See CHANGELOG.md for full details.
Highlights
- New
bad-charactersattack recipe (BadCharacters2021): imperceptible NLP attacks via invisible characters/homoglyphs/reorderings (#817) - New
leapattack recipe (LEAP2023): Levy-flight/adaptive-inertia PSO variant (arXiv:2308.11284) RemoteModelWrapperfor querying models behind a remote HTTP APItrain --dataset-from-filesupport, matchingattack/eval- A round of correctness fixes across
AttackedText,GreedyWordSwapWIR,Augmenter,HuggingFaceModelWrapper,Attackdevice placement,CompositeTransformation, and more — see CHANGELOG for the full list.