Towards Efficient VLMs: Information-Theoretic Driven Compression via Adaptive Structural Pruning

Open in new window