
  </a>
  
</li>
        
      </ul>
    </nav>
  
</li>
      
        <li class="md-nav__item">
  <a href="#indexes" class="md-nav__link">
    <span class="md-ellipsis">
      七、Indexes
    </span>
  </a>
  
    <nav class="md-nav" aria-label="七、Indexes">
      <ul class="md-nav__list">
        
          <li class="md-nav__item">
  <a href="#1_1" class="md-nav__link">
    <span class="md-ellipsis">
      1 文档加载器
    </span>
  </a>
  
</li>
        
          <li class="md-nav__item">
  <a href="#2_3" class="md-nav__link">
    <span class="md-ellipsis">
      2 文档分割器
    </span>
  </a>
  
</li>
        
          <li class="md-nav__item">
  <a href="#3-vectorstores" class="md-nav__link">
    <span class="md-ellipsis">
      3 VectorStores
    </span>
  </a>
  
</li>
        
          <li class="md-nav__item">
  <a href="#4_1" class="md-nav__link">
    <span class="md-ellipsis">
      4 检索器
    </span>
  </a>
  
    <nav class="md-nav" aria-label="4 检索器">
      <ul class="md-nav__list">
        
          <li class="md-nav__item">
  <a href="#41-langchain" class="md-nav__link">
    <span class="md-ellipsis">
      4.1 LangChain的检索器定义
    </span>
  </a>
  
</li>
        
          <li class="md-nav__item">
  <a href="#42" class="md-nav__link">
    <span class="md-ellipsis">
      4.2 检索器的工作原理
    </span>
  </a>
  
</li>
        
          <li class="md-nav__item">
  <a href="#43" class="md-nav__link">
    <span class="md-ellipsis">
      4.3 检索器类型
    </span>
  </a>
  
</li>
        
          <li class="md-nav__item">
  <a href="#44" class="md-nav__link">
    <span class="md-ellipsis">
      4.4 拓展
    </span>
  </a>
  
</li>
        
      </ul>
    </nav>
  
</li>
        
      </ul>
    </nav>
  
</li>
      
        <li class="md-nav__item">
  <a href="#agents" class="md-nav__link">
    <span class="md-ellipsis">
      八、Agents
    </span>
  </a>
  
    <nav class="md-nav" aria-label="八、Agents">
      <ul class="md-nav__list">
        
          <li class="md-nav__item">
  <a href="#1-agent" class="md-nav__link">
    <span class="md-ellipsis">
      1 Agent 概念
    </span>
  </a>
  
</li>
        
          <li class="md-nav__item">
  <a href="#2-agent" class="md-nav__link">
    <span class="md-ellipsis">
      2 Agent关键组成部分
    </span>
  </a>
  
</li>
        
          <li class="md-nav__item">
  <a href="#3-langchainagent" class="md-nav__link">
    <span class="md-ellipsis">
      3 langchain实现Agent
    </span>
  </a>
  
</li>
        
      </ul>
    </nav>
  
</li>
      
        <li class="md-nav__item">
  <a href="#langchain" class="md-nav__link">
    <span class="md-ellipsis">
      九、LangChain使用场景
    </span>
  </a>
  
</li>
      
        <li class="md-nav__item">
  <a href="#_3" class="md-nav__link">
    <span class="md-ellipsis">
      十、本节小结
    </span>
  </a>
  
</li>
      
        <li class="md-nav__item">
  <a href="#_4" class="md-nav__link">
    <span class="md-ellipsis">
      十一、本节作业
    </span>
  </a>
  
    <nav class="md-nav" aria-label="十一、本节作业">
      <ul class="md-nav__list">
        
          <li class="md-nav__item">
  <a href="#1-1" class="md-nav__link">
    <span class="md-ellipsis">
      1 作业1
    </span>
  </a>
  
</li>
        
          <li class="md-nav__item">
  <a href="#2-2" class="md-nav__link">
    <span class="md-ellipsis">
      2 作业2
    </span>
  </a>
  
</li>
        
      </ul>
    </nav>
  
</li>
      
    </ul>
  
</nav>
                  </div>
                </div>
              </div>
            
          
            <div class="md-content" data-md-component="content">
              <article class="md-content__inner md-typeset">
                
                  
  <h1>2.2 LangChain基础知识.md</h1>

<h2 id="_1">学习目标<a class="headerlink" href="#_1" title="Permanent link">&para;</a></h2>
<ul>
<li>熟悉什么是LangChain及其作用</li>
<li>熟悉LangChain中基本组件及使用</li>
</ul>
<h2 id="_2">一、简述<a class="headerlink" href="#_2" title="Permanent link">&para;</a></h2>
<h3 id="1-langchain">1 什么是LangChain<a class="headerlink" href="#1-langchain" title="Permanent link">&para;</a></h3>
<p>LangChain由 Harrison Chase 创建于2022年10月，它是围绕LLMs（大语言模型）建立的一个框架，它的核心理念是为各种LLMs实现通用的接口，把LLMs相关的组件“链接”在一起，简化LLMs应用的开发难度，方便开发者快速地开发复杂的LLMs应用。</p>
<blockquote>
<p>LangChain 最初是于 2022 年 10 月开源的一个项目，在 GitHub 上获得大量关注之后迅速转变为一家初创公司。2017 年 Harrison Chase 还在哈佛上大学，如今已是硅谷的一家热门初创公司的 CEO，这对他来说是一次重大而迅速的跃迁。Insider 独家报道，人工智能初创公司 LangChain 在种子轮一周后，再次获得红杉领投的 2000 万至 2500 万美元融资，估值达到 2 亿美元。</p>
</blockquote>
<p><strong>LangChain官网</strong>：<a href="https://python.langchain.com/docs/introduction/">https://python.langchain.com/docs/introduction/</a></p>
<p><strong>LangChain官网中文版</strong>：<a href="https://www.langchain.com.cn/">https://www.langchain.com.cn/</a></p>
<p>LangChain 简化了 LLM 应用程序生命周期的每个阶段：</p>
<ul>
<li>开发：使用 LangChain 的开源组件和第三方集成构建应用程序。使用 LangGraph 构建具有一流流式和人机交互支持的状态智能体。</li>
<li>生产化：使用 LangSmith 检查、监控和评估应用程序，方便持续优化和部署。</li>
<li>部署：使用 LangGraph平台 将应用程序转变为可用于生产的 API 和助手。</li>
</ul>
<h3 id="2">2 主要组件<a class="headerlink" href="#2" title="Permanent link">&para;</a></h3>
<p><img alt="image-20250703152303288" src="assets/image-20250703152303288.png" /></p>
<ul>
<li>Models：模型，各种类型的模型和模型集成，比如GPT-4</li>
<li>Prompts：提示，包括提示管理、提示优化和提示序列化</li>
<li>Chains：链，一系列对各种组件的调用</li>
<li>Memory：记忆，用来保存和模型交互时的上下文状态</li>
<li>Indexes：索引，用来结构化文档，以便和模型交互</li>
<li>Agents：代理，决定模型采取哪些行动，执行并且观察流程，直到完成为止</li>
</ul>
<h3 id="3-langchain">3 LangChain核心包<a class="headerlink" href="#3-langchain" title="Permanent link">&para;</a></h3>
<ul>
<li><strong>langchain-core</strong>：聊天模型和其他组件的基础抽象。</li>
<li>
<p><strong>集成包（例如 langchain-openai、langchain-anthropic 等）</strong>：重要的集成被拆分为轻量级的独立包，由 LangChain 团队和集成方共同维护。</p>
</li>
<li>
<p><strong>langchain</strong>：包含链（chains）、智能体（agents）和检索策略，这些构成了应用的认知架构。</p>
</li>
<li>
<p><strong>langchain-community</strong>：由社区维护的第三方集成。</p>
</li>
<li>
<p><strong>langgraph</strong>：一个编排框架，用于将 LangChain 组件组合成可用于生产的应用，支持持久化、流式处理及其他关键特性。</p>
</li>
</ul>
<h3 id="4">4 环境准备<a class="headerlink" href="#4" title="Permanent link">&para;</a></h3>
<p>本知识文档以LangChain+Qwen进行学习，需要提前安装</p>
<blockquote>
<p>pip install openai</p>
<p>pip install langchain </p>
<p>pip install modelscope</p>
</blockquote>
<p>借助阿里云-百炼平台（需要申请API Key 以及Secret Key）：</p>
<blockquote>
<p><a href="https://bailian.console.aliyun.com/#/home">https://bailian.console.aliyun.com/#/home</a></p>
</blockquote>
<h2 id="models">二、Models<a class="headerlink" href="#models" title="Permanent link">&para;</a></h2>
<p>现在市面上的模型多如牛毛，各种各样的模型不断出现，LangChain模型组件提供了与各种模型的集成，并为所有模型提供一个精简的统一接口。</p>
<p>LangChain目前支持三种类型的模型：LLMs、Chat Models(聊天模型)、Embeddings Models(嵌入模型）。</p>
<ul>
<li>LLMs：大语言模型接收文本字符作为输入，返回的也是文本字符。</li>
<li>聊天模型：基于LLMs, 不同的是它接收聊天消(一种特定格式的数据)作为输入，返回的也是聊天消息。</li>
<li>文本嵌入模型：文本嵌入模型接收文本作为输入, 返回的是浮点数列表。</li>
</ul>
<p><img alt="image-20250704171332127" src="assets/image-20250704171332127.png" /></p>
<p>LangChain支持的三类模型，它们的使用场景不同，输入和输出不同，开发者需要根据项目需要选择相应的模型。</p>
<h3 id="1-llms">1 LLMs<a class="headerlink" href="#1-llms" title="Permanent link">&para;</a></h3>
<p>LLMs（大语言模型）使用场景最多，常用大模型的下载库：</p>
<blockquote>
<p><a href="https://huggingface.co/models">https://huggingface.co/models</a></p>
<p><a href="https://modelscope.cn/models">https://modelscope.cn/models</a></p>
</blockquote>
<p>下面Qwen为例进行讲解。</p>
<hr />
<p>模型调用有2种方式：</p>
<ul>
<li>通过ChatOpenAI进行调用</li>
</ul>
<div class="highlight"><pre><span></span><code><span class="kn">from</span><span class="w"> </span><span class="nn">langchain_openai</span><span class="w"> </span><span class="kn">import</span> <span class="n">ChatOpenAI</span>
<span class="kn">import</span><span class="w">  </span><span class="nn">os</span>

<span class="c1"># 实例化模型</span>
<span class="n">model</span> <span class="o">=</span> <span class="n">ChatOpenAI</span><span class="p">(</span>
    <span class="n">base_url</span><span class="o">=</span><span class="n">os</span><span class="o">.</span><span class="n">environ</span><span class="o">.</span><span class="n">get</span><span class="p">(</span><span class="s1">&#39;base_url&#39;</span><span class="p">,</span><span class="s1">&#39;https://dashscope.aliyuncs.com/compatible-mode/v1&#39;</span><span class="p">),</span>
    <span class="n">model</span><span class="o">=</span><span class="s1">&#39;qwen-plus&#39;</span><span class="p">,</span>
    <span class="n">openai_api_key</span><span class="o">=</span><span class="n">os</span><span class="o">.</span><span class="n">environ</span><span class="o">.</span><span class="n">get</span><span class="p">(</span><span class="s1">&#39;api_key&#39;</span><span class="p">,</span><span class="s1">&#39;sk-XXXXXX&#39;</span><span class="p">),</span>
    <span class="n">max_tokens</span><span class="o">=</span><span class="mi">1000</span><span class="p">,</span>
    <span class="n">temperature</span><span class="o">=</span><span class="mi">0</span>
<span class="p">)</span>

<span class="c1"># 获取问答结果</span>
<span class="n">result</span> <span class="o">=</span> <span class="n">model</span><span class="o">.</span><span class="n">invoke</span><span class="p">(</span><span class="s2">&quot;帮我讲个笑话吧&quot;</span><span class="p">)</span>
<span class="nb">print</span><span class="p">(</span><span class="n">result</span><span class="o">.</span><span class="n">content</span><span class="p">)</span>
</code></pre></div>
<ul>
<li>通过Ollama进行调用 <a href="https://ollama.com/">Ollama支持模型</a></li>
</ul>
<div class="highlight"><pre><span></span><code><span class="c1"># from langchain_community.llms import Ollama # Langchain 0.x版本使用</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">langchain_ollama</span><span class="w"> </span><span class="kn">import</span> <span class="n">OllamaLLM</span>

<span class="c1"># 实例化模型</span>
<span class="c1"># model = Ollama(model=&quot;qwen2.5:7b&quot;) # Langchain 0.x版本使用</span>
<span class="n">model</span> <span class="o">=</span> <span class="n">OllamaLLM</span><span class="p">(</span><span class="n">model</span><span class="o">=</span><span class="s2">&quot;qwen2.5:7b&quot;</span><span class="p">)</span>

<span class="c1"># 获取问答结果</span>
<span class="n">result</span> <span class="o">=</span> <span class="n">model</span><span class="o">.</span><span class="n">invoke</span><span class="p">(</span><span class="s2">&quot;请给我讲个笑话吧&quot;</span><span class="p">)</span>
<span class="nb">print</span><span class="p">(</span><span class="n">result</span><span class="p">)</span>
</code></pre></div>
<h3 id="2-chat-models">2 Chat Models<a class="headerlink" href="#2-chat-models" title="Permanent link">&para;</a></h3>
<p>聊天模型，聊天消息包含下面几种类型，使用时需要按照约定传入合适的值：</p>
<ul>
<li>
<p><strong>AIMessage</strong>: 就是 AI 输出的消息，可以是针对问题的回答.</p>
</li>
<li>
<p><strong>HumanMessage</strong>: 人类消息就是用户信息，由人给出的信息发送给LLMs的提示信息，比如“实现一个快速排序方法”.</p>
</li>
<li>
<p><strong>SystemMessage</strong>: 可以用于指定模型具体所处的环境和背景，如角色扮演等。你可以在这里给出具体的指示，比如“作为一个代码专家”，或者“返回json格式”.</p>
</li>
<li>
<p><strong>ChatMessage</strong>: Chat 消息可以接受任意角色的参数，但是在大多数时间，我们应该使用上面的三种类型.</p>
</li>
</ul>
<p>LangChain支持大量的chat 模型,可以通过官网查询：</p>
<blockquote>
<p><a href="https://python.langchain.com/docs/integrations/chat/">https://python.langchain.com/docs/integrations/chat/</a></p>
</blockquote>
<p>也可以通过langchain源码查看</p>
<p><img alt="image-20250704174858064" src="assets/image-20250704174858064.png" /></p>
<ul>
<li>SystemMessage+HumanMessage+AIMessage</li>
</ul>
<div class="highlight"><pre><span></span><code><span class="kn">from</span><span class="w"> </span><span class="nn">langchain_core.messages</span><span class="w"> </span><span class="kn">import</span> <span class="n">HumanMessage</span><span class="p">,</span> <span class="n">SystemMessage</span><span class="p">,</span> <span class="n">AIMessage</span>
<span class="c1"># from langchain_community.chat_models import ChatOllama # Langchain 0.x版本使用</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">langchain_ollama</span><span class="w"> </span><span class="kn">import</span> <span class="n">ChatOllama</span>

<span class="c1"># 实例化模型</span>
<span class="n">model</span> <span class="o">=</span> <span class="n">ChatOllama</span><span class="p">(</span><span class="n">model</span><span class="o">=</span><span class="s2">&quot;qwen2.5:7b&quot;</span><span class="p">)</span>

<span class="c1"># 定义提示词</span>
<span class="n">messages</span> <span class="o">=</span> <span class="p">[</span>
        <span class="n">SystemMessage</span><span class="p">(</span><span class="n">content</span><span class="o">=</span><span class="s2">&quot;现在你是一个著名的诗人&quot;</span><span class="p">),</span>
        <span class="n">HumanMessage</span><span class="p">(</span><span class="n">content</span><span class="o">=</span><span class="s2">&quot;给我写一首唐诗&quot;</span><span class="p">)]</span>

<span class="c1"># 获取问答结果</span>
<span class="n">result</span> <span class="o">=</span> <span class="n">model</span><span class="o">.</span><span class="n">invoke</span><span class="p">(</span><span class="n">messages</span><span class="p">)</span>
<span class="c1"># print(result)</span>
<span class="nb">print</span><span class="p">(</span><span class="n">result</span><span class="o">.</span><span class="n">content</span><span class="p">)</span>
</code></pre></div>
<p><img src="assets/1762153476605.png" alt="image-20251004021532441" style="zoom:50%;" /></p>
<h3 id="3-embeddings-models">3 Embeddings Models<a class="headerlink" href="#3-embeddings-models" title="Permanent link">&para;</a></h3>
<p>Embeddings Models(嵌入模型）特点：将字符串作为输入，返回一个浮动数的列表。在NLP中，Embedding的作用就是将数据进行文本向量化。</p>
<div align=center><img src="assets/6-4.png" style="zoom:40%" ><img/></div>

<p>Embeddings Models可以为文本创建向量映射，这样就能在向量空间里去考虑文本，执行诸如语义搜索之类的操作，比如说寻找相似的文本片段。</p>
<blockquote>
<p><a href="https://python.langchain.com/docs/integrations/text_embedding/">https://python.langchain.com/docs/integrations/text_embedding/</a></p>
</blockquote>
<hr />
<p>不同的Embedding模型对多语言支持和文本类型有不同的特点：</p>
<ul>
<li><strong>多语言支持</strong>：<ul>
<li><code>text-embedding-ada-002</code>：支持多种语言，但对中文等亚洲语言的支持相对较弱</li>
<li><code>bge-large-zh</code>：对中文有很好的支持</li>
<li><code>multilingual-e5-large</code>：对多语言都有较好的支持</li>
<li>mxbai-embed-large：</li>
</ul>
</li>
<li><strong>文本类型适用性</strong>：<ul>
<li>代码文本：建议使用专门的代码Embedding模型，如 <code>CodeBERT</code></li>
<li>通用文本：可以使用<code>text-embedding-ada-002</code>或 <code>bge-large-zh</code></li>
<li>专业领域文本：建议使用该领域的专门模型</li>
</ul>
</li>
</ul>
<p>可以参考MTEB（大规模文本嵌入基准）排行榜以获取最新模型效果：<a href="https://huggingface.co/spaces/mteb/leaderboard">https://huggingface.co/spaces/mteb/leaderboard</a></p>
<p>接下来以一个文本嵌入模型的例子进行说明：</p>
<div class="highlight"><pre><span></span><code><span class="c1"># from langchain_community.embeddings import OllamaEmbeddings # Langchain 0.x版本使用</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">langchain_ollama</span><span class="w"> </span><span class="kn">import</span> <span class="n">OllamaEmbeddings</span>

<span class="c1"># 初始化Ollama嵌入模型，使用mxbai-embed-large模型，温度设置为0</span>
<span class="n">model</span> <span class="o">=</span> <span class="n">OllamaEmbeddings</span><span class="p">(</span><span class="n">model</span><span class="o">=</span><span class="s2">&quot;mxbai-embed-large&quot;</span><span class="p">,</span> <span class="n">temperature</span><span class="o">=</span><span class="mi">0</span><span class="p">)</span>

<span class="c1"># 对单个查询文本进行嵌入编码</span>
<span class="n">res1</span> <span class="o">=</span> <span class="n">model</span><span class="o">.</span><span class="n">embed_query</span><span class="p">(</span><span class="s1">&#39;这是第一个测试文档&#39;</span><span class="p">)</span>

<span class="nb">print</span><span class="p">(</span><span class="sa">f</span><span class="s1">&#39;result1-&gt;</span><span class="si">{</span><span class="n">res1</span><span class="si">}</span><span class="s1">&#39;</span><span class="p">)</span>
<span class="nb">print</span><span class="p">(</span><span class="sa">f</span><span class="s1">&#39;result1的长度-&gt;</span><span class="si">{</span><span class="nb">len</span><span class="p">(</span><span class="n">res1</span><span class="p">)</span><span class="si">}</span><span class="s1">&#39;</span><span class="p">)</span>

<span class="c1"># 对多个文档进行批量嵌入编码</span>
<span class="n">res2</span> <span class="o">=</span> <span class="n">model</span><span class="o">.</span><span class="n">embed_documents</span><span class="p">([</span><span class="s1">&#39;这是第一个测试文档&#39;</span><span class="p">,</span> <span class="s1">&#39;这是第二个测试文档&#39;</span><span class="p">])</span>
<span class="nb">print</span><span class="p">(</span><span class="n">res2</span><span class="p">)</span>
</code></pre></div>
<p>运行结果：</p>
<p><img alt="1762164216432" src="assets/1762164216432.png" /></p>
<h2 id="prompts">三、Prompts<a class="headerlink" href="#prompts" title="Permanent link">&para;</a></h2>
<h3 id="1-prompt">1 通用prompt<a class="headerlink" href="#1-prompt" title="Permanent link">&para;</a></h3>
<p>Prompt是指当用户输入信息给模型时加入的提示，这个提示的形式可以是zero-shot或者few-shot等方式，目的是让模型理解更为复杂的业务场景以便更好的解决问题。</p>
<p>提示模板：如果你有了一个起作用的提示，你可能想把它作为一个模板用于解决其他问题，LangChain就提供了PromptTemplates组件，它可以帮助你更方便的构建提示。</p>
<ul>
<li><strong>zero-shot提示方式</strong></li>
</ul>
<div class="highlight"><pre><span></span><code><span class="c1"># from langchain import PromptTemplate # Langchain 0.x版本使用</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">langchain_core.prompts</span><span class="w"> </span><span class="kn">import</span> <span class="n">PromptTemplate</span>
<span class="c1"># from langchain_community.llms import Ollama  # Langchain 0.x版本使用</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">langchain_ollama</span><span class="w"> </span><span class="kn">import</span> <span class="n">OllamaLLM</span>

<span class="c1"># model = Ollama(model=&quot;qwen2.5:7b&quot;) # Langchain 0.x版本使用</span>
<span class="n">model</span> <span class="o">=</span> <span class="n">OllamaLLM</span><span class="p">(</span><span class="n">model</span><span class="o">=</span><span class="s2">&quot;qwen2.5:7b&quot;</span><span class="p">)</span>
<span class="c1"># 定义模板</span>
<span class="n">template</span> <span class="o">=</span> <span class="s2">&quot;我的邻居姓</span><span class="si">{lastname}</span><span class="s2">，他生了个儿子，给他儿子起个名字&quot;</span>

<span class="n">prompt</span> <span class="o">=</span> <span class="n">PromptTemplate</span><span class="p">(</span>
    <span class="n">input_variables</span><span class="o">=</span><span class="p">[</span><span class="s2">&quot;lastname&quot;</span><span class="p">],</span>
    <span class="n">template</span><span class="o">=</span><span class="n">template</span><span class="p">)</span>

<span class="n">prompt_text</span> <span class="o">=</span> <span class="n">prompt</span><span class="o">.</span><span class="n">format</span><span class="p">(</span><span class="n">lastname</span><span class="o">=</span><span class="s2">&quot;王&quot;</span><span class="p">)</span>
<span class="nb">print</span><span class="p">(</span><span class="n">prompt_text</span><span class="p">)</span>
<span class="c1"># result: 我的邻居姓王，他生了个儿子，给他儿子起个名字</span>

<span class="n">result</span> <span class="o">=</span> <span class="n">model</span><span class="o">.</span><span class="n">invoke</span><span class="p">(</span><span class="n">prompt_text</span><span class="p">)</span>
<span class="nb">print</span><span class="p">(</span><span class="n">result</span><span class="p">)</span>
</code></pre></div>
<p>运行结果：</p>
<p><img alt="image-20251002154154332" src="assets/image-20251002154154332.png" /></p>
<ul>
<li><strong>few-shot提示方式</strong></li>
</ul>
<div class="highlight"><pre><span></span><code><span class="c1"># from langchain import PromptTemplate, FewShotPromptTemplate # Langchain 0.x版本使用</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">langchain_core.prompts</span><span class="w"> </span><span class="kn">import</span> <span class="n">PromptTemplate</span><span class="p">,</span> <span class="n">FewShotPromptTemplate</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">langchain_ollama</span><span class="w"> </span><span class="kn">import</span> <span class="n">OllamaLLM</span>

<span class="n">model</span> <span class="o">=</span> <span class="n">OllamaLLM</span><span class="p">(</span><span class="n">model</span><span class="o">=</span><span class="s2">&quot;qwen2.5:7b&quot;</span><span class="p">)</span>

<span class="n">examples</span> <span class="o">=</span> <span class="p">[</span>
    <span class="p">{</span><span class="s2">&quot;word&quot;</span><span class="p">:</span> <span class="s2">&quot;开心&quot;</span><span class="p">,</span> <span class="s2">&quot;antonym&quot;</span><span class="p">:</span> <span class="s2">&quot;难过&quot;</span><span class="p">},</span>
    <span class="p">{</span><span class="s2">&quot;word&quot;</span><span class="p">:</span> <span class="s2">&quot;高&quot;</span><span class="p">,</span> <span class="s2">&quot;antonym&quot;</span><span class="p">:</span> <span class="s2">&quot;矮&quot;</span><span class="p">}]</span>

<span class="n">example_template</span> <span class="o">=</span> <span class="s2">&quot;&quot;&quot;</span>
<span class="s2">单词: </span><span class="si">{word}</span>
<span class="s2">反义词: </span><span class="si">{antonym}</span><span class="se">\\</span><span class="s2">n</span>
<span class="s2">&quot;&quot;&quot;</span>

<span class="n">example_prompt</span> <span class="o">=</span> <span class="n">PromptTemplate</span><span class="p">(</span>
    <span class="n">input_variables</span><span class="o">=</span><span class="p">[</span><span class="s2">&quot;word&quot;</span><span class="p">,</span> <span class="s2">&quot;antonym&quot;</span><span class="p">],</span>
    <span class="n">template</span><span class="o">=</span><span class="n">example_template</span><span class="p">,</span>
<span class="p">)</span>

<span class="n">few_shot_prompt</span> <span class="o">=</span> <span class="n">FewShotPromptTemplate</span><span class="p">(</span>
    <span class="n">examples</span><span class="o">=</span><span class="n">examples</span><span class="p">,</span>
    <span class="n">example_prompt</span><span class="o">=</span><span class="n">example_prompt</span><span class="p">,</span>
    <span class="n">prefix</span><span class="o">=</span><span class="s2">&quot;给出每个单词的反义词&quot;</span><span class="p">,</span>
    <span class="n">suffix</span><span class="o">=</span><span class="s2">&quot;单词: </span><span class="si">{input}</span><span class="se">\\</span><span class="s2">n反义词:&quot;</span><span class="p">,</span>
    <span class="n">input_variables</span><span class="o">=</span><span class="p">[</span><span class="s2">&quot;input&quot;</span><span class="p">],</span>
    <span class="n">example_separator</span><span class="o">=</span><span class="s2">&quot;</span><span class="se">\\</span><span class="s2">n&quot;</span><span class="p">,</span>
<span class="p">)</span>

<span class="n">prompt_text</span> <span class="o">=</span> <span class="n">few_shot_prompt</span><span class="o">.</span><span class="n">format</span><span class="p">(</span><span class="nb">input</span><span class="o">=</span><span class="s2">&quot;粗&quot;</span><span class="p">)</span>

<span class="nb">print</span><span class="p">(</span><span class="n">model</span><span class="o">.</span><span class="n">invoke</span><span class="p">(</span><span class="n">prompt_text</span><span class="p">))</span>
<span class="c1"># 细</span>
</code></pre></div>
<p>运行结果：</p>
<p><img src="assets/image-20251002152859092.png" alt="image-20251002152859092" style="zoom:67%;" /></p>
<h3 id="2-chatprompts">2 ChatPrompts<a class="headerlink" href="#2-chatprompts" title="Permanent link">&para;</a></h3>
<p>适合交互式对话应用，如聊天机器人、智能客服等，这些应用需要处理用户和LLM之间的多轮对话。</p>
<blockquote>
<p>ChatPromptTemplate</p>
<p>SystemMessagePromptTemplate</p>
<p>HumanMessagePromptTemplate</p>
<p>history=[("system","......"),('human',"......"),("ai","......")]</p>
</blockquote>
<ul>
<li><strong>直接提问</strong></li>
</ul>
<p>提示模板就是把一些常见的提示整理成模板，用户只需要修改模板中特定的词语，就能快速准确地告诉模型自己的需求。我们看个例子：</p>
<div class="highlight"><pre><span></span><code><span class="kn">from</span><span class="w"> </span><span class="nn">langchain_core.prompts</span><span class="w"> </span><span class="kn">import</span> <span class="n">ChatPromptTemplate</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">langchain_ollama</span><span class="w"> </span><span class="kn">import</span> <span class="n">OllamaLLM</span>

<span class="c1"># 实例化模型</span>
<span class="n">model</span> <span class="o">=</span> <span class="n">OllamaLLM</span><span class="p">(</span><span class="n">model</span><span class="o">=</span><span class="s2">&quot;qwen2.5:7b&quot;</span><span class="p">)</span>

<span class="c1"># 定义提示词模版</span>
<span class="n">template_str</span> <span class="o">=</span> <span class="s2">&quot;帮我讲个关于</span><span class="si">{name}</span><span class="s2">笑话吧&quot;</span>
<span class="n">prompt_template</span> <span class="o">=</span> <span class="n">ChatPromptTemplate</span><span class="o">.</span><span class="n">from_template</span><span class="p">(</span><span class="n">template_str</span><span class="p">)</span>
<span class="n">prompt</span> <span class="o">=</span> <span class="n">prompt_template</span><span class="o">.</span><span class="n">format_messages</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s2">&quot;气球&quot;</span><span class="p">)</span>
<span class="nb">print</span><span class="p">(</span><span class="sa">f</span><span class="s1">&#39;prompt--&gt;</span><span class="si">{</span><span class="n">prompt</span><span class="si">}</span><span class="s1">&#39;</span><span class="p">)</span>

<span class="c1"># 调用模型</span>
<span class="n">result</span> <span class="o">=</span> <span class="n">model</span><span class="o">.</span><span class="n">invoke</span><span class="p">(</span><span class="n">prompt</span><span class="p">)</span>
<span class="nb">print</span><span class="p">(</span><span class="sa">f</span><span class="s1">&#39;result--&gt;</span><span class="si">{</span><span class="n">result</span><span class="si">}</span><span class="s1">&#39;</span><span class="p">)</span>
</code></pre></div>
<p>运行结果：</p>
<p><img src="assets/image-20251002153543943.png" alt="image-20251002153543943" style="zoom:67%;" /></p>
<ul>
<li><strong>zero-shot提示方式</strong></li>
</ul>
<div class="highlight"><pre><span></span><code><span class="kn">from</span><span class="w"> </span><span class="nn">langchain_core.prompts</span><span class="w"> </span><span class="kn">import</span> <span class="n">ChatPromptTemplate</span><span class="p">,</span> <span class="n">HumanMessagePromptTemplate</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">langchain_core.messages</span><span class="w"> </span><span class="kn">import</span> <span class="n">SystemMessage</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">langchain_ollama</span><span class="w"> </span><span class="kn">import</span> <span class="n">OllamaLLM</span>

<span class="c1"># 实例化模型</span>
<span class="n">model</span> <span class="o">=</span> <span class="n">OllamaLLM</span><span class="p">(</span><span class="n">model</span><span class="o">=</span><span class="s2">&quot;qwen2.5:7b&quot;</span><span class="p">)</span>

<span class="c1"># 系统信息</span>
<span class="n">system_prompt</span> <span class="o">=</span> <span class="n">SystemMessage</span><span class="p">(</span><span class="s2">&quot;你是取名专家。&quot;</span><span class="p">)</span>
<span class="c1"># 用户信息模版</span>
<span class="n">human_str</span> <span class="o">=</span> <span class="s2">&quot;我的邻居姓</span><span class="si">{lastname}</span><span class="s2">，他生了个儿子，给他儿子起个名字。&quot;</span>
<span class="n">human_template</span> <span class="o">=</span> <span class="n">HumanMessagePromptTemplate</span><span class="o">.</span><span class="n">from_template</span><span class="p">(</span><span class="n">human_str</span><span class="p">)</span>
<span class="c1"># 组装</span>
<span class="n">chat_template</span> <span class="o">=</span> <span class="n">ChatPromptTemplate</span><span class="o">.</span><span class="n">from_messages</span><span class="p">([</span><span class="n">system_prompt</span><span class="p">,</span> <span class="n">human_template</span><span class="p">])</span>
<span class="c1"># 生成最终的提示词</span>
<span class="n">prompt</span> <span class="o">=</span> <span class="n">chat_template</span><span class="o">.</span><span class="n">format_messages</span><span class="p">(</span><span class="n">lastname</span><span class="o">=</span><span class="s2">&quot;王&quot;</span><span class="p">)</span>
<span class="nb">print</span><span class="p">(</span><span class="sa">f</span><span class="s1">&#39;prompt--&gt;</span><span class="si">{</span><span class="n">prompt</span><span class="si">}</span><span class="s1">&#39;</span><span class="p">)</span>

<span class="c1"># 调用模型</span>
<span class="n">result</span> <span class="o">=</span> <span class="n">model</span><span class="o">.</span><span class="n">invoke</span><span class="p">(</span><span class="n">prompt</span><span class="p">)</span>  <span class="c1"># 返回结果出了content外，还有元数据信息</span>
<span class="nb">print</span><span class="p">(</span><span class="sa">f</span><span class="s1">&#39;result--&gt;</span><span class="si">{</span><span class="n">result</span><span class="si">}</span><span class="s1">&#39;</span><span class="p">)</span>
</code></pre></div>
<p>运行结果：</p>
<p><img alt="image-20251002200159523" src="assets/image-20251002200159523.png" /></p>
<ul>
<li><strong>few-shot提示方式</strong></li>
</ul>
<div class="highlight"><pre><span></span><code><span class="kn">from</span><span class="w"> </span><span class="nn">langchain_core.prompts</span><span class="w"> </span><span class="kn">import</span> <span class="n">ChatPromptTemplate</span><span class="p">,</span> <span class="n">MessagesPlaceholder</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">langchain_ollama</span><span class="w"> </span><span class="kn">import</span> <span class="n">OllamaLLM</span>

<span class="c1"># 实例化模型</span>
<span class="n">model</span> <span class="o">=</span> <span class="n">OllamaLLM</span><span class="p">(</span><span class="n">model</span><span class="o">=</span><span class="s2">&quot;qwen2.5:7b&quot;</span><span class="p">)</span>

<span class="c1"># 创建 prompt 模版</span>
<span class="n">prompt_template</span> <span class="o">=</span> <span class="n">ChatPromptTemplate</span><span class="o">.</span><span class="n">from_messages</span><span class="p">(</span>
      <span class="p">[</span>
         <span class="p">(</span><span class="s2">&quot;system&quot;</span><span class="p">,</span> <span class="s2">&quot;给出每个单词的反义词&quot;</span><span class="p">),</span>
         <span class="n">MessagesPlaceholder</span><span class="p">(</span><span class="s2">&quot;history&quot;</span><span class="p">),</span>
         <span class="p">(</span><span class="s2">&quot;human&quot;</span><span class="p">,</span> <span class="s2">&quot;</span><span class="si">{question}</span><span class="s2">&quot;</span><span class="p">)</span>
      <span class="p">]</span>
<span class="p">)</span>
<span class="c1"># 创建 few-shot prompt</span>
<span class="c1"># history = [</span>
<span class="c1">#     HumanMessage(content=&quot;开心&quot;),</span>
<span class="c1">#     AIMessage(content=&quot;难过&quot;),</span>
<span class="c1">#     HumanMessage(content=&quot;高&quot;),</span>
<span class="c1">#     AIMessage(content=&quot;矮&quot;)</span>
<span class="c1"># ]</span>
<span class="n">history</span> <span class="o">=</span> <span class="p">[(</span><span class="s2">&quot;human&quot;</span><span class="p">,</span> <span class="s2">&quot;开心&quot;</span><span class="p">),</span> <span class="p">(</span><span class="s2">&quot;ai&quot;</span><span class="p">,</span> <span class="s2">&quot;难过&quot;</span><span class="p">),(</span><span class="s2">&quot;human&quot;</span><span class="p">,</span> <span class="s2">&quot;高&quot;</span><span class="p">),</span> <span class="p">(</span><span class="s2">&quot;ai&quot;</span><span class="p">,</span> <span class="s2">&quot;矮&quot;</span><span class="p">)]</span>
<span class="n">prompt</span> <span class="o">=</span> <span class="n">prompt_template</span><span class="o">.</span><span class="n">format_messages</span><span class="p">(</span><span class="n">history</span><span class="o">=</span><span class="n">history</span><span class="p">,</span> <span class="n">question</span><span class="o">=</span><span class="s2">&quot;富有&quot;</span><span class="p">)</span>
<span class="nb">print</span><span class="p">(</span><span class="sa">f</span><span class="s2">&quot;prompt--&gt;</span><span class="si">{</span><span class="n">prompt</span><span class="si">}</span><span class="s2">&quot;</span><span class="p">)</span>

<span class="c1"># 调用模型</span>
<span class="n">result</span> <span class="o">=</span> <span class="n">model</span><span class="o">.</span><span class="n">invoke</span><span class="p">(</span><span class="n">prompt</span><span class="p">)</span>
<span class="nb">print</span><span class="p">(</span><span class="sa">f</span><span class="s1">&#39;result--&gt;</span><span class="si">{</span><span class="n">result</span><span class="si">}</span><span class="s1">&#39;</span><span class="p">)</span>
</code></pre></div>
<p>运行结果：</p>
<p><img alt="image-20251002205814344" src="assets/image-20251002205814344.png" /></p>
<h2 id="chains">四、Chains<a class="headerlink" href="#chains" title="Permanent link">&para;</a></h2>
<p>在LangChain中，Chains描述了将LLM与其他组件结合起来完成一个应用程序的过程。
针对上一小节的提示模版例子，zero-shot里面，我们可以用链来连接提示模版组件和模型，进而可以实现代码的更改，主要使用LCEL方法。</p>
<p>LCEL（Lang Chain Expression Language） 是一种声明式的方法，用于轻松组合链条。</p>
<p>LCEL的基本语法规则是使用<code>|</code>符号将不同的组件连接起来，形成一个链式结构。<code>|</code>符号类似于Unix的管道操作符，它将一个组件的输出作为下一个组件的输入，从而实现数据的传递和处理。</p>
<p>上一个组件的输出作为下一个组件的输入，输出和输入的类型必须保持一致，否则不能连接。</p>
<div class="highlight"><pre><span></span><code><span class="kn">from</span><span class="w"> </span><span class="nn">langchain</span><span class="w"> </span><span class="kn">import</span> <span class="n">PromptTemplate</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">langchain_core.prompts</span><span class="w"> </span><span class="kn">import</span> <span class="n">PromptTemplate</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">langchain_ollama</span><span class="w"> </span><span class="kn">import</span> <span class="n">OllamaLLM</span>

<span class="c1"># 定义模板</span>
<span class="n">template</span> <span class="o">=</span> <span class="s2">&quot;我的邻居姓</span><span class="si">{lastname}</span><span class="s2">，他生了个儿子，给他儿子起个名字&quot;</span>

<span class="n">prompt</span> <span class="o">=</span> <span class="n">PromptTemplate</span><span class="p">(</span>
    <span class="n">input_variables</span><span class="o">=</span><span class="p">[</span><span class="s2">&quot;lastname&quot;</span><span class="p">],</span>
    <span class="n">template</span><span class="o">=</span><span class="n">template</span><span class="p">)</span>

<span class="c1"># 实例化模型</span>
<span class="n">llm</span> <span class="o">=</span> <span class="n">OllamaLLM</span><span class="p">(</span><span class="n">model</span><span class="o">=</span><span class="s2">&quot;qwen2.5:7b&quot;</span><span class="p">)</span>

<span class="n">chain</span> <span class="o">=</span> <span class="n">prompt</span> <span class="o">|</span> <span class="n">llm</span>
<span class="c1"># 执行链</span>
<span class="nb">print</span><span class="p">(</span><span class="n">chain</span><span class="o">.</span><span class="n">invoke</span><span class="p">(</span><span class="s2">&quot;王&quot;</span><span class="p">))</span>
</code></pre></div>
<p>如果你想将第一个模型输出的结果，直接作为第二个模型的输入，直接使用管道符, 代码如下：</p>
<div class="highlight"><pre><span></span><code><span class="kn">from</span><span class="w"> </span><span class="nn">langchain_core.prompts</span><span class="w"> </span><span class="kn">import</span> <span class="n">PromptTemplate</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">langchain_ollama</span><span class="w"> </span><span class="kn">import</span> <span class="n">OllamaLLM</span>

<span class="c1"># 创建第一条链</span>
<span class="n">template</span> <span class="o">=</span> <span class="s2">&quot;我的邻居姓</span><span class="si">{lastname}</span><span class="s2">，他生了个儿子，给他儿子起个名字&quot;</span>

<span class="n">first_prompt</span> <span class="o">=</span> <span class="n">PromptTemplate</span><span class="p">(</span>
    <span class="n">input_variables</span><span class="o">=</span><span class="p">[</span><span class="s2">&quot;lastname&quot;</span><span class="p">],</span>
    <span class="n">template</span><span class="o">=</span><span class="n">template</span><span class="p">,)</span>

<span class="c1"># 实例化模型</span>
<span class="n">llm</span> <span class="o">=</span> <span class="n">OllamaLLM</span><span class="p">(</span><span class="n">model</span><span class="o">=</span><span class="s2">&quot;qwen2.5:7b&quot;</span><span class="p">)</span>

<span class="n">first_chain</span> <span class="o">=</span> <span class="n">first_prompt</span> <span class="o">|</span> <span class="n">llm</span>

<span class="c1"># 创建第二条链</span>
<span class="n">second_prompt</span> <span class="o">=</span> <span class="n">PromptTemplate</span><span class="p">(</span>
    <span class="n">input_variables</span><span class="o">=</span><span class="p">[</span><span class="s2">&quot;child_name&quot;</span><span class="p">],</span>
    <span class="n">template</span><span class="o">=</span><span class="s2">&quot;邻居的儿子名字叫</span><span class="si">{child_name}</span><span class="s2">，给他起一个小名&quot;</span><span class="p">)</span>

<span class="n">second_chain</span> <span class="o">=</span> <span class="n">second_prompt</span> <span class="o">|</span> <span class="n">llm</span>

<span class="c1"># 链接两条链</span>
<span class="n">overall_chain</span> <span class="o">=</span> <span class="n">first_chain</span> <span class="o">|</span> <span class="n">second_chain</span>

<span class="nb">print</span><span class="p">(</span><span class="n">overall_chain</span><span class="p">)</span>
<span class="nb">print</span><span class="p">(</span><span class="s1">&#39;*&#39;</span><span class="o">*</span><span class="mi">80</span><span class="p">)</span>
<span class="c1"># 执行链，只需要传入第一个参数</span>
<span class="n">catchphrase</span> <span class="o">=</span> <span class="n">overall_chain</span><span class="o">.</span><span class="n">invoke</span><span class="p">(</span><span class="s2">&quot;王&quot;</span><span class="p">)</span>
<span class="nb">print</span><span class="p">(</span><span class="n">catchphrase</span><span class="p">)</span>
</code></pre></div>
<p>运行结果：</p>
<p><img alt="image-20251002212834547" src="assets/image-20251002212834547.png" /></p>
<p><img alt="image-20251002212930745" src="assets/image-20251002212930745.png" /></p>
<h2 id="output_parsers">五、output_parsers<a class="headerlink" href="#output_parsers" title="Permanent link">&para;</a></h2>
<p>LLM 的输出是自然语言文本，但在应用开发中，我们经常需要将这些文本转换为结构化的数据格式，如列表、字典或对象。LangChain 输出解析器负责获取 LLM 的输出并将其转换为更合适的格式。</p>
<p>部分解析器如下：</p>
<table>
<thead>
<tr>
<th>解析器名称</th>
<th>核心功能</th>
<th>输出的 Python 类型</th>
<th>工业级应用场景</th>
</tr>
</thead>
<tbody>
<tr>
<td>StrOutputParser</td>
<td>默认解析器。将 LLM 的输出直接解析为字符串。</td>
<td>str</td>
<td>只需要原始回答时（如问答任务、对话场景）</td>
</tr>
<tr>
<td>CommaSeparatedListParser</td>
<td>将 LLM 输出的、用逗号分隔的文本解析为列表。</td>
<td>list[str]</td>
<td>列表/枚举型输出</td>
</tr>
<tr>
<td>JsonOutputParser</td>
<td>极其常用。将 LLM 输出的 JSON 字符串解析为 Python 字典。</td>
<td>dict</td>
<td>结构化 JSON 输出</td>
</tr>
<tr>
<td>PydanticOutputParser</td>
<td>极其常用。将 LLM 输出解析为预先定义的 Pydantic 对象，提供类型安全和数据验证。</td>
<td>自定义的 pydantic.BaseModel 对象</td>
<td>输出需要严格结构化（JSON-like）数据时</td>
</tr>
<tr>
<td>DatetimeOutputParser</td>
<td>从文本中智能地解析出日期和时间信息。</td>
<td>datetime.datetime</td>
<td>需要时间格式时</td>
</tr>
</tbody>
</table>
<h3 id="1">1 字符串解析器<a class="headerlink" href="#1" title="Permanent link">&para;</a></h3>
<p>StrOutputParser，最简单的解析器，用于提取模型返回的原始文本：</p>
<div class="highlight"><pre><span></span><code><span class="kn">from</span><span class="w"> </span><span class="nn">langchain_core.output_parsers</span><span class="w"> </span><span class="kn">import</span> <span class="n">StrOutputParser</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">langchain_core.prompts</span><span class="w"> </span><span class="kn">import</span> <span class="n">ChatPromptTemplate</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">langchain_ollama</span><span class="w"> </span><span class="kn">import</span> <span class="n">OllamaLLM</span>

<span class="c1"># 实例化模型</span>
<span class="n">model</span> <span class="o">=</span> <span class="n">OllamaLLM</span><span class="p">(</span><span class="n">model</span><span class="o">=</span><span class="s2">&quot;qwen2.5:7b&quot;</span><span class="p">)</span>

<span class="c1"># 创建简单链</span>
<span class="n">prompt</span> <span class="o">=</span> <span class="n">ChatPromptTemplate</span><span class="o">.</span><span class="n">from_template</span><span class="p">(</span><span class="s2">&quot;解释</span><span class="si">{topic}</span><span class="s2">是什么？回答控制20字以内&quot;</span><span class="p">)</span>
<span class="n">chain1</span> <span class="o">=</span> <span class="n">prompt</span> <span class="o">|</span> <span class="n">model</span>
<span class="n">result1</span> <span class="o">=</span> <span class="n">chain1</span><span class="o">.</span><span class="n">invoke</span><span class="p">({</span><span class="s2">&quot;topic&quot;</span><span class="p">:</span> <span class="s2">&quot;ai&quot;</span><span class="p">})</span>
<span class="nb">print</span><span class="p">(</span><span class="sa">f</span><span class="s2">&quot;result1--&gt;</span><span class="si">{</span><span class="n">result1</span><span class="si">}</span><span class="s2">&quot;</span><span class="p">)</span>

<span class="c1"># 添加字符串解析器</span>
<span class="n">parser</span> <span class="o">=</span> <span class="n">StrOutputParser</span><span class="p">()</span>
<span class="n">chain2</span> <span class="o">=</span> <span class="n">prompt</span> <span class="o">|</span> <span class="n">model</span> <span class="o">|</span> <span class="n">parser</span>
<span class="n">result2</span> <span class="o">=</span> <span class="n">chain2</span><span class="o">.</span><span class="n">invoke</span><span class="p">({</span><span class="s2">&quot;topic&quot;</span><span class="p">:</span> <span class="s2">&quot;ai&quot;</span><span class="p">})</span>
<span class="nb">print</span><span class="p">(</span><span class="sa">f</span><span class="s2">&quot;result2--&gt;</span><span class="si">{</span><span class="n">result2</span><span class="si">}</span><span class="s2">&quot;</span><span class="p">)</span>
</code></pre></div>
<p>运行结果：</p>
<p><img alt="image-20251002230930266" src="assets/image-20251002230930266.png" /></p>
<h3 id="2_1">2 列表解析器<a class="headerlink" href="#2_1" title="Permanent link">&para;</a></h3>
<p>CommaSeparatedListOutputParser，将逗号分隔的文本转换为Python列表：</p>
<div class="highlight"><pre><span></span><code><span class="kn">from</span><span class="w"> </span><span class="nn">langchain_core.output_parsers</span><span class="w"> </span><span class="kn">import</span> <span class="n">CommaSeparatedListOutputParser</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">langchain_core.prompts</span><span class="w"> </span><span class="kn">import</span> <span class="n">ChatPromptTemplate</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">langchain_ollama</span><span class="w"> </span><span class="kn">import</span> <span class="n">OllamaLLM</span>

<span class="c1"># 实例化模型</span>
<span class="n">model</span> <span class="o">=</span> <span class="n">OllamaLLM</span><span class="p">(</span><span class="n">model</span><span class="o">=</span><span class="s2">&quot;qwen2.5:7b&quot;</span><span class="p">)</span>

<span class="c1"># 创建列表解析器</span>
<span class="n">parser</span> <span class="o">=</span> <span class="n">CommaSeparatedListOutputParser</span><span class="p">()</span>

<span class="c1"># 创建带格式说明的提示模板</span>
<span class="n">format_instructions</span> <span class="o">=</span> <span class="n">parser</span><span class="o">.</span><span class="n">get_format_instructions</span><span class="p">()</span>
<span class="n">prompt</span> <span class="o">=</span> <span class="n">ChatPromptTemplate</span><span class="o">.</span><span class="n">from_template</span><span class="p">(</span>
    <span class="s2">&quot;用中文列出</span><span class="si">{topic}</span><span class="s2">的五个最重要特点。</span><span class="se">\n</span><span class="si">{format_instructions}</span><span class="s2">&quot;</span>
<span class="p">)</span>

<span class="c1"># 组合组件</span>
<span class="n">chain</span> <span class="o">=</span> <span class="n">prompt</span> <span class="o">|</span> <span class="n">model</span> <span class="o">|</span> <span class="n">parser</span>

<span class="c1"># 调用链</span>
<span class="n">result</span> <span class="o">=</span> <span class="n">chain</span><span class="o">.</span><span class="n">invoke</span><span class="p">({</span>
    <span class="s2">&quot;topic&quot;</span><span class="p">:</span> <span class="s2">&quot;大模型&quot;</span><span class="p">,</span>
    <span class="s2">&quot;format_instructions&quot;</span><span class="p">:</span> <span class="n">format_instructions</span>
<span class="p">})</span>
<span class="nb">print</span><span class="p">(</span><span class="n">result</span><span class="p">)</span>
</code></pre></div>
<p>运行结果：</p>
<p><img src="assets/image-20251002231200108.png" alt="image-20251002231200108" style="zoom: 67%;" /></p>
<h3 id="3-json">3 JSON解析器<a class="headerlink" href="#3-json" title="Permanent link">&para;</a></h3>
<p>JsonOutputParser，将JSON格式文本转换为Python字典或列表：</p>
<div class="highlight"><pre><span></span><code><span class="kn">from</span><span class="w"> </span><span class="nn">langchain_core.output_parsers</span><span class="w"> </span><span class="kn">import</span> <span class="n">JsonOutputParser</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">langchain_core.prompts</span><span class="w"> </span><span class="kn">import</span> <span class="n">ChatPromptTemplate</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">langchain_ollama</span><span class="w"> </span><span class="kn">import</span> <span class="n">OllamaLLM</span>

<span class="c1"># 实例化模型</span>
<span class="n">model</span> <span class="o">=</span> <span class="n">OllamaLLM</span><span class="p">(</span><span class="n">model</span><span class="o">=</span><span class="s2">&quot;qwen2.5:7b&quot;</span><span class="p">)</span>

<span class="c1"># 创建JSON解析器</span>
<span class="n">json_parser</span> <span class="o">=</span> <span class="n">JsonOutputParser</span><span class="p">()</span>

<span class="c1"># 创建带格式说明的提示模板</span>
<span class="n">json_format_instructions</span> <span class="o">=</span> <span class="n">json_parser</span><span class="o">.</span><span class="n">get_format_instructions</span><span class="p">()</span>
<span class="n">json_prompt</span> <span class="o">=</span> <span class="n">ChatPromptTemplate</span><span class="o">.</span><span class="n">from_template</span><span class="p">(</span>
    <span class="s2">&quot;生成一个包含</span><span class="si">{person}</span><span class="s2">基本信息的JSON。应包括姓名、职业、年龄和技能列表, 不要包含任何注释或额外说明。</span><span class="se">\n</span><span class="si">{format_instructions}</span><span class="s2">&quot;</span>
<span class="p">)</span>

<span class="c1"># 组合组件</span>
<span class="n">json_chain</span> <span class="o">=</span> <span class="n">json_prompt</span> <span class="o">|</span> <span class="n">model</span> <span class="o">|</span> <span class="n">json_parser</span>

<span class="c1"># 调用链</span>
<span class="n">result</span> <span class="o">=</span> <span class="n">json_chain</span><span class="o">.</span><span class="n">invoke</span><span class="p">({</span>
    <span class="s2">&quot;person&quot;</span><span class="p">:</span> <span class="s2">&quot;雷军&quot;</span><span class="p">,</span>
    <span class="s2">&quot;format_instructions&quot;</span><span class="p">:</span> <span class="n">json_format_instructions</span>
<span class="p">})</span>

<span class="nb">print</span><span class="p">(</span><span class="n">result</span><span class="p">)</span>
</code></pre></div>
<p>运行结果：</p>
<p><img alt="image-20251003000254030" src="assets/image-20251003000254030.png" /></p>
<h3 id="4-pydantic">4 Pydantic解析器<a class="headerlink" href="#4-pydantic" title="Permanent link">&para;</a></h3>
<p>PydanticOutputParser，使用Pydantic模型定义输出结构：</p>
<div class="highlight"><pre><span></span><code><span class="kn">from</span><span class="w"> </span><span class="nn">langchain_core.prompts</span><span class="w"> </span><span class="kn">import</span> <span class="n">ChatPromptTemplate</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">langchain_core.output_parsers</span><span class="w"> </span><span class="kn">import</span> <span class="n">PydanticOutputParser</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">pydantic</span><span class="w"> </span><span class="kn">import</span> <span class="n">BaseModel</span><span class="p">,</span> <span class="n">Field</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">typing</span><span class="w"> </span><span class="kn">import</span> <span class="n">List</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">langchain_ollama</span><span class="w"> </span><span class="kn">import</span> <span class="n">OllamaLLM</span>

<span class="c1"># 实例化模型</span>
<span class="n">model</span> <span class="o">=</span> <span class="n">OllamaLLM</span><span class="p">(</span><span class="n">model</span><span class="o">=</span><span class="s2">&quot;qwen2.5:7b&quot;</span><span class="p">)</span>

<span class="c1"># 定义Pydantic模型</span>
<span class="k">class</span><span class="w"> </span><span class="nc">Movie</span><span class="p">(</span><span class="n">BaseModel</span><span class="p">):</span>
    <span class="n">title</span><span class="p">:</span> <span class="nb">str</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span><span class="n">description</span><span class="o">=</span><span class="s2">&quot;电影标题&quot;</span><span class="p">)</span>
    <span class="n">director</span><span class="p">:</span> <span class="nb">str</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span><span class="n">description</span><span class="o">=</span><span class="s2">&quot;导演姓名&quot;</span><span class="p">)</span>
    <span class="n">year</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span><span class="n">description</span><span class="o">=</span><span class="s2">&quot;上映年份&quot;</span><span class="p">)</span>
    <span class="n">genre</span><span class="p">:</span> <span class="n">List</span><span class="p">[</span><span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span><span class="n">description</span><span class="o">=</span><span class="s2">&quot;电影类型&quot;</span><span class="p">)</span>
    <span class="n">rating</span><span class="p">:</span> <span class="nb">float</span> <span class="o">=</span> <span class="n">Field</span><span class="p">(</span><span class="n">description</span><span class="o">=</span><span class="s2">&quot;评分（1-10）&quot;</span><span class="p">)</span>


<span class="c1"># 创建Pydantic解析器</span>
<span class="n">pydantic_parser</span> <span class="o">=</span> <span class="n">PydanticOutputParser</span><span class="p">(</span><span class="n">pydantic_object</span><span class="o">=</span><span class="n">Movie</span><span class="p">)</span>

<span class="c1"># 创建带格式说明的提示模板</span>
<span class="n">format_instructions</span> <span class="o">=</span> <span class="n">pydantic_parser</span><span class="o">.</span><span class="n">get_format_instructions</span><span class="p">()</span>
<span class="n">pydantic_prompt</span> <span class="o">=</span> <span class="n">ChatPromptTemplate</span><span class="o">.</span><span class="n">from_template</span><span class="p">(</span>
    <span class="s2">&quot;生成一部</span><span class="si">{genre}</span><span class="s2">电影的信息。</span><span class="se">\n</span><span class="si">{format_instructions}</span><span class="s2">&quot;</span>
<span class="p">)</span>

<span class="c1"># 组合组件</span>
<span class="n">pydantic_chain</span> <span class="o">=</span> <span class="n">pydantic_prompt</span> <span class="o">|</span> <span class="n">model</span> <span class="o">|</span> <span class="n">pydantic_parser</span>

<span class="c1"># 调用链</span>
<span class="n">movie_data</span> <span class="o">=</span> <span class="n">pydantic_chain</span><span class="o">.</span><span class="n">invoke</span><span class="p">({</span>
    <span class="s2">&quot;genre&quot;</span><span class="p">:</span> <span class="s2">&quot;科幻&quot;</span><span class="p">,</span>
    <span class="s2">&quot;format_instructions&quot;</span><span class="p">:</span> <span class="n">format_instructions</span>
<span class="p">})</span>
<span class="nb">print</span><span class="p">(</span><span class="n">movie_data</span><span class="p">)</span>
</code></pre></div>
<p>运行结果：</p>
<p><img alt="image-20251003000720225" src="assets/image-20251003000720225.png" /></p>
<h3 id="5">5 自定义解析器<a class="headerlink" href="#5" title="Permanent link">&para;</a></h3>
<p>创建自定义输出解析器</p>
<div class="highlight"><pre><span></span><code><span class="kn">from</span><span class="w"> </span><span class="nn">langchain_core.output_parsers</span><span class="w"> </span><span class="kn">import</span> <span class="n">BaseOutputParser</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">typing</span><span class="w"> </span><span class="kn">import</span> <span class="n">Dict</span><span class="p">,</span> <span class="n">Any</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">langchain_core.prompts</span><span class="w"> </span><span class="kn">import</span> <span class="n">ChatPromptTemplate</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">langchain_ollama</span><span class="w"> </span><span class="kn">import</span> <span class="n">OllamaLLM</span>

<span class="c1"># 实例化模型</span>
<span class="n">model</span> <span class="o">=</span> <span class="n">OllamaLLM</span><span class="p">(</span><span class="n">model</span><span class="o">=</span><span class="s2">&quot;qwen2.5:7b&quot;</span><span class="p">)</span>

<span class="k">class</span><span class="w"> </span><span class="nc">CustomKeyValueParser</span><span class="p">(</span><span class="n">BaseOutputParser</span><span class="p">[</span><span class="n">Dict</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="n">Any</span><span class="p">]]):</span>
<span class="w">    </span><span class="sd">&quot;&quot;&quot;解析形如&#39;key: value&#39;的文本&quot;&quot;&quot;</span>
    <span class="k">def</span><span class="w"> </span><span class="nf">parse</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">text</span><span class="p">:</span> <span class="nb">str</span><span class="p">)</span> <span class="o">-&gt;</span> <span class="n">Dict</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="n">Any</span><span class="p">]:</span>
<span class="w">        </span><span class="sd">&quot;&quot;&quot;从文本中解析键值对&quot;&quot;&quot;</span>
        <span class="n">result</span> <span class="o">=</span> <span class="p">{}</span>
        <span class="n">lines</span> <span class="o">=</span> <span class="n">text</span><span class="o">.</span><span class="n">strip</span><span class="p">()</span><span class="o">.</span><span class="n">split</span><span class="p">(</span><span class="s1">&#39;</span><span class="se">\n</span><span class="s1">&#39;</span><span class="p">)</span>

        <span class="k">for</span> <span class="n">line</span> <span class="ow">in</span> <span class="n">lines</span><span class="p">:</span>
            <span class="k">if</span> <span class="s1">&#39;:&#39;</span> <span class="ow">in</span> <span class="n">line</span><span class="p">:</span>
                <span class="n">key</span><span class="p">,</span> <span class="n">value</span> <span class="o">=</span> <span class="n">line</span><span class="o">.</span><span class="n">split</span><span class="p">(</span><span class="s1">&#39;:&#39;</span><span class="p">,</span> <span class="mi">1</span><span class="p">)</span>
                <span class="n">result</span><span class="p">[</span><span class="n">key</span><span class="o">.</span><span class="n">strip</span><span class="p">()]</span> <span class="o">=</span> <span class="n">value</span><span class="o">.</span><span class="n">strip</span><span class="p">()</span>

        <span class="k">return</span> <span class="n">result</span>

    <span class="k">def</span><span class="w"> </span><span class="nf">get_format_instructions</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-&gt;</span> <span class="nb">str</span><span class="p">:</span>
<span class="w">        </span><span class="sd">&quot;&quot;&quot;提供格式指导给模型&quot;&quot;&quot;</span>
        <span class="k">return</span> <span class="s2">&quot;&quot;&quot;请以&#39;键: 值&#39;的格式返回信息，每行一个键值对。</span>
<span class="s2">                例如：</span>
<span class="s2">                名称: 爱因斯坦</span>
<span class="s2">                职业: 物理学家</span>
<span class="s2">                贡献: 相对论&quot;&quot;&quot;</span>


<span class="c1"># 使用自定义解析器</span>
<span class="n">custom_parser</span> <span class="o">=</span> <span class="n">CustomKeyValueParser</span><span class="p">()</span>
<span class="n">custom_prompt</span> <span class="o">=</span> <span class="n">ChatPromptTemplate</span><span class="o">.</span><span class="n">from_template</span><span class="p">(</span>
    <span class="s2">&quot;提供关于</span><span class="si">{person}</span><span class="s2">的基本信息。</span><span class="se">\n</span><span class="si">{format_instructions}</span><span class="s2">&quot;</span>
<span class="p">)</span>

<span class="c1"># 组合组件</span>
<span class="n">custom_chain</span> <span class="o">=</span> <span class="n">custom_prompt</span> <span class="o">|</span> <span class="n">model</span> <span class="o">|</span> <span class="n">custom_parser</span>

<span class="c1"># 调用模型</span>
<span class="n">result</span> <span class="o">=</span> <span class="n">custom_chain</span><span class="o">.</span><span class="n">invoke</span><span class="p">({</span>
    <span class="s2">&quot;person&quot;</span><span class="p">:</span> <span class="s2">&quot;屠呦呦&quot;</span><span class="p">,</span>
    <span class="s2">&quot;format_instructions&quot;</span><span class="p">:</span> <span class="n">custom_parser</span><span class="o">.</span><span class="n">get_format_instructions</span><span class="p">()</span>
<span class="p">})</span>
<span class="nb">print</span><span class="p">(</span><span class="n">result</span><span class="p">)</span>
</code></pre></div>
<p>运行结果：</p>
<p><img alt="image-20251003001007923" src="assets/image-20251003001007923.png" /></p>
<h2 id="memory">六、Memory<a class="headerlink" href="#memory" title="Permanent link">&para;</a></h2>
<p>大模型本身不具备上下文的概念，它并不保存上次交互的内容，ChatGPT之所以能够和人正常沟通对话，因为它进行了一层封装，将历史记录回传给了模型。</p>
<p>因此 LangChain 也提供了Memory组件, Memory分为两种类型： <strong>短期记忆和长期记忆</strong> 。短期记忆一般指单一会话时传递数据，长期记忆则是处理多个会话时获取和更新信息。</p>
<ul>
<li>使用 <strong>ChatMessageHistory</strong> 手动添加上下文</li>
</ul>
<div class="highlight"><pre><span></span><code><span class="kn">from</span><span class="w"> </span><span class="nn">langchain_community.chat_message_histories</span><span class="w"> </span><span class="kn">import</span> <span class="n">ChatMessageHistory</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">langchain_core.messages</span><span class="w"> </span><span class="kn">import</span> <span class="n">messages_to_dict</span><span class="p">,</span> <span class="n">messages_from_dict</span>
<span class="kn">import</span><span class="w"> </span><span class="nn">json</span>

<span class="c1"># 1.创建一个ChatMessageHistory对象，用来存储对话信息</span>
<span class="n">history</span> <span class="o">=</span> <span class="n">ChatMessageHistory</span><span class="p">()</span>
<span class="c1"># 添加用户消息</span>
<span class="n">history</span><span class="o">.</span><span class="n">add_user_message</span><span class="p">(</span><span class="s2">&quot;在吗？&quot;</span><span class="p">)</span>
<span class="c1"># 添加大模型消息</span>
<span class="n">history</span><span class="o">.</span><span class="n">add_ai_message</span><span class="p">(</span><span class="s2">&quot;在&quot;</span><span class="p">)</span>
<span class="c1"># 打印所有消息</span>
<span class="nb">print</span><span class="p">(</span><span class="sa">f</span><span class="s1">&#39;history.messages--&gt;</span><span class="si">{</span><span class="n">history</span><span class="o">.</span><span class="n">messages</span><span class="si">}</span><span class="s1">&#39;</span><span class="p">)</span>

<span class="c1"># 2.可以将history.messages中的信息保存到字典中，然后保存到数据库或者文件中，方便后续读取</span>
<span class="c1"># 2.1 messages_to_dict()方法将history.messages中的信息转换成字典</span>
<span class="n">dicts</span> <span class="o">=</span> <span class="n">messages_to_dict</span><span class="p">(</span><span class="n">history</span><span class="o">.</span><span class="n">messages</span><span class="p">)</span>
<span class="nb">print</span><span class="p">(</span><span class="sa">f</span><span class="s1">&#39;dicts--&gt;</span><span class="si">{</span><span class="n">dicts</span><span class="si">}</span><span class="s1">&#39;</span><span class="p">)</span>
<span class="c1"># 2.2 这里将dicts保存到文件中</span>
<span class="k">with</span> <span class="nb">open</span><span class="p">(</span><span class="s1">&#39;history.json&#39;</span><span class="p">,</span> <span class="s1">&#39;w&#39;</span><span class="p">,</span> <span class="n">encoding</span><span class="o">=</span><span class="s1">&#39;utf-8&#39;</span><span class="p">)</span> <span class="k">as</span> <span class="n">f</span><span class="p">:</span>
    <span class="n">f</span><span class="o">.</span><span class="n">write</span><span class="p">(</span><span class="n">json</span><span class="o">.</span><span class="n">dumps</span><span class="p">(</span><span class="n">dicts</span><span class="p">,</span> <span class="n">indent</span><span class="o">=</span><span class="mi">2</span><span class="p">,</span> <span class="n">ensure_ascii</span><span class="o">=</span><span class="kc">False</span><span class="p">))</span>

<span class="c1"># 3.从文件中读取出字典，然后将字典转换成消息</span>
<span class="c1"># 3.1 读取文件</span>
<span class="n">messages</span> <span class="o">=</span> <span class="n">json</span><span class="o">.</span><span class="n">load</span><span class="p">(</span><span class="nb">open</span><span class="p">(</span><span class="s1">&#39;history.json&#39;</span><span class="p">,</span> <span class="s1">&#39;r&#39;</span><span class="p">,</span> <span class="n">encoding</span><span class="o">=</span><span class="s1">&#39;utf-8&#39;</span><span class="p">))</span>
<span class="c1"># 3.2 然后将字典转换成消息</span>
<span class="n">chat_messages</span> <span class="o">=</span> <span class="n">messages_from_dict</span><span class="p">(</span><span class="n">messages</span><span class="p">)</span>
<span class="nb">print</span><span class="p">(</span><span class="sa">f</span><span class="s1">&#39;chat_messages--&gt;</span><span class="si">{</span><span class="n">chat_messages</span><span class="si">}</span><span class="s1">&#39;</span><span class="p">)</span>
</code></pre></div>
<p>运行结果：</p>
<p><img alt="image-20251003001546687" src="assets/image-20251003001546687.png" /></p>
<p>LangChain v1.0 版本对 Memory（记忆） 组件进行了深度优化与工程化重构，使其不仅支持多轮对话上下文管理，更成为构建生产级智能体（Agent）不可或缺的核心能力之一。在 LangChain 从“实验性框架”迈向“企业级平台”的关键转型中，Memory 模块被赋予了更清晰的职责边界、更灵活的策略机制以及更强的可扩展性，从而支撑复杂任务场景下的长期记忆、状态追踪与个性化交互。</p>
<h3 id="1-short-term-memory">1 短期记忆（Short-term Memory）<a class="headerlink" href="#1-short-term-memory" title="Permanent link">&para;</a></h3>
<p>短期记忆的本质是线程级状态管理]。在 LangGraph 中：</p>
<blockquote>
<p>短期记忆 = Agent 状态（State） + 检查点持久化（Checkpointer） + 线程标识（thread_id）。</p>
</blockquote>
<p><strong>（1）状态（State）</strong></p>
<p>通常是一个包含 messages 字段的字典或 Pydantic 模型（如 MessagesState 或自定义 CustomState），用于存储当前会话的所有消息、中间变量、工具调用结果等。</p>
<p><strong>（2）检查点器（Checkpointer）</strong></p>
<p>负责将状态序列化并持久化到内存、SQLite、PostgreSQL 等后端。每次状态变更（如新增一条消息）都会触发一次检查点保存。</p>
<p><strong>（3）会话ID（thread_id）</strong></p>
<p>作为会话的唯一标识符，确保不同用户或不同对话之间的状态完全隔离。即使多个用户并发交互，也不会发生记忆混淆。</p>
<div class="highlight"><pre><span></span><code><span class="kn">from</span><span class="w"> </span><span class="nn">langchain.agents</span><span class="w"> </span><span class="kn">import</span> <span class="n">create_agent</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">langgraph.checkpoint.memory</span><span class="w"> </span><span class="kn">import</span> <span class="n">InMemorySaver</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">langchain.messages</span><span class="w"> </span><span class="kn">import</span> <span class="n">HumanMessage</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">langchain_ollama</span><span class="w"> </span><span class="kn">import</span> <span class="n">OllamaLLM</span>
<span class="kn">import</span><span class="w"> </span><span class="nn">time</span>

<span class="c1"># 初始化大模型，这里使用的是 Qwen 的聊天模型</span>
<span class="n">model</span> <span class="o">=</span> <span class="n">OllamaLLM</span><span class="p">(</span><span class="n">model</span><span class="o">=</span><span class="s2">&quot;qwen2.5:7b&quot;</span><span class="p">)</span>

<span class="c1"># 创建内存保存器 (Checkpointer)</span>
<span class="c1"># 这是 LangGraph 的核心概念之一，用于在对话过程中保存和恢复状态（记忆）</span>
<span class="n">checkpointer</span> <span class="o">=</span> <span class="n">InMemorySaver</span><span class="p">()</span>

<span class="c1"># 定义 Agent 的配置</span>
<span class="c1"># &quot;thread_id&quot; (线程ID) 非常关键：它相当于对话的 ID，用于区分不同的会话记忆</span>
<span class="c1"># 使用时间戳作为 ID，确保每次运行脚本时的线程 ID 不同（但在同一次运行中复用）</span>
<span class="n">timestamp</span> <span class="o">=</span> <span class="n">time</span><span class="o">.</span><span class="n">time</span><span class="p">()</span>
<span class="n">config</span> <span class="o">=</span> <span class="p">{</span>
    <span class="s2">&quot;configurable&quot;</span><span class="p">:</span> <span class="p">{</span><span class="s2">&quot;thread_id&quot;</span><span class="p">:</span> <span class="sa">f</span><span class="s2">&quot;</span><span class="si">{</span><span class="nb">int</span><span class="p">(</span><span class="n">timestamp</span><span class="p">)</span><span class="si">}</span><span class="s2">&quot;</span><span class="p">}</span>
<span class="p">}</span>

<span class="c1"># 创建 Agent 实例</span>
<span class="c1"># model: 指定使用的底层大模型</span>
<span class="c1"># tools: 工具列表，这里为空，表示 Agent 只能进行对话，不能调用外部工具（如搜索、计算器等）</span>
<span class="c1"># checkpointer: 传入上面的内存保存器，赋予 Agent 记忆能力</span>
<span class="c1"># system_prompt: 设定 Agent 的角色和行为模式</span>
<span class="n">agent</span> <span class="o">=</span> <span class="n">create_agent</span><span class="p">(</span>
    <span class="n">model</span><span class="o">=</span><span class="n">model</span><span class="p">,</span>
    <span class="n">tools</span><span class="o">=</span><span class="p">[],</span>
    <span class="n">checkpointer</span><span class="o">=</span><span class="n">checkpointer</span><span class="p">,</span>
    <span class="n">system_prompt</span><span class="o">=</span><span class="s2">&quot;你是一个翻译官，擅长中英互译。&quot;</span>
<span class="p">)</span>

<span class="c1"># 第一轮对话输入</span>
<span class="n">input_1</span> <span class="o">=</span> <span class="p">{</span>
    <span class="s2">&quot;messages&quot;</span><span class="p">:</span> <span class="p">[</span><span class="n">HumanMessage</span><span class="p">(</span><span class="s2">&quot;你好，我叫小呆。&quot;</span><span class="p">)]</span>
<span class="p">}</span>

<span class="c1"># 调用 Agent (Response A)</span>
<span class="c1"># 使用 config，这意味着这次对话会记录在 thread_id 对应的内存中</span>
<span class="c1"># Agent 会记住 &quot;我叫小呆&quot; 这个信息</span>
<span class="n">response_a</span> <span class="o">=</span> <span class="n">agent</span><span class="o">.</span><span class="n">invoke</span><span class="p">(</span>
    <span class="nb">input</span><span class="o">=</span><span class="n">input_1</span><span class="p">,</span>
    <span class="n">config</span><span class="o">=</span><span class="n">config</span>
<span class="p">)</span>

<span class="c1"># 第二轮对话输入（测试是否记得之前的信息）</span>
<span class="n">input_2</span> <span class="o">=</span> <span class="p">{</span>
    <span class="s2">&quot;messages&quot;</span><span class="p">:</span> <span class="p">[</span><span class="n">HumanMessage</span><span class="p">(</span><span class="s2">&quot;你好，我叫什么？&quot;</span><span class="p">)]</span>
<span class="p">}</span>

<span class="c1"># 定义一个新的配置 (config_2)，使用完全不同的 thread_id</span>
<span class="c1"># 这代表开启了一个全新的、独立的对话会话</span>
<span class="n">config_2</span> <span class="o">=</span> <span class="p">{</span>
    <span class="s2">&quot;configurable&quot;</span><span class="p">:</span> <span class="p">{</span><span class="s2">&quot;thread_id&quot;</span><span class="p">:</span> <span class="sa">f</span><span class="s2">&quot;</span><span class="si">{</span><span class="nb">int</span><span class="p">(</span><span class="n">time</span><span class="o">.</span><span class="n">time</span><span class="p">())</span><span class="si">}</span><span class="s2">&quot;</span><span class="p">}</span>  <span class="c1"># 这里生成了一个新的时间戳 ID</span>
<span class="p">}</span>

<span class="c1"># 调用 Agent (Response B)</span>
<span class="c1"># 使用新的 config_2，Agent 看不到 config_1 中的历史记录</span>
<span class="c1"># 因此，Agent 应该不知道用户叫 &quot;小呆&quot;，因为它处于一个新的 &quot;线程&quot; 中</span>
<span class="n">response_b</span> <span class="o">=</span> <span class="n">agent</span><span class="o">.</span><span class="n">invoke</span><span class="p">(</span>
    <span class="nb">input</span><span class="o">=</span><span class="n">input_2</span><span class="p">,</span>
    <span class="n">config</span><span class="o">=</span><span class="n">config_2</span>
<span class="p">)</span>

<span class="c1"># 调用 Agent (Response C)</span>
<span class="c1"># 再次使用最初的 config</span>
<span class="c1"># 这会恢复到第一个会话的记忆中</span>
<span class="c1"># Agent 应该能回想起用户叫 &quot;小呆&quot;，因为它共享同一个 thread_id</span>
<span class="n">response_c</span> <span class="o">=</span> <span class="n">agent</span><span class="o">.</span><span class="n">invoke</span><span class="p">(</span>
    <span class="nb">input</span><span class="o">=</span><span class="n">input_2</span><span class="p">,</span>
    <span class="n">config</span><span class="o">=</span><span class="n">config</span>
<span class="p">)</span>

<span class="c1"># 输出结果进行对比</span>
<span class="nb">print</span><span class="p">(</span><span class="s2">&quot;Response A (第一次对话):&quot;</span><span class="p">)</span>
<span class="nb">print</span><span class="p">(</span><span class="n">response_a</span><span class="p">)</span>

<span class="nb">print</span><span class="p">(</span><span class="s2">&quot;</span><span class="se">\n</span><span class="s2">Response B (新线程，无记忆):&quot;</span><span class="p">)</span>
<span class="nb">print</span><span class="p">(</span><span class="n">response_b</span><span class="p">)</span>

<span class="nb">print</span><span class="p">(</span><span class="s2">&quot;</span><span class="se">\n</span><span class="s2">Response C (回到原线程，有记忆):&quot;</span><span class="p">)</span>
<span class="nb">print</span><span class="p">(</span><span class="n">response_c</span><span class="p">)</span>
</code></pre></div>
<p><strong>代码运行逻辑：</strong></p>
<ul>
<li><strong>checkpointer (记忆核心)</strong>：这是 LangGraph 区别于传统简单 API 调用的关键。它允许 Agent 在多次 invoke 调用之间保存状态。</li>
<li><strong>thread_id (记忆索引)</strong>：你可以把 thread_id 想象成数据库中的主键。</li>
<li>Response A：向 ID 为 100 的记录里写入了名字。</li>
<li>Response B：向 ID 为 200 的新记录提问，因为 200 是空的，所以 Agent 不知道名字。</li>
<li>Response C：又回到 ID 为 100 的记录提问，Agent 读取了历史记录，所以知道名字。</li>
<li><strong>System Prompt</strong>：虽然输入是在对话，但因为设定了 “你是一个翻译官”，Agent 可能会试图在回答的同时进行翻译，或者在回复格式上符合翻译官的身份。</li>
</ul>
<h3 id="2_2">2 长期记忆<a class="headerlink" href="#2_2" title="Permanent link">&para;</a></h3>
<p>对于需要长期运行和可靠记忆的应用，推荐使用数据库进行持久化。详见后面阶段LangGraph部分。</p>
<h2 id="indexes">七、Indexes<a class="headerlink" href="#indexes" title="Permanent link">&para;</a></h2>
<p>Indexes组件的目的是让LangChain具备处理文档处理的能力，包括：文档加载、检索等。注意，这里的文档不局限于txt、pdf等文本类内容，还涵盖email、区块链、视频等内容。</p>
<p>Indexes组件主要包含类型：</p>
<ul>
<li>文档加载器</li>
<li>文本分割器</li>
<li>VectorStores</li>
<li>检索器</li>
</ul>
<h3 id="1_1">1 文档加载器<a class="headerlink" href="#1_1" title="Permanent link">&para;</a></h3>
<blockquote>
<p><a href="https://python.langchain.com/v0.2/docs/introduction/">https://python.langchain.com/v0.2/docs/introduction/</a></p>
</blockquote>
<p>文档加载器主要基于<code>Unstructured</code> 包，<code>Unstructured</code> 是一个python包，可以把各种类型的文件转换成文本。文档加载器使用起来很简单，只需要引入相应的loader工具。</p>
<p>LangChain支持的文档加载器 (部分)：</p>
<p><img alt="image-20250705100227413" src="assets/image-20250705100227413.png" /></p>
<p>示例代码：</p>
<blockquote>
<p>需要安装的包：</p>
<p>pip install unstructured</p>
<p>pip install langchain-unstructured</p>
</blockquote>
<div class="highlight"><pre><span></span><code><span class="kn">from</span><span class="w"> </span><span class="nn">langchain_unstructured</span><span class="w"> </span><span class="kn">import</span> <span class="n">UnstructuredLoader</span>

<span class="c1"># 创建 UnstructuredLoader 对象</span>
<span class="n">loader</span> <span class="o">=</span> <span class="n">UnstructuredLoader</span><span class="p">(</span><span class="s1">&#39;./data/衣服属性.txt&#39;</span><span class="p">,</span> <span class="n">encoding</span><span class="o">=</span><span class="s1">&#39;utf8&#39;</span><span class="p">)</span>
<span class="n">docs</span> <span class="o">=</span> <span class="n">loader</span><span class="o">.</span><span class="n">load</span><span class="p">()</span>
<span class="nb">print</span><span class="p">(</span><span class="sa">f</span><span class="s1">&#39;docs--&gt;</span><span class="si">{</span><span class="n">docs</span><span class="si">}</span><span class="s1">&#39;</span><span class="p">)</span>
<span class="nb">print</span><span class="p">(</span><span class="sa">f</span><span class="s1">&#39;len--&gt;</span><span class="si">{</span><span class="nb">len</span><span class="p">(</span><span class="n">docs</span><span class="p">)</span><span class="si">}</span><span class="s1">&#39;</span><span class="p">)</span>
<span class="nb">print</span><span class="p">(</span><span class="sa">f</span><span class="s1">&#39;第一行数据--&gt;</span><span class="si">{</span><span class="n">docs</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span><span class="o">.</span><span class="n">page_content</span><span class="si">}</span><span class="s1">&#39;</span><span class="p">)</span>
<span class="nb">print</span><span class="p">(</span><span class="s1">&#39;*&#39;</span> <span class="o">*</span> <span class="mi">100</span><span class="p">)</span>

<span class="kn">from</span><span class="w"> </span><span class="nn">langchain_community.document_loaders</span><span class="w"> </span><span class="kn">import</span> <span class="n">TextLoader</span>

<span class="c1"># 创建 TextLoader 对象</span>
<span class="n">loader</span> <span class="o">=</span> <span class="n">TextLoader</span><span class="p">(</span><span class="s1">&#39;./data/衣服属性.txt&#39;</span><span class="p">,</span> <span class="n">encoding</span><span class="o">=</span><span class="s1">&#39;utf8&#39;</span><span class="p">)</span>
<span class="n">docs</span> <span class="o">=</span> <span class="n">loader</span><span class="o">.</span><span class="n">load</span><span class="p">()</span>
<span class="nb">print</span><span class="p">(</span><span class="sa">f</span><span class="s1">&#39;docs--&gt;</span><span class="si">{</span><span class="n">docs</span><span class="si">}</span><span class="s1">&#39;</span><span class="p">)</span>
<span class="nb">print</span><span class="p">(</span><span class="sa">f</span><span class="s1">&#39;len--&gt;</span><span class="si">{</span><span class="nb">len</span><span class="p">(</span><span class="n">docs</span><span class="p">)</span><span class="si">}</span><span class="s1">&#39;</span><span class="p">)</span>
<span class="nb">print</span><span class="p">(</span><span class="s1">&#39;第一行数据--&gt;</span><span class="si">{}</span><span class="s1">&#39;</span><span class="o">.</span><span class="n">format</span><span class="p">(</span><span class="n">docs</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span><span class="o">.</span><span class="n">page_content</span><span class="o">.</span><span class="n">split</span><span class="p">(</span><span class="s1">&#39;</span><span class="se">\n</span><span class="s1">&#39;</span><span class="p">)[</span><span class="mi">0</span><span class="p">]))</span>
</code></pre></div>
<p>运行结果：</p>
<p><img alt="image-20251003010445727" src="assets/image-20251003010445727.png" /></p>
<h3 id="2_3">2 文档分割器<a class="headerlink" href="#2_3" title="Permanent link">&para;</a></h3>
<p>由于模型对输入的字符长度有限制，我们在碰到很长的文本时，需要把文本分割成多个小的文本片段。</p>
<p>文本分割最简单的方式是按照字符长度进行分割，但是这会带来很多问题，比如说如果文本是一段代码，一个函数被分割到两段之后就成了没有意义的字符，所以整体的原则是把语义相关的文本片段放在一起。</p>
<p>LangChain中最基本的文本分割器是<code>CharacterTextSplitter</code> ，它按照指定的分隔符（默认“\n\n”）进行分割，并且考虑文本片段的最大长度。我们看个例子：</p>
<div class="highlight"><pre><span></span><code><span class="kn">from</span><span class="w"> </span><span class="nn">langchain_core.documents</span><span class="w"> </span><span class="kn">import</span> <span class="n">Document</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">langchain_text_splitters</span><span class="w"> </span><span class="kn">import</span> <span class="n">CharacterTextSplitter</span>

<span class="c1"># 创建分词器 separator参数指的是分割的分隔符，chunk_size指的是分割出来的每个块的大小，chunk_overlap指的是每个块之间重叠的大小</span>
<span class="n">text_splitter</span> <span class="o">=</span> <span class="n">CharacterTextSplitter</span><span class="p">(</span><span class="n">separator</span><span class="o">=</span><span class="s2">&quot; &quot;</span><span class="p">,</span> <span class="n">chunk_size</span><span class="o">=</span><span class="mi">5</span><span class="p">,</span> <span class="n">chunk_overlap</span><span class="o">=</span><span class="mi">1</span><span class="p">)</span>

<span class="c1"># 一句话分割</span>
<span class="n">result1</span> <span class="o">=</span> <span class="n">text_splitter</span><span class="o">.</span><span class="n">split_text</span><span class="p">(</span><span class="s2">&quot;a b c d e f&quot;</span><span class="p">)</span>
<span class="nb">print</span><span class="p">(</span><span class="sa">f</span><span class="s1">&#39;result1---&gt;</span><span class="si">{</span><span class="n">result1</span><span class="si">}</span><span class="s1">&#39;</span><span class="p">)</span>

<span class="c1"># 多句话分割</span>
<span class="n">result2</span> <span class="o">=</span> <span class="n">text_splitter</span><span class="o">.</span><span class="n">create_documents</span><span class="p">([</span><span class="s2">&quot;a b c d e f&quot;</span><span class="p">,</span> <span class="s2">&quot;e f g h&quot;</span><span class="p">])</span>
<span class="nb">print</span><span class="p">(</span><span class="sa">f</span><span class="s1">&#39;result2---&gt;</span><span class="si">{</span><span class="n">result2</span><span class="si">}</span><span class="s1">&#39;</span><span class="p">)</span>

<span class="c1"># 多句话分割</span>
<span class="n">result3</span> <span class="o">=</span> <span class="n">text_splitter</span><span class="o">.</span><span class="n">split_documents</span><span class="p">([</span><span class="n">Document</span><span class="p">(</span><span class="n">page_content</span><span class="o">=</span><span class="s2">&quot;a b c d e f&quot;</span><span class="p">,</span> <span class="n">metadata</span><span class="o">=</span><span class="p">{</span><span class="s2">&quot;id&quot;</span><span class="p">:</span> <span class="s2">&quot;1&quot;</span><span class="p">})])</span>
<span class="nb">print</span><span class="p">(</span><span class="sa">f</span><span class="s1">&#39;result3---&gt;</span><span class="si">{</span><span class="n">result3</span><span class="si">}</span><span class="s1">&#39;</span><span class="p">)</span>
</code></pre></div>
<p>运行结果：</p>
<p><img alt="image-20251003011030583" src="assets/image-20251003011030583.png" /></p>
<hr />
<p>除了CharacterTextSplitter分割器，LangChain还支持其他文档分割器 (部分)：</p>
<table>
<thead>
<tr>
<th>分割器名称</th>
<th>功能描述</th>
<th>类型</th>
<th>工业场景应用</th>
</tr>
</thead>
<tbody>
<tr>
<td>CharacterTextSplitter</td>
<td>简单按指定分隔符（如换行、逗号）直接分割。</td>
<td>基础字符解析</td>
<td>简单字符串或 CSV 数据处理，如传感器数据日志。</td>
</tr>
<tr>
<td>RecursiveCharacterTextSplitter</td>
<td>递归按字符分割，先尝试自然边界（如段落、句子），太大则继续细分。</td>
<td>通用字符解析</td>
<td>通用文本处理，如日志、报告、PDF 文档分割，便于 RAG 检索。</td>
</tr>
<tr>
<td>TokenTextSplitter</td>
<td>按 token（词元）分割，支持 LLM token 计数。</td>
<td>Token 基于解析</td>
<td>LLM 输入优化，如处理 API 响应或长查询，控制 token 限制。</td>
</tr>
<tr>
<td>SentenceTextSplitter</td>
<td>按句子边界分割，使用 NLP 识别句子（包括标点）。</td>
<td>语义解析</td>
<td>自然语言文本，如文章或对话分析，保持句子完整。</td>
</tr>
<tr>
<td>SpacyTextSplitter</td>
<td>使用 SpaCy NLP 库按句子或实体分割（需安装 SpaCy）。</td>
<td>语义解析</td>
<td>高级 NLP 场景，如实体提取或生物医学文本。</td>
</tr>
<tr>
<td>NLTKTextSplitter</td>
<td>使用 NLTK 库按句子或词分割（需安装 NLTK）。</td>
<td>语义解析</td>
<td>文本研究或分析，如时间序列数据描述。</td>
</tr>
<tr>
<td>MarkdownHeaderTextSplitter</td>
<td>按 Markdown 结构（如标题、列表）智能分割。</td>
<td>结构化解析</td>
<td>Markdown 文档分割，保留语义结构，用于知识库构建。</td>
</tr>
<tr>
<td>HTMLSplitter</td>
<td>按 HTML 标签（如 、）分割网页内容。</td>
<td>结构化解析</td>
<td>网页数据爬取，如在线技术文档或新闻提取。</td>
</tr>
<tr>
<td>LatexTextSplitter</td>
<td>按 LaTeX 结构（如章节、公式）分割。</td>
<td>结构化解析</td>
<td>学术论文或数学文档处理。</td>
</tr>
<tr>
<td>PythonCodeTextSplitter</td>
<td>按 Python 代码结构（如函数、类）分割。</td>
<td>代码解析</td>
<td>源代码文件分析，如脚本调试或代码库管理。</td>
</tr>
</tbody>
</table>
<p>下面就几个重要的分割器进行讲解。</p>
<ul>
<li>递归字符文本分割器（RecursiveCharacterTextSplitter）</li>
</ul>
<blockquote>
<p>递归字符文本分割器是一种更智能的分割方法，它尝试在特定分隔符处分割文本，以保持更好的语义完整性。 特点：</p>
<ul>
<li>尝试在自然断点处分割文本</li>
<li>比简单的字符分割更能保持语义完整性</li>
<li>适用于结构化程度较高的文本，如 Markdown、HTML 等</li>
</ul>
<p>运行流程：</p>
<ul>
<li>首先尝试使用第一个分隔符（如 "\n\n"）分割文本</li>
<li>如果分割后的块仍然过大，则使用下一个分隔符继续分割</li>
<li>重复此过程，直到达到指定的 chunk_size 或用完所有分隔符</li>
</ul>
<p>示例代码</p>
<div class="highlight"><pre><span></span><code><span class="kn">from</span><span class="w"> </span><span class="nn">langchain_text_splitters</span><span class="w"> </span><span class="kn">import</span> <span class="n">RecursiveCharacterTextSplitter</span>

<span class="n">text_splitter</span> <span class="o">=</span> <span class="n">RecursiveCharacterTextSplitter</span><span class="p">(</span>
    <span class="n">chunk_size</span><span class="o">=</span><span class="mi">20</span><span class="p">,</span>  <span class="c1"># 每个块最多 100 个字符</span>
    <span class="n">chunk_overlap</span><span class="o">=</span><span class="mi">6</span><span class="p">,</span>  <span class="c1"># 相邻分块会共享 20 个字符</span>
    <span class="n">length_function</span><span class="o">=</span><span class="nb">len</span><span class="p">,</span>  <span class="c1"># 用字符数来衡量长度</span>
    <span class="n">separators</span><span class="o">=</span><span class="p">[</span><span class="s2">&quot;</span><span class="se">\n\n</span><span class="s2">&quot;</span><span class="p">,</span> <span class="s2">&quot;</span><span class="se">\n</span><span class="s2">&quot;</span><span class="p">,</span> <span class="s2">&quot; &quot;</span><span class="p">,</span> <span class="s2">&quot;&quot;</span><span class="p">]</span>  <span class="c1"># 会优先尝试按 \n\n 分段，如果太长，再按 \n → 空格 → 逐字符切分。</span>
<span class="p">)</span>

<span class="n">text</span> <span class="o">=</span> <span class="s2">&quot;&quot;&quot;</span>
<span class="s2">人工智能正在快速发展，尤其是大语言模型的应用，正在改变人类的工作方式。</span>
<span class="s2">它们可以帮助人们进行写作、代码生成、甚至是科研探索。</span>
<span class="s2">相比之下，新能源的发展同样重要。</span>
<span class="s2">电动车和太阳能正在逐渐替代传统能源，减少碳排放，对全球环境保护至关重要。</span>
<span class="s2">&quot;&quot;&quot;</span>
<span class="n">docs</span> <span class="o">=</span> <span class="n">text_splitter</span><span class="o">.</span><span class="n">split_text</span><span class="p">(</span><span class="n">text</span><span class="p">)</span>
<span class="nb">print</span><span class="p">(</span><span class="n">docs</span><span class="p">)</span>
</code></pre></div>
<p>运行结果</p>
<p><img alt="1768393012466" src="assets/1768393012466.png" /></p>
</blockquote>
<ul>
<li>语义文档分割器（SemanticChunker）</li>
</ul>
<blockquote>
<p>语义文档分割器使用语义理解来分割文本，这是一种更高级的分割方法。 特点：</p>
<ul>
<li>基于语义相似性分割文本</li>
<li>能够更好地保持语义完整性</li>
<li>计算成本较高，处理大量文本时可能效率较低</li>
<li>适用于需要高度语义理解的场景</li>
</ul>
<p>示例代码</p>
<div class="highlight"><pre><span></span><code><span class="kn">from</span><span class="w"> </span><span class="nn">langchain_experimental.text_splitter</span><span class="w"> </span><span class="kn">import</span> <span class="n">SemanticChunker</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">langchain_ollama</span><span class="w"> </span><span class="kn">import</span> <span class="n">OllamaEmbeddings</span>

<span class="n">embed</span> <span class="o">=</span> <span class="n">OllamaEmbeddings</span><span class="p">(</span><span class="n">model</span><span class="o">=</span><span class="s2">&quot;mxbai-embed-large&quot;</span><span class="p">)</span>

<span class="n">text_splitter</span> <span class="o">=</span> <span class="n">SemanticChunker</span><span class="p">(</span>
     <span class="n">embeddings</span><span class="o">=</span><span class="n">embed</span><span class="p">,</span>
     <span class="n">breakpoint_threshold_type</span><span class="o">=</span><span class="s1">&#39;percentile&#39;</span><span class="p">,</span>
     <span class="c1"># 更低的百分位会更“积极”地切分（数值越小 =&gt; 切得越多）</span>
     <span class="n">breakpoint_threshold_amount</span><span class="o">=</span><span class="mf">70.0</span><span class="p">,</span>
     <span class="c1"># 句子拆分正则为同时识别中/英文终结符</span>
     <span class="n">sentence_split_regex</span><span class="o">=</span><span class="sa">r</span><span class="s1">&#39;(?&lt;=[。！？.!?])\s*&#39;</span><span class="p">,</span>
     <span class="c1"># 每个切分后片段（chunk）至少要有多少个字符</span>
     <span class="n">min_chunk_size</span><span class="o">=</span><span class="mi">10</span>
    <span class="p">)</span>

<span class="n">text</span> <span class="o">=</span> <span class="s2">&quot;&quot;&quot;</span>
<span class="s2">人工智能正在快速发展，尤其是大语言模型的应用，正在改变人类的工作方式。</span>
<span class="s2">它们可以帮助人们进行写作、代码生成、甚至是科研探索。</span>
<span class="s2">相比之下，新能源的发展同样重要。</span>
<span class="s2">电动车和太阳能正在逐渐替代传统能源，减少碳排放，对全球环境保护至关重要。</span>
<span class="s2">&quot;&quot;&quot;</span>

<span class="n">docs</span> <span class="o">=</span> <span class="n">text_splitter</span><span class="o">.</span><span class="n">split_text</span><span class="p">(</span><span class="n">text</span><span class="p">)</span>
<span class="k">for</span> <span class="n">i</span><span class="p">,</span> <span class="n">d</span> <span class="ow">in</span> <span class="nb">enumerate</span><span class="p">(</span><span class="n">docs</span><span class="p">):</span>
     <span class="nb">print</span><span class="p">(</span><span class="sa">f</span><span class="s2">&quot;------ Chunk </span><span class="si">{</span><span class="n">i</span><span class="o">+</span><span class="mi">1</span><span class="si">}</span><span class="s2"> ------&quot;</span><span class="p">)</span>
     <span class="nb">print</span><span class="p">(</span><span class="n">d</span><span class="o">.</span><span class="n">strip</span><span class="p">())</span>
     <span class="nb">print</span><span class="p">()</span>
</code></pre></div>
<p>运行结果</p>
<p><img alt="image-20251003095433705" src="assets/image-20251003095433705.png" /></p>
</blockquote>
<ul>
<li>MarkdownHeaderTextSplitter（Markdown文档切割器）</li>
</ul>
<blockquote>
<p>适用于Markdown文档，按照标题进行拆分</p>
<p>示例代码</p>
<div class="highlight"><pre><span></span><code><span class="kn">from</span><span class="w"> </span><span class="nn">langchain_text_splitters</span><span class="w"> </span><span class="kn">import</span> <span class="n">MarkdownHeaderTextSplitter</span>

<span class="n">headers_to_split_on</span> <span class="o">=</span> <span class="p">[</span>
<span class="p">(</span><span class="s2">&quot;#&quot;</span><span class="p">,</span> <span class="s2">&quot;Header 1&quot;</span><span class="p">),</span>
<span class="p">(</span><span class="s2">&quot;##&quot;</span><span class="p">,</span> <span class="s2">&quot;Header 2&quot;</span><span class="p">),</span>
<span class="p">(</span><span class="s2">&quot;###&quot;</span><span class="p">,</span> <span class="s2">&quot;Header 3&quot;</span><span class="p">),</span>
<span class="p">]</span>

<span class="n">markdown_splitter</span> <span class="o">=</span> <span class="n">MarkdownHeaderTextSplitter</span><span class="p">(</span><span class="n">headers_to_split_on</span><span class="o">=</span><span class="n">headers_to_split_on</span><span class="p">)</span>

<span class="n">markdown_text</span> <span class="o">=</span> <span class="s2">&quot;# Header 1</span><span class="se">\n</span><span class="s2">Some text</span><span class="se">\n</span><span class="s2">## Header 2</span><span class="se">\n</span><span class="s2">More text</span><span class="se">\n</span><span class="s2">### Header 3</span><span class="se">\n</span><span class="s2">Even more text&quot;</span>
<span class="n">docs</span> <span class="o">=</span> <span class="n">markdown_splitter</span><span class="o">.</span><span class="n">split_text</span><span class="p">(</span><span class="n">markdown_text</span><span class="p">)</span>
<span class="nb">print</span><span class="p">(</span><span class="n">docs</span><span class="p">)</span>
</code></pre></div>
<p>运行结果</p>
<p><img alt="image-20251003100510115" src="assets/image-20251003100510115.png" /></p>
</blockquote>
<p>其他拓展知识可以参考：<a href="https://blog.csdn.net/qq_28540861/article/details/149161419">https://blog.csdn.net/qq_28540861/article/details/149161419</a></p>
<h3 id="3-vectorstores">3 VectorStores<a class="headerlink" href="#3-vectorstores" title="Permanent link">&para;</a></h3>
<p>VectorStores是一种特殊类型的数据库，它的作用是存储由嵌入创建的向量，提供相似查询等功能。</p>
<p>LangChain支持的VectorStore有<a href="https://python.langchain.com/docs/integrations/vectorstores/">https://python.langchain.com/docs/integrations/vectorstores/</a>，常见的如下：</p>
<p><img alt="image-20250705114738060" src="assets/image-20250705114738060.png" /></p>
<blockquote>
<p>我们使用其中一个<code>Chroma</code> 组件作为例子：</p>
<p>pip install chromadb</p>
<p>pip install langchain-chroma</p>
</blockquote>
<div class="highlight"><pre><span></span><code><span class="kn">from</span><span class="w"> </span><span class="nn">langchain_text_splitters</span><span class="w"> </span><span class="kn">import</span> <span class="n">CharacterTextSplitter</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">langchain_chroma</span><span class="w"> </span><span class="kn">import</span> <span class="n">Chroma</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">langchain_community.document_loaders</span><span class="w"> </span><span class="kn">import</span> <span class="n">TextLoader</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">langchain_ollama</span><span class="w"> </span><span class="kn">import</span> <span class="n">OllamaEmbeddings</span>

<span class="c1"># 1.加载文档</span>
<span class="c1"># 创建 TextLoader 对象</span>
<span class="n">loader</span> <span class="o">=</span> <span class="n">TextLoader</span><span class="p">(</span><span class="s1">&#39;./data/pku.txt&#39;</span><span class="p">,</span> <span class="n">encoding</span><span class="o">=</span><span class="s1">&#39;utf-8&#39;</span><span class="p">)</span>
<span class="c1"># 加载文档</span>
<span class="n">docs</span> <span class="o">=</span> <span class="n">loader</span><span class="o">.</span><span class="n">load</span><span class="p">()</span>
<span class="c1"># print(f&#39;docs--&gt;{docs}&#39;)</span>

<span class="c1"># 2.将文档进行分块</span>
<span class="c1"># 创建 CharacterTextSplitter 对象</span>
<span class="n">text_splitter</span> <span class="o">=</span> <span class="n">CharacterTextSplitter</span><span class="p">(</span><span class="n">separator</span><span class="o">=</span><span class="s2">&quot;</span><span class="se">\n\n</span><span class="s2">&quot;</span><span class="p">,</span> <span class="n">chunk_size</span><span class="o">=</span><span class="mi">200</span><span class="p">,</span> <span class="n">chunk_overlap</span><span class="o">=</span><span class="mi">30</span><span class="p">)</span>
<span class="c1"># 分块</span>
<span class="n">split_docs</span> <span class="o">=</span> <span class="n">text_splitter</span><span class="o">.</span><span class="n">split_documents</span><span class="p">(</span><span class="n">docs</span><span class="p">)</span>
<span class="nb">print</span><span class="p">(</span><span class="sa">f</span><span class="s1">&#39;split_docs--&gt;</span><span class="si">{</span><span class="n">split_docs</span><span class="si">}</span><span class="s1">&#39;</span><span class="p">)</span>

<span class="c1"># 3.将分割后的文档存储到向量数据库中</span>
<span class="c1"># 加载embedding模型</span>
<span class="n">embedding</span> <span class="o">=</span> <span class="n">OllamaEmbeddings</span><span class="p">(</span><span class="n">model</span><span class="o">=</span><span class="s2">&quot;mxbai-embed-large&quot;</span><span class="p">)</span>
<span class="c1"># 创建向量数据库，需要指定 存储的文档和向量模型名称以及持久化目录</span>
<span class="n">chromadaDB</span> <span class="o">=</span> <span class="n">Chroma</span><span class="o">.</span><span class="n">from_documents</span><span class="p">(</span><span class="n">documents</span><span class="o">=</span><span class="n">split_docs</span><span class="p">,</span>
                                   <span class="n">embedding</span><span class="o">=</span><span class="n">embedding</span><span class="p">,</span>
                                   <span class="n">persist_directory</span><span class="o">=</span><span class="s1">&#39;./chroma_db&#39;</span><span class="p">)</span>

<span class="c1"># 假如你的向量数据库已经存在，那么可以直接加载</span>
<span class="c1"># chromadaDB = Chroma(persist_directory=&#39;./chroma_db&#39;, embedding_function=embedding)</span>

<span class="c1"># 4.使用向量数据库进行查询</span>
<span class="n">query</span> <span class="o">=</span> <span class="s2">&quot;1937年北京大学发生了什么？&quot;</span>
<span class="n">result</span> <span class="o">=</span> <span class="n">chromadaDB</span><span class="o">.</span><span class="n">similarity_search</span><span class="p">(</span><span class="n">query</span><span class="p">,</span> <span class="n">k</span><span class="o">=</span><span class="mi">2</span><span class="p">)</span>
<span class="nb">print</span><span class="p">(</span><span class="sa">f</span><span class="s1">&#39;result--&gt;</span><span class="si">{</span><span class="n">result</span><span class="si">}</span><span class="s1">&#39;</span><span class="p">)</span>
</code></pre></div>
<p>运行结果：</p>
<p><img alt="image-20251003101141993" src="assets/image-20251003101141993.png" /></p>
<h3 id="4_1">4 检索器<a class="headerlink" href="#4_1" title="Permanent link">&para;</a></h3>
<h4 id="41-langchain">4.1 LangChain的检索器定义<a class="headerlink" href="#41-langchain" title="Permanent link">&para;</a></h4>
<p>检索器是 LangChain 中负责信息检索的模块，通常与 索引（Indexes） 模块（如向量存储、嵌入模型）结合使用。它的核心功能是：</p>
<ul>
<li>输入：接收用户查询（通常是文本）。</li>
<li>处理：根据查询从数据源中检索相关内容。</li>
<li>输出：返回一组相关文档或文本片段（通常是 Document 对象列表）。</li>
</ul>
<p>检索器在以下场景中扮演关键角色：</p>
<ul>
<li>问答系统：从文档或知识库中检索答案的上下文。</li>
<li>语义搜索：根据查询的语义返回相关结果。</li>
<li>上下文增强：为语言模型提供外部知识，解决其知识局限。</li>
</ul>
<h4 id="42">4.2 检索器的工作原理<a class="headerlink" href="#42" title="Permanent link">&para;</a></h4>
<p>检索器通常与 向量存储（Vector Stores） 配合，通过嵌入模型（Embedding Models）将查询和文档转为向量，基于相似性进行检索。工作流程可以分为以下步骤：</p>
<ul>
<li>查询嵌入：将用户查询通过嵌入模型（如 OpenAIEmbeddings）转为向量表示</li>
<li>相似性搜索：在向量存储中查找与查询向量最相似的文档向量。</li>
<li>文档返回：返回匹配的文档（包含内容、元数据等）。</li>
<li>后处理（可选）：对检索结果进行排序、过滤或重新排名。</li>
</ul>
<p>检索器的核心依赖：</p>
<ul>
<li>嵌入模型：将文本转为向量（如 OpenAIEmbeddings, HuggingFaceEmbeddings）。</li>
<li>向量存储：存储文档向量（如 Chroma、FAISS、Pinecone）。</li>
<li>相似性度量：如余弦相似度、欧几里得距离</li>
</ul>
<h4 id="43">4.3 检索器类型<a class="headerlink" href="#43" title="Permanent link">&para;</a></h4>
<p>langchain支持很多检索器<a href="https://python.langchain.com/docs/integrations/retrievers/">https://python.langchain.com/docs/integrations/retrievers/</a>，部分如下：</p>
<p><img alt="image-20250705114709874" src="assets/image-20250705114709874-17594578225791.png" /></p>
<p>此处我们讲解VectorStoreRetriever。</p>
<p>在 LangChain 中，as_retriever() 方法的 search_type 参数决定了向量检索的具体算法和行为。</p>
<div class="highlight"><pre><span></span><code><span class="n">retriever</span> <span class="o">=</span> <span class="n">vector_store</span><span class="o">.</span><span class="n">as_retriever</span><span class="p">(</span>
    <span class="n">search_type</span><span class="o">=</span><span class="s2">&quot;similarity&quot;</span><span class="p">,</span>  <span class="c1"># 可选 &quot;similarity&quot;|&quot;mmr&quot;|&quot;similarity_score_threshold&quot;</span>
    <span class="n">search_kwargs</span><span class="o">=</span><span class="p">{</span>
        <span class="s2">&quot;k&quot;</span><span class="p">:</span> <span class="mi">5</span><span class="p">,</span>  <span class="c1"># 返回结果数量</span>
        <span class="s2">&quot;score_threshold&quot;</span><span class="p">:</span> <span class="mf">0.7</span><span class="p">,</span>  <span class="c1"># 仅当search_type=&quot;similarity_score_threshold&quot;时有效,低于阈值的都丢弃。</span>
        <span class="s2">&quot;filter&quot;</span><span class="p">:</span> <span class="p">{</span><span class="s2">&quot;source&quot;</span><span class="p">:</span> <span class="s2">&quot;重要文档.pdf&quot;</span><span class="p">},</span>  <span class="c1"># 元数据过滤，只会检索满足条件的文档。</span>
        <span class="s2">&quot;lambda_mult&quot;</span><span class="p">:</span> <span class="mf">0.25</span>  <span class="c1"># 仅MMR搜索有效(控制多样性)：接近 0 则更强调和查询的相关性；接近 1 则更强调结果之间的差异性</span>
    <span class="p">}</span>
<span class="p">)</span>
</code></pre></div>
<p>以下是三种搜索类型的对比：</p>
<p><img alt="image-20250714160215042" src="assets/image-20250714160215042-17594581318012.png" /></p>
<p>示例代码：</p>
<div class="highlight"><pre><span></span><code><span class="kn">from</span><span class="w"> </span><span class="nn">langchain_chroma</span><span class="w"> </span><span class="kn">import</span> <span class="n">Chroma</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">langchain_ollama</span><span class="w"> </span><span class="kn">import</span> <span class="n">OllamaEmbeddings</span>

<span class="n">embedding</span> <span class="o">=</span> <span class="n">OllamaEmbeddings</span><span class="p">(</span><span class="n">model</span><span class="o">=</span><span class="s2">&quot;mxbai-embed-large&quot;</span><span class="p">)</span>

<span class="c1"># 向量数据库已经存在，那么可以直接加载</span>
<span class="n">chromadaDB</span> <span class="o">=</span> <span class="n">Chroma</span><span class="p">(</span><span class="n">persist_directory</span><span class="o">=</span><span class="s1">&#39;./chroma_db&#39;</span><span class="p">,</span> <span class="n">embedding_function</span><span class="o">=</span><span class="n">embedding</span><span class="p">)</span>

<span class="c1"># 使用向量数据库进行查询</span>
<span class="n">query</span> <span class="o">=</span> <span class="s2">&quot;1937年北京大学发生了什么？&quot;</span>
<span class="c1"># result = chromadaDB.similarity_search(query, k=2)</span>

<span class="c1"># 使用 as_retriever 方法返回 Retriever 对象，然后调用 invoke 方法进行查询</span>
<span class="n">retriever</span> <span class="o">=</span> <span class="n">chromadaDB</span><span class="o">.</span><span class="n">as_retriever</span><span class="p">(</span><span class="n">search_kwargs</span><span class="o">=</span><span class="p">{</span><span class="s2">&quot;k&quot;</span><span class="p">:</span><span class="mi">2</span><span class="p">})</span>
<span class="n">result</span> <span class="o">=</span> <span class="n">retriever</span><span class="o">.</span><span class="n">invoke</span><span class="p">(</span><span class="n">query</span><span class="p">)</span>
<span class="nb">print</span><span class="p">(</span><span class="sa">f</span><span class="s1">&#39;result--&gt;</span><span class="si">{</span><span class="n">result</span><span class="si">}</span><span class="s1">&#39;</span><span class="p">)</span>
</code></pre></div>
<p>运行结果：</p>
<p><img alt="image-20251003102843558" src="assets/image-20251003102843558.png" /></p>
<p><strong>拓展：</strong></p>
<p><code>vectordb.as_retriever()</code> 和 <code>vectordb.similarity_search()</code> 都是用于从向量数据库中检索相关文档的方法，它们有什么异同：</p>
<ul>
<li>
<p>相同</p>
<ul>
<li>核心功能：两者都基于向量相似度（如余弦相似度）从向量数据库中检索与查询最相关的文档。</li>
<li>底层技术：通常使用相同的嵌入模型和相似度计算方式（如 FAISS、Chroma、Pinecone 等）。</li>
</ul>
</li>
<li>
<p>不同</p>
<p><img alt="image-20250705114109276" src="assets/image-20250705114109276.png" /></p>
</li>
</ul>
<h4 id="44">4.4 拓展<a class="headerlink" href="#44" title="Permanent link">&para;</a></h4>
<p>其他几种常用的检索器介绍如下。</p>
<ul>
<li>TFIDFRetriever</li>
</ul>
<blockquote>
<p>功能：基于 TF-IDF（词频-逆文档频率）的检索器。 </p>
<p>特点：</p>
<ul>
<li>使用 TF-IDF 向量表示文档和查询。</li>
<li>适合快速构建原型。</li>
<li>不支持语义搜索。</li>
</ul>
<p>适用场景：</p>
<ul>
<li>文本搜索</li>
<li>关键词提取</li>
</ul>
<p><strong>补充：理解 TF 和 IDF</strong></p>
<p>为了理解后面的 BM25，可以先把 TF-IDF 拆成两个部分来看：</p>
<ul>
<li><strong>TF（Term Frequency，词频）</strong>：回答“这个词在当前文档中出现得多不多？”</li>
<li><strong>IDF（Inverse Document Frequency，逆文档频率）</strong>：回答“这个词在整个知识库中是否稀有？”</li>
</ul>
<p>常见公式如下：</p>
<div class="highlight"><pre><span></span><code>TF(t,d) = 词 t 在文档 d 中出现次数 / 文档 d 总词数
IDF(t) = log(N / df(t))
</code></pre></div>
<ul>
<li><code>N</code>：知识库文档总数。</li>
<li><code>df(t)</code>：包含词 <code>t</code> 的文档数量。</li>
</ul>
<p>一个词越普遍，<code>df(t)</code> 越大，IDF 越小；一个词越稀有，<code>df(t)</code> 越小，IDF 越大。</p>
<p>例如，知识库有 10 篇文档。当前 100 词的文章中，“asyncio”出现 4 次，整个知识库只有 2 篇包含它：</p>
<div class="highlight"><pre><span></span><code>TF = 4 / 100 = 0.04
IDF = ln(10 / 2) = ln(5) ≈ 1.609
TF-IDF = 0.04 × 1.609 ≈ 0.064
</code></pre></div>
<p>如果“使用”出现在全部 10 篇文档中，则 <code>IDF = ln(10 / 10) = 0</code>。即使它在某篇文章里出现很多次，最终 TF-IDF 权重也会很低，因为它缺乏区分主题的能力。</p>
<p>所以，TF-IDF 的核心思想是：<strong>一个词在当前文档中越常见，并且在整个知识库中越稀有，它越能代表当前文档的主题。</strong> BM25 正是在这个思想上继续加入了长度归一化和词频饱和控制。</p>
<p>示例代码：</p>
<div class="highlight"><pre><span></span><code><span class="kn">from</span><span class="w"> </span><span class="nn">langchain_community.retrievers</span><span class="w"> </span><span class="kn">import</span> <span class="n">TFIDFRetriever</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">langchain_core.documents</span><span class="w"> </span><span class="kn">import</span> <span class="n">Document</span>

<span class="n">docs</span> <span class="o">=</span> <span class="p">[</span>
    <span class="n">Document</span><span class="p">(</span><span class="n">page_content</span><span class="o">=</span><span class="s2">&quot;量子计算是一种基于量子力学的计算范式。&quot;</span><span class="p">),</span>
    <span class="n">Document</span><span class="p">(</span><span class="n">page_content</span><span class="o">=</span><span class="s2">&quot;人工智能是模拟人类智能的技术。&quot;</span><span class="p">)</span>
<span class="p">]</span>

<span class="n">retriever</span> <span class="o">=</span> <span class="n">TFIDFRetriever</span><span class="o">.</span><span class="n">from_documents</span><span class="p">(</span><span class="n">docs</span><span class="p">)</span>
<span class="n">retriever</span><span class="o">.</span><span class="n">k</span> <span class="o">=</span> <span class="mi">1</span>  <span class="c1"># 返回 1 个结果</span>
<span class="n">results</span> <span class="o">=</span> <span class="n">retriever</span><span class="o">.</span><span class="n">invoke</span><span class="p">(</span><span class="s2">&quot;量子计算&quot;</span><span class="p">)</span>
<span class="nb">print</span><span class="p">(</span><span class="n">results</span><span class="p">)</span>
</code></pre></div>
<p>运行结果：</p>
<p><img src="assets/image-20251003104151768.png" alt="image-20251003104151768" style="zoom:80%;" /></p>
</blockquote>
<ul>
<li>BM25Retriever</li>
</ul>
<blockquote>
<p>BM25算法：BM25 是对 TF-IDF 的改进版本，在 TF-IDF 基础上做了归一化（防止长文档优势）与饱和控制（防止词频无限增长）。</p>
<p><img src="assets/image-20251003105157106.png" alt="image-20251003105157106" style="zoom: 60%;" /></p>
<p><img src="assets/image-20251006090343544.png" alt="image-20251006090343544" style="zoom:67%;" /></p>
<p>功能：基于 BM25 算法的关键词检索器，适合基于词频的搜索。</p>
<p>特点：</p>
<ul>
<li>不依赖嵌入模型，使用词频和逆文档频率（TF-IDF）计算相关性。</li>
<li>适合关键词匹配场景，计算成本低。</li>
<li>不支持语义搜索，效果依赖文本的字面匹配。</li>
</ul>
<p>适用场景：</p>
<ul>
<li>传统搜索场景。</li>
<li>关键词驱动的问答。</li>
</ul>
<p>环境依赖：</p>
<p>pip install rank_bm25</p>
<p>示例代码：</p>
<div class="highlight"><pre><span></span><code><span class="kn">from</span><span class="w"> </span><span class="nn">langchain_community.retrievers</span><span class="w"> </span><span class="kn">import</span> <span class="n">BM25Retriever</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">langchain_core.documents</span><span class="w"> </span><span class="kn">import</span> <span class="n">Document</span>

<span class="n">docs</span> <span class="o">=</span> <span class="p">[</span>
    <span class="n">Document</span><span class="p">(</span><span class="n">page_content</span><span class="o">=</span><span class="s2">&quot;量子计算是一种基于量子力学的计算范式。&quot;</span><span class="p">),</span>
    <span class="n">Document</span><span class="p">(</span><span class="n">page_content</span><span class="o">=</span><span class="s2">&quot;人工智能是模拟人类智能的技术。&quot;</span><span class="p">)</span>
<span class="p">]</span>

<span class="n">retriever</span> <span class="o">=</span> <span class="n">BM25Retriever</span><span class="o">.</span><span class="n">from_documents</span><span class="p">(</span><span class="n">docs</span><span class="p">)</span>
<span class="n">retriever</span><span class="o">.</span><span class="n">k</span> <span class="o">=</span> <span class="mi">1</span>  <span class="c1"># 返回 1 个结果</span>
<span class="n">results</span> <span class="o">=</span> <span class="n">retriever</span><span class="o">.</span><span class="n">invoke</span><span class="p">(</span><span class="s2">&quot;量子计算&quot;</span><span class="p">)</span>
<span class="nb">print</span><span class="p">(</span><span class="n">results</span><span class="p">)</span>
</code></pre></div>
<p>运行结果：</p>
<p><img src="assets/image-20251003104151768.png" alt="image-20251003104151768" style="zoom:80%;" /></p>
</blockquote>
<ul>
<li>MultiQueryRetriever</li>
</ul>
<blockquote>
<p>功能：它借助 LLM 自动生成多个语义等价的改写查询，把用户的问题扩展成多个角度，然后对每个改写进行检索，最后合并结果，以提高召回率。 </p>
<p>特点：</p>
<ul>
<li>使用语言模型生成查询的多种表达方式</li>
<li>从向量存储中检索所有变体的结果并合并</li>
<li>提高召回率，适合复杂查询</li>
</ul>
<p>适用场景：</p>
<ul>
<li>查询表达不明确或需要覆盖多种语义</li>
<li>提高检索的全面性
</li>
</ul>
</blockquote>
<ul>
<li>EnsembleRetriever</li>
</ul>
<blockquote>
<p>功能：结合多种检索器（如 BM25 和向量存储），融合结果。</p>
<p>特点：</p>
<ul>
<li>结合关键词搜索和语义搜索的优点</li>
<li>支持加权融合，调整不同检索器的权重</li>
<li>提高召回率和精准度</li>
</ul>
<p>适用场景：</p>
<ul>
<li>需要综合关键词和语义的搜索</li>
<li>复杂查询场景
</li>
</ul>
</blockquote>
<ul>
<li>ContextualCompressionRetriever</li>
</ul>
<blockquote>
<p>功能：对检索结果进行压缩，提取最相关的内容 </p>
<p>特点：</p>
<ul>
<li>使用语言模型对检索到的文档进行重新排序或精炼</li>
<li>减少无关内容，提高结果质量</li>
<li>增加计算开销，但提升精准度</li>
</ul>
<p>适用场景：</p>
<ul>
<li>文档内容冗长，需要提取关键信息</li>
<li>提高问答系统的答案质量
</li>
</ul>
</blockquote>
<ul>
<li>Custom Retriever</li>
</ul>
<blockquote>
<p>功能：开发者可以自定义检索逻辑，适配特定数据源或算法 </p>
<p>特点：</p>
<ul>
<li>继承 BaseRetriever 类，实现 get_relevant_documents 方法</li>
<li>支持任意数据源（如数据库、API）</li>
</ul>
<p>适用场景：</p>
<ul>
<li>特定领域的数据源（如内部数据库）</li>
<li>自定义检索算法
</li>
</ul>
</blockquote>
<h2 id="agents">八、Agents<a class="headerlink" href="#agents" title="Permanent link">&para;</a></h2>
<h3 id="1-agent">1 Agent 概念<a class="headerlink" href="#1-agent" title="Permanent link">&para;</a></h3>
<p>Agent（智能体）是一种能够感知环境、进行决策和执行动作的智能实体。不同于传统的人工智能，Agent 具备通过主动思考、调用工具去逐步完成给定目标的能力。</p>
<p>为什么要调用工具？</p>
<ul>
<li>因为大模型虽然非常强大，但是也具备一定的局限性，比如不能回答实时信息、处理数学逻辑问题仍然非常的初级等等。因此，可以借助第三方工具来辅助大模型的应用。</li>
</ul>
<p>从大模型的角度来看，<strong>Agent其实就是基于大模型的语义理解和推理能力，让大模型拥有解决复杂问题时的任务规划能力，并调用外部工具来执行各种任务，利用向量数据库保留“记忆”的一个智能体</strong>。</p>
<blockquote>
<p>Agent = 大模型 + 任务规划（Planning） + 使用外部工具执行任务（Tools&amp;Action） + 记忆（Memory）</p>
<p>把Agent比作一个人的话，<strong>他应该有大脑（语义理解、存储记忆、推理规划、专业知识）、五官（接收文本、视觉输入、听觉输入等）、四肢（使用工具完成各种具体任务）等主要部件</strong></p>
</blockquote>
<hr />
<p><strong>工作流程概述：</strong></p>
<ul>
<li>用户提出任务。</li>
<li>Agent 启动： 将用户输入与预设的“提示词模板”结合，并结合当前的“上下文”和“变量”，形成一个完整的输入发送给大模型。</li>
<li>大模型思考与决策 (循环)：<ul>
<li>大模型接收输入后，根据其内置的逻辑和提示词的指导，进行“思考”。</li>
<li>它会判断完成当前任务是需要继续使用工具来获取更多信息/执行操作，还是已经可以直接生成最终答案。</li>
<li>如果需要工具： 大模型会根据任务需求和对工具的描述，选择合适的工具，并生成执行该工具所需的输入参数。这些参数通常会从上下文或变量中提取。</li>
<li>工具执行： 选定的工具被调用，执行其功能。</li>
<li>结果反馈： 工具执行的结果会返回给 Agent，并被用来更新“上下文”和“变量”。</li>
<li>循环： Agent 将更新后的“上下文”、“变量”以及工具执行结果再次反馈给大模型，大模型继续进行新一轮的“思考-行动”循环，直到任务完成。</li>
<li>如果直接回答： 当大模型判断任务已完成，或无需额外工具即可回答时，它会生成最终的答案。</li>
</ul>
</li>
<li>用户输出： 最终答案被呈现给用户。</li>
</ul>
<h3 id="2-agent">2 Agent关键组成部分<a class="headerlink" href="#2-agent" title="Permanent link">&para;</a></h3>
<p><img alt="image-20250705115745718" src="assets/image-20250705115745718.png" /></p>
<ul>
<li>
<p>任务规划（Planning）</p>
<p>Agent需要提前将一项复杂任务拆解为多个更小、更易于处理的子任务，从而实现对复杂任务的高效处理</p>
</li>
<li>
<p>长短期记忆（Memory）</p>
<p>聊天上下文，长期保留和回忆信息</p>
</li>
<li>
<p>工具&amp;执行（Tools&amp;Action）</p>
<p>根据拆分好的子任务，调用外部提供好的专业API解决专业问题，完成一个个具体的子任务，并把处理结果返回给大模型</p>
</li>
</ul>
<h3 id="3-langchainagent">3 langchain实现Agent<a class="headerlink" href="#3-langchainagent" title="Permanent link">&para;</a></h3>
<p>LangChain 提供了不同类型的代理（主要罗列一下三种）:</p>
<ul>
<li>
<p><strong>Zero-shot ReAct Description</strong></p>
<ul>
<li>
<p>基于 <strong>ReAct 框架</strong>（推理 + 行动），仅依赖工具的 <strong>描述</strong> 来决定调用哪个工具。</p>
</li>
<li>
<p>特点：无需额外示例（zero-shot），但 <strong>不具备记忆能力</strong>，每次推理都独立进行。</p>
</li>
<li>
<p>使用场景：简单任务，工具选择完全靠工具描述即可。</p>
</li>
</ul>
</li>
<li>
<p><strong>Structured Chat Zero-shot ReAct Description</strong></p>
<ul>
<li>
<p>同样基于 <strong>ReAct 框架</strong>，但可以处理 <strong>结构化输入</strong>，即支持带多个参数的工具（类似函数调用）。</p>
</li>
<li>
<p>特点：不仅能像第一种代理那样根据描述选择工具，还能正确组织并传递复杂参数。</p>
</li>
<li>
<p>使用场景：调用接口类工具、需要多参数输入的任务。</p>
</li>
</ul>
</li>
<li>
<p><strong>Conversational ReAct Description</strong></p>
<ul>
<li>
<p>在 ReAct 框架基础上，增强了 <strong>对话记忆能力</strong>。</p>
</li>
<li>
<p>特点：能根据上下文对话历史来做出工具选择和回应，更适合持续性对话场景。</p>
</li>
<li>
<p>使用场景：多轮对话，用户可能引用之前的内容或需要长期上下文跟踪。</p>
</li>
</ul>
</li>
</ul>
<hr />
<p>LangChain 中集成了很多工具，可以通过下面的方式进行查询：</p>
<div class="highlight"><pre><span></span><code><span class="kn">from</span><span class="w"> </span><span class="nn">langchain_community.agent_toolkits.load_tools</span><span class="w"> </span><span class="kn">import</span> <span class="n">get_all_tool_names</span>
<span class="n">results</span> <span class="o">=</span> <span class="n">get_all_tool_names</span><span class="p">()</span>
<span class="nb">print</span><span class="p">(</span><span class="n">results</span><span class="p">)</span>
</code></pre></div>
<p>运行结果：</p>
<div class="highlight"><pre><span></span><code>[&#39;sleep&#39;, &#39;wolfram-alpha&#39;, &#39;google-search&#39;, &#39;google-search-results-json&#39;, &#39;searx-search-results-json&#39;, &#39;bing-search&#39;, &#39;metaphor-search&#39;, &#39;ddg-search&#39;, &#39;google-books&#39;, &#39;google-lens&#39;, &#39;google-serper&#39;, &#39;google-scholar&#39;, &#39;google-finance&#39;, &#39;google-trends&#39;, &#39;google-jobs&#39;, &#39;google-serper-results-json&#39;, &#39;searchapi&#39;, &#39;searchapi-results-json&#39;, &#39;serpapi&#39;, &#39;dalle-image-generator&#39;, &#39;twilio&#39;, &#39;searx-search&#39;, &#39;merriam-webster&#39;, &#39;wikipedia&#39;, &#39;arxiv&#39;, &#39;golden-query&#39;, &#39;pubmed&#39;, &#39;human&#39;, &#39;awslambda&#39;, &#39;stackexchange&#39;, &#39;sceneXplain&#39;, &#39;graphql&#39;, &#39;openweathermap-api&#39;, &#39;dataforseo-api-search&#39;, &#39;dataforseo-api-search-json&#39;, &#39;eleven_labs_text2speech&#39;, &#39;google_cloud_texttospeech&#39;, &#39;read_file&#39;, &#39;reddit_search&#39;, &#39;news-api&#39;, &#39;tmdb-api&#39;, &#39;podcast-api&#39;, &#39;memorize&#39;, &#39;llm-math&#39;, &#39;open-meteo-api&#39;, &#39;requests&#39;, &#39;requests_get&#39;, &#39;requests_post&#39;, &#39;requests_patch&#39;, &#39;requests_put&#39;, &#39;requests_delete&#39;, &#39;terminal&#39;]
</code></pre></div>
<hr />
<p>接下来，通过一个示例来学习Agent的基本使用。</p>
<ul>
<li>问题1：计算一下300的25%是多少？</li>
<li>问题2：请帮我介绍一下故宫</li>
</ul>
<p>示例代码：</p>
<blockquote>
<p>需要提前安装 wikipedia</p>
<p>pip install wikipedia</p>
</blockquote>
<div class="highlight"><pre><span></span><code><span class="kn">from</span><span class="w"> </span><span class="nn">langchain_core.messages</span><span class="w"> </span><span class="kn">import</span> <span class="n">HumanMessage</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">langgraph.checkpoint.memory</span><span class="w"> </span><span class="kn">import</span> <span class="n">MemorySaver</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">langchain.agents</span><span class="w"> </span><span class="kn">import</span> <span class="n">create_agent</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">langchain_community.agent_toolkits.load_tools</span><span class="w"> </span><span class="kn">import</span> <span class="n">load_tools</span>
<span class="kn">from</span><span class="w"> </span><span class="nn">langchain_openai</span><span class="w"> </span><span class="kn">import</span> <span class="n">ChatOpenAI</span>
<span class="kn">import</span><span class="w">  </span><span class="nn">os</span>

<span class="c1"># 实例化模型</span>
<span class="n">model</span> <span class="o">=</span> <span class="n">ChatOpenAI</span><span class="p">(</span>
    <span class="n">base_url</span><span class="o">=</span><span class="n">os</span><span class="o">.</span><span class="n">environ</span><span class="o">.</span><span class="n">get</span><span class="p">(</span><span class="s1">&#39;base_url&#39;</span><span class="p">,</span><span class="s1">&#39;https://dashscope.aliyuncs.com/compatible-mode/v1&#39;</span><span class="p">),</span>
    <span class="n">model</span><span class="o">=</span><span class="s1">&#39;qwen-plus&#39;</span><span class="p">,</span>
    <span class="n">openai_api_key</span><span class="o">=</span><span class="n">os</span><span class="o">.</span><span class="n">environ</span><span class="o">.</span><span class="n">get</span><span class="p">(</span><span class="s1">&#39;api_key&#39;</span><span class="p">,</span><span class="s1">&#39;sk-XXXXXXXX&#39;</span><span class="p">),</span>
    <span class="n">max_tokens</span><span class="o">=</span><span class="mi">1000</span><span class="p">,</span>
    <span class="n">temperature</span><span class="o">=</span><span class="mi">0</span>
<span class="p">)</span>

<span class="c1"># Create the agent</span>
<span class="n">memory</span> <span class="o">=</span> <span class="n">MemorySaver</span><span class="p">()</span>
<span class="n">tools</span> <span class="o">=</span> <span class="n">load_tools</span><span class="p">([</span><span class="s1">&#39;wikipedia&#39;</span><span class="p">,</span> <span class="s1">&#39;llm-math&#39;</span><span class="p">],</span> <span class="n">llm</span><span class="o">=</span><span class="n">model</span><span class="p">)</span>
<span class="n">agent_executor</span> <span class="o">=</span> <span class="n">create_agent</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="n">tools</span><span class="p">,</span> <span class="n">checkpointer</span><span class="o">=</span><span class="n">memory</span><span class="p">)</span>

<span class="c1"># Use the agent</span>
<span class="n">config</span> <span class="o">=</span> <span class="p">{</span><span class="s2">&quot;configurable&quot;</span><span class="p">:</span> <span class="p">{</span><span class="s2">&quot;thread_id&quot;</span><span class="p">:</span> <span class="s2">&quot;abc123&quot;</span><span class="p">}}</span>

<span class="k">for</span> <span class="n">chunk</span> <span class="ow">in</span> <span class="n">agent_executor</span><span class="o">.</span><span class="n">stream</span><span class="p">(</span>
    <span class="p">{</span><span class="s2">&quot;messages&quot;</span><span class="p">:</span> <span class="p">[</span><span class="n">HumanMessage</span><span class="p">(</span><span class="n">content</span><span class="o">=</span><span class="s2">&quot;计算一下300的25%是多少？&quot;</span><span class="p">)]},</span> <span class="n">config</span>
<span class="p">):</span>
    <span class="nb">print</span><span class="p">(</span><span class="n">chunk</span><span class="p">)</span>
    <span class="nb">print</span><span class="p">(</span><span class="s2">&quot;----&quot;</span><span class="p">)</span>

<span class="c1"># for chunk in agent_executor.stream(</span>
<span class="c1">#     {&quot;messages&quot;: [HumanMessage(content=&quot;大熊猫是哪个国家的国宝？&quot;)]}, config</span>
<span class="c1"># ):</span>
<span class="c1">#     print(chunk)</span>
<span class="c1">#     print(&quot;----&quot;)</span>
</code></pre></div>
<p>运行结果：</p>
<p><img alt="1768397211211" src="assets/1768397211211.png" /></p>
<blockquote>
<p>参考官网 <a href="https://python.langchain.com/docs/how_to/custom_tools/">https://python.langchain.com/docs/how_to/custom_tools/</a></p>
</blockquote>
<h2 id="langchain">九、LangChain使用场景<a class="headerlink" href="#langchain" title="Permanent link">&para;</a></h2>
<ul>
<li>个人助手</li>
<li>基于文档的问答系统</li>
<li>聊天机器人</li>
<li>Tabular数据查询</li>
<li>API交互</li>
<li>信息提取</li>
<li>文档总结</li>
</ul>
<h2 id="_3">十、本节小结<a class="headerlink" href="#_3" title="Permanent link">&para;</a></h2>
<p>本章主要介绍了LangChain的基本组件及使用方式。</p>
<h2 id="_4">十一、本节作业<a class="headerlink" href="#_4" title="Permanent link">&para;</a></h2>
<h3 id="1-1">1 作业1<a class="headerlink" href="#1-1" title="Permanent link">&para;</a></h3>
<ul>
<li>题目1：什么是LangChain？请解释其核心组件和设计思想。</li>
<li>题目2：解释LangChain中的链（Chain）概念，并介绍主要的链类型。</li>
<li>题目3：解释LangChain中的Prompts概念，并介绍主要类型。</li>
<li>题目4：实现LangChain中的output_parsers的具体使用方式</li>
</ul>
<h3 id="2-2">2 作业2<a class="headerlink" href="#2-2" title="Permanent link">&para;</a></h3>
<ul>
<li>题目1：什么是Memory组件？请解释LangChain中不同类型的记忆机制。</li>
<li>题目2：解释LangChain中的提示工程及其最佳实践。</li>
<li>题目3：LangChain的Document Loaders？请介绍其功能和主要类型。</li>
<li>题目4 ：什么是LangChain的代理（Agent）？请说明其工作原理和应用。</li>
</ul>


              </article>
            </div>
          
          
<script>var target=document.getElementById(location.hash.slice(1));target&&target.name&&(target.checked=target.name.startsWith("__tabbed_"))</script>
        </div>
        
      </main>
      
        <footer class="md-footer">
  
  <div class="md-footer-meta md-typeset">
    <div class="md-footer-meta__inner md-grid">
      <div class="md-copyright">
  
    <div class="md-copyright__highlight">
      © EnterpriseKB-RAG 2026. Internal knowledge base demo site.
    </div>
  
  
    Made with
    <a href="https://squidfunk.github.io/mkdocs-material/" target="_blank" rel="noopener">
      Material for MkDocs
    </a>
  
</div>
      
    </div>
  </div>
</footer>
      
    </div>
    <div class="md-dialog" data-md-component="dialog">
      <div class="md-dialog__inner md-typeset"></div>
    </div>
    
    
      <script id="__config" type="application/json">{"base": "..", "features": [], "search": "../assets/javascripts/workers/search.973d3a69.min.js", "tags": null, "translations": {"clipboard.copied": "\u5df2\u590d\u5236", "clipboard.copy": "\u590d\u5236", "search.result.more.one": "\u5728\u8be5\u9875\u4e0a\u8fd8\u6709 1 \u4e2a\u7b26\u5408\u6761\u4ef6\u7684\u7ed3\u679c", "search.result.more.other": "\u5728\u8be5\u9875\u4e0a\u8fd8\u6709 # \u4e2a\u7b26\u5408\u6761\u4ef6\u7684\u7ed3\u679c", "search.result.none": "\u6ca1\u6709\u627e\u5230\u7b26\u5408\u6761\u4ef6\u7684\u7ed3\u679c", "search.result.one": "\u627e\u5230 1 \u4e2a\u7b26\u5408\u6761\u4ef6\u7684\u7ed3\u679c", "search.result.other": "# \u4e2a\u7b26\u5408\u6761\u4ef6\u7684\u7ed3\u679c", "search.result.placeholder": "\u952e\u5165\u4ee5\u5f00\u59cb\u641c\u7d22", "search.result.term.missing": "\u7f3a\u5c11", "select.version": "\u9009\u62e9\u5f53\u524d\u7248\u672c"}, "version": null}</script>
    
    
      <script src="../assets/javascripts/bundle.f55a23d4.min.js"></script>
      
        <script src="../javascripts/extra.js"></script>
      
        <script src="../javascripts/mathjax.js"></script>
      
        <!--        <script src="https://polyfill.io/v3/polyfill.min.js?features=es6"></script>-->
      
        <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-mml-chtml.js"></script>
      
    
  </body>
</html>