Ukwakha Amasistimu Esofthiwe Okuziphilisa Nge-Multi-Agent AI Architectures kanye ne-LangChain

Ukwakha Izinhlelo Zesofthiwe Eziziphilisa Nge-Multi-Agent AI Architectures kanye ne-LangChain

Isingeniso

Ezweni lanamuhla elisheshayo lokuthuthukisa isofthiwe, ukwakha izinhlelo zesofthiwe eziziphilisayo sekuyisidingo esibalulekile sokuqinisekisa ukwethembeka, ukuqina, nokuqina kwezinhlelo zesofthiwe eziyinkimbinkimbi. Izakhiwo zendabuko ze-monolithic azisanele ukuhlangabezana nezidingo zokuthuthukiswa kwesofthiwe yesimanje, lapho amasistimu kufanele akwazi ukuzivumelanisa nezimfuneko ezishintshayo, asingathe ukwehluleka okungalindelekile, futhi alulame emaphutheni ngokushesha. Yilapho izakhiwo ze-AI ze-multi-agent kanye ne-LangChain ziqala ukusebenza, zinikeza isisombululo esinamandla sokwakha izinhlelo zesofthiwe zokuziphilisa ezingakwazi ukufunda, ukujwayela, nokusabela ezimweni ezishintshayo.

Iyini i-Multi-Agent AI Architectures?

Izakhiwo ze-AI ze-multi-agent ziwuhlobo lwesistimu ye-AI esabalalisiwe equkethe ama-ejenti amaningi azimele asebenzisana nomunye kanye nendawo yawo ukuze afinyelele umgomo ofanayo. I-ejenti ngayinye inamakhono ayo okwenza izinqumo, ulwazi, kanye nezinsiza, futhi bangaxhumana ukuze babelane ngolwazi, baxhumanise izenzo, futhi bazivumelanise nezimo ezishintshayo.

Ngomongo wezinhlelo zesofthiwe, izakhiwo ze-AI ezinama-agent amaningi zingasetshenziswa ukwakha izinhlelo zokuziphilisa ezikwazi ukubona nokuphendula amaphutha, ukwehluleka, nokunye okudidayo ngesikhathi sangempela. Ngokusabalalisa inqubo yokuthatha izinqumo kuma-ejenti amaningi, lawa masistimu angafunda kokuhlangenwe nakho, azivumelanise nezimo ezishintshayo, futhi alulame emaphutheni ngokushesha.

Iyini i-LangChain?

I-LangChain iwumthombo ovulekile wezincwadi wePython wokwakha ama-multi-agent AI systems. Ihlinzeka ngohlaka olulula, oluguquguqukayo, nolukakayo lokwakha amasistimu e-AI ayinkimbinkimbi angafunda, azivumelanise nezimo, futhi aphendule ezimweni ezishintshayo. I-LangChain iklanyelwe ukusebenza nezinhlaka eziningi ze-AI, okuhlanganisa i-TensorFlow, i-PyTorch, ne-Keras, futhi ingasetshenziswa ukwakha amasistimu ahlanganisa nemithombo yedatha ehlukahlukene, okuhlanganisa imininingwane yolwazi, ama-API, namadivayisi we-IoT.

Ukwakha Amasistimu Esoftware Yokuziphilisa ngeLangChain

Ukuze sakhe isistimu yesofthiwe yokuziphilisa nge-LangChain, sidinga ukuklama i-architecture ye-AI enama-agent amaningi engakwazi ukubona futhi iphendule kumaphutha, ukwehluleka, nokunye okudidayo ngesikhathi sangempela. Nakhu ukubuka konke kwezinga eliphezulu kwezakhiwo:

Idizayini yomenzeli: Sidinga ukuklama isethi yama-ejenti azimele akwazi ukusebenzisana kanye nendawo ezungezile ukuze sifinyelele umgomo ofanayo. I-ejenti ngayinye kufanele ibe namakhono ayo okwenza izinqumo, ulwazi, kanye nezinsiza.
Igrafu yolwazi: Sidinga ukwakha igrafu yolwazi emele ubudlelwano phakathi kwama-ejenti, indawo ezungezile, kanye nezinjongo zohlelo. Le grafu ingasetshenziswa ukubonisana ngesimo sohlelo nokwenza izinqumo.
Ukwenza Izinqumo: Kudingeka senze inqubo yokuthatha izinqumo evumela ama-ejenti ukuthi acabange ngesimo sohlelo futhi enze izinqumo ezisekelwe kugrafu yolwazi.
Ukwenziwa Kwesenzo: Kudingeka senze inqubo yokwenza isenzo esivumela ama-ejenti ukuthi enze izinqumo zawo futhi axhumane nemvelo.

Nasi isibonelo esilula sesistimu yesofthiwe yokuziphilisa eyakhiwe nge-LangChain:

ngenisa i-langchain
kusuka ku-langchain.agents i-ejenti yokungenisa
kusuka ku-langchain.llms ngenisa i-LLM


igrafu = {
    'ama-ejenti': ['i-ejenti1', 'i-ejenti2', 'i-ejenti3'],
    'imvelo': ['database', 'API', 'IoT device'],
    'izinhloso': ['bona amaphutha', 'phendula ukwehluleka', 'lulama emaphutheni']
}

# Chaza inqubo yokwenza izinqumo
def decision_making(umenzeli, ulwazi_graph):
    # Isizathu mayelana nesimo sohlelo
    state = knowledge_graph['agent'][ejenti]
    # Yenza isinqumo esisekelwe kuhulumeni
    decision = 'phendula ukwehluleka'
    buyisela isinqumo

# Chaza inqubo yokwenza isenzo
def action_execution(i-ejenti, isinqumo):
    # Yenza isinqumo
    uma isinqumo == 'phendula ukwehluleka':
        # Phendula ekuhlulekeni
        phrinta('Iphendula ekuhlulekeni...')
    buyela

# Dala i-ejenti ye-LangChain
i-ejenti = i-ejenti (
    igama='i-ejenti yokuziphilisa',
    decision_making=ukwenza_isinqumo,
    isenzo_execution=ukwenza_isenzo,
    ulwazi_igrafu=igrafu
)

# Dala i-LangChain LLM
llm = LLM(
    igama='ukuziphilisa-llm',
    imodeli='davinci'
)

# Dala uhlelo lwe-LangChain
uhlelo = langchain.System(
    i-ejenti=i-ejenti,
    llm=llm
)

# Qalisa uhlelo
system.run()

Lesi sibonelo sibonisa isistimu yesofthiwe yokuziphilisa elula ekwazi ukubona amaphutha, iphendule ekuhlulekeni, futhi ilulame emaphutheni ngokushesha. Uhlelo luqukethe isethi yama-agent azimele asebenzisana namanye kanye nendawo yawo ukuze afinyelele umgomo ofanayo. Ama-ejenti asebenzisa igrafu yolwazi ukuze acabange ngesimo sohlelo futhi enze izinqumo ezisekelwe kugrafu.

Imiyalo/Amazwibela Aphezulu Okuyi-5-10

Nansi imiyalo/ amazwibela aphezulu angu-5-10 e-LangChain:

langchain.Agent(): Idala i-ejenti entsha ye-LangChain.
langchain.LLM(): Idala i-LLM entsha ye-LangChain.
langchain.System(): Idala isistimu entsha ye-LangChain.
umenzeli.ukwenza_isinqumo(): Iqhuba inqubo yokuthatha izinqumo ye-ejenti.
umenzeli.action_execution(): Iqalisa inqubo yokwenza isenzo kumenzeli.
llm.predict(): Wenza isibikezelo esebenzisa i-LLM.
system.run(): Isebenzisa uhlelo lwe-LangChain.
graph.add_node(): Yengeza inodi entsha kugrafu yolwazi.
graph.add_edge(): Yengeza umphetho omusha kugrafu yolwazi.
graph.remove_node(): Ikhipha inodi kugrafu yolwazi.

Ukubona

Nakhu ukuboniswa okulula kwegrafu yolwazi kusetshenziswa i-Chart.js:

{
  "type": "graph",
  "idatha": {
    "izindawo": [
      {"id": "i-ejenti1", "label": "Umenzeli 1"},
      {"id": "i-agent2", "label": "Umenzeli 2"},
      {"id": "i-ejenti3", "label": "Umenzeli 3"},
      {"id": "isizindalwazi", "label": "Isizindalwazi"},
      {"id": "API", "label": "API"},
      {"id": "IoT device", "label": "IoT Device"}
    ],
    "emaphethelweni": [
      {"source": "agent1", "target": "database"},
      {"source": "agent2", "target": "API"},
      {"source": "agent3", "target": "IoT device"},
      {"source": "agent1", "target": "agent2"},
      {"source": "agent2", "target": "ejenti3"}
    ]
  },
  "izinketho": {
    "isakhiwo": {
      "type": "circular"
    },
    "amanodi": {
      "usayizi": 50
    },
    "emaphethelweni": {
      "ububanzi": 2
    }
  }
}

Lokhu kubukwa kukhombisa ubudlelwano phakathi kwama-ejenti, imvelo, kanye nezinhloso zesistimu.

Isiphetho

Ukwakha izinhlelo ze-software zokuziphilisa nge-multi-agent AI architectures kanye ne-LangChain inikeza isisombululo esinamandla sokuqinisekisa ukwethembeka, ukuqina, nokuqina kwezinhlelo zesofthiwe eziyinkimbinkimbi. Ngokusabalalisa inqubo yokuthatha izinqumo kuma-ejenti amaningi, lawa masistimu angafunda kokuhlangenwe nakho, azivumelanise nezimo ezishintshayo, futhi alulame emaphutheni ngokushesha. Nge-LangChain, abathuthukisi bangakha izinhlelo ze-AI eziyinkimbinkimbi ezihlanganisa nenqwaba yemithombo yedatha, okuhlanganisa imininingwane yolwazi, ama-API, namadivayisi we-IoT. Ngokulandela ikhodi yesibonelo nokusebenzisa imiyalo/ amazwibela aphezulu angu-5-10, abathuthukisi bangazakhela izinhlelo zabo zesofthiwe zokuziphilisa nge-LangChain.

Ukwakha Amasistimu Esofthiwe Okuziphilisa Nge-Multi-Agent AI Architectures kanye ne-LangChain

Have Questions?