Find answers from the community

Updated 2 months ago

When executing ...

Plain Text

for doc in documents:
    index.insert(doc)

Why do we get stuck w/ ...

Plain Text

2024-01-11 22:49:04,713:INFO - Retrying request to /chat/completions in 0.974549 seconds
2024-01-11 22:49:04,714:INFO - Retrying request to /chat/completions in 0.958529 seconds
2024-01-11 22:50:05,681:INFO - Retrying request to /chat/completions in 1.951599 seconds
2024-01-11 22:50:05,695:INFO - Retrying request to /chat/completions in 1.896280 seconds

Is there any workaround for this?

17 comments

LLogan M

What kind of index are you using?

MMaverick

I'm using Weaviate at the moment (on free tier)

LLogan M

so a vector store index?

MMaverick

that's right

LLogan M

No idea why index.insert(doc) is calling the completion endpoint.... it should only be calling embeddings. Unless you have some metadata extractor configured somewhere

MMaverick

I do this before that document.metadata['link']

LLogan M

that's fine 🤔

LLogan M

If you ctrl+c while the above is printing, can you send the full traceback?

LLogan M

that would help narrow down where in the framework this is happening

MMaverick

here it is ...

LLogan M

dang, that wasn't as helpful as I wanted 😅

LLogan M

This is running on a fastapi server I'm guessing?

MMaverick

no it's not on fastapi

LLogan M

🤔 mmm what is going on there lol.

Ok, so if thats a dead end, still not sure why its calling the chat/completions endpoint.

Can you share how you setup the index? Did you set a specific embed_model?

MMaverick

just the usual way ... it does work occasionally ... btw, what's the problem w/ fastapi ?

LLogan M

nothing wrong with fastapi, just the threading/async stuff, and lack of proper traceback info, made me think it was running in fastapi

LLogan M

or at least behind some other service (flask, notebook, etc.)

Add a reply