Translated data: The University of Southern California and Google Research have proposed the Channel-wise Lightweight Reencoding (CLR) method to address the issue of catastrophic forgetting in continuous learning for large language models. The CLR method introduces a lightweight module that reprograms channel-wise feature maps for each layer, allowing the model to adapt to new tasks with only an additional 0.6% of parameters. By employing a dynamic network approach, the CLR method achieves continuous learning across multiple tasks, offering enhanced performance and flexibility, and opening new opportunities for future research and applications in continuous learning.