Type hinting / stub files

[yyueniao]

Describe the feature or improvement you're requesting

I'm new to the openai-python library, and I'm really excited to start working on some cool projects with it! 🚀 However, I noticed that the library's type hinting could be improved to make it even better.

I'd like to kindly request some enhancements in the type hinting for the library's methods. Having more comprehensive type annotations would greatly help developers like me avoid bugs and make our code more readable and maintainable.

If possible, it would be awesome if the maintainers could consider adding these improvements. I believe it would enhance the overall developer experience and make the library even more delightful to use.

Additionally, if anyone in the community has already created some useful stub files, sharing them would be incredibly helpful! It's like having a secret guide to navigate the code with ease.

Thank you for your attention, and I really appreciate all the effort put into making this library fantastic.

Additional context

No response

[ldorigo]

+1. At least some stubs/typed dicts for the completion returns would help a lot.

[D1no]

Just for clarification: This is what it looks like when you copy the hello world example into a python project where pylance / pyright (a static type checker for python by microsoft) is set to strict 😕

it completely blows up 💣

If this thin library is supposed to make it easier to add openai to professional production systems (where static typing is part of the testing pyramid) this isn't great.

Ultimately I, like many others, need to invest time now to isolate your static test failure ultimately delaying shipment, and therefore...

...you losing money by having us not spending money earlier on tokens. A lose lose situation for OpenAI/Azure<->Customers 😪 (@logankilpatrick)

Pylance / Pyright

The OpenAI Python SDK currently seems to violate the following general type errors.

• reportUnknownVariableType
• reportUnknownArgumentType
• reportUnknownLambdaType
• reportUnknownMemberType
• reportUnknownParameterType
• reportMissingParameterType
• reportGeneralTypeIssues

Copy this to the top of the file dealing with the openai python sdk if you can't isolate:

# Dependency Type-Risk Note: openai(0.27.9)
# pyright: reportUnknownVariableType=false, reportUnknownArgumentType=false, reportUnknownLambdaType=false, reportUnknownMemberType=false, reportUnknownParameterType=false, reportMissingParameterType=false, reportGeneralTypeIssues=false

---

I know static typing in python is a general, data "science" notebook driven, issue. But there have been a bunch of improvements over the past two years in py land and it would be great if organizations like OpenAI would make life easier for software engineers required to build solid systems (types, testing, solid errors). If I could, I'd use the OpenAI TypeScript SDK which seems to have solid typing.... but can't in this case 😮‍💨

[JoshJarabek7]

+1 on this as well

[nicognaW]

+1, we are considering switching to custom clients instead of adopting this repo in production.

[rattrayalex]

Great news, we've just published a beta for the a fully-rewritten v1 of our Python SDK, which includes static types for all request parameters & response fields.

We'd love your feedback, so please give it a try and share your thoughts (good and bad, major and minor) in the linked discussion!