langgraph/tutorials/extraction/retries/

[giscus[bot]]

langgraph/tutorials/extraction/retries/

Build language agents as graphs

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

[ricardo-casa]

How could I incorporate the PydanticOutputParser into the output? I had tried to incorporate it inside the chain = prompt | bound_llm string definition, as mentioned in the docs in StructuredOutput, but I got errors when using it with OpenAI models.

[vbarda]

@ricardo-casa you can use llm.with_structured_output(PydanticModel). hope this helps!

[ricardo-casa]

@vbarda thanks for the answer, but I am a bit confused. Exactly where in the code llm.with_structured_output(PydanticModel) should it go? since both with_structured_output and bind_tools return RunnablesSequence so I can't add the tools after defining the llm.with_structured_output.

[hinthornw]

with_structured_output creates a parser for you so you wouldn't have to manually provide a PydanticOutputParser

[Glinte]

@vbarda thanks for the answer, but I am a bit confused. Exactly where in the code llm.with_structured_output(PydanticModel) should it go? since both with_structured_output and bind_tools return RunnablesSequence so I can't add the tools after defining the llm.with_structured_output.

I ran into this problem too. I basically noticed that you can also just pass type[BaseModel] as part of the tools list into .bind_tools(), and then pass the AIMessage into ToolNode(tools=[BaseTools and BaseModels]), and then do an isinstance check to see if a BaseModel is returned. I also wrapped around ToolNode to forge the output to be an AIMessage with the original AIMessage's id when a BaseModel is detected.

[vishnudev-k]

Does it support opensource models severed using llama.cpp or something?

[vishnudev-k]

.