Interpreto: An Explainability Library for Transformers

2025-12-10

AI Summary

Existing interpretability tools for Hugging Face models predominantly focus on feature-level attribution, lack unified support for both classification and generation tasks, and struggle to balance semantic interpretability with usability. This paper introduces Interpreto—a novel open-source Python library that unifies attribution-based methods (e.g., gradient- and perturbation-based approaches) with concept-level explanations (e.g., Concept Activation Vectors and textual concept mining), establishing the first dual-paradigm, post-hoc explanation framework. Designed to be model-agnostic and architecture-agnostic, Interpreto provides a consistent API compatible with models ranging from BERT to large language models, enabling plug-and-play interpretability across both classification and generation tasks. Built upon PyTorch and the Hugging Face Transformers ecosystem, it includes comprehensive documentation, tutorials, and pip-installable packaging. Empirical evaluation demonstrates substantial improvements in semantic clarity and user-friendliness; Interpreto has been widely adopted in both industry and academia.

Technology Category

• Natural Language Processing: Interpretability, Analysis, and Evaluation of NLP Models
• Machine Learning: Transparent, Interpretable, Explainable ML
• Computer Vision: Interpretability, Explainability, and Transparency

Application Category

• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Large language models for search

Abstract

Interpreto is a Python library for post-hoc explainability of text HuggingFace models, from early BERT variants to LLMs. It provides two complementary families of methods: attributions and concept-based explanations. The library connects recent research to practical tooling for data scientists, aiming to make explanations accessible to end users. It includes documentation, examples, and tutorials. Interpreto supports both classification and generation models through a unified API. A key differentiator is its concept-based functionality, which goes beyond feature-level attributions and is uncommon in existing libraries. The library is open source; install via pip install interpreto. Code and documentation are available at https://github.com/FOR-sight-ai/interpreto.

Problem

• Provides explainability for Transformer text models
• Offers attribution and concept-based explanation methods
• Supports classification and generation models via unified API

Innovation

• Python library for explainability of Transformer models
• Provides attribution and concept-based explanation methods
• Unified API supports classification and generation models