This repository provides a primary Retrieval-Augmented Generation (RAG) pipeline for answering questions about Chameleon Cloud documentation. The repo includes files that build a vector store by indexing, ask users questions, and retrieve information from provided sources.

Prerequisites

1. Operating System: The code is written for Linux operating systems.
2. Python: Ensure you have Python 3.11 or higher installed.
3. Virtual Environment: It is necessary to use a virtual environment (venv) to manage dependencies.

Setup

1. Clone the Repository:

       git clone <repository_url>
   

2. Create and activate a virtual environment using Conda:

       # Download and install Conda
       wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
       bash Miniconda3-latest-Linux-x86_64.sh
       
       # Create a new environment named 'rag' with Python 3.11
       conda create --name rag python=3.11
   
       # Activate your new environment
       conda activate rag
   

3. Install faiss-gpu from the pytorch channel:

       conda install -c pytorch faiss-gpu
   

4. Run requirements.txt:

       pip install -r requirements.txt
   

5. Install Llama 3.1:

       curl -fsSL https://ollama.com/install.sh | sh
   
       sudo systemctl start ollama
   
       ollama pull llama3.1
   

6. Add a .env file in the root directory with the following content:

   HUGGINGFACE_API_KEY=<your_huggingface_api_key>
   

Execution

You can run rag.py script using the command line:

1. Basic Execusion:

    python rag.py

2. Using streamlit:

    pip install streamlit
 
    streamlit run web_rag.py

Hardware Requirements

We ran rag.py on a Chameleon Cloud KVM-based VM with one NVIDIA H100 GPU.