Fetching the paper…

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