NameError: Incompatibilities with older Pydantic v2 versions causing NameError when importing from openai

[michaelgrewal]

Confirm this is an issue with the Python library and not an underlying OpenAI API

• This is an issue with the Python library

Describe the bug

As soon as the package openai v1.108.0 was released, it broke functionality with older pydantic v2 versions like v2.3.0. There is a NameError caused by the usage of __pydantic_extra__ like from recent changes made to the module openai.types.evals.runs: https://github.com/openai/openai-python/tree/main/src/openai/types/evals/runs

It was always OK in previous versions like openai v1.107.3 was OK.

/usr/local/lib/python3.10/site-packages/llama_index/__init__.py:13: in <module>
    from llama_index.callbacks.global_handlers import set_global_handler
/usr/local/lib/python3.10/site-packages/llama_index/callbacks/__init__.py:7: in <module>
    from .token_counting import TokenCountingHandler
/usr/local/lib/python3.10/site-packages/llama_index/callbacks/token_counting.py:6: in <module>
    from llama_index.utilities.token_counting import TokenCounter
/usr/local/lib/python3.10/site-packages/llama_index/utilities/token_counting.py:6: in <module>
    from llama_index.llms import ChatMessage, MessageRole
/usr/local/lib/python3.10/site-packages/llama_index/llms/__init__.py:26: in <module>
    from llama_index.llms.litellm import LiteLLM
/usr/local/lib/python3.10/site-packages/llama_index/llms/litellm.py:27: in <module>
    from llama_index.llms.litellm_utils import (
/usr/local/lib/python3.10/site-packages/llama_index/llms/litellm_utils.py:4: in <module>
    from openai.resources import Completions
/usr/local/lib/python3.10/site-packages/openai/resources/__init__.py:27: in <module>
    from .evals import (
/usr/local/lib/python3.10/site-packages/openai/resources/evals/__init__.py:3: in <module>
    from .runs import (
/usr/local/lib/python3.10/site-packages/openai/resources/evals/runs/__init__.py:3: in <module>
    from .runs import (
/usr/local/lib/python3.10/site-packages/openai/resources/evals/runs/runs.py:16: in <module>
    from .output_items import (
/usr/local/lib/python3.10/site-packages/openai/resources/evals/runs/output_items.py:17: in <module>
    from ....types.evals.runs import output_item_list_params
/usr/local/lib/python3.10/site-packages/openai/types/evals/runs/__init__.py:6: in <module>
    from .output_item_list_response import OutputItemListResponse as OutputItemListResponse
/usr/local/lib/python3.10/site-packages/openai/types/evals/runs/output_item_list_response.py:14: in <module>
    class Result(BaseModel):
/usr/local/lib/python3.10/site-packages/pydantic/_internal/_model_construction.py:98: in __new__
    private_attributes = inspect_namespace(
/usr/local/lib/python3.10/site-packages/pydantic/_internal/_model_construction.py:327: in inspect_namespace
    raise NameError(
E   NameError: Fields must not use names with leading underscores; e.g., use 'pydantic_extra__' instead of '__pydantic_extra__'.

To Reproduce

1. Install an older pydantic v2 like pydantic==2.3.0 and openai==1.108.0
2. Create a mytest.py file and import from openai.resources import Completions
3. Run mytest.py. Notice the NameError.
4. Then, install openai==1.107.3 and run mytest.py again. Notice there is no more NameError.

Code snippets

pip install pydantic==2.3.0 openai==1.108.0

from openai.resources import Completions

OS

Windows

Python version

v3.10.12

Library version

openai v1.108.0

[RobertCraigie]

Adding __pydantic_extra__ prevents a separate deserialisation bug, so I would recommend updating to a newer pydantic version if you can.

[michaelgrewal]

I updated to pydantic==2.11.9 and it fixed it.

Should openai update the minimum required pydantic version?

I encountered this bug because my project had pinned pydantic==2.3.0 which still falls into the range of supported versions defined by openai.

[alexmojaki]

The minimum would have to be updated to 2.8.0:

➜  ~ echo 'from openai.resources import Completions' > /tmp/test.py
➜  ~ uv run --with pydantic==2.8.0 --with openai /tmp/test.py      
➜  ~ uv run --with pydantic==2.7.4 --with openai /tmp/test.py      
Traceback (most recent call last):
  File "/tmp/test.py", line 1, in <module>
    from openai.resources import Completions
  File "/Users/alex/Library/Caches/uv/archive-v0/g5fn3MuSE2LcsLLzgAqHL/lib/python3.10/site-packages/openai/resources/__init__.py", line 27, in <module>
    from .evals import (
  File "/Users/alex/Library/Caches/uv/archive-v0/g5fn3MuSE2LcsLLzgAqHL/lib/python3.10/site-packages/openai/resources/evals/__init__.py", line 3, in <module>
    from .runs import (
  File "/Users/alex/Library/Caches/uv/archive-v0/g5fn3MuSE2LcsLLzgAqHL/lib/python3.10/site-packages/openai/resources/evals/runs/__init__.py", line 3, in <module>
    from .runs import (
  File "/Users/alex/Library/Caches/uv/archive-v0/g5fn3MuSE2LcsLLzgAqHL/lib/python3.10/site-packages/openai/resources/evals/runs/runs.py", line 16, in <module>
    from .output_items import (
  File "/Users/alex/Library/Caches/uv/archive-v0/g5fn3MuSE2LcsLLzgAqHL/lib/python3.10/site-packages/openai/resources/evals/runs/output_items.py", line 17, in <module>
    from ....types.evals.runs import output_item_list_params
  File "/Users/alex/Library/Caches/uv/archive-v0/g5fn3MuSE2LcsLLzgAqHL/lib/python3.10/site-packages/openai/types/evals/runs/__init__.py", line 6, in <module>
    from .output_item_list_response import OutputItemListResponse as OutputItemListResponse
  File "/Users/alex/Library/Caches/uv/archive-v0/g5fn3MuSE2LcsLLzgAqHL/lib/python3.10/site-packages/openai/types/evals/runs/output_item_list_response.py", line 14, in <module>
    class Result(BaseModel):
  File "/Users/alex/Library/Caches/uv/archive-v0/g5fn3MuSE2LcsLLzgAqHL/lib/python3.10/site-packages/pydantic/_internal/_model_construction.py", line 93, in __new__
    private_attributes = inspect_namespace(
  File "/Users/alex/Library/Caches/uv/archive-v0/g5fn3MuSE2LcsLLzgAqHL/lib/python3.10/site-packages/pydantic/_internal/_model_construction.py", line 381, in inspect_namespace
    raise NameError(
NameError: Fields must not use names with leading underscores; e.g., use 'pydantic_extra__' instead of '__pydantic_extra__'.

The current minimum is 1.9.0:

openai-python/pyproject.toml

Line 12 in 0d85ca0

"pydantic>=1.9.0, <3",

[RobertCraigie]

@alexmojaki any suggestions fo preventing this error on the pydantic side? we could update __pydantic_extra__ to be conditionally added but that would not be ideal, as we use that information at runtime.

To be clear I would like to maintain support for older pydantic versions, I just suggested updating as a workaround in the meantime.

[alexmojaki]

@Viicos any ideas?

For now I'm working around this in pydantic/logfire#1416 with sys.modules['openai.resources.evals'] = unittest.mock.MagicMock() in conftest.py

[mdrxy]

Some libraries (like ours) requires older Pydantic versions for backward compat. Would appreciate any workarounds or a fix to this.

[Viicos]

Seems like before Pydantic 2.8.0, a bug would not allow assigning any value (at class definition) to __pydantic_extra__. I would suggest implementing the affected models this way:

class Model(BaseModel):
    if TYPE_CHECKING:
        __pydantic_extra__: dict[str, object] = Field(init=False)
    else:
        __pydantic_extra__: dict[str, object]

[RobertCraigie]

Thanks @Viicos! This will be fixed in the next release #2664