Cannot see Azure provider models

[TariniBindu03]

Hi

I am not able to see the AZURE providers model sin the dropdown list . Can you tell me why?

[krassowski]

You may need to install langchain-openai and restart Jupyter server. AzureChatOpenAI is implemented as part of OpenAI provider which is not documented very well indeed. For reference

https://jupyter-ai.readthedocs.io/en/latest/users/index.html#model-providers

jupyter-ai/packages/jupyter-ai-magics/jupyter_ai_magics/partner_providers/openai.py

Lines 84 to 90 in 4b45ec2

	class AzureChatOpenAIProvider(BaseProvider, AzureChatOpenAI):
	id = "azure-chat-openai"
	name = "Azure OpenAI"
	models = ["*"]
	model_id_key = "azure_deployment"
	model_id_label = "Deployment name"
	pypi_package_deps = ["langchain_openai"]

I think that instead of hiding providers, these should be still included but disabled with an informative message.

[TariniBindu03]

Hi krassowski

Thanks for responding , this suggestion helped me .I am now able to see Azure completion & embedded models.

Having a follow-up question regarding /learn & /ask . I am using Azure embedded model - text-embedding-ada-002. It always responds I don't know . Can you help me on this please.

jovyan

7:08 PM

/learn /home/jovyan/pdf-files/

Jupyternaut

7:08 PM

🎉 I have learned documents at /home/jovyan/pdf-files/ and I am ready to answer questions about them. You can ask questions about these docs by prefixing your message with /ask.

jovyan

7:08 PM

/ask what did you learn

Jupyternaut

7:08 PM

I don't know.

jovyan

7:09 PM

/ask how many pages does the pdf file contain

Jupyternaut

7:09 PM

I don't know.

[srdas]

@TariniBindu03 This is not an issue for jupyter-ai -- behavior varies with embedding model. I tried the question "What did you learn?" with Bedrock Titan embedding v2 and also with Ollama nomic-embed-text and both return clear and complete answers to this question. On the question about how many pages the pdf file contains, I tried different LLMs and embedding models and none of them can return an answer -- it may be that this is a question you cannot answer from shards-based embeddings in a vector store.

[TariniBindu03]

HI @srdas

Thanks for responding . Currently i have only access to Azure embedded models . which will you think will be more useful in learning local data

text-embedding-3-large text-embedding-3-small text-embedding-ada-002

[srdas]

HI @srdas

Thanks for responding . Currently i have only access to Azure embedded models . which will you think will be more useful in learning local data

text-embedding-3-large text-embedding-3-small text-embedding-ada-002

Difficult to say, as I do not have access to Azure embedding models. I would also suggest using Ollama with embedding models in jupyter-ai as access is free. See: https://ollama.com/blog/embedding-models. Also: https://jupyter-ai.readthedocs.io/en/latest/users/index.html#ollama-usage. I tried embedding models from AWS, such as Titan-v2, and nomic-embed-text from Ollama and both worked well in explaining what was learned. However, asking about how many pages there are in a PDF is not going to elicit a proper answer as this is not explicit in sharded embeddings. I did try different prompts to see if this could provide a good answer, and it did not, FYI.