Picture for Jan Bronec

Jan Bronec

Charles University, Faculty of Mathematics and Physics, Institute of Formal and Applied Linguistics

Atyaephyra at SemEval-2025 Task 4: Low-Rank NPO

Add code
Mar 17, 2025
Jan Bronec, Jindřich Helcl
Figure 1 for Atyaephyra at SemEval-2025 Task 4: Low-Rank NPO
Figure 2 for Atyaephyra at SemEval-2025 Task 4: Low-Rank NPO
Figure 3 for Atyaephyra at SemEval-2025 Task 4: Low-Rank NPO
Figure 4 for Atyaephyra at SemEval-2025 Task 4: Low-Rank NPO

We present a submission to the SemEval 2025 shared task on unlearning sensitive content from LLMs. Our approach employs negative preference optimization using low-rank adaptation. We show that we can utilize this combination to cheaply compute additional regularization terms, which help with unlearning stabilization. The results of our approach significantly exceed the shared task baselines.

* 5 pages, 1 figure, 1 table, submitted to SemEval proceedings for ACL Anthology 
Viaarxiv icon
Github IconAccess Paper or Ask Questions
Twitter IconFacebook IconLinkedin IconWhatsapp IconMessenger Icon