Interpreto: An Explainability Library for Transformers

TL;DR

Interpreto is an open-source Python library that offers comprehensive interpretability tools for HuggingFace language models, including attribution and concept-based explanations, with an end-to-end pipeline for deeper insights.

Contribution

It introduces a unified API for explanation workflows and features an end-to-end concept-based pipeline, advancing interpretability methods for transformer models.

Findings

Supports a wide range of models from BERT to LLMs

Provides a unified API for classification and text generation

Includes an end-to-end concept-based explanation pipeline

Abstract

Interpreto is an open-source Python library for interpreting HuggingFace language models, from early BERT variants to LLMs. It provides two complementary families of methods: attribution methods and concept-based explanations. The library bridges recent research and practical tooling by exposing explanation workflows through a unified API for both classification and text generation. A key differentiator is its end-to-end concept-based pipeline (from activation extraction to concept learning, interpretation, and scoring), which goes beyond feature-level attributions and is uncommon in existing libraries.