Efficient Fine-Tuning for Llama-v2-7b on a Single GPU