langgraph/how-tos/extraction/retries/

[giscus[bot]]

langgraph/how-tos/extraction/retries/

Build language agents as graphs

https://langchain-ai.github.io/langgraph/how-tos/extraction/retries/

[Jaid844]

Hi William, can you make a Youtube video of it? It will be much better if this was explained by you

[hinthornw]

What do you mean by version validators?

[ricardo-casa]

Yes, @Jaid844, I’m referring to Pydantic validators. I apologize for the confusion. After examining the JSON Patch code, I noticed that the Pydantic validators were removed. In my code, I retained the Pydantic validators, but whenever I attempt extraction, I consistently receive BadRequest errors from OpenAI:
BadRequestError: Error code: 400 - {'error': {'message': "Invalid parameter: 'tool_call_id' of 'call_3eze1wLN7FF6NAE1zD6Ji2oT' not found in 'tool_calls' of previous message.", 'type': 'invalid_request_error', 'param': 'messages.[5].tool_call_id', 'code': None}}.

After debugging and not finding any issues, I tried removing the Pydantic validators and it worked well. I’m having difficulty adapting these validators to the JSON Patch Code Section.

[Jaid844]

Can you give me your pydantic model here ,I think there might be issue with the serialization of the pydantic models.
for instance my pydantic model looks like this

class To_Loan_tool_1(BaseModel):
"""
This function will be able to calculate the loan amount for the customer ,Initially the rate will be 5 %
but if the customer is felling a bit steep pay,A 10 % rate will be calculated

  """

rate: Union[loan_amount_5, loan_amount_10] = Field(discriminator="rate")

class Config:
    schema_extra = {
        "example 1": {
            "rate": 5,
            "name": "jake",
            "dialogue": "I would like a loan adjustment",
        },
        "example 2": {
            "rate": 10,
            "name": "shela",
            "dialogue": "The loan adjustment is too steep for me",
        }
    }

[ricardo-casa]

Yes, no problem. This is my full pydantic schema:

`class ContactResponse(BaseModel):
"""
Utiliza esta herramienta para generar una respuesta, SIEMPRE USALA
PARA CONTESTARLE AL USUARIO.
"""
requirement: str = Field(
description="""
Informacion relacionada lo solicitado por parte del cliente
esto incluye:
1) Modelos, Series y Lineas (Equipos de Laboratorios)
2) Especificaciones sobre equipos
3) Aplicaciones
4) Materiales
5) Consumibles
6) Cantidades
7) Servicios ['calificacion', 'mantenimiento', 'etc']
8) Marcas
""",

    example=(
        "Refrigerador para vacunas entre 300 a 600 L ",
        "Centrífuga para 24 tubos, para muestras sanguíneas, de 7 mililitro",
        "Campana de extracción de humos para manejo de sustancias corrosivas, como por ejemplo, ácido sulfúrico, anhídrido acético, ácido acético glaciar, salida solo para servicio de electricidad, ancho de 145 cm de preferencia.",
        "calificacion  del sistema computarizado de la cámaras de estabilidad. Software acoplado Newtronic ICDAS 2.1 lite versión 2.1.1. "
        "Analizador, autoclave, fotoespectro, centrífuga"
    )
)

interest: Optional[List[InterestEnum]] = Field(
    description="""
        "Interes del cliente basado en las líneas de producto que "
        "vende Inolab Especialistas"
    """
)

comments: Optional[str] = Field(
    description="""
        Comentarios sobre la solicitud del cliente, incluyen:
            1) Horarios
            2) Localidades y Lugares
            3) Detalles de entrega
            4) Presupuestos
            5) Condiciones de contacto
            6) Otros Contactos
    """,
    example=(
        "Contactar únicamente por WhatsApp antes de las 2 p.m. ",
        "Enviar la cotización al primer correo y con copia para [email protected], ",
        "contactar solo por WhatsApp o correo ya que no puede recibir llamadas, ",
        "Esterilizar material de curación, espejos vaginales o material sólido que se requiera para la atención médica en una clínica médica. CDMX, contactar de 2 a 6 WhatsApp, ",
        "Tlalnepantla, Edo.Mex Presupuesto de 30mil Atendido por Carlos",
        "Quieren equipar por completo un laboratorio de análisis clínico "
    )
)

@validator('requirement')
def validate_requirement(cls, v):
    if not None:
        raise ValueError('Requerimento incompleto, se necesita extraer el requerimento')
    return v

@validator('interest')
def validate_interest(cls, v):
    if v not in InterestEnum.__members__.values() or not None:
        raise ValueError('Interés no válido')
    return v

`

Overall, this is for fields extraction for some kind of emails.

[Jaid844]

I am guessing instead of using examples in your models .I think that is what might be causing issue in your serialization.
Can you use class config like this in your examples of your code
class Config: schema_extra = { "example": { "reason": "User changed their mind about the current task.", }, "example 2": { "reason": """The calculated loan amount for this month, after applying a 5% discount, is $16.67. Would you like to proceed with this amount, James?""", }, "example 3": { "reason": "I have fully calculated the loan amount", } }