Search papers, labs, and topics across Lattice.
University of Modena and Reggio Emilia
1
0
3
Get up to 20% faster ViT inference by hot-swapping certain attention heads for depthwise convolutions – without tanking accuracy.
