Fatal error during indexing: CustomError: Orama database not found -Vault QA won't connect to LM STUDIO, loads model multiple times,

[AdamGman]

Copilot version: 2.7.15 + LM Studio 0.3.6 Build 8

Chat opens just fine with whatever model I choose. When switching to Vault QA it simply does not think its connected to the model and tries to load it multiple times. Perhaps I'm not using the correct embedding model, but it would still connect first to it to say that correct?

Describe how to reproduce
Click Vault QA or force rebuild index, with LM studio running model with API. Does not create index or connect to LLM.

Expected behavior
Create index and build, then chat in Vault QA

[AdamGman]

Is this true? we can't run both at the same time? If this is true how is it not stated anywhere.

[AdamGman]

Hello, I got this to work with the snowflake-arctic-embed-l-v2.0 model. You guys should really have a list of working embedding models. I tried so many trial and error.

[logancyang]

@AdamGman thanks for reporting. Which ones weren't working for you? This is an open-source project and your trial and errors could be helpful to others.

And where was that warning message "LM studio cannot be both your LLM and embedded" from? Can you share the link to it pls.

[AdamGman]

Ahh very cool! Makes sense. Would def like to see the list of everyone else's working LM studio embeddings.

I started going down the list of the top embeddings leaderboard here https://huggingface.co/spaces/mteb/leaderboard.
Unless I'm wrong, most of the top ones don't exist as GGUF for LM Studio yet or have errors in opening. (I have a 4090.)
Does it have to be a "text" only embedding? Is that where I went wrong going for the multi embeddings?

1-4 -- Couldn't find GGUF
5. stella_en_1.5B_v5 -- Exists but error loading in LM Studio.
7. gte-Qwen2-7B-instruct -- The one I was using that loaded over and over but wasn't recognized by Copilot.

I settled on 113 on the list, snowflake-arctic-embed-l-v2.0 this rebuilt my index perfectly.

Then I used qwen2.5-32b-agi for my chat model, it actually thinks pretty slowly in Copilot but a very smart model.

I wonder the combination what everyone else is using?

As for the warning message it was actually from the AnythingLLM Doc, so probably a totally different context.