PydanticUserError when instantiating SyntheticDataGenerator

[zakhikhan]

I am attempting to use this library to generate tabular synthetic data. Any time I try to instantiate the SyntheticDataGenerator class from tabular_synthetic_data (either through the create_openai_data_generator function or directly), I get an error like the below:

PydanticUserError: SyntheticDataGenerator is not fully defined; you should define BaseCache, then call SyntheticDataGenerator.model_rebuild().

For further information visit https://errors.pydantic.dev/2.10/u/class-not-fully-defined

versions (all installed via pip)
langchain_experimental: 0.3.3
langchain: 0.3.11
langchain_openai 0.2.12
langchain_core: 0.3.24

Here is simplified example code that causes the error:

from langchain_openai import ChatOpenAI
from langchain_core.prompts import FewShotPromptTemplate
from langchain_experimental.tabular_synthetic_data.base import SyntheticDataGenerator
from pydantic import BaseModel

OPENAI_TEMPLATE = PromptTemplate(input_variables=["example"], template="{example}")

examples = [
    {"example": "toothbrush_name: Crest Brush, price: 10"},
    {"example": "toothbrush_name: Generic toothbrush, price: 5"}
]

class ToothbrushSchema(BaseModel):
    toothbrush_name: str
    price: int
    
output_schema=ToothbrushSchema

llm=ChatOpenAI(temperature=1)

prompt_template = FewShotPromptTemplate(
    examples=examples,
    suffix=("Generate synthetic data about {subject}."),
    input_variables=["subject"],
    example_prompt=OPENAI_TEMPLATE,
)

prompt=prompt_template

chain = llm.with_structured_output(ToothbrushSchema)

sdg = SyntheticDataGenerator(template=prompt_template,llm_chain=chain)

[zakhikhan]

I got it to work by creating a dev install of langchain_experimental and editing line 27 of tabular_synthetic_data/base.py in the SyntheticDataGenerator class

BEFORE:
llm: Optional[BaseLanguageModel] = None

AFTER:
llm: Optional[BaseChatModel] = None

I am not sure why this worked. BaseLanguageModel is the parent class of BaseChatModel.

[shub7389]

Hi,
Please can you share the git repo link or collab link with the updated code?

[JeremyDeng-xin]

i got the same error
this is code
`from langchain_core.prompts import PromptTemplate, FewShotPromptTemplate
from langchain_experimental.tabular_synthetic_data.openai import create_openai_data_generator
from langchain_experimental.tabular_synthetic_data.prompts import SYNTHETIC_FEW_SHOT_PREFIX, SYNTHETIC_FEW_SHOT_SUFFIX
from langchain_openai import ChatOpenAI, OpenAIEmbeddings
import os

from pydantic import BaseModel, Field

os.environ['LANGCHAIN_TRACING_V2'] = "true"
os.environ['LANGCHAIN_API_KEY'] = "lsv2_pt_xxx"

model = ChatOpenAI(model='gpt-3.5-turbo', temperature=0.8)

class MedicalBilling(BaseModel):
patient_id: int = Field(..., alias='patientId')
patient_name: str = Field(..., alias='patientName')
diagnosis: str = Field(..., alias='diagnosis')
procedure: str = Field(..., alias='procedure')
total_charge: float = Field(..., alias='totalCharge')
insurance_claim_amount: float = Field(..., alias='insurance claim amount')

examples = [
{
"example": "Patient ID: 123456, Patient name: 刘娜, Diagnosis: J20.9, Procedure: z456, Total Charge: 653.8, insurance claim amount: 1.00",
},
{
"example": "Patient ID: 674357, Patient name: 马钢, Diagnosis: J10.6, Procedure: z496, Total Charge: 5653.8, insurance claim amount: 5000.00",
},
{
"example": "Patient ID: 746565, Patient name: 冯亮, Diagnosis: E28.9, Procedure: z056, Total Charge: 6993.8, insurance claim amount: 1000.00",
}
]
openai_template = PromptTemplate(input_variables=['example'], template="{example}")

prompt_template = FewShotPromptTemplate(
prefix=SYNTHETIC_FEW_SHOT_PREFIX,
suffix=SYNTHETIC_FEW_SHOT_SUFFIX,
examples=examples,
example_prompt=openai_template,
input_variables=['subject','extra'],
)
generator = create_openai_data_generator(
output_schema=MedicalBilling,
llm=model,
prompt=prompt_template,
)
result = generator.generate(
subject='医疗帐单',
extra='名字可以是随机的，最好使用比较生僻的人名。账单金额大一点',
runs=10 # records
)
print(result)`