Search papers, labs, and topics across Lattice.
1
0
3
2
Overlapping communication with computation in distributed Diffusion Transformer inference can yield up to 8.4x speedups, challenging the assumption that collectives are a fixed bottleneck.
