Contextual Compression Encoding for Large Language Models: A Novel Framework for Multi-Layered Parameter Space Pruning