Search papers, labs, and topics across Lattice.
College of Computer Science and Artificial Intelligence, Fudan University
1
0
2
10
Selectively transferring KV caches can accelerate LLM inference by over 4x without sacrificing performance.