Transformer Explainability: Visualize Transformer Classifications

Understand the project

Practical overview

Use it when you want to inspect class-specific explanations for a Transformer prediction or work through the provided ViT and BERT examples.

Researchers and developers analyzing predictions from Transformer classifiers.
PyTorch users exploring vision or NLP model explanations through Jupyter notebooks.

Visualize an image classifier’s explanation for a selected class or its top class.
Explore BERT or ViT explainability through the repository’s example notebooks.
Reproduce documented ViT segmentation and perturbation evaluations or BERT results when the required data and weights are available.

Before you choose

The project is a code repository and PyTorch implementation, not a hosted analysis service.
The README describes an image workflow based on the samples folder and generate_visualization function.
Reproducing evaluations requires task-specific data; the BERT reproduction instructions also require pretrained classifier weights.
The repository identifies a separate follow-on project for broader attention-model explainability, including multimodal and encoder-decoder Transformers.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Use the documented image-input workflow to request an explanation for a chosen class or the top class.

The repository’s code and an image input.
A selected class index if you want an explanation for a class other than the top class.

A visualization for the selected class, or for the model’s top class when no class index is specified.

The README describes this workflow for image inputs; it does not provide installation or environment setup instructions in the available text.

Evidence and transparency

Source record details

Source availability: Active
Sources checked: 16 October 2026
Original website title: hila-chefer/Transformer-Explainability · Official homepage title · retained in its source language