Support for Azure OpenAI Completion Extensions

[raffertyuy]

Azure OpenAI released Completion Extensions which adds a dataSources property in the payload, where we can specify Azure cognitive search.

It will be great if LangChain can support this extension API. I believe OpenAI SDK doesn't support it yet either.

[Falven]

From the Azure side it seems to involve modifying the url and body format slightly. Does anybody have any details on implementing this from a Langchain perspective? Would it be extending AzureChatOpenAI to support the completions extensions endpoint?
Note that @Azure offers a client here. It could be as simple as integrating this with langchain.

[minven2]

In case you want to use the dataSources property please try to add the suffix extensions to the deployment name when the correct endpoint is constructed. My environment is openai 1.8.0 and langchain 0.1.1

import os
from langchain.chat_models import AzureChatOpenAI
from langchain.schema import HumanMessage

chat_llm = AzureChatOpenAI(
    azure_endpoint=os.getenv("AZURE_OPENAI_ENDPOINT"), 
    openai_api_key=os.getenv("AZURE_OPENAI_API_KEY"),
    deployment_name=f"{os.getenv('OPENAI_MODEL_NAME')}/extensions",
    openai_api_version=os.getenv("AZURE_OPENAI_API_VERSION")
)
msg = HumanMessage(content="Talk with your data")
print(
    chat_llm(
        messages=[msg],
        extra_body={
            "dataSources": MyDataSources
        },
    )
)