Llama2-13b-based NEFT fine-tuning for financial sentiment classification

This study introduces a groundbreaking approach in financial sentiment analysis using the Llama2-13b model, fine-tuned with NEFT (Neural Fine-Tuning) and QLora, an innovative fine-tuning technique. Applied to a financial sentiment dataset containing Positive, Negative, and Neutral categories, our adapted Llama2-13b model demonstrates superior performance in sentiment classification. It surpasses other advanced models including Roberta-Base, Roberta-Large, Deberta-v2-Base, and Deberta-v2-Large. The fine-tuned Llama2-13b model achieves an accuracy of 0.8392, outperforming the next best model, Deberta-v2-Large, at 0.8105. This study showcases the potential of specialized fine-tuning in enhancing language model performance for specific domain tasks, particularly in financial sentiment analysis.