I m on 0 6 5 and trying to use the

vvampir

I'm on 0.6.5 and trying to use the Composable graph, I pass it custom_query_engines with matching index id but it seems the subsequent query doesn't use the service context at all and complains about missing engine or deployment id.. any ideas what I might be doing wrong?

Plain Text

custom_query_engines = {
        i.index_id: i.as_query_engine(
            mode="embedding", 
            text_qa_template=_get_prompt_template(lang), 
            similarity_top_k=k, 
            # https://github.com/jerryjliu/llama_index/blob/main/docs/guides/primer/usage_pattern.md#configuring-response-synthesis
            response_mode="tree_summarize", # other modes are default and compact 
            refine_template=_get_refined_prompt(lang), service_context=service_context
        )    
        for i in indices.values()
    }

    query_engine = graph.as_query_engine(custom_query_engines=custom_query_engines)
    
    response = query_engine.query(query)

15 comments

LLogan M

I see the tutorial page also sets the query engine for graph.root_id, maybe you need that as well?

LLogan M

Attachment

vvampir

Thanks! I have tried that has well since then to no avail. It seems tied to AzureOpenAI usage and embeddings. While i have changed nothing in that area I did a pip upgrade this morning.

vvampir

I pin pointed it to the version of llama-index. Back with 0.6.0 this code works. Not finding exactly what changed. Its definitely missing some service context

LLogan M

Hmmm good find... I know between 6.0 and now there was some changes related to azure

LLogan M

https://github.com/jerryjliu/llama_index/pull/3140

https://github.com/jerryjliu/llama_index/pull/3071

LLogan M

Maybe some clues in there

LLogan M

Maybe this notebook will help confirm your setup too
https://github.com/jerryjliu/llama_index/blob/main/docs/examples/customization/llms/AzureOpenAI.ipynb

vvampir

Cheers!!! I have to look at this on Monday!

vvampir

Sadly no fix @Logan M , I tried various ways and it still complains. i assume it's a similar fix id need for the AzureChatOpenAi class. Otherwise I am using all calls for embedding

vvampir

Error is specific to Embedding engine or deployment id.

LLogan M

Dang, sorry about that!

Can you make a github issue for this? 🙏 I guess all the examples don't use the chat class

vvampir

Before posting the issue I searched and found issue 3264 (sorry on phone, unable to use discord from work laptop 😔) and I will double check the suggestions there

vvampir

No luck, i submitted issue 3372

LLogan M

Sounds good, thanks!

Add a reply

Find answers from the community

I m on 0 6 5 and trying to use the