No mapped relationships
ReDiPrune: Training-Free Token Pruning Before Projection Boosts MLLM Efficiency 6x, Gains 2% Accuracy