Search papers, labs, and topics across Lattice.
1
0
3
LLM privacy on shared accelerators doesn't have to break the bank: GELO achieves strong obfuscation with only 20-30% latency overhead, defeating common attacks.
