RAG (Retrieval-Augmented Generation)

[jarroddavis68]

I have started the implementation of Retrieval-Augmented Generation (RAG) in LMEngine. This feature will utilize an SQLite database for vector embeddings. The upcoming update will include support for LocalDb, providing local SQLite database functionality. Once the implementation is completed, thoroughly tested, and proven stable, I will proceed with the development of vector embedding capabilities. Stay tuned. 😎

[ElCondor1969]

Hi Jarrod.
Thanks a lot for your work.
Is there a roadmap for this project? Is there a release date, even an approximate one?
Thanks in advance.

[jarroddavis68]

Hi Jarrod. Thanks a lot for your work. Is there a roadmap for this project? Is there a release date, even an approximate one? Thanks in advance.

Hi, thanks for reaching out.

It's been challenging to keep up with the updates to llama.cpp, which LMEngine relies on for GenAI. I've already had to change how I build and integrate llama.cpp with Delphi twice because of its rapid evolution. I'm not sure when or if vision support for the latest LLMs will be added. Right now, I'm exploring potential alternatives. Unfortunately, these and other related issues have slowed down development as I work through them.

[ElCondor1969]

Thank you Jarrod for your reply.
LMEngine is really a nice project and I really hope you will be able to overcome the problems you described and evolve the library because it's really useful.

[jarroddavis68]

It looks like I may have resolved the issues mentioned earlier. This marks the third major architectural redesign in an effort to keep up with the rapid advancements in AI. I've developed some scripts to automate the process of converting llama.cpp C++ code into a format that can be used with Delphi. So much has changed that I've decided to build an entirely new product, while still maintaining the same ease of use. I'm in the early stages, but if you're interested in testing these early builds, join my Discord and reach out about AI testing. I'll make them available to anyone who wants to be part of the initial testing phase.