fix: assert_has_numpy before using np.from_buffer

[zzstoatzz]

it seems Embeddings.acreate should call assert_has_numpy before attempting to use np.frombuffer, as is done on L42 of Embeddings.create here

I encountered the following on 3.11, openai==0.27.8 using Embeddings.acreate in an env where numpy was not installed:

File ~/src/open-source/marvin/src/marvin/utilities/embeddings.py:11, in create_openai_embeddings(texts)
      8 async def create_openai_embeddings(texts: List[str]) -> List[List[float]]:
      9     """Create OpenAI embeddings for a list of texts."""
---> 11     embeddings = await openai.Embedding.acreate(
     12         input=[text.replace("\n", " ") for text in texts],
     13         engine=marvin.settings.embedding_engine,
     14     )
     16     return [
     17         r["embedding"] for r in sorted(embeddings["data"], key=lambda x: x["index"])
     18     ]

File ~/micromamba/envs/marvin/lib/python3.11/site-packages/openai/api_resources/embedding.py:82, in Embedding.acreate(cls, *args, **kwargs)
     78         for data in response.data:
     79
     80             # If an engine isn't using this optimization, don't do anything
     81             if type(data["embedding"]) == str:
---> 82                 data["embedding"] = np.frombuffer(
     83                     base64.b64decode(data["embedding"]), dtype="float32"
     84                 ).tolist()
     86     return response
     87 except TryAgain as e:

AttributeError: 'NoneType' object has no attribute 'frombuffer'

which appears to be because the "helper" is setting the module to None like

try:
    import numpy
except ImportError:
    numpy = None

calling assert_has_numpy before attempting to use np.frombuffer in acreate raises a more helpful error

File ~/micromamba/envs/marvin/lib/python3.11/site-packages/openai/api_resources/embedding.py:82, in Embedding.acreate(cls, *args, **kwargs)
     78     for data in response.data:
     79
     80         # If an engine isn't using this optimization, don't do anything
     81         if type(data["embedding"]) == str:
---> 82             assert_has_numpy()
     83             data["embedding"] = np.frombuffer(
     84                 base64.b64decode(data["embedding"]), dtype="float32"
     85             ).tolist()
     87 return response

File ~/micromamba/envs/marvin/lib/python3.11/site-packages/openai/datalib/numpy_helper.py:15, in assert_has_numpy()
     13 def assert_has_numpy():
     14     if not HAS_NUMPY:
---> 15         raise MissingDependencyError(NUMPY_INSTRUCTIONS)

MissingDependencyError:

OpenAI error:

    missing `numpy`

This feature requires additional dependencies:

    $ pip install openai[datalib]

[rattrayalex]

Thanks for this!

We've since rewritten the library entirely, so I believe this change is no longer relevant. I'm sorry we didn't get to it sooner.