Search papers, labs, and topics across Lattice.
University of Victoria
1
0
2
Co-located speculative decoding consistently beats distributed variants in latency, challenging the assumption that edge-cloud setups always yield faster inference.
