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

Symptons: Creating multiple embeddings with one api call is very slow (seconds vs miliseconds)
Where to look:
The bug is in https://github.com/openai/openai-python/blob/main/src/openai/lib/_parsing/_embeddings.py line 27
The code calls has_numpy() in a loop for every embedding. The function tries to import numpy, catches the exeption and returns a bool. If numpy isnt installed this is very slow.
Fix:
call has_numpy() either as global on instantiation, outside of the loop once or wrap has_numpy in functools.cache.

To Reproduce

1. create embeddings for 1500 strings with one api call. Leave the encoding_format option empty
2. meassure the time the api call takes
3. do the same with encoding_format="base64", decode the entries with numpy manually and compare the time, it should be identical

Code snippets

OS

windows

Python version

Python v3.12.12

Library version

v3.5.0