Search papers, labs, and topics across Lattice.
Peking University
1
0
3
Achieve practical FHE inference for Llama-3-8B with sub-100 second token generation by cleverly integrating KV caching, leaving prior art in the dust.
