Text this: A LLaMA-Based Efficient Fine-Tuning Method for Image Captioning Using Multi-Feature Dynamic Prompts.