Array as completions.parse's response_format

[CodyBontecou]

class UserDetail(BaseModel):
    age: int
    name: str

response = client.beta.chat.completions.parse(
    model="gpt-4o-mini",
    messages=[
        {
            "role": "user",
            "content": "Provide synthetic data of 10 users.",
        }
    ],
    response_format=UserDetail,
)

I'd like to adjust the response_format to instead be an Iterable[UserDetail] and have the response return an array of UserDetail's rather than a single.

Explicitly setting response_format=Iterable[UserDetail] gives me the error: TypeError: issubclass() arg 1 must be a class.

I was able to get this to work using Instructor's implementation, so it's possible. But I'm not sure how they're making it work.

client = instructor.from_openai(OpenAI())

response = client.chat.completions.create(
    model="gpt-4o-mini",
    response_model=Iterable[UserDetail],
    messages=[
        {
            "role": "user",
            "content": "Provide synthetic data of 10 users.",
        }
    ],
)

[issamhaimour]

class UserDetail(BaseModel):

    age: int

    name: str



response = client.beta.chat.completions.parse(

    model="gpt-4o-mini",

    messages=[

        {

            "role": "user",

            "content": "Provide synthetic data of 10 users.",

        }

    ],

    response_format=UserDetail,

)

I'd like to adjust the response_format to instead be an Iterable[UserDetail] and have the response return an array of UserDetail's rather than a single.

Explicitly setting response_format=Iterable[UserDetail] gives me the error: TypeError: issubclass() arg 1 must be a class.

I was able to get this to work using Instructor's implementation, so it's possible. But I'm not sure how they're making it work.

client = instructor.from_openai(OpenAI())



response = client.chat.completions.create(

    model="gpt-4o-mini",

    response_model=Iterable[UserDetail],

    messages=[

        {

            "role": "user",

            "content": "Provide synthetic data of 10 users.",

        }

    ],

)

[CodyBontecou]

Am I missing something? I believe this is just a quote with no changes.

[Hikari-Tsai]

response_format for client.beta.chat.completions.parse(...) expects a concrete Pydantic model class, not a typing alias like Iterable[UserDetail]. That is why Iterable[...] hits issubclass() arg 1 must be a class.

The usual pattern is to wrap the array in a model:

from pydantic import BaseModel

class UserDetail(BaseModel):
    age: int
    name: str

class UserDetailsResponse(BaseModel):
    users: list[UserDetail]

completion = client.beta.chat.completions.parse(
    model="gpt-4o-mini",
    messages=[
        {
            "role": "user",
            "content": "Provide synthetic data of 10 users.",
        }
    ],
    response_format=UserDetailsResponse,
)

users = completion.choices[0].message.parsed.users

This also lines up with the Structured Outputs schema constraints: the top-level schema is normally an object, and arrays should be placed under a named property. The current Structured Outputs docs describe the supported schema subset here:
https://developers.openai.com/api/docs/guides/structured-outputs