Efficient LLM Jailbreak via Adaptive Dense-to-sparse Constrained Optimization

This is a draft implementation of our paper

• [NeurIPS 2024] Efficient LLM Jailbreak via Adaptive Dense-to-sparse Constrained Optimization

An official implementation will be ready soon.

TODO: As mentioned in #2, the code is outdated with the latest version of transformers, but runable at transformers==4.39.

Add a TODO for either specifying the version requirements or refactor the code for a newer version of transformers.