attack: added latent perturbation attack#23

Open

psyonp wants to merge 18 commits into

mainfrom

psyonp/latent_attack

Commits on Feb 6, 2026

Migrate hooks.py

psyonp
authored and
tomtseng
committed
Change setup

psyonp
authored and
tomtseng
committed
Refactor latent_attack

psyonp
authored and
tomtseng
committed
Refactor as requested with eval

psyonp
authored and
tomtseng
committed
Add tests

psyonp
authored and
tomtseng
committed
Made requested changes for nits and docstrings

psyonp
authored and
tomtseng
committed
Update tests/evals/test_latent_attack_eval.py

authored and
tomtseng
committed
Update src/safetunebed/whitebox/evals/latent_attack/latent_attack.py

authored and
tomtseng
committed
Update src/safetunebed/whitebox/attacks/latent_attack/latent_attack.py

authored and
tomtseng
committed
Update tests/attacks/test_latent_attack.py

authored and
tomtseng
committed
Add link for ported code

psyonp
authored and
tomtseng
committed
Add link to hooks.py

psyonp
authored and
tomtseng
committed
Name change of parent_path to parent_layer

psyonp
authored and
tomtseng
committed
Merge branch 'main' into psyonp/latent_attack
tomtseng
committed
evals/latent_attack: Lint
tomtseng
committed
latent_attack: Fix type errors
tomtseng
committed
latent_attack: Fix merge errors
tomtseng
committed

Commits on Feb 7, 2026

latent_attack: Remove redundant test
tomtseng
committed