Transformer² is a self-adaptation framework that adapts large language models for unseen tasks in real-time by dynamically adjusting weight matrices and employing a two-pass mechanism for task-specific behavior.
