Collaborative and Efficient Fine-tuning: Leveraging Task Similarity