SimpleRAG

SimpleRAG is a repository designed to demonstrate the use of Retrieval-Augmented Generation (RAG) with Milvus and LangChain.

Prerequisites

Before setting up SimpleRAG, ensure you have the following:

1. LLamaParse API Key:

   • Sign up and obtain your API key from LLamaParse.

2. Milvus Installation:

   • Follow the official Milvus installation guide to set up a standalone Milvus instance using Docker.
   • Linux and Mac OS is recommended.

3. Environment Requirements:

   • Anaconda
   • (Recommended) A GPU server for hosting the embedding model & LLM.

Installation

1. Clone the Repository:

   git clone https://github.com/haozhuang0000/SimpleRAG.git
   cd SimpleRAG

2. Create a Conda Environment:

   conda create -n simplerag python=3.11
   conda activate simplerag

3. Install Dependencies:

   pip install -r requirements.txt

Configuration

.env File Setup

Create a .env file in the root directory of the project and configure the following variables:

VDB_HOST=YOUR_MILVUS_IP_ADDRESS
VDB_PORT=YOUR_MILVUS_PORT
EMBEDDING_HOST=YOUR_EMBEDDING_MODEL_IP_ADDRESS
EMBEDDING_PORT=YOUR_EMBEDDING_MODEL_PORT
OLLAMA_HOST=YOUR_OLLAMA_IP_ADDRESS
OLLAMA_PORT=YOUR_OLLAMA_PORT
LLAMAPARSER_API_KEY=your_llamaparse_api_key

Usage

1. Create a folder _static & Put your PDF files under _static
2. python main.py

SimpleRAG Workflow

The Retrieval-Augmented Generation (RAG) process in SimpleRAG follows these steps:

1. Create a Vector Database Collection

• Initialize Collection: Create a new collection in Milvus to store document embeddings.

2. Parse PDF Documents and insert Data into Vector Database

2.1 Parse Documents

• Process PDFs: Use LLamaParse to process all PDF files located in the _static directory.

2.2 Split Documents

• Document Chunks: Break down parsed documents into chunks for processing.

2.3 Embed Chunks

• Generate Embeddings: Use the embedding model to generate embeddings for each document chunk.

2.4 Insert Embeddings

• Store in Milvus: Insert the generated embeddings into the Milvus vector database for future retrieval.

3. Query the Vector Database and Get a response from LLM

3.1 Embed Query

• Query Embedding: Convert the user's query into an embedding using the same embedding model.

3.2 Search Database

• Similarity Search: Perform a similarity search in Milvus to find chunks that are most relevant to the query embedding.

3.3 Retrieve Chunks

• Fetch Results: Retrieve the most relevant document chunks based on the similarity search results.

3.4 Generate Response with LLM

• Contextual Response: Utilize LangChain to generate a response from the language model, incorporating the retrieved context for response.