Overview
Captum is an open-source model interpretability library for PyTorch. It provides tools to understand and attribute the predictions of PyTorch models across various modalities like vision and text. Built directly on PyTorch, Captum supports most PyTorch model types and integrates with them with minimal modification. It is designed to be extensible, allowing researchers and developers to easily implement and benchmark new interpretability algorithms. Captum offers a generic framework for attributing the importance of inputs, features, or layers to the output of a neural network. It is intended for use by machine learning practitioners and researchers who want to gain insights into their model's behavior and improve its transparency.