Search papers, labs, and topics across Lattice.
1
0
3
BinaryAttention proves you can more than halve the runtime of attention in vision and diffusion transformers without sacrificing accuracy, simply by using the sign of queries and keys.
