Mask the Target: A Plug-and-Play Regularizer Against LoRA Forgetting 事件

Mask the Target: A Plug-and-Play Regularizer Against LoRA Forgetting arXiv:2605.29498v1 Announce Type: cross Abstract: Low-Rank Adaptation (LoRA) has become one of the most widely used fine-tuning mechanisms for adapting large language models to new domains, tasks, and users. Yet adaptation performance alone can obscure an important failure mode: LoRA updates may improve performance on the target distribution while degrading prior capabilities learned during pretraining and alignment. We show t