GTPme - Private conversations with your docs

GPTme is totally free aiming to make local RAG pipes as easy to integrate AND as modular as possible!

Main features

1. Total privicy & security - your data doesn't leave your machine!
2. LLMs infrastructure modularity - Currently supporting LlamaCPP + HF & Ollama wraped with Langchain - working on the non free services intergration right now
3. Work with the latest models HuggingFace
4. Utilize LangGraph to run corrective RAG on your documents - Awesome blog post & videos can be found here
5. Create your own vector database - Credit to PromtEngineer & his repo localGPT

Install steps - Colab

One click set up !

1. Corrective RAG example - checkout notebook -

2. Simple RAG example - checkout notebook -

Install steps - Linux

1. Pull the repo & cd to GPTme folder
2. Create a virtual env (or conda env)

   python venv -m GPTme_env
   source GPTme_env/bin/activate

3. Run the setup script for the repo - this will - install required pip packages, build and install GPTme, install LlamaCPP with GPU support -

   source setup.sh

Examples - once GPTme is setup

1. To run the streamlit app for local discussion with your docs run -

   streamlit run GPTme/streamlit_app/crag_app.py

2. Corrective RAG example - checkout this script -

   python GPTme/Examples/crag_example.py

Installation of LlamaCPP - included in the llama_cpp_setup.sh script

I didn't have much luck with installing the package and getting the GPU to work, using the default Linux installation instructions i.e. If you have issues with utilizing your GPU, please try the llamaCPP setup script.

The install directions are described in this github issue This works every time, but the main drawback is that it takes ~5 min to build and install

checkout the notebook for more experimentation