CLAIDec 9, 2023

Using Captum to Explain Generative Language Models

Meta AI
arXiv:2312.05491v1149 citationsh-index: 13NLPOSS
Originality Synthesis-oriented
AI Analysis

This work addresses the need for interpretability tools for generative language models, but it is incremental as it builds on an existing library.

The paper tackles the problem of explaining generative language models by introducing new features in the Captum library, providing an overview of functionalities and example applications for understanding learned associations.

Captum is a comprehensive library for model explainability in PyTorch, offering a range of methods from the interpretability literature to enhance users' understanding of PyTorch models. In this paper, we introduce new features in Captum that are specifically designed to analyze the behavior of generative language models. We provide an overview of the available functionalities and example applications of their potential for understanding learned associations within generative language models.

Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes