Instruction-Based Fine-tuning of Open-Source LLMs for Predicting Customer Purchase Behaviors