Global Prompt Cell: A Portable Control Module for Effective Prompt Tuning
This addresses the problem of effectively training and utilizing prompt embeddings in prompt tuning for NLP researchers and practitioners, representing an incremental advancement.
The paper tackles the limitation of prompt tuning methods that focus primarily on initialization by introducing the Global Prompt Cell (GPC), a portable control module that selectively preserves prompt information across encoder layers, achieving a 5.8% improvement on SuperGLUE datasets compared to vanilla prompt tuning.
As a novel approach to tuning pre-trained models, prompt tuning involves freezing the parameters in downstream tasks while inserting trainable embeddings into inputs in the first layer. However, previous methods have mainly focused on the initialization of prompt embeddings. The strategy of training and utilizing prompt embeddings in a reasonable way has become a limiting factor in the effectiveness of prompt tuning. To address this issue, we introduce the Global Prompt Cell (GPC), a portable control module for prompt tuning that selectively preserves prompt information across all encoder layers. Our experimental results demonstrate a 5.8% improvement on SuperGLUE datasets compared to vanilla prompt tuning.