code-unlearning

Code Unlearning is a PyTorch implementation designed to erase sensitive memorization in Code Language Models (CLMs) as highlighted in the authors' ICSE'26 paper. The tool focuses on detecting sensitive memorization risks and enabling selective unlearning of specific sensitive information while preserving the model's overall performance. Key features include a systematic approach to curate a Sensitive Memorization Dataset and the ability for CLMs to forget targeted information efficiently.