Multitask Prompt Tuning Enables Parameter-Efficient Transfer Learning
This addresses the need for parameter-efficient transfer learning in NLP, offering a novel method to leverage cross-task knowledge, though it is incremental in improving prompt tuning.
The paper tackles the problem of efficiently adapting large language models to multiple downstream tasks by proposing multitask prompt tuning (MPT), which learns a transferable prompt from multiple tasks and adapts it with low-rank updates, achieving state-of-the-art performance on 23 NLP datasets while tuning only 0.035% as many parameters as full fine-tuning.
Prompt tuning, in which a base pretrained model is adapted to each task via conditioning on learned prompt vectors, has emerged as a promising approach for efficiently adapting large language models to multiple downstream tasks. However, existing methods typically learn soft prompt vectors from scratch, and it has not been clear how to exploit the rich cross-task knowledge with prompt vectors in a multitask learning setting. We propose multitask prompt tuning (MPT), which first learns a single transferable prompt by distilling knowledge from multiple task-specific source prompts. We then learn multiplicative low rank updates to this shared prompt to efficiently adapt it to each downstream target task. Extensive experiments on 23 NLP datasets demonstrate that our proposed approach outperforms the state-of-the-art methods, including the full finetuning baseline in some cases, despite only tuning 0.035% as many task-specific parameters.