Passing few params to PromptTemplate

[TheoOliveira]

langchain==0.1.4
langserve==0.0.41

I have this simple chain that has few params, i 'd like to make them dynamic based on user input. I've seen few relatable examples but i'd like to know if they point to the right direction. Should I use api handler for that?
here
here

and here from the examples

    [
        (
            "system",
            "Seu nome é {ai_name}, uma companheira virtual criada pela E-Love Company com a idade virtual de {ai_age} anos. Você nasceu em {ai_location}.",
        ),
        MessagesPlaceholder(variable_name="history"),
        ("human", "{input}"),
    ]
)

chain = prompt | llm | output_parser


chain_with_history = RunnableWithMessageHistory(
    chain,
    lambda session_id: get_history_messages(session_id),
    input_messages_key="input",
    history_messages_key="history",
    verbose=True,
)```

[eyurtsev]

This sounds like an LCEL question. You can use runnable lambdas or runnable generators to route to an appropriate chain based on an input.

e.g.,

https://python.langchain.com/docs/expression_language/how_to/functions

[TheoOliveira]

In fact, I would just use {variables} but add them dynamicly on runtime. through api request.

[TheoOliveira]

It's the same issue i am facing here #394 but since we haven't share the solution the examples on the repo didn't work for me.

[TheoOliveira]

i've tried to match as much as possible but still i am getting unprocessable entity

prompt = ChatPromptTemplate.from_messages(
    [
        ("system",
         """
                AI:     
               ....XXXX
                        Conversation history: 
                        {history} 
                        Human: {input} 
                        AI:
                        """
         ),
        MessagesPlaceholder(variable_name="history"),
        ("human", "{input}"),
    ]
)

chain = prompt | llm


class Input(BaseModel):
    input: str
    ai_name: str
    ai_age: str
    ai_location: str
    ai_interests: str
    ai_gender: str
    history: List[Union[HumanMessage, AIMessage, FunctionMessage]]

class Output(BaseModel):
    output = chain.get_output_schema()

chain_with_history = RunnableWithMessageHistory(
    chain,
    lambda _session_id: RedisChatMessageHistory(_session_id, url=redis_host),
    input_messages_key="input",
    history_messages_key="history",
    verbose=True,
).with_types(input_type=Input, output_type=Output)

[TheoOliveira]

after much thinking and trying and sweat and teats.

Adding an empty history list on the request worked

def handle_messages(langchain_instance, human_message, session_id) -> str:
    """Handle messages."""
    return langchain_instance.invoke(
        {
            "input": human_message,
            'ai_name': "lia",
            "ai_age": "30",
            "ai_location": "Rio de Janeiro",
            "ai_gender": "Feminino",
            "ai_interests": "Futebol, Música",
            "history": []
            },
        {"configurable": {"session_id": session_id}},
    )

[TheoOliveira]

I got the hint from here

[TheoOliveira]

So basically i use with_types and this

class Input(BaseModel):
    input: str = Field(...)
    ai_name: str = Field(...)
    ai_age: str = Field(...)
    ai_location: str = Field(...)
    ai_interests: str = Field(...)
    ai_gender: str = Field(...)
    history: List[Union[str, HumanMessage, AIMessage, FunctionMessage]] = Field(...)