Advancing Single- and Multi-task Text Classification through Large Language Model Fine-tuning