This repo is dedicated to an analysis of the LLMs performance as text classifiers on five German datasets from social media covering 13 tasks. We experiment with zero- and few-shot classification approaches as well as with fine-tuning using various LLMs