TextAttack

Why we included this project

TextAttack is a Python framework for probing NLP models with adversarial inputs. It bundles dozens of published attack recipes, including TextFooler and DeepWordBug, into one command-line tool and Python API, so you can feed malformed text to a classifier and see exactly where it fails. The same framework also handles data augmentation and model training, letting you generate paraphrased training examples or train a model end to end with a single command. That makes it useful for both researchers testing new attack methods and engineering teams hardening a model before production. Because it's modular, you can also pull out a single piece, like a word-embedding wrapper or constraint checker, and drop it into your own pipeline without adopting the whole framework.