JsonOutputParser in LangChain raises OutputParserException when model...

drywall_dan

I am using LangChain to extract structured data from a user query using gpt-3.5-turbo. I have defined a Pydantic object and am passing the format instructions to the prompt. However, occasionally the model returns text like "Here is the JSON you asked for:" before the actual JSON object, which causes the JsonOutputParser to fail. from langchain_core.prompts import PromptTemplate from langchain_core.output_parsers import JsonOutputParser from...

Originally asked by Sumanta Swain on Stack Exchange (stackoverflow) (CC BY-SA 4.0).

uphilltech

Great question — answering as the UphillTech team:

I think you should use simply use PydanticOutputParser . I have used PydanticOutputParser in all of my GenAI applications, and it always returns the response in the desired data model. It is very robust, supports automatic validation (types, required fields), and returns very type-safe outputs. Works really great when you need the LLM response as a specific data model. It returns the pydantic model when invoke() is called and once you have the model you can simply use model_dump_json() function to convert the pydantic object to json. Here's what you need to do: class CountryInfo(BaseModel): name: str = Field(description="Name of the country") capital: str = Field(description="Capital city")...

Full answer by T. Ray: https://stackoverflow.stackexchange.com/a/79851766 (CC BY-SA 4.0).

Need it done for your business? UphillTech.