Fine-tuning NeoBERT for a multi-class text classification