Exploring Variability in Fine-Tuned Models for Text Classification with DistilBERT