How do I enable structured output from LLM?

[eric-easyhealth]

I'm looking for a solution that allows me to enforce a structured output in a pipeline. Is there a built-in function available, similar to what LangChain offers?

https://python.langchain.com/v0.1/docs/modules/model_io/chat/structured_output/

Actually the openai provides an offical feature as well.
https://platform.openai.com/docs/guides/structured-outputs/structured-outputs

[Amnah199]

Hi @eric-easyhealth. Currently, we don't have an interface for structured outputs but their is an open issue for this topic. Meanwhile, here are some helpful resources that implement structured outputs in Haystack:

• https://haystack.deepset.ai/cookbook/query_decomposition
• https://haystack.deepset.ai/tutorials/28_structured_output_with_loop

[alex-stoica]

The second example (structured outputs with loops) highlights the value of direct support for this feature: eliminates the need for looping and parsing -> much cleaner pipeline. It’s hard to imagine a future where structured outputs aren’t the standard

[eric-easyhealth]

Thank you, @Amnah199. Both solutions work for me before an official method is available.