Ukwakha iiNkqubo zeSoftware zokuZiphilisa nge-Multi-Agent AI Architectures kunye neLangChain

Ukwakha iiNkqubo zeSoftware zokuZiphilisa nge-Multi-Agent AI Architectures kunye neLangChain

Intshayelelo

Kwimeko yanamhlanje ekhawulezayo yophuhliso lwesoftware, ulwakhiwo lweenkqubo zesoftware zokuziphilisa lube yimfuneko ebalulekileyo ekuqinisekiseni ukuthembeka, ukuncipha, kunye nokomelela kweenkqubo zesoftware ezintsonkothileyo. Izakhiwo zemveli ze-monolithic azisekho ngokwaneleyo ukuhlangabezana neemfuno zophuhliso lwesofthiwe yanamhlanje, apho iinkqubo kufuneka zikwazi ukulungelelanisa neemfuno eziguqukayo, ukuphatha ukusilela okungalindelekanga, kunye nokubuyisela kwiimpazamo ngokukhawuleza. Yilapho i-AI yezakhiwo ezininzi ze-AI kunye neLangChain zidlala, zinika isisombululo esinamandla sokwakha iinkqubo zesofthiwe zokuziphilisa ezinokuthi zifunde, zilungelelanise, kwaye ziphendule kwiimeko eziguqukayo.

Yintoni i-Multi-Agent AI Architectures?

Izakhiwo ezininzi ze-AI luhlobo lwenkqubo ye-AI esasazwayo equlethwe ngama-agent azimeleyo amaninzi asebenzisana kunye kunye nokusingqongileyo ukuze afezekise injongo efanayo. I-arhente nganye inamandla ayo okwenza izigqibo, ulwazi, kunye nezibonelelo, kwaye banokunxibelelana omnye nomnye ukwabelana ngolwazi, ukulungelelanisa izenzo, kunye nokuziqhelanisa neemeko eziguqukayo.

Kumxholo weenkqubo zesoftware, i-AI yezakhiwo ezininzi ze-AI zingasetyenziselwa ukwakha iinkqubo zokuziphilisa ezinokuthi zibone kwaye ziphendule kwiimpazamo, ukungaphumeleli, kunye nezinye izinto ezingaqhelekanga ngexesha langempela. Ngokusasaza inkqubo yokwenza izigqibo kwii-arhente ezininzi, ezi nkqubo zinokufunda kumava, zilungelelanise iimeko eziguqukayo, kwaye ziphinde zibuyele kwiimpazamo ngokukhawuleza.

Yintoni iLangChain?

ILangChain yilayibrari yePython evulekileyo yomthombo wokwakha iinkqubo ze-AI ezininzi. Ibonelela ngesakhelo esilula, esiguquguqukayo, kunye nesakhelekileyo sokwakha iinkqubo ze-AI ezintsonkothileyo ezinokufunda, ukuziqhelanisa, kunye nokusabela kwiimeko eziguqukayo. I-LangChain yenzelwe ukusebenza kunye noluhlu olubanzi lwezicwangciso ze-AI, kuquka i-TensorFlow, i-PyTorch, kunye ne-Keras, kwaye ingasetyenziselwa ukwakha iinkqubo ezidibanisa nemithombo eyahlukeneyo yedatha, kubandakanywa i-database, i-APIs, kunye nezixhobo ze-IoT.

Ukwakha iiNkqubo zeSoftware zokuZiphilisa ngeLangChain

Ukwakha inkqubo ye-software yokuziphilisa kunye neLangChain, kufuneka siyile i-architecture ye-AI ye-multi-agent enokuthi ibone kwaye iphendule iimpazamo, ukungaphumeleli kunye nezinye izinto ezingaqhelekanga ngexesha langempela. Nantsi isishwankathelo somgangatho ophezulu woyilo:

** Uyilo lwe-Agent**: Sidinga ukuyila isethi yee-arhente ezizimeleyo ezinokusebenzisana kunye kunye nokusingqongileyo ukufezekisa injongo efanayo. I-arhente nganye kufuneka ibe namandla ayo okwenza izigqibo, ulwazi, kunye nezibonelelo.
Igrafu yoLwazi: Kufuneka senze igrafu yolwazi emele ubudlelwane phakathi kwee-arhente, okusingqongileyo, kunye neenjongo zesistim. Le grafu ingasetyenziselwa ukuqiqa malunga nemeko yenkqubo kunye nokwenza izigqibo.
Ukwenziwa kwezigqibo: Kufuneka siphumeze inkqubo yokwenza izigqibo evumela ii-arhente ukuba ziqiqe ngemeko yenkqubo kwaye zenze izigqibo ezisekelwe kwigrafu yolwazi.
Ukuphunyezwa kweAction: Kufuneka siphumeze inkqubo yokwenziwa kwentshukumo evumela ii-arhente ukuba zenze izigqibo zazo kwaye zisebenzisane nokusingqongileyo.

Nanku umzekelo olula wenkqubo yesoftware yokuziphilisa eyakhiwe ngeLangChain:

ngenisa i-langchain
ukusuka langchain.agents yokungenisa Agent
ukusuka langchain.llms ukungenisa LLM


igrafu = {
    'ii-arhente': ['i-arhente1', 'i-arhente2', 'i-arhente3'],
    'indalo': ['isiseko sedatha', 'API', 'Isixhobo se-IoT'],
    'iinjongo': ['bona iimpazamo', 'phendula kwiintsilelo', 'buyela kwiimpazamo']
}

# Chaza inkqubo yokwenza izigqibo
def decision_making(iarhente, ulwazi_igraph):
    # Isizathu malunga nemeko yenkqubo
    ilizwe = ulwazi_igraph['iiarhente'][arhente]
    # Yenza isigqibo esisekwe kurhulumente
    isigqibo = 'phendula ukusilela'
    isigqibo sokubuya

# Chaza inkqubo yokwenza isenzo
def action_execution(iarhente, isigqibo):
    # Yenza isigqibo
    ukuba isigqibo == 'phendula ukusilela':
        # Phendula ukusilela
        shicilela('Ukuphendula ukusilela...')
    ukubuya

# Yenza iarhente yeLangChain
iarhente = iarhente(
    igama='iarhente yokuziphilisa',
    decision_making=ukwenza_isigqibo,
    action_execution=ukwenziwa_kwesenzo,
    ulwazi_igraph=igrafu
)

# Yenza iLangChain LLM
llm = LLM(
    igama='ukuziphilisa-llm',
    imodeli='davinci'
)

# Yenza inkqubo yeLangChain
inkqubo = langchain.System(
    iarhente=iarhente,
    llm=llm
)

# Qhuba inkqubo
inkqubo.run()

Lo mzekelo ubonisa inkqubo ye-software yokuziphilisa elula ekwazi ukubona iimpazamo, ukuphendula ukungaphumeleli, kunye nokubuyisela kwiimpazamo ngokukhawuleza. Inkqubo iqulethe isethi yee-agent ezizimeleyo ezisebenzisana nomnye kunye nokusingqongileyo ukufezekisa injongo efanayo. Iiarhente zisebenzisa igrafu yolwazi ukuqiqa malunga nemeko yenkqubo kwaye zenze izigqibo ezisekelwe kwigrafu.

Ephezulu 5-10 Eyeyona Isetyenziswayo Imiyalelo/iziqwengana

Nantsi eyona miyalelo/iziqwengana eziphezulu ezi-5-10 zeLangChain:

langchain.Agent(): Yenza iarhente entsha yeLangChain.
langchain.LLM(): Yenza iLangChain LLM entsha.
langchain.System(): Yenza inkqubo entsha yeLangChain.
ukwenza_i-arhente(): Iqhuba inkqubo yokwenziwa kwesigqibo se-arhente.
i-arhente.action_execution(): Iqhuba inkqubo yophumezo ye-arhente.
llm.predict(): Yenza uqikelelo usebenzisa iLLM.
inkqubo.run(): Iqhuba inkqubo yeLangChain.
igrafu.yongeza_inodi(): Yongeza indawo entsha kwigrafu yolwazi.
Igrafu.yongeza_ungqameko(): Yongeza isiphelo esitsha kwigrafu yolwazi.
igrafu.remove_node(): Ikhupha inqaku kwigrafu yolwazi.

Umbono

Nantsi imbonakalo elula yegrafu yolwazi usebenzisa iChart.js:

{
  "type": "graph",
  "idatha": {
    "iindawo": [
      {"id": "arhente1", "ileyibhile": "Ummeli 1"},
      {"id": "arhente2", "ileyibhile": "Ummeli 2"},
      {"id": "iarhente3", "ileyibhile": "Ummeli 3"},
      {"id": "isiseko sedatha", "ileyibhile": "Isiseko sedatha"},
      {"id": "API", "label": "API"},
      {"id": "IoT isixhobo", "label": "IoT Isixhobo"}
    ],
    "imiphetho": [
      {"umthombo": "agent1", "target": "database"},
      {"source": "agent2", "target": "API"},
      {"source": "agent3", "target": "IoT device"},
      {"umthombo": "iarhente1", "ithagethi": "iarhente2"},
      {"source": "agent2", "target": "agent3"}
    ]
  },
  "iinketho": {
    "uyilo": {
      "uhlobo": "setyhula"
    },
    "iindawo": {
      "ubukhulu": 50
    },
    "imiphetho": {
      "ububanzi": 2
    }
  }
}

Lo mboniso ubonisa ubudlelwane phakathi kwee-arhente, indawo engqongileyo, kunye neenjongo zenkqubo.

Isiphelo

Ukwakha iinkqubo ze-software zokuziphilisa kunye ne-multi-agent AI architectures kunye ne-LangChain inika isisombululo esinamandla sokuqinisekisa ukuthembeka, ukulinganisa, kunye nokuqina kweenkqubo ze-software eziyinkimbinkimbi. Ngokusasaza inkqubo yokwenza izigqibo kwii-arhente ezininzi, ezi nkqubo zinokufunda kumava, zilungelelanise iimeko eziguqukayo, kwaye ziphinde zibuyele kwiimpazamo ngokukhawuleza. NgeLangChain, abaphuhlisi banokwakha iinkqubo ze-AI eziyinkimbinkimbi ezidibanisa kunye noluhlu olubanzi lwemithombo yedatha, kuquka i-database, i-APIs, kunye nezixhobo ze-IoT. Ngokulandela ikhowudi yomzekelo kunye nokusebenzisa i-5-10 esetyenziswa kakhulu imiyalelo / iziqwenga, abaphuhlisi banokuzakhela iinkqubo zabo zesoftware zokuziphilisa ngeLangChain.

Ukwakha iiNkqubo zeSoftware zokuZiphilisa nge-Multi-Agent AI Architectures kunye neLangChain

Have Questions?