City University of Hong Kong

Step-Level Sparse Autoencoder for Reasoning Process Interpretation

Diversity-Preserved Distribution Matching Distillation for Fast Visual Synthesis