Finetune-RAG: Fine-Tuning Language Models to Resist Hallucination in Retrieval-Augmented Generation