Differentiable Prompt Makes Pre-trained Language Models Better Few-shot Learners

TL;DR

This paper introduces DART, a differentiable prompt method that enhances pre-trained language models' few-shot learning capabilities without prompt engineering, by jointly optimizing prompts and labels through backpropagation.

Contribution

The paper presents DART, a novel, plug-and-play, differentiable prompt approach that improves few-shot learning in pre-trained language models without requiring prompt engineering.

Findings

DART improves few-shot performance across standard NLP tasks.

The approach is compatible with various pre-trained models.

It extends to multiple classification tasks.

Abstract

Large-scale pre-trained language models have contributed significantly to natural language processing by demonstrating remarkable abilities as few-shot learners. However, their effectiveness depends mainly on scaling the model parameters and prompt design, hindering their implementation in most real-world applications. This study proposes a novel pluggable, extensible, and efficient approach named DifferentiAble pRompT (DART), which can convert small language models into better few-shot learners without any prompt engineering. The main principle behind this approach involves reformulating potential natural language processing tasks into the task of a pre-trained language model and differentially optimizing the prompt template as well as the target label with backpropagation. Furthermore, the proposed approach can be: (i) Plugged to any pre-trained language models; (ii) Extended to widespread classification tasks. A comprehensive evaluation of standard NLP tasks demonstrates that the proposed approach achieves a better few-shot performance. Code is available in https://github.com/zjunlp/DART.