Small Language Models Can Effectively Perform Zero-Shot Text Classification, Challenging the Dominance of Large Language Models

Small language models can effectively classify texts in a zero-shot setting, matching or surpassing the performance of their larger counterparts across a diverse set of datasets.