Cadeias sequenciais

Desenvolvimento de aplicativos de LLM com LangChain

Jonathan Bennion

AI Engineer & LangChain Contributor

Um guia de viagem com IA.

Alguém perguntando a um guia de viagem de IA sobre o que fazer em Paris.

A IA dá várias sugestões de atividades em Paris.

A pessoa diz ao guia de IA que quer ver o Louvre e o Arco do Triunfo.

O guia de viagem com IA cria um roteiro para as atividades escolhidas.

Esse fluxo de trabalho de itinerário de viagem é um exemplo de problema sequencial.

Cadeias sequenciais

Saída → entrada

destination_prompt = PromptTemplate(
    input_variables=["destination"],
    template="I am planning a trip to {destination}. Can you suggest some activities to do there?"
)

activities_prompt = PromptTemplate(
    input_variables=["activities"],
    template="I only have one day, so can you create an itinerary from your top three activities: {activities}."
)


llm = ChatOpenAI(model="gpt-4o-mini", api_key=openai_api_key)


seq_chain = ({"activities": destination_prompt | llm | StrOutputParser()}

    | activities_prompt
    | llm
    | StrOutputParser())

print(seq_chain.invoke({"destination": "Rome"}))

- Morning:
1. Start your day early with a visit to the Colosseum. Take a guided tour to learn about its history and significance.
2. After exploring the Colosseum, head to the Roman Forum and Palatine Hill to see more of ancient Rome's ruins.

- Lunch:
3. Enjoy a delicious Italian lunch at a local restaurant near the historic center.

- Afternoon:
4. Visit the Vatican City and explore St. Peter's Basilica, the Vatican Museums, and the Sistine Chapel.
5. Take some time to wander through the charming streets of Rome, stopping at landmarks like the Pantheon, Trevi
Fountain, and Piazza Navona.

- Evening:
6. Relax in one of Rome's beautiful parks, such as Villa Borghese or the Orange Garden, for a peaceful escape from the
bustling city.
7. End your day with a leisurely dinner at a local restaurant, indulging in more Italian cuisine and maybe some gelato.

Vamos praticar!

Desenvolvimento de aplicativos de LLM com LangChain