X-LLM: Explainable Large Language Models

A tool for understanding and exploring the internal mechanisms of Large Language Models (LLMs) through multiple explanation methods.

Project Overview

X-LLM provides comprehensive explanations of how different language models (GPT-2, OPT, Pythia) process and understand information. By analyzing both transformer attention patterns and feed-forward network activations, this tool enables researchers and enthusiasts to gain insights into these complex systems.

Features

• Multiple Model Support: Currently supports GPT-2, OPT, and Pythia models
• Dual Explanation Methods:
  • Transformer-based Explanations: Analyze how attention heads focus on word relationships
  • FFN-based Explanations: Understand feed-forward network activations and word importance
• Interactive UI: Built with Streamlit for easy exploration
• Retrieval-Augmented Generation (RAG): Uses semantic search to find relevant explanations based on user queries
• Conversational Interface: Chat-based interaction with the system

Technical Architecture

• Database: PostgreSQL for storing model explanations
• Embedding Model: SentenceTransformer for semantic similarity search
• LLM Agent: GPT-4o-mini for generating human-friendly responses
• Web Framework: Streamlit for the user interface

Setup Instructions

Prerequisites

• Python 3.11+
• PostgreSQL server
• OpenAI API key

Environment Setup

1. Clone the repository:

   git clone https://github.com/Secrisx/XLLM_app.git
   cd XLLM_app

2. Install dependencies:

   pip install -r requirements.txt

3. Create a .env file with your OpenAI API key:

   OPENAI_API_KEY=your_api_key_here
   

Running the Application

Start the Streamlit app:

streamlit run app.py