Enhancing Low-Resource LLMs Classification with PEFT and Synthetic Data

arXiv:2404.02422v1 Announce Type: cross
Abstract: Large Language Models (LLMs) operating in 0-shot or few-shot settings achieve competitive results in Text Classification tasks. In-Context Learning (ICL) typically achieves better accuracy than the 0-shot setting, but it pays in terms of efficiency, due to the longer input prompt. In this paper, we propose a strategy to make LLMs as efficient as 0-shot text classifiers, while getting comparable or better accuracy than ICL. Our solution targets the low resource setting, i.e., when only …

abstract accuracy arxiv classification context cs.cl cs.lg data efficiency few-shot in-context learning language language models large language large language models llms low paper peft prompt results synthetic synthetic data tasks terms text text classification type