Kumanga Mapulogalamu Odzichiritsa Okha ndi Multi-Agent AI Architectures ndi LangChain

Kumanga Mapulogalamu Odzichiritsa Odzichiritsa okhala ndi Multi-Agent AI Architectures ndi LangChain

Chiyambi

M'mawonekedwe amakono amakono opanga mapulogalamu, kumanga machitidwe odzichiritsa okha kwakhala chofunikira kwambiri powonetsetsa kudalirika, scalability, ndi kulimba kwa mapulogalamu ovuta. Zomangamanga zachikhalidwe za monolithic sizilinso zokwanira kuti zikwaniritse zofuna za mapulogalamu amakono amakono, kumene machitidwe ayenera kukhala okonzeka kusintha kusintha, kuthana ndi zolephera zosayembekezereka, ndikuchira mwamsanga. Apa ndipamene zomangamanga za AI zambiri ndi LangChain zimalowa, kupereka njira yothetsera mphamvu yopangira mapulogalamu odzichiritsa okha omwe angaphunzire, kusintha, ndi kuyankha kusintha kwa zinthu.

Kodi Multi-Agent AI Architectures ndi chiyani?

Zomangamanga zamitundu yambiri za AI ndi mtundu wamakina ogawidwa a AI omwe amakhala ndi othandizira angapo odziyimira pawokha omwe amalumikizana wina ndi mnzake komanso chilengedwe chawo kuti akwaniritse cholinga chimodzi. Wothandizira aliyense ali ndi luso lake lopanga zisankho, chidziwitso, ndi zida zake, ndipo amatha kulumikizana wina ndi mnzake kugawana zambiri, kulinganiza zochita, ndikusintha momwe zinthu zikuyendera.

Pankhani ya mapulogalamu a mapulogalamu, zomangamanga zambiri za AI zingagwiritsidwe ntchito popanga machitidwe odzichiritsa okha omwe amatha kuzindikira ndi kuyankha zolakwika, zolephera, ndi zovuta zina mu nthawi yeniyeni. Pogawira njira yopangira zisankho kwa othandizira angapo, makinawa amatha kuphunzira kuchokera pazomwe akumana nazo, kusintha momwe zinthu zikuyendera, ndikuchira msanga.

LangChain ndi chiyani?

LangChain ndi laibulale yotseguka ya Python yomanga makina a AI amitundu yambiri. Amapereka njira yosavuta, yosinthika, komanso yowonjezereka yomanga machitidwe ovuta a AI omwe amatha kuphunzira, kusintha, ndi kuyankha kusintha kwa zinthu. LangChain yapangidwa kuti igwire ntchito ndi machitidwe osiyanasiyana a AI, kuphatikizapo TensorFlow, PyTorch, ndi Keras, ndipo angagwiritsidwe ntchito pomanga machitidwe omwe amaphatikizana ndi magwero osiyanasiyana a deta, kuphatikizapo ma database, APIs, ndi zipangizo za IoT.

Kumanga Mapulogalamu Odzichiritsa Odzichiritsa ndi LangChain

Kupanga pulogalamu yodzichiritsa yokha ndi LangChain, tifunika kupanga mapangidwe a AI amitundu yambiri omwe amatha kuzindikira ndikuyankha zolakwika, zolephera, ndi zolakwika zina munthawi yeniyeni. Nayi chithunzithunzi chapamwamba cha zomangamanga:

1. Kupanga Ma Agent: Tiyenera kupanga gulu la othandizira odziyimira pawokha omwe amatha kulumikizana wina ndi mnzake komanso chilengedwe chawo kuti akwaniritse cholinga chimodzi. Wothandizira aliyense ayenera kukhala ndi luso lake lopangira zisankho, chidziwitso, ndi zida zake.
2. Chidziwitso Chachidziwitso: Tiyenera kupanga chithunzi cha chidziwitso chomwe chikuyimira maubwenzi pakati pa othandizira, chilengedwe, ndi zolinga za dongosolo. Chithunzichi chingagwiritsidwe ntchito kulingalira za momwe dongosololi likukhalira ndikupanga zisankho.
3. Kupanga zisankho: Tiyenera kukhazikitsa njira yopangira zisankho zomwe zimalola othandizira kulingalira za momwe dongosololi lilili ndikupanga zisankho motengera chidziwitso cha graph.
4. Kuchitapo kanthu: Tiyenera kukhazikitsa njira yochitira zomwe zimalola othandizira kuchita zisankho zawo ndikuyanjana ndi chilengedwe.

Nachi chitsanzo chosavuta cha pulogalamu yodzichiritsa yokha yomangidwa ndi LangChain:

import langchain
kuchokera ku langchain.agents import Agent
kuchokera ku langchain.llms import LLM


chithunzi = {
    'othandizira': ['agent1', 'agent2', 'agent3'],
    'environment': ['database', 'API', 'IoT device'],
    'zolinga': ['zindikira zolakwika', 'yankhani zolephera', 'chirani ku zolakwika']
}

# Fotokozani njira yopangira zisankho
def decision_making(wothandizira, knowledge_graph):
    # Chifukwa cha momwe dongosololi lilili
    state = knowledge_graph['agents'][wothandizira]
    # Pangani chisankho potengera dziko
    decision = 'kuyankha kulephera'
    chisankho chobwezera

# Fotokozani njira yochitira
def action_execution (wothandizira, chisankho):
    # Pangani chisankho
    ngati chisankho == 'kuyankha kulephera':
        # Yankhani kulephera
        print('Kuyankha kulephera...')
    kubwerera

# Pangani wothandizira wa LangChain
wothandizira = Agent (
    dzina='wodzichiritsa yekha',
    kusankha_kupanga=kupanga_zisankho,
    action_execution=kuchita_kuchita,
    chidziwitso_graph=graph
)

# Pangani LangChain LLM
LLM = LLM(
    dzina='kudzichiritsa-llm',
    model='davinci'
)

# Pangani dongosolo la LangChain
dongosolo = langchain.System(
    wothandizira = wothandizira,
    llm=m
)

# Pangani ndondomekoyi
system.run ()

Chitsanzochi chikuwonetsa pulogalamu yosavuta yodzichiritsa yokha yomwe imatha kuzindikira zolakwika, kuyankha zolephera, ndikuchira msanga. Dongosololi lili ndi gulu laodziyimira pawokha omwe amalumikizana wina ndi mnzake komanso chilengedwe chawo kuti akwaniritse cholinga chimodzi. Othandizira amagwiritsa ntchito graph yodziwa kulingalira za momwe dongosololi likukhalira ndikupanga zisankho motengera graph.

**Malamulo/Zidutswa Zapamwamba za 5-10 **

Nawa maulamuliro/zidutswa 5-10 zapamwamba kwambiri za LangChain:

1. langchain.Agent(): Amapanga wothandizira LangChain watsopano.
2. langchain.LLM(): Imapanga LangChain LLM yatsopano.
3. langchain.System(): Amapanga dongosolo latsopano la LangChain.
4. agent.decision_making(): Amayendetsa zisankho za wothandizira.
5. agent.action_execution(): Imayendetsa zochita kwa wothandizira.
6. llm.predict(): Amalosera pogwiritsa ntchito LLM.
7. system.run (): Imayendetsa dongosolo la LangChain.
8. graph.add_node(): Imawonjezera nodi yatsopano ku graph ya chidziwitso.
9. graph.add_edge(): Imawonjezera m'mphepete mwachidziwitso.
10. graph.remove_node(): Imachotsa mfundo pa graph ya chidziwitso.

**Kuwona **

Nayi chithunzithunzi chosavuta cha graph ya chidziwitso pogwiritsa ntchito Chart.js:

{
  "type": "graph",
  "data": {
    "mfundo": [
      {"id": "agent1", "label": "Mthandizi 1"},
      {"id": "agent2", "label": "Mthandizi 2"},
      {"id": "agent3", "label": "Mthandizi 3"},
      {"id": "database", "label": "Database"},
      {"id": "API", "label": "API"},
      {"id": "IoT device", "label": "IoT Chipangizo"}
    ],
    "mbali": [
      {"source": "agent1", "target": "database"},
      {"source": "agent2", "target": "API"},
      {"source": "agent3", "target": "IoT device"},
      {"source": "agent1", "target": "agent2"},
      {"source": "agent2", "target": "agent3"}
    ]
  },
  "zosankha": {
    "mapangidwe": {
      "type": "zozungulira"
    },
    "mfundo": {
      "kukula": 50
    },
    "mbali": {
      "m'lifupi": 2
    }
  }
}

Chiwonetserochi chikuwonetsa mgwirizano pakati pa othandizira, chilengedwe, ndi zolinga za dongosolo.

Mapeto

Kumanga mapulogalamu odzichiritsa okha omwe ali ndi zomangamanga zambiri za AI ndi LangChain amapereka yankho lamphamvu kuti atsimikizire kudalirika, scalability, ndi kukhazikika kwa machitidwe ovuta a mapulogalamu. Pogawira njira yopangira zisankho kwa othandizira angapo, makinawa amatha kuphunzira kuchokera pazomwe akumana nazo, kusintha momwe zinthu zikuyendera, ndikuchira msanga. Ndi LangChain, opanga mapulogalamu amatha kupanga machitidwe ovuta a AI omwe amaphatikizana ndi magwero ambiri a deta, kuphatikizapo ma database, APIs, ndi zipangizo za IoT. Potsatira kachidindo kachitsanzo ndikugwiritsa ntchito 5-10 yogwiritsidwa ntchito kwambiri malamulo / zidule, omanga akhoza kupanga mapulogalamu awo odzichiritsa okha ndi LangChain.