Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Koboldcpp would really benefit from RAG and Websearch #1246

Open
adrian52999 opened this issue Dec 1, 2024 · 3 comments
Open

Koboldcpp would really benefit from RAG and Websearch #1246

adrian52999 opened this issue Dec 1, 2024 · 3 comments

Comments

@adrian52999
Copy link

So, let me start by saying that Koboldcpp is truly the most pleasant to work with out of all the endpoints. It is simultaneously the easiest to install and work with, and it is also one of the fastest ones. The RP possibilities it opens are great, however I think that by adding RAG and web search it would branch out and possibly compete with the likes of open webui.

How likely would this be?

@Asherathe
Copy link

Why have so many people been asking about RAG and web search lately? Those are frontend features, and there are tons of LLM frontends that implement them already.

@adrian52999
Copy link
Author

Yes, you are right. I could use Kobold as a backend in frontends like open-webui or tl/dw, and I did - it's really nice. Yet, I still feel that RAG and websearch would help push Kobold in the one-size-fits-all direction it seems to be going...

@esolithe
Copy link

Yes, you are right. I could use Kobold as a backend in frontends like open-webui or tl/dw, and I did - it's really nice. Yet, I still feel that RAG and websearch would help push Kobold in the one-size-fits-all direction it seems to be going...

I presume this would be one to ask for here then? Kobold CPP is the back end - focused specifically on the generation, whereas the front end you see in the web browser is Kobold Lite.

While it is early days, there's a "sort of" context searching feature which attempts to pull relevant snippets from your whole history (it does not use vectorising or similar, more of an enhanced text search), and if successful might be possible to expand to documents as a separate PR in the future (mostly if there's interest and the base mechanism works well enough for most people).

The link to the PR is here if you're interested to follow / try it out.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

3 participants