RAG Chatbot with LangChain and Hugging Face

This project implements a Retrieval-Augmented Generation (RAG) chatbot using LangChain and Hugging Face models. It fetches data from a Wikipedia article, creates a vector store using ChromaDB, and generates responses with a CodeLlama model.

Features

• RAG Implementation: Uses LangChain to create a RAG pipeline.
• Wikipedia Data: Fetches data from a specified Wikipedia article.
• Chroma Vector Store: Stores document embeddings for efficient retrieval.
• Hugging Face Models: Uses Hugging Face embeddings and a CodeLlama language model.
• Gradio Interface: Provides a user-friendly chat interface.
• Customizable Parameters: Allows adjusting max output tokens and creativeness.

Prerequisites

• Python 3.1x
• Pip
• A Hugging Face API key (set as the HUGGINGFACE_API_KEY environment variable)

Installation

1. Clone the repository:

   git clone <repository_url>
   cd <repository_directory>

   (Replace <repository_url> and <repository_directory> with your repository's URL and directory.)

2. Install dependencies:

   pip install gradio langchain langchain-huggingface langchain-chroma python-dotenv requests sentence-transformers

3. Set the Hugging Face API key:

   • Create a .env file in the project's root directory.

   • Add your Hugging Face API key to the .env file:

     HUGGINGFACE_API_KEY=your_huggingface_api_key
     

     (Replace your_huggingface_api_key with your actual API key.)

4. Run the application:

   python app.py

Usage

1. Run the app.py script.
2. The Gradio interface will launch in your web browser.
3. Enter your question in the chat input.
4. The chatbot will generate a response based on the fetched Wikipedia data.
5. Use the sliders to adjust the Max output tokens and Creativeness of the responses

Code Explanation

• Data Loading: The script loads data from a Wikipedia article using WebBaseLoader.
• Text Splitting: The loaded data is split into chunks using RecursiveCharacterTextSplitter.
• Vector Store: ChromaDB is used to create a vector store from the document chunks, using HuggingFaceEmbeddings.
• Language Model: A CodeLlama model from Hugging Face is used to generate responses.
• RAG Chain: LangChain's RunnablePassthrough, PromptTemplate, and StrOutputParser are used to create the RAG pipeline.
• Gradio Interface: Gradio's ChatInterface is used to create the user interface.

Notes

• Remember to set your Hugging Face API key as an environment variable or in a .env file.
• The Chroma database is persisted to the ./chroma_langchain_db directory.
• You can change the Wikipedia URL to use a different source.
• The all-MiniLM-L6-v2 and BAAI/bge-small-zh-v1.5 models are used for embeddings.
• The codellama/CodeLlama-34b-Instruct-hf model is used for text generation.

.gitignore

Make sure to add the following to your .gitignore file to prevent unnecessary files from being committed: