About this document

P P: C L L M: Rogressive Rompts Ontinual Earning For Anguage Odels by Dunhua You is a document available to read on EtoBox.

The paper presents Progressive Prompts, a novel approach for continual learning in language models that mitigates catastrophic forgetting while allowing forward transfer without requiring data replay or excessive task-specific parameters. By learning a new soft prompt for each task and concatenating it with previously learned prompts, the method significantly improves performance, achieving over 20% higher average test accuracy compared to state-of-the-art methods on the T5 model. Extensive experiments demo

Author

Dunhua You

Language

EN