Skip to content

Latest commit

 

History

History
185 lines (147 loc) · 7.11 KB

README.md

File metadata and controls

185 lines (147 loc) · 7.11 KB
Raw

LLM RAG - Streamlit RAG Language Model App 🤖

🌟 Overview

This Streamlit app leverages Retrieval-Augmented Generation (RAG) by using OpenAI's Large Language Model (LLM) in conjunction with FAISS, a vector database. The app allows users to upload markdown files 📂, ask questions related to the content of these files ❓, and receive AI-generated answers based on the uploaded content 📚.

❓ How It Works

The LLM RAG Streamlit app is structured into several key areas, each serving a specific function within the application:

  • Setup Knowledge Base 📂: Upload markdown documents to establish the knowledge base.
  • Explore Knowledge Base 🔍: Browse and manage the uploaded documents.
  • RAG Query 💡: Pose questions to receive answers referencing the knowledge base and the model's knowledge.

Additionally, the app offers advanced settings for customization based on user needs:

  • OpenAI Embedding Model Settings: Select the embedding model for document vectorization.
  • OpenAI LLM Settings: Choose the OpenAI language model variant for generating answers.
  • LLM Temperature: Adjust the creativity of the model’s responses.
  • Max Completion Tokens: Define the maximum length of the generated response.
  • Drop All Documents in Knowledge Base: Clear the database by typing a confirmatory command.

🛠️ System Architecture

The following diagram illustrates the flow of data through the system:

graph TD
    A[Markdown Documents] -->|Data Cleaning &<br>Splitting in Chunks| B[Cleaned Text]
    B -->|OpenAI Model<br>Embedding| C[Document Embeddings]
    C -->|Store| D[(Vectorstore)]
    D -->|Similarity Search| E[Relevant Documents]
 
    F[User Question] -->|OpenAI Model<br>Embedding| G[Query Embedding]
    G -->|Fetch| D
 
    F --> J[Contextualized Prompt]
    E --> J
    J -->|OpenAI LLM Model| L[Answer]
 
    subgraph Data Preparation
        A
        B
    end
 
    subgraph Vectorization
        C
        G
    end
 
    subgraph Relevant Documents Retrieval
        D
        E
    end
 
    subgraph LLM Querying
        J
        L
    end
 
    %% Styles
    style A fill:#7f7f7f,stroke:#fff,stroke-width:2px
    style B fill:#8fa1b3,stroke:#fff,stroke-width:2px
    style C fill:#8fa1b3,stroke:#fff,stroke-width:2px
    style D fill:#8fa1b3,stroke:#fff,stroke-width:2px
    style E fill:#8fa1b3,stroke:#fff,stroke-width:2px
    style F fill:#7f7f7f,stroke:#fff,stroke-width:2px
    style G fill:#8fa1b3,stroke:#fff,stroke-width:2px
    style J fill:#e07b53,stroke:#fff,stroke-width:2px
    style L fill:#e07b53,stroke:#fff,stroke-width:2px
Loading

Project Structure 🏗️

The project's main directories are laid out as follows:

LLM-RAG/
├── .github/workflows/          # CI/CD pipeline definitions
├── configs/                    # Configuration files for the model (model names, pricing..)
├── data/                       # Data and indices used by the app (FAISS Knowledge Base)
├── docker/                     # Docker related files 
├── notebooks/                  # Jupyter notebooks for experiments
├── secrets/                    # API keys and other secrets (excluded from version control)
├── src/                        # Source code for the LLM RAG logic
├── streamlit_app/              # Streamlit app files for the Web Interface
├── tests/                      # Test cases for the application
├── .dockerignore               # Specifies ignored files in Docker builds
├── .gitignore                  # Specifies untracked files ignored by git
├── Dockerfile                  # Dockerfile for building the Docker image
├── Makefile                    # Make commands for building and running the app 🧑‍💻
├── README.md                   # Documentation and instructions
├── requirements.txt            # Python dependencies for the project
└── (additional project files and scripts)

🚀 Getting Started

To begin using the LLM RAG app, follow these simple steps:

  1. Clone the Repository:

    git clone https://github.com/labrijisaad/LLM-RAG.git

  2. Create the Environment: Set up your virtual environment using either venv or conda:

    # Using venv
python -m venv env_llm_rag
source env_llm_rag/bin/activate

# Using conda
conda create --name env_llm_rag
conda activate env_llm_rag

  3. Install Dependencies: Install the required dependencies by

running:

pip install -r requirements.txt

  1. Set Up OpenAI API: Rename the example credentials file to secrets/credentials.yml and replace the placeholder key ('sk-xxx') with your actual OpenAI API key. You can obtain your API key by following the instructions provided in the OpenAI documentation.

    rename secrets/credentials-example.yml secrets/credentials.yml

  2. Run the Streamlit App: Launch the Streamlit app using either the provided Makefile command or directly via the Streamlit CLI:

    # Using Makefile
make stream

# Or directly
streamlit run streamlit_app/main.py

🐳 Docker Version

The application is available as a Docker container and can be easily set up and run with a few commands. If you want to run the application using the Docker image from the public registry, ensure that you have a secrets directory with the necessary API keys as specified in the secrets/credentials.yml.

To pull and run the Docker container:

  1. Pull the Docker Image: You can pull the image directly from Google Artifact Registry using the following command:

    docker pull europe-west1-docker.pkg.dev/llm-rag-application/llm-rag/llm_rag_app:latest

  2. Run the Docker Container: After pulling the image, you can run it with:

    docker run -p 8501:8501 -v $(pwd)/secrets:/app/secrets europe-west1-docker.pkg.dev/llm-rag-application/llm-rag/llm_rag_app:latest

    This command will start the container and mount your secrets directory for the application to use.

If you prefer to use the Makefile, the equivalent commands are provided for convenience:

# To pull the Docker image
make docker-pull

# To run the pulled Docker image
make docker-run-pulled

The Streamlit app will be available at http://localhost:8501 once the container is running.

🌐 Connect with me