diagNNose: A Library for Neural Activation Analysis

TL;DR

diagNNose is an open-source library that offers various interpretability tools for analyzing neural network activations, demonstrated through a case study on language models' subject-verb agreement.

Contribution

The paper introduces diagNNose, a comprehensive library for neural activation analysis, providing new tools for interpretability in deep learning models.

Findings

Effective analysis of language model activations

Insights into subject-verb agreement mechanisms

Open source availability for community use

Abstract

In this paper we introduce diagNNose, an open source library for analysing the activations of deep neural networks. diagNNose contains a wide array of interpretability techniques that provide fundamental insights into the inner workings of neural networks. We demonstrate the functionality of diagNNose with a case study on subject-verb agreement within language models. diagNNose is available at https://github.com/i-machine-think/diagnnose.