Azure connection for embedding models broken

[TeemuSo]

Hey,

I'm using OpenAI API from Azure. It seems that querying index with LangChain OpenAIEmbeddings model is broken. I have model for embedding documents and model for querying, and I'm specifying them with OpenAIEmbeddings. Direct invocation of OpenAIEmbedding.get_query_embedding and OpenAIEmbedding.get_text_embedding works, so the problem is at GPTListIndex.query. Error message is

InvalidRequestError: Must provide an 'engine' or 'deployment_id' parameter to create a <class 'openai.api_resources.completion.Completion'>

but neither deployment_id nor engine can be passed to query, and it shouldn't be. I suppose query should use query_model_name and open.Embeddings API, but it's trying to call openai.Completion.

index = GPTListIndex.. # init list index
em_model = LangchainEmbedding(OpenAIEmbeddings(document_model_name='text-search-ada-doc-001', query_model_name='text-search-ada-query-001'))
index.query(
    "<query_text>",
    mode="embedding",
    embed_model=em_model
)

[Disiok]

Hey @TeemuSo, the index.query call actually calls 2 APIs under the hood: 1) the embeddings API to embed and retrieve relevant documents, and 2) the completion API for synthesizing final response from retrieved documents.

In order to change configure 2), you can pass a custom llm_predictor when constructing the index.

Let me know if this addresses your question/issue.

[TeemuSo]

It was my lack of knowledge, now it works. Thank you :)