Beyond Attention or Similarity: Maximizing Conditional Diversity for Token Pruning in MLLMs

Open in new window