How to describe fields when using structured output

API
1

I actually have two similar questions, both related to structured output. One for using structured output for the LLM response, and the other when using it for function calling.

When using “response_format” to specify the response schema, is it possible to include field descriptions? Or is the only option to describe them in the prompt? And if it’s possible, is it possible both when using a JSON schema and a pydantic object?

And a similar question goes for function calling. If I use a JSON schema to describe my function arguments, I can use the “description” field to describe them. Can I also describe them, other than in the prompt, when using a pydantic model?

1 Like
2

Yes you can:

class TranslationChunk(BaseModel):
                improved_sentences: List[str] = Field(
                    ...,
                    description=f"List of translated sentences in the target language (which is given in {target_language}).",
                )

However, I noticed that the description tends to overwrite the instructions. Could you let me know if this is also the case for you?