Azure OpenAI methods for embeddings and chat completions don't have a deployment parameter

[tonybaloney]

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

The Azure AsyncAzureOpenAI class has a single parameter to set the deployment name azure_deployment.

A common situation is to have an embeddings model deployment and a chat model deployment. In the calls for embeddings.create() and chat.completions.create() there is no parameter to set the deployment name.

What this means is you need to create multiple clients for each deployment you want to use.

To Reproduce

openai_client = AsyncAzureOpenAI(
            api_version="2023-07-01-preview",
            azure_endpoint=f"https://{AZURE_OPENAI_SERVICE}.openai.azure.com",
            azure_ad_token_provider=token_provider,
            organization=OPENAI_ORGANIZATION,
            azure_deployment='?', # A single parameter for the whole client?
        )

# No parameter for the chat deployment name?
await openai_client.chat.completions.create(
            model='gpt-35-turbo',
            messages=messages,
            temperature=0.0,
            max_tokens=100,
            n=1,
            functions=functions,
            function_call="auto",
        )
# No deployment name parameter?
await openai_client.embeddings.create(model=embedding_model, input=query_text)

Without the deployment name being correct, calls to the embeddings API give an error message

Code snippets

No response

OS

macOS

Python version

Python 3.11.1

Library version

1.3.6

[kristapratico]

@tonybaloney in the v1 library, the model parameter accepts the deployment name for the Azure API: https://github.com/openai/openai-python/blob/main/examples/azure.py#L15

You can also create a deployment client like in your example, if you intend to tie the client to a certain deployment.

[pamelafox]

Ah interesting, previously we'd set both deployment_id and model, with the understanding that some internal SDK code might change based on the model. So presumably that is no longer the case. Thanks!

[tonybaloney]

@tonybaloney in the v1 library, the model parameter accepts the deployment name for the Azure API: https://github.com/openai/openai-python/blob/main/examples/azure.py#L15

You can also create a deployment client like in your example, if you intend to tie the client to a certain deployment.

thanks, this worked for me