JShollaj/awesome-llm-interpretability

What it solves

This project addresses the "black box" nature of Large Language Models (LLMs). Because these models are highly complex, it is often difficult to understand why they make specific decisions, how they store knowledge, or how they process information internally. This repository provides a centralized collection of resources to help researchers and developers demystify these processes.

How it works

The repository acts as a curated directory of tools, academic papers, and articles. It categorizes these resources to help users find specific methods for analyzing neural networks, such as visualizing internal activations, identifying specific neurons responsible for certain behaviors, or using causal intervention techniques to edit model knowledge.

Who it’s for

This list is designed for AI researchers, machine learning engineers, and developers who are building or studying LLMs and need to improve model transparency, debug performance issues, or ensure safety and reliability.

Highlights

• A comprehensive list of software libraries for visualizing and debugging transformer models.
• A deep collection of academic research papers covering topics like mechanistic interpretability, hallucination detection, and bias mitigation.
• Links to interactive tools that allow users to explore model internals, such as attention maps and latent space representations.
• Resources for understanding how models evolve during training and how they store factual information.