← back to paper
arxiv: 2604.15408 · 2 revisions
Dispatch-Aware Ragged Attention for Pruned Vision Transformers