This is official model collection for paper "Leak@$k$: Unlearning Does Not Make LLMs Forget Under Probabilistic Decoding"