Accelerating Inference of Retrieval-Augmented Generation via Sparse Context Selection