NeuroSurgeon: A Toolkit for Subnetwork Analysis

TL;DR

NeuroSurgeon is a Python toolkit designed to identify and manipulate subnetworks within neural network models, aiding in the interpretability and understanding of learned representations.

Contribution

The paper introduces NeuroSurgeon, a novel Python library that enables subnetwork discovery and manipulation in Huggingface Transformer models, facilitating explainability research.

Findings

Enables analysis of functional circuits in neural networks

Supports manipulation of subnetworks for interpretability

Accessible as open-source Python library

Abstract

Despite recent advances in the field of explainability, much remains unknown about the algorithms that neural networks learn to represent. Recent work has attempted to understand trained models by decomposing them into functional circuits (Csord\'as et al., 2020; Lepori et al., 2023). To advance this research, we developed NeuroSurgeon, a python library that can be used to discover and manipulate subnetworks within models in the Huggingface Transformers library (Wolf et al., 2019). NeuroSurgeon is freely available at https://github.com/mlepori1/NeuroSurgeon.