Small but Mighty: How Fine-Tuned Small Language Models Can Outperform Large Language Models