[Bug]: When using AzureOpenAI for embedding `azure_deployment` needs to be provided as well

[pavlovmilen]

What happened?

As a chromadb user, when I use Azure Open AI, I want to be able to use my embedding funciton with collection creation but currently it does not let me specify azure_deployment so it throws DeploymentNotFound exception (see logs):

my_azure_openai_ef= embedding_functions.OpenAIEmbeddingFunction(
            api_key=OPEN_AI_API_KEY,
            model_name="text-embedding-ada-002",
            api_base="https://my_azure_open_ai_endpoint.openai.azure.com",
            api_type="azure",
            api_version="2023-05-15",
            )
`client.get_or_create_collection("my_collection", embedding_function=my_azure_openai_ef)`

In embedding_functions.py there is a code execution path that gets triggered when api_type="azure"

        self._client = openai.AzureOpenAI(
            api_key=api_key,
            api_version=api_version,
            azure_endpoint=api_base,
            default_headers=default_headers,
        ).embeddings

The issue is that it is missing azure_deployment="your_deployment_name_here" Whithout it my azure deployment cannot be reached. See attached log for more details

To test it I have this setup:

client = AzureOpenAI(
    api_key=OPEN_AI_API_KEY, 
    azure_endpoint="https://my_azure_open_ai_endpoint.openai.azure.com", 
    azure_deployment="my_azure_open_ai_deployment",
    api_version="2023-05-15")

def text_embedding(text):
    response = client.embeddings.create(model="text-embedding-ada-002", input=[text])
    return response.data[0].embedding

When I skip azure_deployment="my_azure_open_ai_deployment", my text embedding function throws not found exception. Azure Open AI endpoints need the deployment name in order for it to work.

See https://github.com/openai/openai-python?tab=readme-ov-file#microsoft-azure-openai for more details Also can you update docs to include link to api versioning in Azure: https://learn.microsoft.com/en-us/azure/ai-services/openai/reference#rest-api-versioning

Versions

ChromaDB 0.4.21, Python 3.11.5

Relevant log output

Cell In[2], line 14
     13 def text_embedding(text):
---> 14     response = client.embeddings.create(model="text-embedding-ada-002", input=[text])
     15     return response.data[0].embedding

File c:\Python\Lib\site-packages\openai\resources\embeddings.py:113, in Embeddings.create(self, input, model, dimensions, encoding_format, user, extra_headers, extra_query, extra_body, timeout)
    107         embedding.embedding = np.frombuffer(  # type: ignore[no-untyped-call]
    108             base64.b64decode(data), dtype="float32"
    109         ).tolist()
    111     return obj
--> 113 return self._post(
    114     "/embeddings",
    115     body=maybe_transform(params, embedding_create_params.EmbeddingCreateParams),
    116     options=make_request_options(
...
   (...)
    987     stream_cls=stream_cls,
    988 )

Error code: 404 - {'error': {'code': 'DeploymentNotFound', 'message': 'The API deployment for this resource does not exist. If you created the deployment within the last 5 minutes, please wait a moment and try again.'}}

[beggers]

hey @pavlovmilen, sorry this got dropped!

This should be a very easy change to make. I'd be happy to review a PR if you want to add this to the embedding function (https://github.com/chroma-core/chroma/blob/main/chromadb/utils/embedding_functions.py#L103-L180) or I may be able to get to it next week.

[sumaiyah]

I have a PR ready for review