RIFF: Rephrasing Inputs for Few-shot Fine-tuning of Language Models

Anonymous

TL;DR: We propose an efficient idea for Rephrasing Inputs for Few-shot Fine-tuning of language models (RIFF) with prompt optimization and efficient tuning methods.

Abstract: Pre-trained Language Models (PLMs) have shown remarkable performance when fine-tuned for downstream text processing tasks. Recently, researchers have effectively guided PLMs to perform specific tasks by optimizing input prompts (prompt optimization) or adjusting a small number of model parameters (efficient tuning). In this study, we explore the impact of altering the input text of the original task while fine-tuning language models with these recent prompt optimization and efficient tuning methods. To most effectively rewrite the input text, we apply a learning objective based on Maximum-Marginal Likelihood estimation in a few-shot setting. Experimenting with seven few-shot text classification datasets, we show that enriching training examples with the input text’s paraphrases at train and test time significantly enhances the performance of recent prompt optimization and efficient tuning techniques.

Paper Type: long

Research Area: Efficient/Low-Resource Methods for NLP

Contribution Types: Approaches to low-resource settings, Approaches low compute settings-efficiency

Languages Studied: English