Abstract page for arXiv paper 2608.13596: Training-Free Knowledge Transfer Across Model Scales through Activation-Guided Pruning