Comparative Study of Pre-Trained BERT and Large Language Models for Code-Mixed Named Entity Recognition