Why Training-Free Token Reduction Collapses: The Inherent Instability of Pairwise Scoring Signals

Create a diagnostic visualization library for Vision Transformers that computes ranking consistency and off-diagonal correlations. This tool would help researchers identify exactly when their token compression models will collapse during inference.

Suggested repo: CollapseCheck

"Visualize the exact moment your ViT compression falls apart."

Estimated effort: 20h