[Bug]: Error when using a generic datamodel with structured output

[schelv]

Bug Description

Using a generic BaseModel with as_structured_llm causes a 400 BadRequest from OpenAI.
This is probably due to the generated JSON schema name.

Version

0.14.8

Steps to Reproduce

The following code works correctly with a non-generic model (DataModel) but fails when using the generic version (GenericDataModel[int]):

from typing import Sequence
from llama_index.core.base.llms.types import ChatMessage
from llama_index.llms.openai import OpenAI
from pydantic import BaseModel

class DataModel(BaseModel):
    answer: int

class GenericDataModel[T](BaseModel):
    answer: T

def structured_output_existential_chat(datamodel):
    llm = OpenAI(model="gpt-4o")
    sllm = llm.as_structured_llm(output_cls=datamodel)

    messages: Sequence[ChatMessage] = [ChatMessage(content="What is the meaning of life?")]
    response = sllm.chat(messages)
    print(response)

structured_output_existential_chat(DataModel)

print(GenericDataModel[int].model_json_schema())
structured_output_existential_chat(GenericDataModel[int])

Observed behavior

assistant: {"answer":42}

{'properties': {'answer': {'title': 'Answer', 'type': 'integer'}}, 'required': ['answer'], 'title': 'GenericDataModel[int]', 'type': 'object'}

Then:

openai.BadRequestError: Error code: 400 - {'error': {'message': "Invalid 'response_format.json_schema.name': string does not match pattern. Expected a string that matches the pattern '^[a-zA-Z0-9_-]+$'.", 'type': 'invalid_request_error', 'param': 'response_format.json_schema.name', 'code': 'invalid_value'}}

[dosubot]

This happens because generic Pydantic models like GenericDataModel[int] generate a schema title with brackets (e.g., GenericDataModel[int]), but OpenAI's API requires the schema name to match ^[a-zA-Z0-9_-]+$—so brackets and similar characters aren't allowed. LlamaIndex passes the schema title directly to OpenAI without sanitizing it, which causes the 400 error you see when using generics with as_structured_llm see code reference.

To work around this, set a valid schema title on your generic model using Pydantic's config, for example:

class GenericDataModelInt(BaseModel):
    answer: int

    class Config:
        title = "GenericDataModelInt"

Or, for Pydantic v2:

from pydantic import BaseModel, ConfigDict

class GenericDataModelInt(BaseModel):
    answer: int
    model_config = ConfigDict(title="GenericDataModelInt")

This will ensure the schema name is valid for OpenAI and avoid the error.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}

   

[vivek12345]

How about if we add sanitizer function for name
function = {
"name": schema["title"],
"description": schema_description,
"parameters": pydantic_class.schema(),
}

so instead of just doing schema["title"] we first pass this through our regex check function which can strip out the "[" and then the safe name would come out to be "GenericDataModel" which is accepted by open ai

@logan-markewich if this is a decent approach to this problem, I would like to pick this item