Class 3: RAG (Retrieval-Augmented Generation)

Introduction

Master a technique that integrates relevant documents into the generation process, enhancing the quality and contextuality of LLM responses. This repo demonstrates a simple RAG example and more advanced implementations.

Class Materials

• Link to Class Slides

Prerequisites

• Docker (for containerized usage - recommended)
• Python 3.11 (for local setup)
• Pinecone account (free tier is sufficient)

Setup

1. Create a Pinecone Index:

Before running the application, you need to set up a Pinecone index:

1. Log in to your Pinecone account at https://app.pinecone.io/
2. Navigate to the "Indexes" section and click "Create Index"
3. Fill in the following details:
   • Name: Choose a name for your index (e.g., "rag-project-index")
   • Dimensions: Set to 3072
   • Metric: Choose cosine
4. For Environment and Region:
   • Cloud Provider: Select aws
   • Region: Choose us-east-1 (Note: These last two options are required for a free instance)
5. Click "Create Index" to confirm
6. If you wish to run RAG_fusion_101.ipynb you must create a separate Pinecone index
7. For this second index, follow the above procedure but set the Dimensions to 1536, and give the index a name of your choice
8. Make sure and enter this second index into the .env file

Remember the names you gave to your indices, as you'll need them for the .env file.

2. Set up environment variables:

• Copy the sample environment file:

  cp .env.sample .env
  

• Edit the .env file and add your API keys and other required variables:

   OPENAI_API_KEY=your-secret-key
   LANGCHAIN_API_KEY=your-secret-key
   LANGCHAIN_TRACING_V2=true
   LANGCHAIN_PROJECT=gauntlet_class_3
   PINECONE_API_KEY=your-secret-key
   PINECONE_INDEX=your-3072-dimension-index-name
   PINECONE_INDEX_TWO=your-1536-dimension-index-name
  

  Make sure to use the relevant index names you created for PINECONE_INDEX and PINECONE_INDEX_TWO, respectfully.

Quick Start with Docker

1. Run the upload script:

   docker compose run --rm upload_service
   

2. Run the main RAG example:

   docker compose run --rm rag_app
   

3. Start Jupyter for notebook examples:

   docker compose up jupyter
   

4. Run a specific script (any new .py file you may add):

   docker compose run --rm rag_app python <script_name.py>
   

Note: The Docker setup will automatically use the environment variables from your .env file. You don't need to export them to your system environment when using Docker.

Running Different Scripts

You can use the provided run.sh script for easier execution. Make sure to make the script executable with chmod +x run.sh before using:

./run.sh upload
./run.sh main
./run.sh jupyter
./run.sh <your_new_py_file>

Local Setup (Alternative to Docker)

If you prefer to run the examples locally:

1. Ensure you have Python 3.10+ installed.
2. Clone the repository:

   git clone [repository-url]
   cd [repository-name]
   

3. Set up the environment:

   python3 -m venv .venv
   source .venv/bin/activate  # On Windows use `.venv\Scripts\activate`
   pip install -r requirements.txt
   

4. Configure environment variables as described in the Setup section.
5. Export the environment variables (python-dotenv should handle this automatically in the ):

   export $(cat .env | xargs)
   

6. Run an example:

   python3 upload.py
   

Troubleshooting

• Ensure you're using Python 3.11.0 or later for local setup.
• For Docker issues, check your Docker installation and version.
• If you encounter package issues, try updating pip: pip install --upgrade pip
• Make sure all required environment variables are set in your .env file.
• If you're having issues with environment variables in Docker, ensure your .env file is in the same directory as your docker-compose.yml file.
• If you encounter issues with Pinecone, double-check that your index is created correctly and that you're using the correct API key and index name(s) in your .env file.

Need Help?

Refer to the project documentation or reach out to the learning assistant