What is ferret?

ferret is Python library for benchmarking interpretability techniques. The library is deeply integrated with the transformers library, models, and code.

Concretely, ferret lets you:

• visualize and compare Transformers-based models output explanations using state-of-the-art XAI methods

• benchmark explanations with ad-hoc XAI metrics for faithfulness and plausibility

• streamline the evaluation on existing XAI corpora

• work both on text-based data and audio-based data (speech classification tasks)